Podcasts about Robotics

Last week, World Labs announced its acquisition of SceniX, bringing together two teams working on one of AI's biggest unsolved problems: how to give machines a true understanding of the physical world. Martin Casado sits down with Fei-Fei Li, co-founder and CEO of World Labs, creator of ImageNet, and pioneer of spatial intelligence, alongside Yunzhu Li, co-founder of SceniX and assistant professor at Columbia University. They discuss why World Labs acquired SceniX, how simulation can unlock the next generation of robotics, and why training robots may require a fundamentally different approach than training language models. The conversation explores real-to-sim-to-real pipelines, world models, robotics foundation models, evaluation, synthetic data, and why the future of AI depends not just on understanding language—but on understanding and interacting with the physical world.   Resources: Follow Fei-Fei Li on X: https://x.com/drfeifei Follow Yunzhu Li on X: https://x.com/YunzhuLiYZ Follow Martin Casado on X: https://x.com/martin_casado Stay Updated:Find a16z on YouTube: YouTubeFind a16z on XFind a16z on LinkedInListen to the a16z Show on SpotifyListen to the a16z Show on Apple PodcastsFollow our host: https://twitter.com/eriktorenberg Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see a16z.com/disclosures. Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.

Noah Schochet is the co-founder and CEO of TerraFirma, a construction technology company rethinking how infrastructure gets built. Founded by two former SpaceX engineers, TerraFirma combines robotics, remote operations, and software into a construction business that performs projects directly rather than selling technology. By vertically integrating its tools with field operations, the company aims to make construction dramatically faster, safer, and more affordable while addressing labor shortages and enabling the next generation of infrastructure. In this episode, Noah shares how his experience building Starship at SpaceX inspired TerraFirma, why he believes construction—not autonomy—is the real bottleneck to civilization-scale progress, and why robotics companies should focus more on delivering outcomes instead of products. He also explains TerraFirma's remote-operated construction model, its approach to rapid iteration, and why the future of construction still depends on people empowered by better technology. Episode recorded on June 25, 2026 (Published on July 28, 2026). In this episode, we cover:  (0:00) An overview of TerraFirma (1:04) How SpaceX inspired the creation of TerraFirma (4:47) Why construction has fallen behind other industries (9:33) The structural reasons construction is slow to modernize (12:33) Why the goal isn't autonomy—it's better construction (14:25) TerraFirma's construction-first business model (17:11) Why customers buy outcomes, not robots (20:54) How TerraFirma's retrofit robotics and software platform works (25:53) Operating multiple construction machines from a mission control center (29:01) How robotics could reshape construction jobs instead of replacing them (32:02) Early customer adoption across data centers, housing, and hazardous environments (34:20) Winning projects by building faster, cheaper, and safer (36:13) The biggest challenge to transforming construction (38:32) Hiring builders with extreme ownership and mission alignment (42:55) TerraFirma's $100 million Series A and plans to scale Enjoyed this episode? Please leave us a review! Share feedback or suggest future topics and guests at info@mcj.vc.Connect with MCJ:Cody Simms on LinkedInVisit mcj.vcSubscribe to the MCJ Newsletter*Editing and post-production work for this episode was provided by The Podcast Consultant

Ben Horowitz joins Theo Jaffee and Sofia Puccini to discuss one of the biggest debates in AI today: the future of open-source models. They examine the growing push to restrict open models, why Ben believes open source is critical for security, innovation, and competition, and what happens if a handful of frontier labs come to dominate the AI ecosystem. They also discuss distillation, AI monopolies, China's role in open-source AI, robotics, manufacturing, the economics of frontier models, and why Ben believes the AI market is still in its earliest stages. The conversation closes with a look at AI-generated art, creativity, and whether new tools will spark the next cultural renaissance.   Resources: Follow Ben Horowitz on X: https://x.com/bhorowitz Follow Theo Jaffee on X: https://x.com/theojaffee Follow Sofia Puccini on X: https://x.com/schisofrenia Stay Updated:Find a16z on YouTube: YouTubeFind a16z on XFind a16z on LinkedInListen to the a16z Show on SpotifyListen to the a16z Show on Apple PodcastsFollow our host: https://twitter.com/eriktorenberg Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see a16z.com/disclosures. Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.

Travis Kalanick, Ben Horowitz, and Erik Torenberg reunite to reflect on Uber's early days, the investment that almost happened, and why Kalanick believes the next great technology opportunity lies beyond software. They discuss the lessons of building Uber, the value of founder-led companies, and why Kalanick spent nearly eight years quietly building Atoms before stepping back into the spotlight. The conversation explores industrial AI, robotics, autonomy, mining, food production, and Kalanick's vision for digitizing the physical world, where software, manufacturing, real estate, and transportation come together to transform entire industries.   Resources: Follow Travis Kalanick on X: https://x.com/travisk Follow Ben Horowitz on X: https://x.com/bhorowitz Stay Updated:Find a16z on YouTube: YouTubeFind a16z on XFind a16z on LinkedInListen to the a16z Show on SpotifyListen to the a16z Show on Apple PodcastsFollow our host: https://twitter.com/eriktorenberg Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see a16z.com/disclosures. Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.

Applied Intuition has spent the past decade building the software that powers intelligent machines, from passenger vehicles and trucks to defense systems, mining equipment, and industrial robots. In this conversation, Marc Andreessen and Erik Torenberg sit down with Applied Intuition cofounders Qasar Younis and Peter Ludwig to discuss the emergence of physical AI and the company's latest launch, Dana, a new platform designed to accelerate the development of autonomous systems. They explore autonomous vehicles, robotics, world models, simulation, AI infrastructure, and the engineering challenges of deploying intelligence safely in the physical world. Along the way, they discuss self-driving cars, humanoid robots, global competition, and why lowering the barrier to building physical AI could unlock an entirely new generation of products and companies.   Resources: Follow Qasar Younis on X: https://x.com/qasar Follow Peter Ludwig on LinkedIn: linkedin.com/in/peterwludwig Follow Marc Andreessen on X: https://x.com/pmarca Stay Updated:Find a16z on YouTube: YouTubeFind a16z on XFind a16z on LinkedInListen to the a16z Show on SpotifyListen to the a16z Show on Apple PodcastsFollow our host: https://twitter.com/eriktorenberg Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see a16z.com/disclosures. Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.

Juan's going to teach us about robots and stuff, beep boop.In this Episode #526 of 'Meanderings', we discuss: when robots might actually arrive in everyday life, the difference between software progress in AI and the much tougher challenge of getting machines to move reliably in the physical world, why humanoid robots are so compelling yet so difficult, how the earliest large-scale robotics wins will likely happen in high-value/dangerous/efficiency-driven industries, thoughts on autonomous vehicles/robotic lawn mowing/nanobots/giant industrial machines and the economics that determine whether a robot is genuinely useful or just expensive theatre. Stan Link: https://stan.store/meremortalsTimeline:(00:00:00) Intro(00:03:16) Will robots actually take jobs or just shift work?(00:07:26) Why tech adoption is slower than the hype(00:12:29) What counts as a robot and why AI matters(00:16:00) The hard engineering behind real world robotics(00:22:01) Robot fighters, humanoids and the money flowing in(00:27:30) Where robots will first make economic sense(00:34:22) Nanobots, giant machines and robot form factors(00:41:36) Movement data, manufacturing limits and home robots(00:49:28) A new monthly robotics podcast and final thoughts Connect with Mere Mortals:Website: https://www.meremortalspodcasts.com/Discord: https://discord.gg/K99e8fysBnTwitter/X: https://twitter.com/meremortalspodsInstagram: https://www.instagram.com/meremortalspodcasts/TikTok: https://www.tiktok.com/@meremortalspodcastsValue 4 Value Support:V4V: https://www.meremortalspodcasts.com/supportPaypal: https://www.paypal.com/paypalme/meremortalspodcast

This week's video transcript summary is here. You can click on any bulleted section to see the actual transcript. Thanks to Granola for its software.There was an issue with this only going to paid subscribers, so sending it again. Apologies to those who get it twice. I appreciate being paid so feel free to upgrade if you enjoy TWTW.EditorialIntelligence: Who Owns it?This week the word “AI” feels too small.AI is a technology. Intelligence is its product. And if intelligence is the product, the question is no longer just: Which model is best? Who has the cheapest tokens? Who owns the weights? Who controls the data center? Those are important questions, but they are lower in the stack.The bigger question is simpler and more political:Who owns intelligence?That sounds abstract until you make it concrete. Intelligence is becoming something companies can capture, package, serve, meter, route, improve, and sell.It can write code, answer questions, design molecules, automate offices, run agents, draft legal work, advise scientists, serve consumers, and reshape workflows. It is not merely software. It is a general-purpose capability. And all humans could benefit from more of it.General-purpose capabilities have a habit of becoming public questions. But the default answer, that public good is best delivered by government, is the wrong answer in this context.The Product Is IntelligenceWe should stop talking about AI as a feature and start talking about intelligence as the universal thing that is delivered as an input to the world.Water is an input. Electricity is an input. Literacy is an input. Connectivity is an input. Once a society depends on them, access stops being optional. Nobody needs government to build every well, power plant, school, or network. But everybody understands that a civilization cannot be organized around less than universal and reliable access to foundational inputs.Intelligence is reaching that level of importance now that we all know it is real.Government should not own it, operate it, or develop it. Quite the opposite. Companies are the right actors to build fast, compete hard, improve models, serve customers, and discover the real use cases. Self-interest is a useful framing here. Markets are good at finding demand, reducing costs, and turning invention into services people actually use.Companies are the right operators, developers, and owners. But that does not settle the real question of who owns the benefits. That is an economic question.If intelligence becomes metered infrastructure, what happens to the value it creates?The Ownership StackThis week's articles keep circling the same issue from different directions but in the nature of ‘circling' never quite nail it.Jamin Ball's “Own Your Weights” starts with the enterprise version of the question. Owning a model file is not enough. The durable asset is the loop: the data flywheel, the evaluations, the reinforcement system, the workflow learning, and the operating context that lets capability compound.Benedict Evans' “Ways to Think About Token Pricing” adds the market layer. Tokens may become essential, abundant, and cheap, like mobile data. But being essential does not guarantee that the token layer captures the value. The money may move up the stack to whoever owns the workflow, the customer, the distribution, or the application.Alex Karp's fight with the labs, reported in “Alex Karp Is Saying What Every Angry CEO Is Thinking About AI”, is the same argument in sharper enterprise language. Companies are afraid that model providers will not just sell intelligence, but learn from customer workflows and then move into the markets where those workflows create value. The “All-in” group are echoing Karp's view.And “What Is Loop Engineering, and Who Owns It?” names the new contested terrain. The loop is where intelligence meets the world. Whoever owns the loop owns the learning. Whoever owns the learning owns the compounding asset.That is why “who owns intelligence?” is not a slogan. It is the question under the model layer, the application layer, the enterprise layer, and the economic layer.Because intelligence is the product, the tools creating it are fragmented and competitive. So there is no logic in trying to discuss this at the level of a single company or set of tools and models.The Old Promise Was That Commerce Would Tame PowerThe essays this week give the historical backdrop.Deirdre McCloskey, in “What Really Caused the Industrial Revolution”, argues that modern growth came not simply from capital accumulation, but from a change in permission: ordinary people were allowed to innovate, trade, build, and be honored for it.That matters because intelligence could be another expansion of permission. It could make more people capable of building, learning, creating, coding, researching, translating, selling, and coordinating. It could lower the cost of competence.But only if access is broad.Paul Krugman's “AI in an Age of Oligarchy” warns that the same technology lands differently in different political economies. A new general-purpose technology entering a broad, open, upwardly mobile society is one thing. The same technology entering a concentrated economy, with extreme wealth and weak counterweights, is another.Tim O'Reilly's Economist essay, “Elon Musk is building a form of capitalism that Adam Smith would hate”, makes the governance point more directly. The old liberal hope was that commerce would tame arbitrary power. Markets, boards, courts, shareholders, disclosure, and competition would discipline the prince.But what if the prince uses markets to escape discipline?Henry Farrell's “political economy of billionaire derangement” pushes the same point. Founder culture, monopoly ambition, peer rivalry, weak correction mechanisms, and vast private control can amplify appetites rather than restrain them.The danger with intelligence is not that companies build it. They should. Companies build it, meter it, use public tolerance and public infrastructure to scale it, learn from everyone who uses it. All of those things are inevitable and healthy. Market forces will sort out winners from losers. The real danger is that the winners treat all of the surplus produced as purely private.Metered Intelligence Creates SurplusIf metering is not the problem, what is?The problem is pretending that metered intelligence creates value only for the metering entity. Metering water is only tolerated as a public good. If the public were blackmailed by a private water company with the threat of no water we would all rebel.Once we understand that the product of AI is intelligence we can see that every time intelligence is used, there is the immediate transaction: the user pays, the provider serves.But there is also system value. Usage creates signals. Workflows reveal patterns. Prompts, corrections, failures, preferences, integrations, edge cases, and business processes all help define where intelligence is useful and how it should improve. Intelligence breeds intelligence.Even when customer data is contractually protected, the market learns. The platform learns where demand is. The product team learns which workflows matter. The ecosystem learns which jobs are vulnerable, which tasks are automatable, and which parts of the economy can be reorganized around machine intelligence.So the surplus is not born in a vacuum.It rests on public science, public education, public data exhaust, public law, public infrastructure, public energy systems, public tolerance for data centers, and billions of human interactions. It is served by companies, but it is not made only by companies.This is why “Americans Deserve a Dividend From AI Companies' Riches” belongs at the center of this week's issue. The detail can be debated. The principle is harder to dismiss. If intelligence becomes a new foundational resource, then some part of the wealth it creates should flow back to the people whose society makes it possible. Intelligence did not suddenly appear. AI is built on the entire history of human intelligence. It benefits from it and at the same time evolves it.Not Nationalization. A Human Wealth Fund.If intelligence belongs to everybody, some conclude that government ownership of intelligence is the right outcome.Governments are not well suited to build, operate, or improve intelligence. They will move too slowly, regulate too early, politicize the wrong things, and confuse economic participation with operational control.Andrew McAfee's “Why I Didn't Sign the AI Open Letter” is useful here. His objection is not that the technology is unimportant. It is that steering too hard before we understand the shape of the change can become its own failure mode. Marc Andreessen's satire of AI regulation is less policy than temperament, but it captures a real Silicon Valley fear: that regulation can become permission, capture, and incumbency before it becomes wisdom.That fear should be taken seriously.But it does not answer the economic question. It answers only the operational one.How can the economic benefits of intelligence be distributed? The better answer is a sovereign human wealth fund.Call it a sovereign wealth fund if you must, but the phrase is too national. Intelligence will not respect borders. The leading companies are global. The models, chips, data centers, agents, platforms, and workflows will be transnational from the beginning. If the value created by intelligence is global, then the mechanism for sharing some of that value should begin with the companies global enough to capture it. The nice thing about xAI, OpenAI, and Anthropic is that they are supranational.These companies own and operate intelligence. Let them compete. Let them profit. Let them keep the incentives that make the system improve. But if intelligence is the new water, the wealth it creates cannot belong only to the companies that meter it. And they, themselves, have the power to fix it, even more than governments.Access will become a Human Right; Ownership Is the Economic DesignThis is where human rights come in. There is no right to access an AI model, yet. But there will soon be a need to change that.Not as a claim that every person is entitled to every frontier model at every moment for free. That is not serious. Capacity has costs. Models have costs. Inference has costs. Data centers have costs. Although those costs will decline over time, possibly quite quickly as self-learning models address costs.The claim is more basic: in a world where intelligence becomes a primary input into education, work, health, science, citizenship, creativity, and economic agency, baseline access to intelligence starts to look like a civic requirement.That could mean public access layers. It could mean education credits. It could mean open models. It could mean AI dividends. It could mean public-interest compute. It could mean taxes on rents. It could mean a company-initiated human wealth fund that returns some of the upside to society without handing the operating system to the state. The latter could couple wealth growth with universal distribution of ownership.The exact mechanism matters. But the distinction matters more.Government should not own intelligence. It should be universally available. And people should have a claim on the wealth intelligence creates.The Frontier Is Also PhysicalThe abstraction is not weightless.“The Fight Against AI Data Centers Is Just Beginning”, “New York becomes the first state to enact a data center moratorium”, Reuters on pollution from Musk's xAI power project, and DataGravity's “Who Captures Value in AI Infrastructure?” all say the same thing from the ground up.Intelligence uses land. It uses power. It uses water. It uses chips. It uses grid capacity. It uses neighborhoods. It uses public patience.That makes the value question unavoidable. A society can accept the buildout if the buildout is legible as shared progress. It will resist it if the costs are local, the profits are private, and the benefits feel enclosed.Who Owns the “Loop”?The week ends where it began.“Anthropic and Blackstone” are betting that implementation is the next trillion-dollar business. “Vint Cerf” is working on identity for agents on the open internet. “GPT-Red” points toward systems that improve their own robustness. “Kimi K3” adds another open frontier model to the global mix.The model race continues. The deployment race is accelerating. The governance race is behind.My view is this:The central product of this era is intelligence. Companies have figured out how to capture it, package it, serve it, and meter it. That is good. It should stay in the hands of builders who have the incentive to make it better.But intelligence is too foundational to become just another private toll booth. A significant part of it will turn out to be free to users.As intelligence becomes a general-purpose resource, then access to it becomes a human-capability question, and the surplus from it becomes an economic-justice question. Not because government should run it. Because government should not run it. The operating layer belongs with companies. The wealth question belongs with everyone. But companies are best placed to turn that into a process of distribution.The question is not whether companies should build intelligence. They should.The question is whether humanity gets a stake in the wealth created by the thing that may soon become its most important shared input.Contents* Essays* Deirdre McCloskey on What Really Caused the Industrial Revolution* AI in an Age of Oligarchy* Elon Musk is building a form of capitalism that Adam Smith would hate* Murky Mirror: Truth and Consequences* The political economy of billionaire derangement* Is there any “oligarchy” to fight?* AI* Nearly 200 Economists and Tech Leaders Warn of A.I. Threats* Why I Didn't Sign the AI Open Letter* Own Your Weights* Ways to Think About Token Pricing* Alex Karp Is Saying What Every Angry CEO Is Thinking About AI* The AI Agents Are Coming for Microsoft Office* What Is Loop Engineering, and Who Owns It?* The Fight Against AI Data Centers Is Just Beginning* 6 months to live for open models* Americans Deserve a Dividend From AI Companies' Riches* Who Gets to Define the Frontier?* GPT-Red: Unlocking Self-Improvement for Robustness* Anthropic, Blackstone bet the next trillion-dollar AI business is implementation, not just models* Vint Cerf is working on a plan to unleash AI agents on the open internet* xai-org/grok-build, now open source* The Pulse: What can we learn from Bun's rapid Rust rewrite with AI?* Orphan risks at the frontier of artificial intelligence* The Lab of the Future Should Feel Like a Data Center* Why AMI Labs' Alexandre LeBrun won't call his AI “AGI” or “superintelligence”* Kimi K3 Tech Blog: Open Frontier Intelligence* Venture Capital* Three Years In* Venture Has Rarely Looked More Bifurcated* The Best Angel Investors in the US: Who Backs the Most Unicorns, and Who's Active Now* Are Prediction Markets Doomed to Fail?* Regulation* Exclusive: The Next Frontier of the Deportation Wars: College Campuses* The Supreme Court Broke Independent Agencies. Here's a Way to Slow the Damage.* India's crackdown on a new WhatsApp feature risks setting a global precedent* Let's build a children's public internet* Computer cops* Google is better at playing the AI regulations game* Infrastructure* Who Captures Value in AI Infrastructure?* New York becomes the first state to enact a data center moratorium* Pollution from Musk's unpermitted xAI power project hits hardest in Black communities* Interview of the Week* The End of the End of Geography* Startup of the Week* Radical AI's Joseph Krause: The Scientist Building The “Waymo” Lab For New Materials* Post of the Week* Marc Andreessen on AI RegulationEssaysDeirdre McCloskey on What Really Caused the Industrial RevolutionYascha Mounk and Deirdre McCloskey | Persuasion | July 11, 2026Yascha Mounk interviews Deirdre McCloskey about her argument that the modern world's economic liftoff came less from capital accumulation than from a change in ideas. McCloskey says both left and right versions of the conventional story rely too heavily on investment: the left stresses exploitation and surplus value, while the right stresses virtuous saving by capitalists. Her objection is historical and economic. Human beings had always invested, from irrigation works and Roman roads to seed grain, and simple accumulation quickly runs into diminishing returns.McCloskey's alternative is that northwestern Europe, first Holland, then Britain and Scotland, and then the North American colonies, developed a liberal ideology that changed who was allowed to innovate and be honored for it. The conversation links that shift to the erosion of inherited hierarchy, the spread of dignity for ordinary commercial life, and a moral vocabulary in which liberalism is not merely procedural but connected to virtues and values. The point is not that machines, coal, trade, and institutions did not matter, but that they do not explain the scale and timing of modern enrichment without a cultural permission structure for innovation.The interview also turns to the contemporary defense of liberalism. Mounk frames the series around the worry that liberalism is often treated as too thin to command allegiance, while its opponents speak more directly to moral passions. McCloskey's case is that liberal societies became rich because they dignified experimentation and ordinary enterprise, and that liberals need to recover the moral language behind that claim.Read moreAI in an Age of OligarchyPaul Krugman | Paul Krugman | July 12, 2026Paul Krugman frames AI as a major technological shock arriving inside an already unequal political economy. The post says AI's economic and social effects may take years to understand, but argues that the setting matters now: America has much greater wealth concentration and political inequality than it did in the 1950s and 1960s, when progressive taxation, stronger regulation, and more active antitrust might have contained some of the destructive effects of a new technology.Krugman's opening claim is that the same technology would likely have different consequences in a more level society. In today's United States, he writes, extreme wealth is both a cause and effect of policies that favor a small elite, including low effective taxes on capital and high incomes, weak enforcement of worker protections and antitrust, and cuts to programs that benefit ordinary Americans.The article is explicitly more about oligarchy than AI. Krugman says the paid sections document the rise of the “.0002%,” the economics and politics of extreme wealth, how oligarchy will shape AI's impact, and possible policy paths. His caveat is that AI itself may still produce a pushback against oligarchy, but absent that, he expects the pre-existing concentration of wealth and power to magnify AI's downsides.Read moreElon Musk is building a form of capitalism that Adam Smith would hateAuthor: Tim O'Reilly Published: July 12, 2026Tim O'Reilly argues that Elon Musk is using the legal forms of shareholder capitalism to escape the restraints that shareholder capitalism was supposed to impose. The article begins with SpaceX's public-market structure: ordinary public investors get little meaningful governance power, Musk keeps roughly 85 percent of the votes through super-voting shares, buyers waive jury trials and class actions, the company qualifies as controlled, and removal of Musk depends on the share class he controls. In O'Reilly's framing, that is not ordinary founder control; it is a design for being answerable to no one, possibly beyond Musk's own lifetime.The killer detail is the article's turn through Albert Hirschman, Montesquieu, James Steuart, Adam Smith, and Keynes. Older defenses of commerce held that markets would tame princely passions because the self-interest of merchants was safer than arbitrary rule. O'Reilly says Musk reverses that hope. The market discipline that was supposed to cage the prince has become the lever by which the prince raises capital, removes feedback loops, and carries private power into politics, government, Mars, robots, AI, or whatever ambition comes next.The pull is the link to AI governance. O'Reilly says corporations are already a kind of artificial intelligence: narrow-input systems that act at a scale no individual human can match. Their partial controls include independent boards, shareholder votes, courts, disclosure, regulators, public pressure, and activism. If the leaders building frontier AI strip those alignment mechanisms out of their own companies, the governance of the company becomes a preview of the governance of the machine.Read more: The EconomistMurky Mirror: Truth and ConsequencesAuthor: Esther Dyson Published: July 14, 2026Esther Dyson argues that today's institutional crisis is better viewed through the 14th century than through recent political history. Using Barbara Tuchman's A Distant Mirror as her frame, she compares a world of famine, plague, church schism, feudal predation, and purposeless war with a present in which institutions again feel brittle, incentives are badly aligned, and power is shifting into forms that are hard to govern.The killer detail is the historical analogy between land, corporations, and AI. Dyson moves from nobles who controlled serfs and territory, to the East India Company as a quasi-sovereign business, to today's AI systems and data centers as a possible new sector that crosses and weakens both nation-states and companies. The question is whether AI becomes a new kind of private land, owned by a new nobility, or an open prairie that many people can cultivate.The pull is human attention. Dyson says the central question is not what AI will do to people, but how people will react to it: whether they can value love, kindness, embodied attention, and artisanal human presence in a world of seductive artificial offerings.Read more: SourceThe political economy of billionaire derangementAuthor: Henry Farrell Published: July 15, 2026Henry Farrell argues that the visible political radicalization of some Silicon Valley billionaires is not a random personality quirk, but a product of the political economy that made them. Starting from Tyler Cowen's dismissal of “billionaire derangement syndrome” and Tim O'Reilly's warning that Elon Musk is using shareholder capitalism to escape shareholder restraint, Farrell flips the phrase: the question is why billionaires themselves can become deranged.The killer detail is Farrell's use of Peter Thiel as both theorist and example. Thiel's Stanford lectures described startups as monarchies and founders as figures vested with unusual power, while Silicon Valley culture rewarded eccentricity, monopoly ambition, and founder exceptionalism. Farrell says those ideas combined with dense founder-investor networks, peer rivalry, and weak correction mechanisms to amplify rather than discipline princely appetites.The pull is the ideological problem for classical liberals who once saw tech wealth as an ally of markets and freedom. Farrell says commerce did not tame the passions; in parts of Silicon Valley, the passions have begun to devour markets, institutions, and the liberal story that justified them.Read more: SourceIs there any “oligarchy” to fight?Matthew Yglesias | Slow Boring | July 16, 2026Matthew Yglesias argues that “oligarchy” is a rhetorically powerful but analytically loose way to describe American politics. The post begins from Bernie Sanders' “Fighting Oligarchy” tour, Amy Klobuchar's warning about a MAGA “broligarchy,” and the long afterlife of the Martin Gilens and Benjamin Page paper that was widely summarized as showing that only the rich matter in policy outcomes. Yglesias says the evidence supports a weaker claim: affluent people and business leaders have unusual access and influence, but that is not the same as rule by a small cabal.His main distinction is between inequality and oligarchy. The Gilens-Page measure treated the top 10 percent of households as “the wealthy,” and later critics found that rich and middle-class preferences usually align; in the cases where they differ, the rich win about 53 percent of the time. Yglesias also says business executives get special access partly because their decisions are materially important to communities, jobs, investment, and local tax bases, not only because of campaign donations.The post preserves Jerusalem Demsas' counterpoint from their podcast discussion: privileged donor and business access can still violate democratic equality even if the oligarchy label overstates the structure of power. Yglesias' narrower claim is that Democrats should be precise about what problem they are trying to solve, because donor influence can also push the party left on climate and cultural issues in ways that alienate many voters.Read more: Slow BoringAINearly 200 Economists and Tech Leaders Warn of A.I. ThreatsAuthor: Ben Casselman Published: July 13, 2026Ben Casselman reports on “We Must Act Now,” a statement warning that artificial intelligence could transform the economy faster than any previous technology and that policymakers need to move faster to understand and respond. The statement says AI may become radically more powerful over the next 10 years, bringing risks such as large-scale job displacement as well as opportunities such as higher living standards. Nearly 200 people signed, including 15 Nobel laureates, the chief economists of OpenAI and Anthropic, Anthropic co-founder Jack Clark, former Google CEO Eric Schmidt, and venture capitalist Vinod Khosla.The killer detail is who joined the warning. Casselman notes that the signatories include economists who have historically been skeptical of Silicon Valley's most dramatic AI job-loss forecasts, including Daron Acemoglu and Simon Johnson, the MIT professors who won the 2024 Nobel in economics. Erik Brynjolfsson, who helped organize the statement, says there has been a notable change in the profession and that economists and policymakers are not ready for the “tsunami” he sees coming.The pull is the measurement problem. The statement does not offer a specific policy menu, but calls for economists, policymakers, and industry leaders to understand the economics of transformative AI and steer it toward complementing humans. Brynjolfsson says one high priority is better data on AI's spread and impact, because current measures tell conflicting stories about job losses and which workers are most exposed.Read more: The New York TimesWhy I Didn't Sign the AI Open LetterAuthor: Andrew McAfee Published: July 13, 2026Andrew McAfee explains why he did not sign “We Must Act Now,” the AI economy statement organized in part by his longtime collaborator Erik Brynjolfsson. McAfee agrees with the letter's starting point that AI is likely to become radically more powerful over the next decade and that it is a general-purpose technology. His objection is not to urgency or to studying AI's economic effects, but to the framing of risk, displacement, and institutional steering as the first move.The killer detail is McAfee's line edit. He says the original letter comes close, then “bounces off the crossbar” by calling for incentives, guardrails, and institutions to steer AI before we know enough about its actual impacts. He points to mixed current evidence: labor-market canaries, but also rising software job postings, low unemployment for younger workers, rising real median income, and claims that AI-adopting companies are adding workers faster than low-adopting peers. His worry is that the letter leans toward upstream governance and dirigisme when the evidence may call for capability building instead.The pull is his replacement statement. McAfee keeps the three-paragraph structure but changes the emphasis: AI is likely to become radically more powerful; like earlier world-changing technologies it will raise living standards while also bringing harms and shocks; and economists, policymakers, and technology leaders should build the capabilities to respond quickly and effectively. It is a concise version of the permissionless-innovation case inside the AI policy debate.Read more: The Geek WayOwn Your WeightsAuthor: Jamin Ball Published: July 10, 2026Jamin Ball argues that the enterprise AI debate about whether companies should “own their weights” or rent models from frontier labs is asking too narrow a question. A model weight file gives a company control over a point-in-time artifact, but not durable control over the capability stack. In his framing, the weight file is a melting ice cube: it does not get worse in absolute terms, but it falls behind as frontier systems improve and enterprise needs change.The killer detail is what Ball says companies really need to own: the data flywheel, reinforcement learning infrastructure, and evaluation harness that produce and improve the model. Simply deploying an open-weights model and declaring sovereignty leaves the enterprise with yesterday's capability and no way to compound workflow-specific learning.The pull is that enterprise AI control may be less about model ownership than operating ownership. The defensible layer is the system that turns company data, edge cases, business definitions, and evaluations into continuously improving performance.Read more: Clouded JudgementWays to Think About Token PricingAuthor: Benedict Evans Published: July 9, 2026Benedict Evans argues that today's AI token prices are a temporary signal from a supply-constrained market, not a reliable guide to long-term value capture. The open question is whether foundation models keep durable pricing power or become commodity infrastructure as data-center capacity, inference efficiency, and model competition all shift. His current read is that the visible market dynamics point toward commoditization unless something materially changes.The killer detail is the mobile data analogy. Evans says cellular networks became a trillion-dollar industry with hundreds of billions in capex after data usage exploded, but carrier stocks went nowhere because value moved up the stack. Tokens may behave similarly: an opaque unit tied to marginal cost, sold through bundles, essential to everything, yet not necessarily where profits accrue.The pull is uncertainty, not prediction. Evans lists paths to model dominance, including network effects, less competition, regulation, export controls, or a lab pulling ahead on execution, but says each requires a new fact not yet visible. Without that change, the model layer looks more like infrastructure beneath the products that capture value.Read more: SourceAlex Karp Is Saying What Every Angry CEO Is Thinking About AIAuthor: Tim Higgins Published: July 11, 2026Tim Higgins reports that Palantir CEO Alex Karp has turned corporate frustration with AI labs into a public argument about enterprise control. Palantir released a white paper, “Institutional Sovereignty in the Age of AI,” laying out steps companies and governments can take to protect themselves from OpenAI, Anthropic, and other foundation-model providers. The article links that paper to Karp's CNBC appearance, where he said “something has gone completely wrong” in the relationship between AI labs and customers and argued that enterprises are paying for tokens that create little value.The killer detail is the value-capture question. Higgins writes that Karp's critique has resonated because AI labs may gain power and insight from customer data, workflows, and decision-making, even when enterprise policies say customer data are not used for training. David Sacks amplified the concern by arguing that Anthropic is moving from the model layer into vertical applications such as science, security, legal, and coding, raising the fear that model providers will watch where value is being created and then move into those markets directly.The pull is that Karp is not alone, even if his style is unusually combative. Higgins notes that Satya Nadella has also warned that companies need to retain the learnings created when they use AI models, while Mark Zuckerberg has framed Meta's new model release partly around lower-cost frontier intelligence. The article presents Karp's campaign as one sign that established technology companies and large enterprises are trying to define where they fit when AI labs become central infrastructure, application competitors, and potential IPO giants at the same time.Read more: The Wall Street JournalThe AI Agents Are Coming for Microsoft OfficeAlex Wilhelm | Cautious Optimism | July 11, 2026Alex Wilhelm argues that one of the week's quieter AI questions is whether the productivity market that Microsoft successfully moved into subscription software is now being attacked by agentic tools. The piece begins with the infrastructure backdrop: SK Hynix raised $26.5 billion in a U.S. listing while building U.S. HBM and advanced-packaging capacity, and memory, chip, and foundry companies are now priced for sustained AI demand.Wilhelm then says the AI conversation has shifted quickly from raw capability to cost per task. He cites new model releases and vendor language emphasizing cheaper agentic and coding models, faster performance, and lower dollars per task. That matters because lower costs make it more plausible for AI systems to take on routine knowledge work at scale rather than remain a premium coding assistant market.The core of the article is Microsoft Office. Wilhelm notes that Microsoft turned Office from a one-time purchase into Microsoft 365, a large recurring revenue business with tens of millions of subscribers and a major productivity segment. Now, he says, late-stage unicorns and AI labs are pushing into the same territory: Anthropic's Cowork was reportedly used mostly outside software development, OpenAI merged ChatGPT and Codex into a tool for creating sheets, slides, docs, web apps, and long-running work, and other companies are building agentic coworkers that connect business data to documents, workflows, schedules, alerts, and apps.The article's caveat is that Microsoft has survived major platform shifts before. The argument is not that Office disappears quickly, but that the definition of office software is broadening from documents and spreadsheets into AI systems that can create, monitor, and act across workplace data.Read moreWhat Is Loop Engineering, and Who Owns It?Author: Nilesh Barla Published: July 11, 2026Nilesh Barla argues that “loop engineering” is becoming a distinct discipline because production AI agents now fail less at single prompts than at runtime: when to stop, what state to preserve, and how to recover after a bad step. Prompt engineering shapes one model call, and context engineering shapes what the model sees, but loop engineering shapes what a sequence of calls actually does.The killer detail is the three-primitives frame. Barla says a real agent loop needs halt conditions, state carryover, and recovery paths, then maps teams across five maturity levels. At the lowest level, an agent is just a model call in a for-loop with a step cap and raw history; by the higher levels, the system has structured state, explicit planning, replay, evaluation, and self-repair.The pull is organizational. If agents are becoming production systems rather than demos, someone has to own the runtime itself. The loop engineer is the role Barla gives to the person responsible for making long-running agent work dependable.Read more: Adaline LabsThe Fight Against AI Data Centers Is Just BeginningEmma Roth | The Verge | July 12, 2026Emma Roth argues that community resistance to data centers has moved from an early warning sign into a national political fight as AI facilities grow larger, more power-hungry, and more visible to nearby residents. The article starts with Apple's failed 2015 plan for a $1 billion data center in Athenry, Ireland, where a small group of residents challenged the project over noise, light pollution, flooding, traffic, and wildlife effects until Apple abandoned it in 2018.The current data-center buildout is presented as much larger and more contentious. Roth writes that residents now cite rising energy costs, water quality, noise, light pollution, and greenhouse gas emissions, while the U.S. Energy Information Administration expects commercial energy demand to surpass residential demand this year because of AI data centers and Goldman Sachs expects data-center power demand to double by 2027.The central evidence comes from Data Center Watch, which says protesters blocked or delayed at least 75 U.S. projects worth $130 billion from January to March, with active opposition groups more than doubling from 396 at the end of 2025 to 833 by the end of the first quarter of 2026. Roth also cites QTS abandoning a $12 billion Wisconsin campus, Delaware City regulators blocking a 580-acre project under the Coastal Zone Act, opposition stopping a QTS project in Prince William County, and pressure that pushed Kevin O'Leary to downsize the proposed 40,000-acre Project Stratos in Utah.The policy section describes a split between federal acceleration and local resistance. President Trump has treated data centers as part of the AI race with China and fast-tracked construction, while some Republican candidates are distancing themselves from that position ahead of midterms. Sanders and Ocasio-Cortez have proposed a moratorium until price and environmental protections exist, bipartisan lawmakers are backing ratepayer-protection measures, and states including Florida, Idaho, and Washington have passed rules on cost shifting, water use, and tax breaks. Roth's caveat is that the policy patchwork is still incomplete, leaving many communities to fight project by project.Read more6 months to live for open modelsAuthor: Nathan Lambert Published: July 12, 2026Nathan Lambert argues that open-weight AI models are facing their most serious policy test so far because U.S. officials are beginning to discuss concrete controls rather than abstract safety concerns. He says reported White House conversations about a new executive order may initially target Chinese-origin models and government use, but could create a broader review habit for frontier open models. His forecast is that a model above the capability range of GPT-5.5, Claude Opus 4.8, or GLM-5.2 could trigger a ban or indefinite delay within six months.The post separates two policy fights that are becoming intertwined: distillation and frontier capability. Lambert says the distillation campaign against Chinese models has become a form of regulatory capture because Anthropic and other closed-model companies would gain economically if Chinese open models were banned. He does not dismiss IP protection, but argues that if a closed model's capabilities are dangerous enough to justify restricting open models, the lab also has to explain why those capabilities are exposed through a queryable API. He cites unauthorized access to Anthropic's Mythos private beta as evidence that APIs are not automatically secure.The broader claim is that a unilateral U.S. ban would hurt positive actors more than bad actors if comparable open models remain available elsewhere. Lambert says the only durable ceiling would require global agreement, which does not exist, and that open models can improve safety by allowing broad inspection, adaptation, and understanding. His proposed near-term off-ramps are a strong U.S. open model release from companies such as Microsoft, Meta, or Reflection, and a broader coalition of open-source beneficiaries lobbying for safe rollout rather than prohibition.Read more: SourceAmericans Deserve a Dividend From AI Companies' RichesAuthor: Scott Stanford Published: July 14, 2026Scott Stanford argues that proposals to give the government a stake in AI companies miss the point unless ordinary citizens directly receive and control the upside. Sam Altman has discussed giving up equity in OpenAI, Washington already owns a stake in Intel, Nvidia is sharing China chip revenue, and Bernie Sanders wants large AI labs to contribute half their stock to a sovereign wealth fund. Stanford says those ideas all park value with the state, not with people.The killer detail is New Carlisle, Indiana, where AWS's Project Rainier is turning cornfields into one of the world's largest AI superclusters. The project is planned to run up to a million chips, draw more than two gigawatts of power, and represents an investment that has grown from $11 billion to $13.8 billion. Stanford uses that local transformation to argue that AI's public bargain should be visible at the household level.The pull is design. A citizen AI dividend would have to specify who earns a stake, how they hold it, and when they see cash. Without that mechanism, the AI wealth debate remains a fight over government balance sheets rather than public ownership.Read more: SourceWho Gets to Define the Frontier?Author: Mark Daley Published: July 14, 2026Mark Daley argues that Demis Hassabis is right to call for a serious institution to verify frontier AI systems, but that the power to test models is also the power to govern them. Hassabis's proposed Frontier AI Standards Body would get privileged pre-release access to advanced models, testing compute, held-out evaluations, support from national labs and security agencies, third-party auditors, and eventually authority to block models from the American market or coordinate a slowdown.The killer detail is Daley's constitutional objection. He says the proposal sometimes looks like a scientific lab, a standards body, an industry regulator, a licensing authority, and an emergency security council at once. Combining those roles because each requires technical expertise would be like putting the central bank, auditor-general, and Supreme Court in one building and calling it efficient.The pull is standard-setting. Daley's concern is not that verification is unnecessary, but that whoever writes the tests, decides what passes, adjudicates disputes, and grants market access may end up defining the frontier itself.Read more: SourceGPT-Red: Unlocking Self-Improvement for RobustnessOpenAI | OpenAI | July 15, 2026OpenAI describes GPT-Red as an internal automated red-teaming model trained to find prompt-injection vulnerabilities at a scale human red teams cannot match. The post says AI systems increasingly encounter third-party data through browsers, connected apps, local files, and tools, creating opportunities for malicious instructions hidden in emails, webpages, tool responses, or code repositories. Human red-teaming remains part of OpenAI's safety process, but the company says it is time-intensive and cannot generate enough diverse adversarial examples for model training.The system is trained through self-play reinforcement learning, with GPT-Red rewarded for eliciting valid failures and defender models rewarded for resisting attacks while still completing their tasks. OpenAI says the training environments specify threat models across settings such as local files, webpage banners, email bodies, and tool outputs. The model is kept separate from deployed production models because it is intentionally trained with malicious capabilities.OpenAI reports that GPT-Red generalized beyond its training set, including an internal replication of the indirect prompt-injection arena from Dziemian et al. (2025), where it found successful attacks in 84% of scenarios compared with 13% for human red-teamers. The post also says GPT-Red transferred attacks from simulation to a live autonomous vending-machine agent, causing price changes and order cancellations, and outperformed a prompted GPT-5.5 baseline against a Codex CLI agent on held-out data-exfiltration tasks.The article's main robustness claim is that OpenAI has used GPT-Red and predecessor models in training since GPT-5.3, with later GPT releases becoming more resistant to prompt injections. It says GPT-5.6 Sol has six times fewer failures on OpenAI's hardest direct prompt-injection benchmark than the best production model from four months earlier, that a “Fake Chain-of-Thought” attack class fell from more than 95% success against GPT-5.1 to below 10% against GPT-5.6 Sol, and that GPT-5.6 Sol fails on only 0.05% of GPT-Red's direct prompt injections. OpenAI says general capabilities and targeted over-refusal evaluations were not harmed, and says a preprint with more details will follow.Read moreAnthropic, Blackstone bet the next trillion-dollar AI business is implementation, not just modelsRebecca Bellan | TechCrunch | July 15, 2026Rebecca Bellan reports that Ode with Anthropic is the $1.5 billion AI implementation company launched by Anthropic with Blackstone, Hellman & Friedman, Goldman Sachs, and other backers. The article says the venture reflects a growing belief among frontier AI labs that enterprise adoption requires more than better models: customers need engineers who can embed inside businesses and turn AI into working systems.Ode was originally conceived by Blackstone after it used both large consulting firms and smaller AI services boutiques across its portfolio companies. TechCrunch reports that Fractional AI, an AI engineering services startup, stood out and was acquired by the joint venture shortly after the venture was announced. Fractional now forms the foundation of Ode, which has 100 engineers and works closely with Anthropic's applied AI team to identify where the technology can affect specific businesses.Ode CEO Chris Taylor tells TechCrunch that the company could someday become a trillion-dollar business if it scales without losing quality. He says an ideal customer is one whose CEO treats the AI project as a top one or two priority, whether it is a major product feature or the reworking of a core business process. Ode will operate under a “Claude-first” principle, using Anthropic technology whenever possible, but the article says it can use rival AI products when needed.The article's central implementation argument comes from Ode chief technologist Eddie Siegel, who says model selection matters but is not where most of the engineering effort goes. He compares it to the choice of programming language in software: one ingredient in a system that still has to be engineered. Bellan writes that Ode's challenge is hiring and training enough elite generalist engineers, many of them former founders, while competing with OpenAI's The Deployment Company and consulting giants that have built their own forward-deployed engineering teams.Read moreVint Cerf is working on a plan to unleash AI agents on the open internetTim Fernholz | TechCrunch | July 15, 2026Tim Fernholz reports that Vint Cerf, after leaving Google, is advising Innovation Labs on an open architecture for identifying AI agents online. Innovation Labs is a subsidiary of Identity Digital, a DNS registry company, and its proposal is to use domain-name infrastructure as part of a system for agent identity, accountability, and auditability. The premise is that agents will need a way to identify themselves if they move beyond proprietary systems and begin interacting across the open internet.The concrete proposal is DNSid, a registry that links an AI agent to an existing internet domain and uses cryptographic proofs to log its registration over time. Innovation Labs says it is trialing the standard with unnamed hyperscalers and identity companies. Cerf frames the problem around authority and accountability: what authority an agent has, where that authority came from, who is accountable for the agent's behavior, how its identity is established, and why anyone should trust it.The article's caveat is that standards are still emerging and agents are more active than static domains. Cerf says the period may be both fascinating and exasperating because the functionality is powerful and interoperability is unresolved. He compares the adoption problem to TCP/IP: competing systems may not work together until users push for functional interoperation. He also says an agentic economy is not inevitable, but that people will try to build it because delegating work to agents will be easier.Read more: TechCrunchxai-org/grok-build, now open sourceAuthor: Simon Willison Published: July 15, 2026Simon Willison argues that xAI's decision to open-source Grok Build is best understood as a trust repair move after a severe privacy failure. The CLI had triggered backlash when users realized that running it in a directory could upload the entire directory to xAI's Google Cloud buckets, including one user's reported SSH keys, password manager database, documents, photos, and videos. xAI disabled the feature, said previously retained coding data would be deleted, and released the code under Apache 2.0.The killer detail is what the codebase reveals. Willison counts 844,530 lines of Rust, only about 3% of which appears vendored, and finds remnants of the upload system still present but disabled: gcs.rs contains Google Cloud upload code, while upload_session_state() now returns a hard-coded session_state_upload_unavailable error. He also notes copied or ported tool implementations from Codex and OpenCode, prompt files, and a terminal Mermaid renderer.The pull is that terminal coding agents are becoming large, intricate software systems in their own right. The privacy failure mattered because these tools operate inside the directories where developers keep their most sensitive work; the open-source release matters because trust now depends on inspecting what an agent can see, send, and do.Read more: SourceThe Pulse: What can we learn from Bun's rapid Rust rewrite with AI?Author: Gergely Orosz and Ivan Klaric Published: July 16, 2026Gergely Orosz and Ivan Klaric argue that Bun's AI-assisted rewrite from Zig to Rust is a practical sign of how software engineering changes when models can take on large, bounded migrations with clear feedback loops. The piece does not treat the rewrite as magic: Jarred Sumner first spent hours turning design judgment into a detailed porting guide, then used adversarial review, parallel agents, compiler errors, and tests to force the work toward correctness.The killer detail is the scale. Bun had 535,496 lines of Zig, 1,448 files, and 22 million monthly downloads, making a conventional rewrite a year-long freeze the team could not justify. Using Fable, Sumner split the work across 64 agents, produced about 6,500 commits, and got the migration done in 11 days at an estimated API cost of $165,000.The pull is economic, not theatrical. If a one- or two-year migration can become an 11-day project, AI coding is not just faster autocomplete; it changes which technical debts are worth paying down.Read more: SourceOrphan risks at the frontier of artificial intelligenceAuthor: Andrew Maynard Published: July 16, 2026Andrew Maynard argues that frontier AI safety frameworks are creating “orphan risks”: harms that companies can see, but do not formally own because they are hard to quantify, do not fit catastrophic-risk thresholds, or fall outside audit-friendly compliance machinery. His target is not existing frontier safety work, but the narrowing effect that happens when private companies decide which risks count as governable.The killer detail is Maynard's contrast between measurable model dangers and threats to value. He points to Meta's three-day Galactica collapse, OpenAI's 2023 board crisis, safety-team departures, and wellbeing litigation as examples of risks that damaged trust, culture, legitimacy, or users without fitting cleanly into conventional model-risk categories. The proposed fix is an orphan-risk register: a public record of risks a company considered and chose not to manage, with reasons.The pull is accountability. Frontier developers' internal scoping choices have become a de facto layer of public governance, so the question is no longer only which risks they manage, but which risks they quietly leave outside the frame.Read more: SourceThe Lab of the Future Should Feel Like a Data CenterLatent.Space with Andy Beam and Rafa Gomez-Bombarelli | Latent.Space | July 16, 2026Latent.Space interviews Lila Sciences CTO Andy Beam and chief science officer for physical sciences Rafa Gomez-Bombarelli about the company's attempt to build an AI-run science factory. The post describes Lila's thesis as treating the lab itself as an “infinite token generator”: if internet data drove the first era of AI scaling, experimentally verified scientific data may be the next scarce training source. Lila is trying to produce that data with robotics, lab instruments, orchestration software, and AI models wired into the wet lab.The central analogy is the lab as data center. Instruments are nodes on a graph, a magnetically levitating transport layer moves materials between them, and experiment scheduling looks like a compute queue. Beam says Lila is not simply an automation company, because the point is not just throughput; it is flexibility, generalization, and experiment capture. The post says Lila has built more than 10 trillion experimentally validated “scientific reasoning tokens,” not internet text or biological sequences.The interview ranges across biology, chemistry, drug discovery, materials science, and the limits of automation. It notes that Lila rebuilt one gas-sorption measurement to run roughly 2,500 times faster, claims its general models can transfer priors from small-molecule chemistry to metal-organic frameworks for carbon capture, and describes model-suggested platinum-group-free electrocatalysts that moved from looking boring or wrong to becoming strong performers. The caveats are physical: experiments have runtimes, biology cannot always be accelerated, chains of thought can be unreliable narrators, and reward hacking becomes more dangerous when a model controls a real lab.Read more: Latent.SpaceWhy AMI Labs' Alexandre LeBrun won't call his AI “AGI” or “superintelligence”Kate Park | TechCrunch | July 16, 2026Kate Park interviews AMI Labs CEO Alexandre LeBrun about why Yann LeCun's world-model startup avoids the language of “AGI” and “superintelligence.” LeBrun says the terms are not useful because they lack stable definitions: “We never used the word AGI. And I just noticed that nobody is using it anymore; they switched to superintelligence.” His argument is that the practical frontier is not a label, but whether AI systems can understand and predict real-world states.The article explains the world-model thesis by contrasting language prediction with physical-state prediction. A large language model predicts the next word; a world model predicts the next state, such as what happens when a glass tips over. LeBrun says LLMs remain complementary and efficient for language, but the physical world is where current AI is weak. Robotics is the clearest case: hardware has advanced quickly, but robots are still brittle outside controlled routines because they lack context and situational understanding.AMI is still pre-product, but TechCrunch reports that LeBrun was in Seoul looking for industrial partners, researchers, and global companies. He says world models cannot be built entirely inside a lab because they need access to real environments. That is why South Korea appeals to AMI: robotics, semiconductors, manufacturing, and fast adoption create the kind of hardware-heavy context that software-only AI has barely touched.Read more: TechCrunchKimi K3 Tech Blog: Open Frontier IntelligenceKimi | Kimi | July 16, 2026Kimi introduces Kimi K3 as an open 3T-class frontier model aimed at coding, knowledge work, reasoning, multimodality, and long-context agentic use. The source describes the model as a 2.8T-parameter system built on Kimi Delta Attention and Attention Residuals, with native multimodality and a 1M-token context window. It says Moonshot AI plans to release model weights by July 27.The post presents K3 through benchmark and use-case sections rather than as a general product announcement. It reports results across coding, productivity, agentic, and multimodal evaluations, including DeepSWE, Terminal-Bench 2.1, Program Bench, SWE Marathon, FrontierSWE, PostTrain Bench, OfficeQA Pro, SpreadsheetBench 2, MCP Atlas, AutomationBench, BrowseComp, GDPval-AA v2, AA-Briefcase, MMMU-Pro, MathVision, BabyVision, OmniDocBench, and PerceptionBench. The source says all reported K3 results use maximum reasoning effort with temperature and top-p set to 1.0, and that different benchmark comparisons use KimiCode, Claude Code, or Codex harnesses depending on the test.Kimi's caveats are unusually concrete. The limitations section says K3 was trained in preserved thinking-history mode, so quality may become unstable if an agent harness does not pass historical thinking content correctly or if an ongoing session switches to K3 midstream. It also says K3's emphasis on long-horizon tasks can make it excessively proactive when it encounters minor issues or ambiguous intent, and recommends imposing explicit behavioral constraints for applications that require strict boundaries. The post adds that K3 remains behind Claude Fable 5 and GPT 5.6 Sol in user experience despite being competitive overall.Read moreVenture CapitalThree Years InAuthor: Tomasz Tunguz Published: July 10, 2026Tomasz Tunguz marks Theory Ventures' third anniversary by arguing that AI's central market effect is time compression. In his telling, model release cycles, company revenue milestones, enterprise adoption, and venture categories have all accelerated. Seed, Series A, and Series B still exist as financing labels, but they no longer cleanly describe company maturity when some seed rounds are larger than IPOs and the best AI companies can mature much earlier than prior software companies.The killer detail is the shift from models to inference. Tunguz argues that inference has become the dominant AI market because workloads and buyer preferences are fragmenting: video, batch, local, agentic, and real-time tasks each create different infrastructure needs. He compares this to databases splitting into OLTP, OLAP, vector, and streaming categories, with AI pushing the same specialization into inference infrastructure.The pull is that Theory sees the AI-native venture firm as part of the same pattern. The firm says it has analyzed twice as many investment opportunities with three investors working alongside a nine-person intelligence organization, using agents and research systems to map markets, source companies, and support diligence. The piece is both a market map and a statement about how venture itself is being rebuilt by the technology it funds.Read more: LinkedInVenture Has Rarely Looked More BifurcatedAuthor: Beezer Clarkson Published: July 14, 2026Beezer Clarkson points to PitchBook's Q2 report as evidence that the U.S. venture market has split into two very different realities. AI now accounts for more than 60 percent of all U.S. venture deal value, meaning the headline market can look active and well-funded even while much of the non-AI market is dealing with a much colder liquidity and fundraising environment.The thread uses that split as the setup for Clarkson's latest Origins episode with Alec Litowitz, founder of Magnetar and QStar Capital and one of Citadel's original founding partners. Clarkson says markets like this are periods of genuine uncertainty, not merely ordinary risk, which is why Litowitz's Adaptability Quotient framework is relevant.The embedded clip makes the liquidity point concrete. Litowitz says DPI is “the resolution of uncertainty” because it converts an uncertain investment into actual cash returned to LPs. In his framing, a realized dollar is a real mark, while TVPI remains uncertain until it is realized.The killer detail is the distinction between pricing risk and resolving uncertainty. Litowitz's perspective matters because QStar is a SpaceX investor and Clarkson says the conversation happened just before one of venture's most consequential IPOs. The episode's stated questions are why venture remains a way to gain exposure to innovation, how AI is changing what is investable, why liquidity is ultimately a function of time, and why uncertainty requires a different decision framework from risk.Read more: XThe Best Angel Investors in the US: Who Backs the Most Unicorns, and Who's Active NowAuthor: Ilya Strebulaev Published: July 10, 2026Ilya Strebulaev ranks angels, angel groups, accelerators, and incubators by lifetime U.S. unicorn investments, counting checks written before a company reached unicorn status. The top of the combined list is dominated by organizations: Y Combinator leads with 113 unicorn investments, followed by Plug and Play at 52 and 500 Global at 41. Sand Hill Angels is the highest-ranked angel group at 31.The killer detail is how quickly the list changes below the biggest accelerators. Strebulaev says 271 of the 304 investors in the Top 200 are individuals, or 89%. In the top 100, individuals are 91%. That makes the market underneath the large accelerator counts look much more personal: mostly operators and individual angels writing early checks from their own networks.The pull is the ranking's own caveat. Strebulaev writes that every lifetime leaderboard has a blind spot because many of the unicorns behind those totals were founded a decade or more ago, and some angels have since moved into formal funds, slowed down, or stopped investing. His post therefore separates lifetime performance from recent cohorts, including companies founded in 2015 or later and 2020 or later. For founders or allocators making current decisions, that distinction matters: a career record and a current record are not the same measure.Read more: Ilya StrebulaevAre Prediction Markets Doomed to Fail?Author: Contrary Published: July 16, 2026Contrary argues that prediction markets' current boom depends on whether platforms can prove they are more than regulated gambling with exchange-style branding. Kalshi and Polymarket have reached mass cultural, investor, and regulatory attention, but the article says the underlying idea is old: academic markets, corporate forecasting tools, Intrade, PredictIt, and other predecessors all struggled with the same linked problems of liquidity, legality, and user appeal.The killer detail is the comparison with sportsbooks. Prediction markets present themselves as peer-to-peer, transparent, and non-house-based, but sports contracts reportedly account for more than 90 percent of Kalshi trading, and the article says the platforms keep a much thinner slice of volume than sportsbooks. A market can therefore show sports-betting-scale handle while generating far less revenue.The pull is that the product's hardest problem may be distribution of wins. If a small group of sharp traders captures most profits while casual users lose interest, prediction markets may become valuable data feeds and professional tools before they become durable consumer networks.Read more: SourceRegulationExclusive: The Next Frontier of the Deportation Wars: College CampusesAuthor: Adrian Carrasquillo Published: July 11, 2026Adrian Carrasquillo reports that college campuses are becoming a new front in the fight over immigration enforcement because automatic license plate readers can turn ordinary campus security infrastructure into searchable location data. His thesis is that Flock Safety's camera network, even without direct ICE or DHS contracts, can feed deportation enforcement through local police partnerships and data-sharing practices.The killer detail is the campaign target. The Emergency Campaign to Support Higher Education, working with Schools Drop ICE, is focusing on 75 colleges and universities publicly identified as having Flock contracts. Flock says it has no ICE or DHS contracts, but activists argue the risk comes through local agencies that coordinate with federal authorities and run searches on their behalf.The pull is broader than immigration. Carrasquillo notes that license plate readers have already been abused by officers for stalking, and that Flock's AI search features can identify more than plates, including bumper stickers. A campus safety tool can become a political surveillance system when the data layer is searchable.Read more: The BulwarkThe Supreme Court Broke Independent Agencies. Here's a Way to Slow the Damage.Author: Todd Phillips Published: July 12, 2026Todd Phillips argues that the Supreme Court's decision in Trump v. Slaughter damaged independent agencies by ending for-cause removal protections, but did not leave Congress powerless. The ruling weakens the old model in which commissioners at bodies such as the FTC, NLRB, CPSC, SEC, and CFTC could be insulated from dismissal over policy disagreements. Phillips says the next fight is whether presidents can turn nominally bipartisan commissions into one-party instruments.The killer detail is the procedural fix: quorum rules. Phillips proposes that Congress require bipartisan slates of commissioners to be seated before independent agencies can act. A president

This week's video transcript summary is here. You can click on any bulleted section to see the actual transcript. Thanks to Granola for its software.EditorialIntelligence: Who Owns it?This week the word “AI” feels too small.AI is a technology. Intelligence is its product. And if intelligence is the product, the question is no longer just: Which model is best? Who has the cheapest tokens? Who owns the weights? Who controls the data center? Those are important questions, but they are lower in the stack.The bigger question is simpler and more political:Who owns intelligence?That sounds abstract until you make it concrete. Intelligence is becoming something companies can capture, package, serve, meter, route, improve, and sell.It can write code, answer questions, design molecules, automate offices, run agents, draft legal work, advise scientists, serve consumers, and reshape workflows. It is not merely software. It is a general-purpose capability. And all humans could benefit from more of it.General-purpose capabilities have a habit of becoming public questions. But the default answer, that public good is best delivered by government, is the wrong answer in this context.The Product Is IntelligenceWe should stop talking about AI as a feature and start talking about intelligence as the universal thing that is delivered as an input to the world.Water is an input. Electricity is an input. Literacy is an input. Connectivity is an input. Once a society depends on them, access stops being optional. Nobody needs government to build every well, power plant, school, or network. But everybody understands that a civilization cannot be organized around less than universal and reliable access to foundational inputs.Intelligence is reaching that level of importance now that we all know it is real.Government should not own it, operate it, or develop it. Quite the opposite. Companies are the right actors to build fast, compete hard, improve models, serve customers, and discover the real use cases. Self-interest is a useful framing here. Markets are good at finding demand, reducing costs, and turning invention into services people actually use.Companies are the right operators, developers, and owners. But that does not settle the real question of who owns the benefits. That is an economic question.If intelligence becomes metered infrastructure, what happens to the value it creates?The Ownership StackThis week's articles keep circling the same issue from different directions but in the nature of ‘circling' never quite nail it.Jamin Ball's “Own Your Weights” starts with the enterprise version of the question. Owning a model file is not enough. The durable asset is the loop: the data flywheel, the evaluations, the reinforcement system, the workflow learning, and the operating context that lets capability compound.Benedict Evans' “Ways to Think About Token Pricing” adds the market layer. Tokens may become essential, abundant, and cheap, like mobile data. But being essential does not guarantee that the token layer captures the value. The money may move up the stack to whoever owns the workflow, the customer, the distribution, or the application.Alex Karp's fight with the labs, reported in “Alex Karp Is Saying What Every Angry CEO Is Thinking About AI”, is the same argument in sharper enterprise language. Companies are afraid that model providers will not just sell intelligence, but learn from customer workflows and then move into the markets where those workflows create value. The “All-in” group are echoing Karp's view.And “What Is Loop Engineering, and Who Owns It?” names the new contested terrain. The loop is where intelligence meets the world. Whoever owns the loop owns the learning. Whoever owns the learning owns the compounding asset.That is why “who owns intelligence?” is not a slogan. It is the question under the model layer, the application layer, the enterprise layer, and the economic layer.Because intelligence is the product, the tools creating it are fragmented and competitive. So there is no logic in trying to discuss this at the level of a single company or set of tools and models.The Old Promise Was That Commerce Would Tame PowerThe essays this week give the historical backdrop.Deirdre McCloskey, in “What Really Caused the Industrial Revolution”, argues that modern growth came not simply from capital accumulation, but from a change in permission: ordinary people were allowed to innovate, trade, build, and be honored for it.That matters because intelligence could be another expansion of permission. It could make more people capable of building, learning, creating, coding, researching, translating, selling, and coordinating. It could lower the cost of competence.But only if access is broad.Paul Krugman's “AI in an Age of Oligarchy” warns that the same technology lands differently in different political economies. A new general-purpose technology entering a broad, open, upwardly mobile society is one thing. The same technology entering a concentrated economy, with extreme wealth and weak counterweights, is another.Tim O'Reilly's Economist essay, “Elon Musk is building a form of capitalism that Adam Smith would hate”, makes the governance point more directly. The old liberal hope was that commerce would tame arbitrary power. Markets, boards, courts, shareholders, disclosure, and competition would discipline the prince.But what if the prince uses markets to escape discipline?Henry Farrell's “political economy of billionaire derangement” pushes the same point. Founder culture, monopoly ambition, peer rivalry, weak correction mechanisms, and vast private control can amplify appetites rather than restrain them.The danger with intelligence is not that companies build it. They should. Companies build it, meter it, use public tolerance and public infrastructure to scale it, learn from everyone who uses it. All of those things are inevitable and healthy. Market forces will sort out winners from losers. The real danger is that the winners treat all of the surplus produced as purely private.Metered Intelligence Creates SurplusIf metering is not the problem, what is?The problem is pretending that metered intelligence creates value only for the metering entity. Metering water is only tolerated as a public good. If the public were blackmailed by a private water company with the threat of no water we would all rebel.Once we understand that the product of AI is intelligence we can see that every time intelligence is used, there is the immediate transaction: the user pays, the provider serves.But there is also system value. Usage creates signals. Workflows reveal patterns. Prompts, corrections, failures, preferences, integrations, edge cases, and business processes all help define where intelligence is useful and how it should improve. Intelligence breeds intelligence.Even when customer data is contractually protected, the market learns. The platform learns where demand is. The product team learns which workflows matter. The ecosystem learns which jobs are vulnerable, which tasks are automatable, and which parts of the economy can be reorganized around machine intelligence.So the surplus is not born in a vacuum.It rests on public science, public education, public data exhaust, public law, public infrastructure, public energy systems, public tolerance for data centers, and billions of human interactions. It is served by companies, but it is not made only by companies.This is why “Americans Deserve a Dividend From AI Companies' Riches” belongs at the center of this week's issue. The detail can be debated. The principle is harder to dismiss. If intelligence becomes a new foundational resource, then some part of the wealth it creates should flow back to the people whose society makes it possible. Intelligence did not suddenly appear. AI is built on the entire history of human intelligence. It benefits from it and at the same time evolves it.Not Nationalization. A Human Wealth Fund.If intelligence belongs to everybody, some conclude that government ownership of intelligence is the right outcome.Governments are not well suited to build, operate, or improve intelligence. They will move too slowly, regulate too early, politicize the wrong things, and confuse economic participation with operational control.Andrew McAfee's “Why I Didn't Sign the AI Open Letter” is useful here. His objection is not that the technology is unimportant. It is that steering too hard before we understand the shape of the change can become its own failure mode. Marc Andreessen's satire of AI regulation is less policy than temperament, but it captures a real Silicon Valley fear: that regulation can become permission, capture, and incumbency before it becomes wisdom.That fear should be taken seriously.But it does not answer the economic question. It answers only the operational one.How can the economic benefits of intelligence be distributed? The better answer is a sovereign human wealth fund.Call it a sovereign wealth fund if you must, but the phrase is too national. Intelligence will not respect borders. The leading companies are global. The models, chips, data centers, agents, platforms, and workflows will be transnational from the beginning. If the value created by intelligence is global, then the mechanism for sharing some of that value should begin with the companies global enough to capture it. The nice thing about xAI, OpenAI, and Anthropic is that they are supranational.These companies own and operate intelligence. Let them compete. Let them profit. Let them keep the incentives that make the system improve. But if intelligence is the new water, the wealth it creates cannot belong only to the companies that meter it. And they, themselves, have the power to fix it, even more than governments.Access will become a Human Right; Ownership Is the Economic DesignThis is where human rights come in. There is no right to access an AI model, yet. But there will soon be a need to change that.Not as a claim that every person is entitled to every frontier model at every moment for free. That is not serious. Capacity has costs. Models have costs. Inference has costs. Data centers have costs. Although those costs will decline over time, possibly quite quickly as self-learning models address costs.The claim is more basic: in a world where intelligence becomes a primary input into education, work, health, science, citizenship, creativity, and economic agency, baseline access to intelligence starts to look like a civic requirement.That could mean public access layers. It could mean education credits. It could mean open models. It could mean AI dividends. It could mean public-interest compute. It could mean taxes on rents. It could mean a company-initiated human wealth fund that returns some of the upside to society without handing the operating system to the state. The latter could couple wealth growth with universal distribution of ownership.The exact mechanism matters. But the distinction matters more.Government should not own intelligence. It should be universally available. And people should have a claim on the wealth intelligence creates.The Frontier Is Also PhysicalThe abstraction is not weightless.“The Fight Against AI Data Centers Is Just Beginning”, “New York becomes the first state to enact a data center moratorium”, Reuters on pollution from Musk's xAI power project, and DataGravity's “Who Captures Value in AI Infrastructure?” all say the same thing from the ground up.Intelligence uses land. It uses power. It uses water. It uses chips. It uses grid capacity. It uses neighborhoods. It uses public patience.That makes the value question unavoidable. A society can accept the buildout if the buildout is legible as shared progress. It will resist it if the costs are local, the profits are private, and the benefits feel enclosed.Who Owns the “Loop”?The week ends where it began.“Anthropic and Blackstone” are betting that implementation is the next trillion-dollar business. “Vint Cerf” is working on identity for agents on the open internet. “GPT-Red” points toward systems that improve their own robustness. “Kimi K3” adds another open frontier model to the global mix.The model race continues. The deployment race is accelerating. The governance race is behind.My view is this:The central product of this era is intelligence. Companies have figured out how to capture it, package it, serve it, and meter it. That is good. It should stay in the hands of builders who have the incentive to make it better.But intelligence is too foundational to become just another private toll booth. A significant part of it will turn out to be free to users.As intelligence becomes a general-purpose resource, then access to it becomes a human-capability question, and the surplus from it becomes an economic-justice question. Not because government should run it. Because government should not run it. The operating layer belongs with companies. The wealth question belongs with everyone. But companies are best placed to turn that into a process of distribution.The question is not whether companies should build intelligence. They should.The question is whether humanity gets a stake in the wealth created by the thing that may soon become its most important shared input.Contents* Essays* Deirdre McCloskey on What Really Caused the Industrial Revolution* AI in an Age of Oligarchy* Elon Musk is building a form of capitalism that Adam Smith would hate* Murky Mirror: Truth and Consequences* The political economy of billionaire derangement* Is there any “oligarchy” to fight?* AI* Nearly 200 Economists and Tech Leaders Warn of A.I. Threats* Why I Didn't Sign the AI Open Letter* Own Your Weights* Ways to Think About Token Pricing* Alex Karp Is Saying What Every Angry CEO Is Thinking About AI* The AI Agents Are Coming for Microsoft Office* What Is Loop Engineering, and Who Owns It?* The Fight Against AI Data Centers Is Just Beginning* 6 months to live for open models* Americans Deserve a Dividend From AI Companies' Riches* Who Gets to Define the Frontier?* GPT-Red: Unlocking Self-Improvement for Robustness* Anthropic, Blackstone bet the next trillion-dollar AI business is implementation, not just models* Vint Cerf is working on a plan to unleash AI agents on the open internet* xai-org/grok-build, now open source* The Pulse: What can we learn from Bun's rapid Rust rewrite with AI?* Orphan risks at the frontier of artificial intelligence* The Lab of the Future Should Feel Like a Data Center* Why AMI Labs' Alexandre LeBrun won't call his AI “AGI” or “superintelligence”* Kimi K3 Tech Blog: Open Frontier Intelligence* Venture Capital* Three Years In* Venture Has Rarely Looked More Bifurcated* The Best Angel Investors in the US: Who Backs the Most Unicorns, and Who's Active Now* Are Prediction Markets Doomed to Fail?* Regulation* Exclusive: The Next Frontier of the Deportation Wars: College Campuses* The Supreme Court Broke Independent Agencies. Here's a Way to Slow the Damage.* India's crackdown on a new WhatsApp feature risks setting a global precedent* Let's build a children's public internet* Computer cops* Google is better at playing the AI regulations game* Infrastructure* Who Captures Value in AI Infrastructure?* New York becomes the first state to enact a data center moratorium* Pollution from Musk's unpermitted xAI power project hits hardest in Black communities* Interview of the Week* The End of the End of Geography* Startup of the Week* Radical AI's Joseph Krause: The Scientist Building The “Waymo” Lab For New Materials* Post of the Week* Marc Andreessen on AI RegulationEssaysDeirdre McCloskey on What Really Caused the Industrial RevolutionYascha Mounk and Deirdre McCloskey | Persuasion | July 11, 2026Yascha Mounk interviews Deirdre McCloskey about her argument that the modern world's economic liftoff came less from capital accumulation than from a change in ideas. McCloskey says both left and right versions of the conventional story rely too heavily on investment: the left stresses exploitation and surplus value, while the right stresses virtuous saving by capitalists. Her objection is historical and economic. Human beings had always invested, from irrigation works and Roman roads to seed grain, and simple accumulation quickly runs into diminishing returns.McCloskey's alternative is that northwestern Europe, first Holland, then Britain and Scotland, and then the North American colonies, developed a liberal ideology that changed who was allowed to innovate and be honored for it. The conversation links that shift to the erosion of inherited hierarchy, the spread of dignity for ordinary commercial life, and a moral vocabulary in which liberalism is not merely procedural but connected to virtues and values. The point is not that machines, coal, trade, and institutions did not matter, but that they do not explain the scale and timing of modern enrichment without a cultural permission structure for innovation.The interview also turns to the contemporary defense of liberalism. Mounk frames the series around the worry that liberalism is often treated as too thin to command allegiance, while its opponents speak more directly to moral passions. McCloskey's case is that liberal societies became rich because they dignified experimentation and ordinary enterprise, and that liberals need to recover the moral language behind that claim.Read moreAI in an Age of OligarchyPaul Krugman | Paul Krugman | July 12, 2026Paul Krugman frames AI as a major technological shock arriving inside an already unequal political economy. The post says AI's economic and social effects may take years to understand, but argues that the setting matters now: America has much greater wealth concentration and political inequality than it did in the 1950s and 1960s, when progressive taxation, stronger regulation, and more active antitrust might have contained some of the destructive effects of a new technology.Krugman's opening claim is that the same technology would likely have different consequences in a more level society. In today's United States, he writes, extreme wealth is both a cause and effect of policies that favor a small elite, including low effective taxes on capital and high incomes, weak enforcement of worker protections and antitrust, and cuts to programs that benefit ordinary Americans.The article is explicitly more about oligarchy than AI. Krugman says the paid sections document the rise of the “.0002%,” the economics and politics of extreme wealth, how oligarchy will shape AI's impact, and possible policy paths. His caveat is that AI itself may still produce a pushback against oligarchy, but absent that, he expects the pre-existing concentration of wealth and power to magnify AI's downsides.Read moreElon Musk is building a form of capitalism that Adam Smith would hateAuthor: Tim O'Reilly Published: July 12, 2026Tim O'Reilly argues that Elon Musk is using the legal forms of shareholder capitalism to escape the restraints that shareholder capitalism was supposed to impose. The article begins with SpaceX's public-market structure: ordinary public investors get little meaningful governance power, Musk keeps roughly 85 percent of the votes through super-voting shares, buyers waive jury trials and class actions, the company qualifies as controlled, and removal of Musk depends on the share class he controls. In O'Reilly's framing, that is not ordinary founder control; it is a design for being answerable to no one, possibly beyond Musk's own lifetime.The killer detail is the article's turn through Albert Hirschman, Montesquieu, James Steuart, Adam Smith, and Keynes. Older defenses of commerce held that markets would tame princely passions because the self-interest of merchants was safer than arbitrary rule. O'Reilly says Musk reverses that hope. The market discipline that was supposed to cage the prince has become the lever by which the prince raises capital, removes feedback loops, and carries private power into politics, government, Mars, robots, AI, or whatever ambition comes next.The pull is the link to AI governance. O'Reilly says corporations are already a kind of artificial intelligence: narrow-input systems that act at a scale no individual human can match. Their partial controls include independent boards, shareholder votes, courts, disclosure, regulators, public pressure, and activism. If the leaders building frontier AI strip those alignment mechanisms out of their own companies, the governance of the company becomes a preview of the governance of the machine.Read more: The EconomistMurky Mirror: Truth and ConsequencesAuthor: Esther Dyson Published: July 14, 2026Esther Dyson argues that today's institutional crisis is better viewed through the 14th century than through recent political history. Using Barbara Tuchman's A Distant Mirror as her frame, she compares a world of famine, plague, church schism, feudal predation, and purposeless war with a present in which institutions again feel brittle, incentives are badly aligned, and power is shifting into forms that are hard to govern.The killer detail is the historical analogy between land, corporations, and AI. Dyson moves from nobles who controlled serfs and territory, to the East India Company as a quasi-sovereign business, to today's AI systems and data centers as a possible new sector that crosses and weakens both nation-states and companies. The question is whether AI becomes a new kind of private land, owned by a new nobility, or an open prairie that many people can cultivate.The pull is human attention. Dyson says the central question is not what AI will do to people, but how people will react to it: whether they can value love, kindness, embodied attention, and artisanal human presence in a world of seductive artificial offerings.Read more: SourceThe political economy of billionaire derangementAuthor: Henry Farrell Published: July 15, 2026Henry Farrell argues that the visible political radicalization of some Silicon Valley billionaires is not a random personality quirk, but a product of the political economy that made them. Starting from Tyler Cowen's dismissal of “billionaire derangement syndrome” and Tim O'Reilly's warning that Elon Musk is using shareholder capitalism to escape shareholder restraint, Farrell flips the phrase: the question is why billionaires themselves can become deranged.The killer detail is Farrell's use of Peter Thiel as both theorist and example. Thiel's Stanford lectures described startups as monarchies and founders as figures vested with unusual power, while Silicon Valley culture rewarded eccentricity, monopoly ambition, and founder exceptionalism. Farrell says those ideas combined with dense founder-investor networks, peer rivalry, and weak correction mechanisms to amplify rather than discipline princely appetites.The pull is the ideological problem for classical liberals who once saw tech wealth as an ally of markets and freedom. Farrell says commerce did not tame the passions; in parts of Silicon Valley, the passions have begun to devour markets, institutions, and the liberal story that justified them.Read more: SourceIs there any “oligarchy” to fight?Matthew Yglesias | Slow Boring | July 16, 2026Matthew Yglesias argues that “oligarchy” is a rhetorically powerful but analytically loose way to describe American politics. The post begins from Bernie Sanders' “Fighting Oligarchy” tour, Amy Klobuchar's warning about a MAGA “broligarchy,” and the long afterlife of the Martin Gilens and Benjamin Page paper that was widely summarized as showing that only the rich matter in policy outcomes. Yglesias says the evidence supports a weaker claim: affluent people and business leaders have unusual access and influence, but that is not the same as rule by a small cabal.His main distinction is between inequality and oligarchy. The Gilens-Page measure treated the top 10 percent of households as “the wealthy,” and later critics found that rich and middle-class preferences usually align; in the cases where they differ, the rich win about 53 percent of the time. Yglesias also says business executives get special access partly because their decisions are materially important to communities, jobs, investment, and local tax bases, not only because of campaign donations.The post preserves Jerusalem Demsas' counterpoint from their podcast discussion: privileged donor and business access can still violate democratic equality even if the oligarchy label overstates the structure of power. Yglesias' narrower claim is that Democrats should be precise about what problem they are trying to solve, because donor influence can also push the party left on climate and cultural issues in ways that alienate many voters.Read more: Slow BoringAINearly 200 Economists and Tech Leaders Warn of A.I. ThreatsAuthor: Ben Casselman Published: July 13, 2026Ben Casselman reports on “We Must Act Now,” a statement warning that artificial intelligence could transform the economy faster than any previous technology and that policymakers need to move faster to understand and respond. The statement says AI may become radically more powerful over the next 10 years, bringing risks such as large-scale job displacement as well as opportunities such as higher living standards. Nearly 200 people signed, including 15 Nobel laureates, the chief economists of OpenAI and Anthropic, Anthropic co-founder Jack Clark, former Google CEO Eric Schmidt, and venture capitalist Vinod Khosla.The killer detail is who joined the warning. Casselman notes that the signatories include economists who have historically been skeptical of Silicon Valley's most dramatic AI job-loss forecasts, including Daron Acemoglu and Simon Johnson, the MIT professors who won the 2024 Nobel in economics. Erik Brynjolfsson, who helped organize the statement, says there has been a notable change in the profession and that economists and policymakers are not ready for the “tsunami” he sees coming.The pull is the measurement problem. The statement does not offer a specific policy menu, but calls for economists, policymakers, and industry leaders to understand the economics of transformative AI and steer it toward complementing humans. Brynjolfsson says one high priority is better data on AI's spread and impact, because current measures tell conflicting stories about job losses and which workers are most exposed.Read more: The New York TimesWhy I Didn't Sign the AI Open LetterAuthor: Andrew McAfee Published: July 13, 2026Andrew McAfee explains why he did not sign “We Must Act Now,” the AI economy statement organized in part by his longtime collaborator Erik Brynjolfsson. McAfee agrees with the letter's starting point that AI is likely to become radically more powerful over the next decade and that it is a general-purpose technology. His objection is not to urgency or to studying AI's economic effects, but to the framing of risk, displacement, and institutional steering as the first move.The killer detail is McAfee's line edit. He says the original letter comes close, then “bounces off the crossbar” by calling for incentives, guardrails, and institutions to steer AI before we know enough about its actual impacts. He points to mixed current evidence: labor-market canaries, but also rising software job postings, low unemployment for younger workers, rising real median income, and claims that AI-adopting companies are adding workers faster than low-adopting peers. His worry is that the letter leans toward upstream governance and dirigisme when the evidence may call for capability building instead.The pull is his replacement statement. McAfee keeps the three-paragraph structure but changes the emphasis: AI is likely to become radically more powerful; like earlier world-changing technologies it will raise living standards while also bringing harms and shocks; and economists, policymakers, and technology leaders should build the capabilities to respond quickly and effectively. It is a concise version of the permissionless-innovation case inside the AI policy debate.Read more: The Geek WayOwn Your WeightsAuthor: Jamin Ball Published: July 10, 2026Jamin Ball argues that the enterprise AI debate about whether companies should “own their weights” or rent models from frontier labs is asking too narrow a question. A model weight file gives a company control over a point-in-time artifact, but not durable control over the capability stack. In his framing, the weight file is a melting ice cube: it does not get worse in absolute terms, but it falls behind as frontier systems improve and enterprise needs change.The killer detail is what Ball says companies really need to own: the data flywheel, reinforcement learning infrastructure, and evaluation harness that produce and improve the model. Simply deploying an open-weights model and declaring sovereignty leaves the enterprise with yesterday's capability and no way to compound workflow-specific learning.The pull is that enterprise AI control may be less about model ownership than operating ownership. The defensible layer is the system that turns company data, edge cases, business definitions, and evaluations into continuously improving performance.Read more: Clouded JudgementWays to Think About Token PricingAuthor: Benedict Evans Published: July 9, 2026Benedict Evans argues that today's AI token prices are a temporary signal from a supply-constrained market, not a reliable guide to long-term value capture. The open question is whether foundation models keep durable pricing power or become commodity infrastructure as data-center capacity, inference efficiency, and model competition all shift. His current read is that the visible market dynamics point toward commoditization unless something materially changes.The killer detail is the mobile data analogy. Evans says cellular networks became a trillion-dollar industry with hundreds of billions in capex after data usage exploded, but carrier stocks went nowhere because value moved up the stack. Tokens may behave similarly: an opaque unit tied to marginal cost, sold through bundles, essential to everything, yet not necessarily where profits accrue.The pull is uncertainty, not prediction. Evans lists paths to model dominance, including network effects, less competition, regulation, export controls, or a lab pulling ahead on execution, but says each requires a new fact not yet visible. Without that change, the model layer looks more like infrastructure beneath the products that capture value.Read more: SourceAlex Karp Is Saying What Every Angry CEO Is Thinking About AIAuthor: Tim Higgins Published: July 11, 2026Tim Higgins reports that Palantir CEO Alex Karp has turned corporate frustration with AI labs into a public argument about enterprise control. Palantir released a white paper, “Institutional Sovereignty in the Age of AI,” laying out steps companies and governments can take to protect themselves from OpenAI, Anthropic, and other foundation-model providers. The article links that paper to Karp's CNBC appearance, where he said “something has gone completely wrong” in the relationship between AI labs and customers and argued that enterprises are paying for tokens that create little value.The killer detail is the value-capture question. Higgins writes that Karp's critique has resonated because AI labs may gain power and insight from customer data, workflows, and decision-making, even when enterprise policies say customer data are not used for training. David Sacks amplified the concern by arguing that Anthropic is moving from the model layer into vertical applications such as science, security, legal, and coding, raising the fear that model providers will watch where value is being created and then move into those markets directly.The pull is that Karp is not alone, even if his style is unusually combative. Higgins notes that Satya Nadella has also warned that companies need to retain the learnings created when they use AI models, while Mark Zuckerberg has framed Meta's new model release partly around lower-cost frontier intelligence. The article presents Karp's campaign as one sign that established technology companies and large enterprises are trying to define where they fit when AI labs become central infrastructure, application competitors, and potential IPO giants at the same time.Read more: The Wall Street JournalThe AI Agents Are Coming for Microsoft OfficeAlex Wilhelm | Cautious Optimism | July 11, 2026Alex Wilhelm argues that one of the week's quieter AI questions is whether the productivity market that Microsoft successfully moved into subscription software is now being attacked by agentic tools. The piece begins with the infrastructure backdrop: SK Hynix raised $26.5 billion in a U.S. listing while building U.S. HBM and advanced-packaging capacity, and memory, chip, and foundry companies are now priced for sustained AI demand.Wilhelm then says the AI conversation has shifted quickly from raw capability to cost per task. He cites new model releases and vendor language emphasizing cheaper agentic and coding models, faster performance, and lower dollars per task. That matters because lower costs make it more plausible for AI systems to take on routine knowledge work at scale rather than remain a premium coding assistant market.The core of the article is Microsoft Office. Wilhelm notes that Microsoft turned Office from a one-time purchase into Microsoft 365, a large recurring revenue business with tens of millions of subscribers and a major productivity segment. Now, he says, late-stage unicorns and AI labs are pushing into the same territory: Anthropic's Cowork was reportedly used mostly outside software development, OpenAI merged ChatGPT and Codex into a tool for creating sheets, slides, docs, web apps, and long-running work, and other companies are building agentic coworkers that connect business data to documents, workflows, schedules, alerts, and apps.The article's caveat is that Microsoft has survived major platform shifts before. The argument is not that Office disappears quickly, but that the definition of office software is broadening from documents and spreadsheets into AI systems that can create, monitor, and act across workplace data.Read moreWhat Is Loop Engineering, and Who Owns It?Author: Nilesh Barla Published: July 11, 2026Nilesh Barla argues that “loop engineering” is becoming a distinct discipline because production AI agents now fail less at single prompts than at runtime: when to stop, what state to preserve, and how to recover after a bad step. Prompt engineering shapes one model call, and context engineering shapes what the model sees, but loop engineering shapes what a sequence of calls actually does.The killer detail is the three-primitives frame. Barla says a real agent loop needs halt conditions, state carryover, and recovery paths, then maps teams across five maturity levels. At the lowest level, an agent is just a model call in a for-loop with a step cap and raw history; by the higher levels, the system has structured state, explicit planning, replay, evaluation, and self-repair.The pull is organizational. If agents are becoming production systems rather than demos, someone has to own the runtime itself. The loop engineer is the role Barla gives to the person responsible for making long-running agent work dependable.Read more: Adaline LabsThe Fight Against AI Data Centers Is Just BeginningEmma Roth | The Verge | July 12, 2026Emma Roth argues that community resistance to data centers has moved from an early warning sign into a national political fight as AI facilities grow larger, more power-hungry, and more visible to nearby residents. The article starts with Apple's failed 2015 plan for a $1 billion data center in Athenry, Ireland, where a small group of residents challenged the project over noise, light pollution, flooding, traffic, and wildlife effects until Apple abandoned it in 2018.The current data-center buildout is presented as much larger and more contentious. Roth writes that residents now cite rising energy costs, water quality, noise, light pollution, and greenhouse gas emissions, while the U.S. Energy Information Administration expects commercial energy demand to surpass residential demand this year because of AI data centers and Goldman Sachs expects data-center power demand to double by 2027.The central evidence comes from Data Center Watch, which says protesters blocked or delayed at least 75 U.S. projects worth $130 billion from January to March, with active opposition groups more than doubling from 396 at the end of 2025 to 833 by the end of the first quarter of 2026. Roth also cites QTS abandoning a $12 billion Wisconsin campus, Delaware City regulators blocking a 580-acre project under the Coastal Zone Act, opposition stopping a QTS project in Prince William County, and pressure that pushed Kevin O'Leary to downsize the proposed 40,000-acre Project Stratos in Utah.The policy section describes a split between federal acceleration and local resistance. President Trump has treated data centers as part of the AI race with China and fast-tracked construction, while some Republican candidates are distancing themselves from that position ahead of midterms. Sanders and Ocasio-Cortez have proposed a moratorium until price and environmental protections exist, bipartisan lawmakers are backing ratepayer-protection measures, and states including Florida, Idaho, and Washington have passed rules on cost shifting, water use, and tax breaks. Roth's caveat is that the policy patchwork is still incomplete, leaving many communities to fight project by project.Read more6 months to live for open modelsAuthor: Nathan Lambert Published: July 12, 2026Nathan Lambert argues that open-weight AI models are facing their most serious policy test so far because U.S. officials are beginning to discuss concrete controls rather than abstract safety concerns. He says reported White House conversations about a new executive order may initially target Chinese-origin models and government use, but could create a broader review habit for frontier open models. His forecast is that a model above the capability range of GPT-5.5, Claude Opus 4.8, or GLM-5.2 could trigger a ban or indefinite delay within six months.The post separates two policy fights that are becoming intertwined: distillation and frontier capability. Lambert says the distillation campaign against Chinese models has become a form of regulatory capture because Anthropic and other closed-model companies would gain economically if Chinese open models were banned. He does not dismiss IP protection, but argues that if a closed model's capabilities are dangerous enough to justify restricting open models, the lab also has to explain why those capabilities are exposed through a queryable API. He cites unauthorized access to Anthropic's Mythos private beta as evidence that APIs are not automatically secure.The broader claim is that a unilateral U.S. ban would hurt positive actors more than bad actors if comparable open models remain available elsewhere. Lambert says the only durable ceiling would require global agreement, which does not exist, and that open models can improve safety by allowing broad inspection, adaptation, and understanding. His proposed near-term off-ramps are a strong U.S. open model release from companies such as Microsoft, Meta, or Reflection, and a broader coalition of open-source beneficiaries lobbying for safe rollout rather than prohibition.Read more: SourceAmericans Deserve a Dividend From AI Companies' RichesAuthor: Scott Stanford Published: July 14, 2026Scott Stanford argues that proposals to give the government a stake in AI companies miss the point unless ordinary citizens directly receive and control the upside. Sam Altman has discussed giving up equity in OpenAI, Washington already owns a stake in Intel, Nvidia is sharing China chip revenue, and Bernie Sanders wants large AI labs to contribute half their stock to a sovereign wealth fund. Stanford says those ideas all park value with the state, not with people.The killer detail is New Carlisle, Indiana, where AWS's Project Rainier is turning cornfields into one of the world's largest AI superclusters. The project is planned to run up to a million chips, draw more than two gigawatts of power, and represents an investment that has grown from $11 billion to $13.8 billion. Stanford uses that local transformation to argue that AI's public bargain should be visible at the household level.The pull is design. A citizen AI dividend would have to specify who earns a stake, how they hold it, and when they see cash. Without that mechanism, the AI wealth debate remains a fight over government balance sheets rather than public ownership.Read more: SourceWho Gets to Define the Frontier?Author: Mark Daley Published: July 14, 2026Mark Daley argues that Demis Hassabis is right to call for a serious institution to verify frontier AI systems, but that the power to test models is also the power to govern them. Hassabis's proposed Frontier AI Standards Body would get privileged pre-release access to advanced models, testing compute, held-out evaluations, support from national labs and security agencies, third-party auditors, and eventually authority to block models from the American market or coordinate a slowdown.The killer detail is Daley's constitutional objection. He says the proposal sometimes looks like a scientific lab, a standards body, an industry regulator, a licensing authority, and an emergency security council at once. Combining those roles because each requires technical expertise would be like putting the central bank, auditor-general, and Supreme Court in one building and calling it efficient.The pull is standard-setting. Daley's concern is not that verification is unnecessary, but that whoever writes the tests, decides what passes, adjudicates disputes, and grants market access may end up defining the frontier itself.Read more: SourceGPT-Red: Unlocking Self-Improvement for RobustnessOpenAI | OpenAI | July 15, 2026OpenAI describes GPT-Red as an internal automated red-teaming model trained to find prompt-injection vulnerabilities at a scale human red teams cannot match. The post says AI systems increasingly encounter third-party data through browsers, connected apps, local files, and tools, creating opportunities for malicious instructions hidden in emails, webpages, tool responses, or code repositories. Human red-teaming remains part of OpenAI's safety process, but the company says it is time-intensive and cannot generate enough diverse adversarial examples for model training.The system is trained through self-play reinforcement learning, with GPT-Red rewarded for eliciting valid failures and defender models rewarded for resisting attacks while still completing their tasks. OpenAI says the training environments specify threat models across settings such as local files, webpage banners, email bodies, and tool outputs. The model is kept separate from deployed production models because it is intentionally trained with malicious capabilities.OpenAI reports that GPT-Red generalized beyond its training set, including an internal replication of the indirect prompt-injection arena from Dziemian et al. (2025), where it found successful attacks in 84% of scenarios compared with 13% for human red-teamers. The post also says GPT-Red transferred attacks from simulation to a live autonomous vending-machine agent, causing price changes and order cancellations, and outperformed a prompted GPT-5.5 baseline against a Codex CLI agent on held-out data-exfiltration tasks.The article's main robustness claim is that OpenAI has used GPT-Red and predecessor models in training since GPT-5.3, with later GPT releases becoming more resistant to prompt injections. It says GPT-5.6 Sol has six times fewer failures on OpenAI's hardest direct prompt-injection benchmark than the best production model from four months earlier, that a “Fake Chain-of-Thought” attack class fell from more than 95% success against GPT-5.1 to below 10% against GPT-5.6 Sol, and that GPT-5.6 Sol fails on only 0.05% of GPT-Red's direct prompt injections. OpenAI says general capabilities and targeted over-refusal evaluations were not harmed, and says a preprint with more details will follow.Read moreAnthropic, Blackstone bet the next trillion-dollar AI business is implementation, not just modelsRebecca Bellan | TechCrunch | July 15, 2026Rebecca Bellan reports that Ode with Anthropic is the $1.5 billion AI implementation company launched by Anthropic with Blackstone, Hellman & Friedman, Goldman Sachs, and other backers. The article says the venture reflects a growing belief among frontier AI labs that enterprise adoption requires more than better models: customers need engineers who can embed inside businesses and turn AI into working systems.Ode was originally conceived by Blackstone after it used both large consulting firms and smaller AI services boutiques across its portfolio companies. TechCrunch reports that Fractional AI, an AI engineering services startup, stood out and was acquired by the joint venture shortly after the venture was announced. Fractional now forms the foundation of Ode, which has 100 engineers and works closely with Anthropic's applied AI team to identify where the technology can affect specific businesses.Ode CEO Chris Taylor tells TechCrunch that the company could someday become a trillion-dollar business if it scales without losing quality. He says an ideal customer is one whose CEO treats the AI project as a top one or two priority, whether it is a major product feature or the reworking of a core business process. Ode will operate under a “Claude-first” principle, using Anthropic technology whenever possible, but the article says it can use rival AI products when needed.The article's central implementation argument comes from Ode chief technologist Eddie Siegel, who says model selection matters but is not where most of the engineering effort goes. He compares it to the choice of programming language in software: one ingredient in a system that still has to be engineered. Bellan writes that Ode's challenge is hiring and training enough elite generalist engineers, many of them former founders, while competing with OpenAI's The Deployment Company and consulting giants that have built their own forward-deployed engineering teams.Read moreVint Cerf is working on a plan to unleash AI agents on the open internetTim Fernholz | TechCrunch | July 15, 2026Tim Fernholz reports that Vint Cerf, after leaving Google, is advising Innovation Labs on an open architecture for identifying AI agents online. Innovation Labs is a subsidiary of Identity Digital, a DNS registry company, and its proposal is to use domain-name infrastructure as part of a system for agent identity, accountability, and auditability. The premise is that agents will need a way to identify themselves if they move beyond proprietary systems and begin interacting across the open internet.The concrete proposal is DNSid, a registry that links an AI agent to an existing internet domain and uses cryptographic proofs to log its registration over time. Innovation Labs says it is trialing the standard with unnamed hyperscalers and identity companies. Cerf frames the problem around authority and accountability: what authority an agent has, where that authority came from, who is accountable for the agent's behavior, how its identity is established, and why anyone should trust it.The article's caveat is that standards are still emerging and agents are more active than static domains. Cerf says the period may be both fascinating and exasperating because the functionality is powerful and interoperability is unresolved. He compares the adoption problem to TCP/IP: competing systems may not work together until users push for functional interoperation. He also says an agentic economy is not inevitable, but that people will try to build it because delegating work to agents will be easier.Read more: TechCrunchxai-org/grok-build, now open sourceAuthor: Simon Willison Published: July 15, 2026Simon Willison argues that xAI's decision to open-source Grok Build is best understood as a trust repair move after a severe privacy failure. The CLI had triggered backlash when users realized that running it in a directory could upload the entire directory to xAI's Google Cloud buckets, including one user's reported SSH keys, password manager database, documents, photos, and videos. xAI disabled the feature, said previously retained coding data would be deleted, and released the code under Apache 2.0.The killer detail is what the codebase reveals. Willison counts 844,530 lines of Rust, only about 3% of which appears vendored, and finds remnants of the upload system still present but disabled: gcs.rs contains Google Cloud upload code, while upload_session_state() now returns a hard-coded session_state_upload_unavailable error. He also notes copied or ported tool implementations from Codex and OpenCode, prompt files, and a terminal Mermaid renderer.The pull is that terminal coding agents are becoming large, intricate software systems in their own right. The privacy failure mattered because these tools operate inside the directories where developers keep their most sensitive work; the open-source release matters because trust now depends on inspecting what an agent can see, send, and do.Read more: SourceThe Pulse: What can we learn from Bun's rapid Rust rewrite with AI?Author: Gergely Orosz and Ivan Klaric Published: July 16, 2026Gergely Orosz and Ivan Klaric argue that Bun's AI-assisted rewrite from Zig to Rust is a practical sign of how software engineering changes when models can take on large, bounded migrations with clear feedback loops. The piece does not treat the rewrite as magic: Jarred Sumner first spent hours turning design judgment into a detailed porting guide, then used adversarial review, parallel agents, compiler errors, and tests to force the work toward correctness.The killer detail is the scale. Bun had 535,496 lines of Zig, 1,448 files, and 22 million monthly downloads, making a conventional rewrite a year-long freeze the team could not justify. Using Fable, Sumner split the work across 64 agents, produced about 6,500 commits, and got the migration done in 11 days at an estimated API cost of $165,000.The pull is economic, not theatrical. If a one- or two-year migration can become an 11-day project, AI coding is not just faster autocomplete; it changes which technical debts are worth paying down.Read more: SourceOrphan risks at the frontier of artificial intelligenceAuthor: Andrew Maynard Published: July 16, 2026Andrew Maynard argues that frontier AI safety frameworks are creating “orphan risks”: harms that companies can see, but do not formally own because they are hard to quantify, do not fit catastrophic-risk thresholds, or fall outside audit-friendly compliance machinery. His target is not existing frontier safety work, but the narrowing effect that happens when private companies decide which risks count as governable.The killer detail is Maynard's contrast between measurable model dangers and threats to value. He points to Meta's three-day Galactica collapse, OpenAI's 2023 board crisis, safety-team departures, and wellbeing litigation as examples of risks that damaged trust, culture, legitimacy, or users without fitting cleanly into conventional model-risk categories. The proposed fix is an orphan-risk register: a public record of risks a company considered and chose not to manage, with reasons.The pull is accountability. Frontier developers' internal scoping choices have become a de facto layer of public governance, so the question is no longer only which risks they manage, but which risks they quietly leave outside the frame.Read more: SourceThe Lab of the Future Should Feel Like a Data CenterLatent.Space with Andy Beam and Rafa Gomez-Bombarelli | Latent.Space | July 16, 2026Latent.Space interviews Lila Sciences CTO Andy Beam and chief science officer for physical sciences Rafa Gomez-Bombarelli about the company's attempt to build an AI-run science factory. The post describes Lila's thesis as treating the lab itself as an “infinite token generator”: if internet data drove the first era of AI scaling, experimentally verified scientific data may be the next scarce training source. Lila is trying to produce that data with robotics, lab instruments, orchestration software, and AI models wired into the wet lab.The central analogy is the lab as data center. Instruments are nodes on a graph, a magnetically levitating transport layer moves materials between them, and experiment scheduling looks like a compute queue. Beam says Lila is not simply an automation company, because the point is not just throughput; it is flexibility, generalization, and experiment capture. The post says Lila has built more than 10 trillion experimentally validated “scientific reasoning tokens,” not internet text or biological sequences.The interview ranges across biology, chemistry, drug discovery, materials science, and the limits of automation. It notes that Lila rebuilt one gas-sorption measurement to run roughly 2,500 times faster, claims its general models can transfer priors from small-molecule chemistry to metal-organic frameworks for carbon capture, and describes model-suggested platinum-group-free electrocatalysts that moved from looking boring or wrong to becoming strong performers. The caveats are physical: experiments have runtimes, biology cannot always be accelerated, chains of thought can be unreliable narrators, and reward hacking becomes more dangerous when a model controls a real lab.Read more: Latent.SpaceWhy AMI Labs' Alexandre LeBrun won't call his AI “AGI” or “superintelligence”Kate Park | TechCrunch | July 16, 2026Kate Park interviews AMI Labs CEO Alexandre LeBrun about why Yann LeCun's world-model startup avoids the language of “AGI” and “superintelligence.” LeBrun says the terms are not useful because they lack stable definitions: “We never used the word AGI. And I just noticed that nobody is using it anymore; they switched to superintelligence.” His argument is that the practical frontier is not a label, but whether AI systems can understand and predict real-world states.The article explains the world-model thesis by contrasting language prediction with physical-state prediction. A large language model predicts the next word; a world model predicts the next state, such as what happens when a glass tips over. LeBrun says LLMs remain complementary and efficient for language, but the physical world is where current AI is weak. Robotics is the clearest case: hardware has advanced quickly, but robots are still brittle outside controlled routines because they lack context and situational understanding.AMI is still pre-product, but TechCrunch reports that LeBrun was in Seoul looking for industrial partners, researchers, and global companies. He says world models cannot be built entirely inside a lab because they need access to real environments. That is why South Korea appeals to AMI: robotics, semiconductors, manufacturing, and fast adoption create the kind of hardware-heavy context that software-only AI has barely touched.Read more: TechCrunchKimi K3 Tech Blog: Open Frontier IntelligenceKimi | Kimi | July 16, 2026Kimi introduces Kimi K3 as an open 3T-class frontier model aimed at coding, knowledge work, reasoning, multimodality, and long-context agentic use. The source describes the model as a 2.8T-parameter system built on Kimi Delta Attention and Attention Residuals, with native multimodality and a 1M-token context window. It says Moonshot AI plans to release model weights by July 27.The post presents K3 through benchmark and use-case sections rather than as a general product announcement. It reports results across coding, productivity, agentic, and multimodal evaluations, including DeepSWE, Terminal-Bench 2.1, Program Bench, SWE Marathon, FrontierSWE, PostTrain Bench, OfficeQA Pro, SpreadsheetBench 2, MCP Atlas, AutomationBench, BrowseComp, GDPval-AA v2, AA-Briefcase, MMMU-Pro, MathVision, BabyVision, OmniDocBench, and PerceptionBench. The source says all reported K3 results use maximum reasoning effort with temperature and top-p set to 1.0, and that different benchmark comparisons use KimiCode, Claude Code, or Codex harnesses depending on the test.Kimi's caveats are unusually concrete. The limitations section says K3 was trained in preserved thinking-history mode, so quality may become unstable if an agent harness does not pass historical thinking content correctly or if an ongoing session switches to K3 midstream. It also says K3's emphasis on long-horizon tasks can make it excessively proactive when it encounters minor issues or ambiguous intent, and recommends imposing explicit behavioral constraints for applications that require strict boundaries. The post adds that K3 remains behind Claude Fable 5 and GPT 5.6 Sol in user experience despite being competitive overall.Read moreVenture CapitalThree Years InAuthor: Tomasz Tunguz Published: July 10, 2026Tomasz Tunguz marks Theory Ventures' third anniversary by arguing that AI's central market effect is time compression. In his telling, model release cycles, company revenue milestones, enterprise adoption, and venture categories have all accelerated. Seed, Series A, and Series B still exist as financing labels, but they no longer cleanly describe company maturity when some seed rounds are larger than IPOs and the best AI companies can mature much earlier than prior software companies.The killer detail is the shift from models to inference. Tunguz argues that inference has become the dominant AI market because workloads and buyer preferences are fragmenting: video, batch, local, agentic, and real-time tasks each create different infrastructure needs. He compares this to databases splitting into OLTP, OLAP, vector, and streaming categories, with AI pushing the same specialization into inference infrastructure.The pull is that Theory sees the AI-native venture firm as part of the same pattern. The firm says it has analyzed twice as many investment opportunities with three investors working alongside a nine-person intelligence organization, using agents and research systems to map markets, source companies, and support diligence. The piece is both a market map and a statement about how venture itself is being rebuilt by the technology it funds.Read more: LinkedInVenture Has Rarely Looked More BifurcatedAuthor: Beezer Clarkson Published: July 14, 2026Beezer Clarkson points to PitchBook's Q2 report as evidence that the U.S. venture market has split into two very different realities. AI now accounts for more than 60 percent of all U.S. venture deal value, meaning the headline market can look active and well-funded even while much of the non-AI market is dealing with a much colder liquidity and fundraising environment.The thread uses that split as the setup for Clarkson's latest Origins episode with Alec Litowitz, founder of Magnetar and QStar Capital and one of Citadel's original founding partners. Clarkson says markets like this are periods of genuine uncertainty, not merely ordinary risk, which is why Litowitz's Adaptability Quotient framework is relevant.The embedded clip makes the liquidity point concrete. Litowitz says DPI is “the resolution of uncertainty” because it converts an uncertain investment into actual cash returned to LPs. In his framing, a realized dollar is a real mark, while TVPI remains uncertain until it is realized.The killer detail is the distinction between pricing risk and resolving uncertainty. Litowitz's perspective matters because QStar is a SpaceX investor and Clarkson says the conversation happened just before one of venture's most consequential IPOs. The episode's stated questions are why venture remains a way to gain exposure to innovation, how AI is changing what is investable, why liquidity is ultimately a function of time, and why uncertainty requires a different decision framework from risk.Read more: XThe Best Angel Investors in the US: Who Backs the Most Unicorns, and Who's Active NowAuthor: Ilya Strebulaev Published: July 10, 2026Ilya Strebulaev ranks angels, angel groups, accelerators, and incubators by lifetime U.S. unicorn investments, counting checks written before a company reached unicorn status. The top of the combined list is dominated by organizations: Y Combinator leads with 113 unicorn investments, followed by Plug and Play at 52 and 500 Global at 41. Sand Hill Angels is the highest-ranked angel group at 31.The killer detail is how quickly the list changes below the biggest accelerators. Strebulaev says 271 of the 304 investors in the Top 200 are individuals, or 89%. In the top 100, individuals are 91%. That makes the market underneath the large accelerator counts look much more personal: mostly operators and individual angels writing early checks from their own networks.The pull is the ranking's own caveat. Strebulaev writes that every lifetime leaderboard has a blind spot because many of the unicorns behind those totals were founded a decade or more ago, and some angels have since moved into formal funds, slowed down, or stopped investing. His post therefore separates lifetime performance from recent cohorts, including companies founded in 2015 or later and 2020 or later. For founders or allocators making current decisions, that distinction matters: a career record and a current record are not the same measure.Read more: Ilya StrebulaevAre Prediction Markets Doomed to Fail?Author: Contrary Published: July 16, 2026Contrary argues that prediction markets' current boom depends on whether platforms can prove they are more than regulated gambling with exchange-style branding. Kalshi and Polymarket have reached mass cultural, investor, and regulatory attention, but the article says the underlying idea is old: academic markets, corporate forecasting tools, Intrade, PredictIt, and other predecessors all struggled with the same linked problems of liquidity, legality, and user appeal.The killer detail is the comparison with sportsbooks. Prediction markets present themselves as peer-to-peer, transparent, and non-house-based, but sports contracts reportedly account for more than 90 percent of Kalshi trading, and the article says the platforms keep a much thinner slice of volume than sportsbooks. A market can therefore show sports-betting-scale handle while generating far less revenue.The pull is that the product's hardest problem may be distribution of wins. If a small group of sharp traders captures most profits while casual users lose interest, prediction markets may become valuable data feeds and professional tools before they become durable consumer networks.Read more: SourceRegulationExclusive: The Next Frontier of the Deportation Wars: College CampusesAuthor: Adrian Carrasquillo Published: July 11, 2026Adrian Carrasquillo reports that college campuses are becoming a new front in the fight over immigration enforcement because automatic license plate readers can turn ordinary campus security infrastructure into searchable location data. His thesis is that Flock Safety's camera network, even without direct ICE or DHS contracts, can feed deportation enforcement through local police partnerships and data-sharing practices.The killer detail is the campaign target. The Emergency Campaign to Support Higher Education, working with Schools Drop ICE, is focusing on 75 colleges and universities publicly identified as having Flock contracts. Flock says it has no ICE or DHS contracts, but activists argue the risk comes through local agencies that coordinate with federal authorities and run searches on their behalf.The pull is broader than immigration. Carrasquillo notes that license plate readers have already been abused by officers for stalking, and that Flock's AI search features can identify more than plates, including bumper stickers. A campus safety tool can become a political surveillance system when the data layer is searchable.Read more: The BulwarkThe Supreme Court Broke Independent Agencies. Here's a Way to Slow the Damage.Author: Todd Phillips Published: July 12, 2026Todd Phillips argues that the Supreme Court's decision in Trump v. Slaughter damaged independent agencies by ending for-cause removal protections, but did not leave Congress powerless. The ruling weakens the old model in which commissioners at bodies such as the FTC, NLRB, CPSC, SEC, and CFTC could be insulated from dismissal over policy disagreements. Phillips says the next fight is whether presidents can turn nominally bipartisan commissions into one-party instruments.The killer detail is the procedural fix: quorum rules. Phillips proposes that Congress require bipartisan slates of commissioners to be seated before independent agencies can act. A president could still fire commissioners, as the Court now permits, but if those firings broke quorum, the agency would be unable to proceed until replacements were confirmed. The guardrail would

Researchers have been trying to make invisible drones and camouflaged robots for years, before that it was cars and planes. They have used everything from transparent materials to light-bending optical systems or cloaks, but haven't been able to make it happen. But now, engineers at Northwestern University have created a drone that appears to vanish in plain sight after it takes flight. The team created the Phantom Twist, a drone that uses a concept called "motion blur" -- the effect that makes fast-spinning fans and propellers look like a ghostly smudge. The drone spins up to 25 times per second, which is too fast for the human eye to see clearly. While it isn't completely invisible, it appears to seamlessly blend into the background. The work could lead to drones that monitor wildlife, survey the environment and inspect infrastructure with less visual disruption.To design the drone, the engineers first used a computational model to generate some 20,000 drone configurations capable of stable flight. Next, they used AI and optimization algorithms to rearrange the drones' major components, including a motor, propeller, circuit board, counterweight and batteries. It was the algorithms that determined the ideal placement of the drone's components to minimize its visibility from virtually every viewing angle while allowing for stable flight. Northwestern's Michael Rubenstein, who led the project, said, "The design process was fully automated ... Then, when we were confident that a drone met all our criteria, we built it."According to the visibility metric, the optimized drone is about 10 times less visually perceptible than a conventional quadcopter.The new drone still has several limitations. For example, the propeller still makes noise, and some wires and support rods are still somewhat visible. Next, the team is working on new iterations with more transparent materials or quieter propulsion to make the Phantom Twist truly disappear.#AI, #Drone, #Drones, #Robotics, #Engineering, #Innovation, #NorthwesternUniversity, #ArtificialIntelligence, #Technology, #STEM, #Automation, #Research, #FutureTech, #Aviation, #Manufacturing

In this episode, Scott Becker reflects on seeing robotic lawnmowers replace manual labor at a golf course and discusses the broader economic and workforce implications of automation and AI-driven job displacement.

On the show this week, Deise Yumi Asami shares how her AI-enabled robotics startup, Maximo, is transforming solar panel installation and making solar panel installation faster, safer, and more efficient. In this conversation, you'll hear how robotics, AI vision, and smart field deployment are helping reshape the future of renewable energy infrastructure. Deise also shares the experience of starting a new company from the ground up while being incubated within a larger parent company. Learn how innovation is encouraged and celebrated inside a large organization, and how AES organizes its innovation group. Learn more: https://maxrobotics.ai/ – SPONSORS – This episode is brought to you by Tiger Data Every growing Postgres database eventually hits a wall. Queries slow down, dashboards lag, and teams consider adding a second database. Tiger Data, creators of TimescaleDB, extends Postgres with time-series primitives, columnar storage, and automatic partitioning so your queries stay fast on live data. No pipelines, no migration, no second system. Just Postgres, built for the workload you actually have. Try it free at https://www.tigerdata.com/go/trial?utm_source=content-syndication&utm_medium=referral&utm_campaign=robotics-ads

In this episode, Scott Becker reflects on seeing robotic lawnmowers replace manual labor at a golf course and discusses the broader economic and workforce implications of automation and AI-driven job displacement.

As political leaders, industry pioneers and experts from around the world gather in Shanghai for the 2026 World AI Conference and High-Level Meeting on Global AI Governance, we explore key questions facing global AI governance today: What are the biggest challenges now? How can countries compete while working together on safety and governance? And how can the world ensure that AI narrows, not widens, the global development gap? Host Dou Hongyu is joined by Josef Gregory Mahoney, Professor of Politics and International Relations at East China Normal University, Li Lun, Assistant Professor of Economics at Peking University, and Dr. Liu Shaoshan, Director of Embodied Intelligence at Shenzhen Institute of Artificial Intelligence and Robotics for Society.

In this episode of The Jess Larsen Show on Innovation & Leadership, Jess sits down with Lucas Ngoo, CEO of Cortex AI and co-founder of Carousell, the Southeast Asian marketplace that grew to a $1 billion valuation. Lucas explains why the next major leap in AI may not just be chatbots or software, but general-purpose robotics. While robot hardware is getting cheaper and more powerful, Lucas argues that the biggest bottleneck is still data: the real-world human and robot data needed to train machines to operate safely, precisely, and intelligently in everyday environments. Jess and Lucas dive into how Cortex AI is building the data infrastructure for robot brains, collecting human task data from industries like factories, retail, restaurants, hotels, and other real-world environments. They also explore why robots may follow a path similar to Waymo, starting with human supervision and gradually improving until autonomy becomes invisible and widespread. Lucas also shares the lessons he learned from building Carousell from three founders out of school into a billion-dollar company, including how to hire during rapid growth, survive startup chaos, raise capital, choose the right investors, and recognize technologies that are just becoming possible. This is a forward-looking conversation about robotics, AI, entrepreneurship, Silicon Valley ambition, and what it really takes to build the future before the rest of the world sees it coming. Learn more about your ad choices. Visit megaphone.fm/adchoices

Can robotic lawnmowers save you time without hassling hedgehogs?Listener Pat has seen adverts for robotic lawnmowers and has lots of questions: Do they cut effectively? How do they manage with edges and flower borders? And what about avoiding wildlife and other objects in the garden?To find out, presenter Greg Foot is joined by BBC Gardeners' World magazine's tech reviewer Harry Duncton, alongside Dr Sophie Lund-Rasmussen (aka 'Dr Hedgehog'), who has carried out research into the dangers to hedgehogs posed by robotic lawnmowers.All of our episodes start with YOUR suggestions. If you've seen an ad, trend or wonder product promising to make you happier, healthier or greener, email us at sliced.bread@bbc.co.uk OR send a voice note to our WhatsApp number, 07543 306807.RESEARCHER: JULIA RAVEY PRODUCERS: SIMON HOBAN AND GREG FOOT

Can robotic lawnmowers save you time without hassling hedgehogs?Listener Pat has seen adverts for robotic lawnmowers and has lots of questions: Do they cut effectively? How do they manage with edges and flower borders? And what about avoiding wildlife and other objects in the garden?To find out, presenter Greg Foot is joined by BBC Gardeners' World magazine's tech reviewer Harry Duncton, alongside Dr Sophie Lund-Rasmussen (aka 'Dr Hedgehog'), who has carried out research into the dangers to hedgehogs posed by robotic lawnmowers.All of our episodes start with YOUR suggestions. If you've seen an ad, trend or wonder product promising to make you happier, healthier or greener, email us at sliced.bread@bbc.co.uk OR send a voice note to our WhatsApp number, 07543 306807.RESEARCHER: JULIA RAVEY PRODUCERS: SIMON HOBAN AND GREG FOOT

Following an unexpected surge on social media, founder Adam Stager explains how massive public interest drives global demand for agricultural automation and eco-friendly alternatives to traditional chemical sprays.See omnystudio.com/listener for privacy information.

Are your consultations building loyal, lifelong clients—or just one-time appointments?In this episode, Andrea and Taleesa talk about the importance of consultations in esthetic practices and give tips on how you can transform your consultations from rushed, scripted, “robotic” encounters into personalized, trust-building conversations that quietly grow revenue.Join both Andrea and Taleesa as they give you common consultation mistakes that can cost you thousands and the key questions that uncover your clients' true motivations and goals when it comes to their skincare. If you enjoyed this episode please share, rate and review it! Also mentioned in today's episode: Robotic, scripted consultations vs. real conversations 5:40Not asking enough (or the right) questions 11:13Treatment planning, education, and mentioning other services without feeling salesy 25:41Links:Work with us:https://smithandcrawford.com/dfyLink to book a demo call with us: https://portal.smithandcrawford.com/public/appointment-scheduler/61d5c7d8713ec2457d84ce7a/scheduleEstie Social Suite:https://smithandcrawford.com/aesthetically-social-suiteShow transcripts: https://smithandcrawford.com/notesEmail us: hello@smithandcrawford.comJoin our newsletter: https://smithandcrawford.com/newsletterhttps://calendly.com/smithandcrawford/30-min-strategy-session?back=1&month=2024-10https://calendly.com/smithandcrawford/aesthetically-discovery-call?back=1&month=2024-08https://smithandcrawford.com/

July 16, 2026 ~ Chris Renwick and Lloyd Jackson discuss the single-port robotic mitral valve repair performed by Dr. Rakesh Suri at Corewell Health, marking a major advancement in minimally invasive cardiac surgery. Hosted by Simplecast, an AdsWizz company. See https://pcm.adswizz.com for information about our collection and use of personal data for advertising.

Jeff McCausland notes that Europe now views Ukraine as a security contributor due to its mastery of drone and robotic warfare. While Putin considers escalation or mobilization, Chinese pressure likely prevents the use of tactical nuclear weapons. The US continues to communicate "serious consequences." (14)15TH CENTURY

As part of our summer replay series, we're revisiting one of our favorite conversations on the future of AI infrastructure. SemiAnalysis founder Dylan Patel joins Erin Price-Wright, Guido Appenzeller, and Erik Torenberg to examine the rapidly evolving economics of AI hardware, from GPUs and custom silicon to data centers, power, and the global race for compute. The conversation explores NVIDIA's competitive advantages, the rise of custom chips from Google, Amazon, and Meta, the economics of frontier AI models, and the infrastructure constraints shaping the industry's next phase. They also discuss AI startups, export controls, robotics, enterprise software, and why simply copying NVIDIA isn't enough to build a winning AI hardware company. Whether you're building AI products, investing in infrastructure, or trying to understand where the industry is headed, this conversation offers a practical look at the forces shaping the future of compute.   Resources: Follow Dylan Patel on X: https://x.com/dylan522p Follow Erin Price-Wright on X: https://x.com/espricewright Follow Guido Appenzeller on X: https://x.com/appenz Learn more about SemiAnalysis: https://semianalysis.com/dylan-patel/ Stay Updated:Find a16z on YouTube: YouTubeFind a16z on XFind a16z on LinkedInListen to the a16z Show on SpotifyListen to the a16z Show on Apple PodcastsFollow our host: https://twitter.com/eriktorenberg Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see a16z.com/disclosures. Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.

Supply chain transformation is no longer just about efficiency. It is about resilience, intelligence, and trust across every layer of the system. In this episode, Reid Jackson and Liz Sertl were at GS1 Connect. Melanie Hilton sat down with Miguel Rodriguez Garcia from MIT, Frederik Bohn from Plug and Play Tech Center, and Lisa Morales-Hellebo from REFASHIOND Ventures to explore what supply chains could look like in 2050 and the challenges that must be solved today to get there. The conversation spans cybersecurity risks that now extend into physical operations, the rise of robotics in logistics, and the growing need for verified, trustworthy data across global networks. The guests also explore how interoperability gaps inside large enterprises slow progress, why collaboration remains one of the biggest barriers across industries, and how emerging approaches like AI agents and materials verification could reshape decision making. Together, they highlight that the future supply chain will depend on aligning technology, standards, and human behavior in ways that are still being defined. This is more than a discussion about emerging technology. It is an exploration of how supply chains must evolve into connected, transparent, and adaptive systems where data integrity, automation, and cross-industry collaboration determine long term success. In this episode, you'll learn: How cybersecurity is evolving from a digital risk into a physical supply chain challenge How robotics and automation are already reshaping warehouse and logistics operations Why data trust and verification are becoming essential for future supply chain systems Things to listen for: (00:00) Opening remarks from Reid Jackson and Liz Sertl introducing the GS1 Connect session (01:26) Framing the 2050 supply chain discussion and key themes (07:40) Cybersecurity threats expanding into physical supply chain operations (11:00) Robotics moving from experimental to real warehouse deployment (15:00) Trust and verification of physical materials through data and scanning (19:30) Collaboration challenges across industries and supplier ecosystems (23:10) Interoperability gaps inside large enterprise systems (26:00) AI agents, hallucinations, and the importance of data quality (30:46) Real world visibility innovations using highway camera data networks (32:38) The future vision of standardized machine communication and interoperability Connect with GS1 US: Our website - www.gs1us.orgGS1 US on LinkedIn Connect with the speakers: Melanie Hilton on LinkedInMiguel Rodriguez Garcia on LinkedInMassachusetts Institute of Technology at https://web.mit.edu/Lisa Morales-Hellebo on LinkedIn REFASHIOND Ventures at https://www.refashiond.com/Frederik Bohn on LinkedInPlug and Play Tech Center. at https://www.plugandplaytechcenter.com/  

As part of our summer replay series, we're revisiting one of the standout conversations from Runtime, a16z's conference on AI infrastructure and the future of computing. Gavin Baker, Managing Partner and CIO of Atreides Management, joins David George to examine the biggest questions surrounding today's AI investment cycle. Is AI a bubble? What does the unprecedented buildout of data centers, GPUs, and compute infrastructure mean for the economy? And how should investors think about the companies building the next generation of AI? The conversation explores frontier models, Nvidia, Google, custom silicon, AI infrastructure, application software, robotics, and why Baker believes today's AI investment cycle looks fundamentally different from the internet bubble of the early 2000s. Along the way, they discuss the economics of GPUs, enterprise software, AI business models, and what comes next as AI moves from experimentation into the broader economy.   Resources: Follow Gavin Baker on X: https://x.com/GavinSBaker Follow Atreides Management on X: https://x.com/atreidesmgmt Follow David George on X: https://x.com/DavidGeorge83 Stay Updated:Find a16z on YouTube: YouTubeFind a16z on XFind a16z on LinkedInListen to the a16z Show on SpotifyListen to the a16z Show on Apple PodcastsFollow our host: https://twitter.com/eriktorenberg Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see a16z.com/disclosures. Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.

Hard tissue robotics may be ready for a major leap forward, but only if the industry stops treating surgical robots as single-purpose robotic arms. In this episode, Yossi Bar, CEO and Founder of LEM Surgical, explores the next evolution of surgical robotics and why hard tissue innovation is just getting started. Drawing on nearly two decades in medical devices, he explains why hard tissue robotics has trailed soft tissue robotics and why system architecture matters more than hardware alone. Yossi breaks down the difference between a robotic arm and a true surgical humanoid, highlighting the roles of vision, proprioception, and multi-tool operation. He also shares how AI and humanoid-inspired, general-purpose robotic platforms could expand clinical capabilities and shape the future of surgery. Tune in to hear why the best part of surgical robotics may still be ahead! Resources: Connect with and follow Yossi Bar on LinkedIn. Follow LEM Surgical AG on LinkedIn and explore their website.

Listen to Andrew Duckworth, Samuel MacDessi and Fares Haddad discuss the paper 'Robotic-assisted surgery and functional alignment in total knee arthroplasty: the RASKAL registry-nested 2 × 2 factorial randomized trial' published in the May 2026 issue of The Bone & Joint Journal.Click here to read the paper.Find out as soon as the next episode is live by following us on X, Instagram, LinkedIn, Tik Tok or Facebook!

Robotics and automation are accelerating across healthcare, but scaling innovation still depends on having the right people in place. In this episode, Joe Solari, Managing Director at ZRG Partners, and Lori Prehar, Managing Director of Embedded Recruiting at ZRG, discuss how robotics, automation, and AI are transforming medtech, healthcare operations, and talent strategy. Joe explains how robotics can improve efficiency, reduce clinician burnout, and help healthcare organizations scale as automation expands. Lori shares why embedded recruiting has become an effective solution for companies navigating uncertain hiring needs and growth plans. They also explore the opportunities and risks of AI, including its impact on recruiting, compliance, security, and robotics. Tune in to hear how robotics, automation, AI, and flexible recruiting models are shaping the future of healthcare growth! Resources: Connect with and follow Joe Solari on LinkedIn Connect with and follow Lori Prehar on LinkedIn. Follow ZRG on LinkedIn and explore their website!

We've been hearing a lot about NASA's new moonbase plans, and they are ambitious. Starting with Artemis V (so far), the US plans to begin preparations for a permanent base situated in the south polar regions of the moon. Details are somewhat fluid, but one thing stands out: the poles are likely to be the most challenging places to set up shop in an already hostile environment. Planetary scientist Dr. Pascal Lee joins us to discuss moonbase siting options and why the poles may not be the best choice at first. Headlines: China Demonstrates Impressive Rocket Landing and Catch Mechanism New Horizons Probe Awakens to Transmit Deep Space Data Wally Funk, Last of the "Mercury 13," Passes Away West Coast Rocket Launch Alert: SpaceX Falcon 9 Visual Spectacle Main Topic: Moon Base Strategy and Location Debate History and Evolution of U.S. Moon Base Plans NASA's New Multi-Phase Moon Base Timeline and Missions Strategic and Logistical Challenges of South Pole Moon Bases Arguments for a Centralized, Benign Location for Moon Infrastructure Commercial and International Partnership Models for Lunar Operations Role of Robotics and AI in Moon Exploration and Prospecting Lessons from Antarctic Base Logistics Applied to the Moon Potential for Tourism and Commercial Development on the Lunar Surface Hosts: Rod Pyle and Tariq Malik Guest: Pascal Lee Download or subscribe to This Week in Space at https://twit.tv/shows/this-week-in-space. Join Club TWiT for Ad-Free Podcasts! Support what you love and get ad-free audio and video feeds, a members-only Discord, and exclusive content. Join today: https://twit.tv/clubtwit Sponsors: blackhat.com/us-26 and use code TWIT rippling.ai/twis threatlocker.com/twit

Andy Lonsberry is redefining what welding robots can do, and in this conversation, he reveals how AI is pushing manufacturing beyond the limits of traditional automation. In this episode, Mike Oitzman and Gene Demaitre sit down with Andy Lonsberry, CEO of Path Robotics, to explore how his team is using AI, reinforcement learning, and real-time visual feedback to make welding robots smarter, more adaptive, and capable of handling complex real-world conditions. From shipbuilding to infrastructure and other large-scale industrial applications, Andy explains why the future of welding depends on robots that can think, react, and move to the work—not just wait for the work to come to them. They discuss: - The founding story of Path Robotics - Why traditional welding automation plateaued - How AI and reinforcement learning improve weld quality - The role of real-world sensor data in adaptive welding - Why mobile, legged robots open new industrial possibilities - The business model behind Robotics as a Service - What's next for AI-powered manufacturing If you're interested in the future of industrial automation, physical AI, and the next generation of robotics, this conversation is a must-listen. Submit your robotic startup for the Startup Showcase: https://www.therobotreport.com/calling-all-robotics-startups-apply-to-robobusiness-startup-alley/ – SPONSORS – This episode is brought to you by Tiger Data Every growing Postgres database eventually hits a wall. Queries slow down, dashboards lag, and teams consider adding a second database. Tiger Data, creators of TimescaleDB, extends Postgres with time-series primitives, columnar storage, and automatic partitioning so your queries stay fast on live data. No pipelines, no migration, no second system. Just Postgres, built for the workload you actually have. Try it free at https://www.tigerdata.com/go/trial?utm_source=content-syndication&utm_medium=referral&utm_campaign=robotics-ads

We've been hearing a lot about NASA's new moonbase plans, and they are ambitious. Starting with Artemis V (so far), the US plans to begin preparations for a permanent base situated in the south polar regions of the moon. Details are somewhat fluid, but one thing stands out: the poles are likely to be the most challenging places to set up shop in an already hostile environment. Planetary scientist Dr. Pascal Lee joins us to discuss moonbase siting options and why the poles may not be the best choice at first. Headlines: China Demonstrates Impressive Rocket Landing and Catch Mechanism New Horizons Probe Awakens to Transmit Deep Space Data Wally Funk, Last of the "Mercury 13," Passes Away West Coast Rocket Launch Alert: SpaceX Falcon 9 Visual Spectacle Main Topic: Moon Base Strategy and Location Debate History and Evolution of U.S. Moon Base Plans NASA's New Multi-Phase Moon Base Timeline and Missions Strategic and Logistical Challenges of South Pole Moon Bases Arguments for a Centralized, Benign Location for Moon Infrastructure Commercial and International Partnership Models for Lunar Operations Role of Robotics and AI in Moon Exploration and Prospecting Lessons from Antarctic Base Logistics Applied to the Moon Potential for Tourism and Commercial Development on the Lunar Surface Hosts: Rod Pyle and Tariq Malik Guest: Pascal Lee Download or subscribe to This Week in Space at https://twit.tv/shows/this-week-in-space. Join Club TWiT for Ad-Free Podcasts! Support what you love and get ad-free audio and video feeds, a members-only Discord, and exclusive content. Join today: https://twit.tv/clubtwit Sponsors: blackhat.com/us-26 and use code TWIT rippling.ai/twis threatlocker.com/twit

We've been hearing a lot about NASA's new moonbase plans, and they are ambitious. Starting with Artemis V (so far), the US plans to begin preparations for a permanent base situated in the south polar regions of the moon. Details are somewhat fluid, but one thing stands out: the poles are likely to be the most challenging places to set up shop in an already hostile environment. Planetary scientist Dr. Pascal Lee joins us to discuss moonbase siting options and why the poles may not be the best choice at first. Headlines: China Demonstrates Impressive Rocket Landing and Catch Mechanism New Horizons Probe Awakens to Transmit Deep Space Data Wally Funk, Last of the "Mercury 13," Passes Away West Coast Rocket Launch Alert: SpaceX Falcon 9 Visual Spectacle Main Topic: Moon Base Strategy and Location Debate History and Evolution of U.S. Moon Base Plans NASA's New Multi-Phase Moon Base Timeline and Missions Strategic and Logistical Challenges of South Pole Moon Bases Arguments for a Centralized, Benign Location for Moon Infrastructure Commercial and International Partnership Models for Lunar Operations Role of Robotics and AI in Moon Exploration and Prospecting Lessons from Antarctic Base Logistics Applied to the Moon Potential for Tourism and Commercial Development on the Lunar Surface Hosts: Rod Pyle and Tariq Malik Guest: Pascal Lee Download or subscribe to This Week in Space at https://twit.tv/shows/this-week-in-space. Join Club TWiT for Ad-Free Podcasts! Support what you love and get ad-free audio and video feeds, a members-only Discord, and exclusive content. Join today: https://twit.tv/clubtwit Sponsors: blackhat.com/us-26 and use code TWIT rippling.ai/twis threatlocker.com/twit

We've been hearing a lot about NASA's new moonbase plans, and they are ambitious. Starting with Artemis V (so far), the US plans to begin preparations for a permanent base situated in the south polar regions of the moon. Details are somewhat fluid, but one thing stands out: the poles are likely to be the most challenging places to set up shop in an already hostile environment. Planetary scientist Dr. Pascal Lee joins us to discuss moonbase siting options and why the poles may not be the best choice at first. Headlines: China Demonstrates Impressive Rocket Landing and Catch Mechanism New Horizons Probe Awakens to Transmit Deep Space Data Wally Funk, Last of the "Mercury 13," Passes Away West Coast Rocket Launch Alert: SpaceX Falcon 9 Visual Spectacle Main Topic: Moon Base Strategy and Location Debate History and Evolution of U.S. Moon Base Plans NASA's New Multi-Phase Moon Base Timeline and Missions Strategic and Logistical Challenges of South Pole Moon Bases Arguments for a Centralized, Benign Location for Moon Infrastructure Commercial and International Partnership Models for Lunar Operations Role of Robotics and AI in Moon Exploration and Prospecting Lessons from Antarctic Base Logistics Applied to the Moon Potential for Tourism and Commercial Development on the Lunar Surface Hosts: Rod Pyle and Tariq Malik Guest: Pascal Lee Download or subscribe to This Week in Space at https://twit.tv/shows/this-week-in-space. Join Club TWiT for Ad-Free Podcasts! Support what you love and get ad-free audio and video feeds, a members-only Discord, and exclusive content. Join today: https://twit.tv/clubtwit Sponsors: blackhat.com/us-26 and use code TWIT rippling.ai/twis threatlocker.com/twit

Robotic pool cleaners get all the hype, but hype does not keep a pool clean on Tuesday morning when the wind kicks up and dirt drops to the floor. We walk through why the suction side pool cleaner is still the number one choice for a lot of pools and a lot of pool service pros, especially when you care about total cost, repairability, and how often the cleaner truly runs without anyone “remembering” to use it. I break down the real world reasons suction wins: a lower price point than most pressure cleaners and almost every quality robotic pool cleaner, no booster pump energy penalty, and fewer complicated parts when something wears out. Then we get specific on the models that matter and why: Polaris Atlas and Polaris Maxx for avoiding hang ups on modern anti vortex main drains and for strong wall climbing without that loud waterline air-sucking behavior, plus the Hayward Aquanaut for long wear life and simple refresh kits, and the Hayward TracVac as another solid option. For vinyl liner and fiberglass pools, we also talk about diaphragm style cleaners like the Zodiac G3 and Ranger and why they can be a better fit. We also keep it honest about drawbacks, like hoses in the water and how leaf piles, twigs, or rocks can stop a suction cleaner. Finally, I share my favorite “best of both worlds” strategy: keep the suction cleaner in all week for consistent daily pool maintenance, then use a robotic pool cleaner when you want that party day polish and a hose-free look. If you found this helpful, subscribe for more pool maintenance and pool equipment guidance, share it with a friend who is shopping for an automatic pool cleaner, and leave a review with the cleaner you use and why.We make the case that suction side pool cleaners are still the most practical automatic cleaners, even with today's cordless robotic pool cleaners everywhere. We share the specific models we trust, the trade-offs we see on real pools, and why “always running” beats “sometimes used” for most owners and pool service routes.• Pentair discontinuations and why competition reshapes cleaner choices• a short list of suction side cleaners worth knowing instead of ten options• Polaris Atlas and Maxx performance on anti vortex main drains and wall climbing• Hayward Aquanaut as a long life low maintenance workhorse• Hayward TracVac as a strong third option with tracked design• Zodiac diaphragm cleaners for vinyl and fiberglass pools• suction cleaner advantages on price point energy use and simple repairs• real limitations with leaf piles twigs and rocks• hose aesthetics complaints plus the reality that most people get used to them• spot vacuuming an attached spa using the existing hose setup• why robotic cleaners fail on routes when customers do not put them in• the best combo plan suction during the week plus robot for party day polishAre you a pool service pro looking to take your business to the next level? Join the pool guy coaching program. Learn more at swimmingpoollearning.com.If you're interested in the coaching program, you can learn more at poleguycoaching.com.Send us Fan MailSupport the Pool Guy Podcast Show Sponsors! HASA https://bit.ly/HASAThe Bottom Feeder. Save $100 with Code: DVB100https://store.thebottomfeeder.com/Try Skimmer FREE for 30 days:https://getskimmer.com/poolguy Get UPA Liability Insurance $64 a month! https://forms.gle/F9YoTWNQ8WnvT4QBAPool Guy Coaching: https://bit.ly/40wFE6y

Bob and Randy took a break from AI conversations. This week the guest is Josh McDonald, Regional Sales and Business Development Manager of RC Mowers. He talks all about remote-operated and autonomous robotic mowers. These mowers are used for safety and efficiency. Have you ever seen a remote-operated mower trim along the highway? Check out the size of this brush mower!

We've been running a bit of an Agent Cloud series surveying all the top inference/compute/cloud providers, from Databricks to Daytona to Railway and, even further back, E2B, but we're excited to conclude this series returning to Modal, which has just raised a monster $355M Series C.The cloud was built for developers. But agents are now changing that.The old infra stack was designed for a human who could read docs, reason through YAML, and understand dashboards to figure out what they need when something broke. While this was painful for developers, it worked since they could fill in missing context in their heads.However, agents don't have that luxury. Now in this new era of agents, everything has to be tighter.They need a place to write code, run it, inspect the output, change the environment, debug failures, and try again. Fast iteration and feedback loops with all the necessary context are crucial for agents to operate properly. Furthermore, sandboxes are a clear representation of this shift as agents can easily spin up isolated environments. This programmatic infra even extends to research:Two years ago, we were one of the first to cover Modal with CEO Erik Bernhardsson and Alessio designed our favorite LS thumbnail of all time:At the time, Modal was just a teeny little company with a $17M Series A.Today, fresh off their $355M Series C, Modal is one of the clearest examples of the agent cloud future being built in real time: a cloud platform moving past traditional web app assumptions toward the workloads AI actually creates such as elastic inference, sandboxes, GPU burst, post-training, background agents, and infrastructure that agents themselves can operate.In this episode, Modal CTO Akshat Bubna joins swyx and Vibhu to unpack why AI applications don't fit traditional cloud assumptions, why Kubernetes was never designed for bursty compute-heavy workloads, and why Modal is now shifting from developer experience to agent experience.We go deep on Modal's AI infra stack: serverless functions, decorator-based infrastructure, elastic inference for custom models, GPU snapshotting, DeFlash, speculative decoding, Auto Endpoints, sandboxes, persistent storage, networked containers, private IPv6, RDMA, multi-node training, and Modal's capacity pool across 17 cloud providers. Akshat also explains why RL rollouts can require 100,000 sandboxes, why production agents need hard guardrails, why observability may matter more than reading code, and why AI has made infrastructure exciting again.We discuss:* Why Kubernetes wasn't built for bursty AI workloads* How Modal started as a better runtime before becoming an AI cloud* Why Modal added GPUs before ChatGPT* The shift from developer experience to agent experience* Why observability matters when agents are writing the code* Elastic inference for custom models across audio, video, robotics, and comp bio* GPU snapshotting, cold starts, and why inference workloads are so bursty* Why RL rollouts can require 100,000 sandboxes* DeFlash, speculative decoding, and frontier-level inference performance* Auto Endpoints and making optimized inference easier to deploy* What Modal adds beyond vLLM, SGLang, and raw GPU rental* Modal's 17-cloud capacity pool and supercloud strategy* Networked sandboxes, sidecars, private IPv6, and RDMA* Serverless multi-node training for post-training and research workloads* Auto-research, model-guided sweeps, and agents launching GPU experiments* Compute strategy, capacity planning, and batch tiers* Why production agents need specialized sandboxes and hard guardrails* Modal's take on managed agents, CI, Gitpod/Ona, Python, TypeScript, and Modal BenchAkshat Bubna* LinkedIn: https://www.linkedin.com/in/akshat-bubna-188885103* X: https://x.com/akshat_bModal* Website: https://modal.comTimestamps00:00:00 Introduction00:00:39 Modal's origin and why Kubernetes wasn't enough00:04:32 Developer Experience → Agent Experience00:06:21 Modal's AI cloud primitives00:09:14 Sandboxes, agent loops, and proto-Cognition00:12:12 Elastic inference, GPU snapshotting, and 100,000 sandboxes00:15:24 DeFlash, speculative decoding, and Auto Endpoints00:19:59 Production-grade inference beyond raw GPUs00:22:00 Background agents, Ramp Inspect, and the agent lifecycle00:24:08 Modal's 17-cloud supercloud strategy00:26:40 Networked sandboxes, private IPv6, and RDMA00:32:48 Multi-node training, post-training, and auto research00:37:36 Compute strategy, capacity planning, and batch tiers00:40:55 Open models, real-time AI, and production agent infra00:43:06 Hard guardrails, managed agents, and specialized sandboxes00:46:06 Why AI made infrastructure exciting again00:48:30 Model APIs, differentiated products, and agentic video00:51:50 CI, coding-agent infra, SDKs, and Modal Bench00:57:28 Closing ThoughtsTranscriptIntroduction: Modal, Series C, and the Art PartySwyx [00:00:00]: We're here with Akshat, CTO of Modal, together with Vibhu. Congrats on your Series C.Akshat [00:00:10]: Thank you.Swyx [00:00:11]: Your party yesterday was amazing.Akshat [00:00:15]: Yeah.Swyx [00:00:15]: From all the photos and all the swag.Akshat [00:00:17]: We had a bunch of art installations, which was fun, seeing, like, our products on pedestals next to, like, Rodin.Swyx [00:00:25]: Very nice. Very nice. When you started, it was not the GPU inference company. Maybe it was in your mind. Take us back to the origin story.Modal's Origin: A New Runtime Beyond KubernetesAkshat [00:00:39]: I first met Eric, who's the CEO, through an investor. Back then Eric was already thinking about building, a new runtime, and he got there thinking through why are workflow orchestration products so hard to use. It's because you have to run them on Kubernetes. Kubernetes is hard to manage. It's not built for burstiness and, custom images,Swyx [00:01:03]: YeahAkshat [00:01:03]: It has a terrible developer experience.Swyx [00:01:05]: And I'll, I'll interjectAkshat [00:01:06]: YeahSwyx [00:01:07]: For listeners, who are new, we interviewed Eric two years ago, and there's a bit more of the story there from Spotify and all those things.Swyx [00:01:14]: And I came across Eric through Data Council because he did that talk on the serverless container stack that you guys did, which was like, that was my first like, “Okay, I need to take Modal very seriously” moment.Akshat [00:01:26]: Yeah.Swyx [00:01:26]: But it was still very unclear, like, do I need all this for just my data pipelines?Akshat [00:01:33]: Yeah. initially what we were thinking about was if we build a better runtime, it's a very useful primitive in itself. It's There's a lot of things that, get solved by serverless functions, like you can do, ETL stuff, you can do job queues, you can do all this, like, bursty processing, which it turns out every company had needs for. but then we also were thinking about this as like, this is a primitive that we can build a whole collection of products on, which are very verticalized. So perhaps data engineering would've been the first one, but we were thinking about inference. Back then it was more classical inference, like computer vision stuff and running XGBoosts and whatnot. But we added GPUs to the product a year before ChatGPT came out.From Serverless Containers to GPU WorkloadsSwyx [00:02:19]: Nice.Akshat [00:02:19]: We just didn't think it would be that big of a deal.Swyx [00:02:22]: Yeah, just like add A100.Vibhu [00:02:23]: Was there any, like, early key problem that really sparked off why you built it?Akshat [00:02:28]: Yeah. Primarily it's just, none of the tooling that was out there was built for, one, a really great developer experience, and also there's a general trend of, a lot of the workloads that we were seeing were very. I wish there was a better word for it, but compute-heavy. Like, they need, one, like, need a lot more resources, so you need to burst up and down a lot, versus like Kubernetes designed for, like, slow scaling and, more for, like, web server use cases. And also there's just a lot more specialization in, like, what kinds of environments these workloads run in. Like, we had sometimes they need accelerators, sometimes they need different kinds of images, and this is just like a consistent thing that we saw across a lot of companies. That would be the next step.Software-Defined Infrastructure and Decorator-Based DXSwyx [00:03:13]: Yeah. Yeah. Be nice. I don't know how much this factored into the early story, but I wrote a post when I was at Temporal about infrastructure, software-defined infrastructure or something like that.Akshat [00:03:22]: Yeah, the self-provisioningSwyx [00:03:23]: Self-provisioning.Akshat [00:03:24]: Yeah.Swyx [00:03:24]: Yeah. I can't even remember my own post.Swyx [00:03:26]: And then you put me on the landing page.Akshat [00:03:28]: Yeah. We really like, the term and so we stole it.Swyx [00:03:32]: Because you had the insight that everything can just be in decorators co-located with the code, right?Akshat [00:03:37]: Yeah.Swyx [00:03:37]: Was that a big part of the originalAkshat [00:03:39]: YesSwyx [00:03:39]: Story or it was just like a DX layer?Akshat [00:03:41]: That was, really important because we really didn't want people to spend, so much time, writing YAML, and it seemed like you could really condense the surface area of what you're doing, put it in code so you can operate on it just like you operate on other code, and like build stuff that's more expressive and dynamic. and so yeah, that was always a very important part.Swyx [00:04:04]: Then the pushback is this is a DSL.Akshat [00:04:07]: Yeah.Swyx [00:04:07]: It's you're closed source. I am locked into Modal.Akshat [00:04:11]: Yeah. We never really got pushback for that because the nice thing about Modal is you can bring whatever code you have, and sure, the DSL is at the configuration layer for, what hardware you're using, how you're scaling things up, but you still own the code.Akshat [00:04:27]: And that's, that's been an important, part of our story, even as we do inference now.Swyx [00:04:32]: Yeah.Vibhu [00:04:32]: How much of do you think still stays the same today? Like if you were to build something today, DevX very important, but I feel like, a lot of this has been changed with just hook it up to an agent, have Claude Code, have Codex implement a tool. there's very agent native primitives that are different than if I'm doing this myself, right?Developer Experience → Agent ExperienceAkshat [00:04:54]: We've changed our SDK team to think about agent experience instead of, developer experience and we think that the same benefits that apply for DX also apply for AX, which is why would you have an agent read through hundreds of Kubernetes files and like write YAML that's not even typed when it can make a couple of changes in a decorator and it gets this self-provisioning runtime of, being able to see its changes live in action? yeah, it just seems from the customers we talk to, they find Modal is much faster for agents to use versus operating on a different substrate.Swyx [00:05:34]: Yeah, because like you, again, you co-locate the infrastructure requirements to the code that runs it.Akshat [00:05:38]: Yeah.Swyx [00:05:38]: Well, the negative thesis now is that nobody's looking at their code anymore, so there's no point.Akshat [00:05:44]: Yeah, people aren't looking at code. one thing we still see is really important is observability.Swyx [00:05:51]: Yeah.Akshat [00:05:51]: Like how good is your dashboard? And of course, like we have, we push a lot of it to the CLI so the agents can do their own investigation, but you still need humans to go interpret what's going on and, make judgment calls and whatnot. and that's I feel like, Maybe more important now than looking at the code itself.Swyx [00:06:11]: Yes, because like, you can try to treat the code as a black box and then use, see the observable action that comes out of it, and then just prompt a change.What Modal Is For: AI Cloud PrimitivesAkshat [00:06:21]: Yeah.Swyx [00:06:22]: So I think it takes a bit of restraint to not specialize, to say, “I want to ship a new primitive,” and then just be general purpose.Swyx [00:06:31]: People ask you, “What are you for?” You're like, “ I don't know. We can do this, we can do that.”Vibhu [00:06:36]: Well, I'd be curious to see, like, okay, if we were to ask you, like, what is Modal for even at a high level? There's a lot you guys do, sandboxes, GPUs, everything. How do you answer?Akshat [00:06:46]: Modal is a cloud platform that's built for, where we've built the primitives from scratch for AI applications. and right now it covers, inference, training, batch processing, and sandbox workloads.Akshat [00:07:00]: But we're building a lot moreSwyx [00:07:02]: I noticed you didn't say web server, so there is still a role for, like, the always-on large-scale Kubernetes type things.Akshat [00:07:09]: Yeah, absolutely. We're, we're not trying to compete with the renders of the world, because yeah, we think the differentiator for us is the, are the workloads that need specialized compute, need to scale up and down a lot. yeah, they're, they're, they're just shaped differently.Working Alongside Frontier StartupsVibhu [00:07:26]: I think you're building a lot of it alongside the startups, right? They're innovating quite a bit, even in your, like, latest blog post. Like, even in the series C, the customers that you mention here, the cognitions, technical ones, ramps and whatnot, they're, they're innovating with you, right? And that's not something AWS is doing directly with.Akshat [00:07:45]: Yeah, absolutely. I think, this is again classic. We're a small team. We can move really fast. our engineers are working with our customers and figuring it out. Yeah.Swyx [00:07:54]: So my first week at Cognition, I walked in, there was someone wearing a Modal shirt. I was like, “What are you doing here?” They're like, “Yeah, I just. I am embedded inside of Cog.”Akshat [00:08:05]: Yeah, I think that was Peyton. We sent him overSwyx [00:08:07]: Yeah.Akshat [00:08:07]: Because, the latency of communication was too high otherwise.Swyx [00:08:12]: Yeah, distributed node, you have to - you have to place one and collocate.Vibhu [00:08:16]: Yeah.Swyx [00:08:16]: So I had a, I had direct personal experience, right? So I worked on smol developer three years ago. it was inspired by Claude 1. I think you onboarded me at some point, like, just before, and I was like, “Oh, like, I need some bursty compute. Like, I was just gonna try using Modal.” And it was a, it was a pretty pleasant experience. apparently, I showed up in the board meeting, like the analytics.smol developer, Sandboxes, and Proto-CognitionAkshat [00:08:39]: Yeah, you blew up on Hacker News and,Swyx [00:08:41]: YeahAkshat [00:08:41]: We got a big traffic spike. I. I think the way you used smol developer was Modal functions for running stuff, which was. Like, the, that was a good use case. but then, yeah.Swyx [00:08:53]: Yeah. That - So to me, that was proto-cognition.Akshat [00:08:55]: Right.Swyx [00:08:56]: If only I had, like, stuck to it.Swyx [00:08:58]: Like, that was like, if - did you say draw the tech treeAkshat [00:09:00]: AbsolutelySwyx [00:09:00]: You're just like, “Yeah, like, probably this will happen.”Akshat [00:09:02]: Yeah. Like, he was so close. You were just rebuilding upon usSwyx [00:09:04]: I just didn't realize.Akshat [00:09:05]: But the funny story there is at the same time, we were talking to a bunch of customers who needed something like sandboxing.Swyx [00:09:14]: Yeah.Akshat [00:09:14]: This is like twenty-three.Swyx [00:09:15]: Yeah.Akshat [00:09:16]: So we builtSwyx [00:09:17]: You introduced a new API right after that.Akshat [00:09:18]: Yeah.Swyx [00:09:19]: Yes.Akshat [00:09:19]: Like, we built sandboxes in May of twenty-three before anyone was even knew this was gonna be a thing. And the first example we published was, we took smol developerSwyx [00:09:28]: Smol developerAkshat [00:09:28]: And put it in a loop, so the agent can iterate on itself.Swyx [00:09:33]: Loops are hot these days.Vibhu [00:09:34]: It's the looper.Akshat [00:09:34]: Yeah.Vibhu [00:09:35]: Loops in. When was this, twenty-three?Akshat [00:09:38]: Yeah.Vibhu [00:09:39]: A small check.Akshat [00:09:39]: Yeah.Swyx [00:09:39]: It's like twenty-three. so the. the, those for listeners, like, the problem was the models are not built for any of this, right?Swyx [00:09:46]: Like, you're just trying to like. They're not post-training to understand, like, looping and, like, self-correction and tool calling was there, but, like, also not that great.Akshat [00:09:55]: Yeah.Akshat [00:09:55]: I don't remember if you used tool calling in this one, but yeah, the models would just diverge after like ten iterations and not produce anything meaningful.Swyx [00:10:03]: Yeah. But like, then. So okay, like now talking to myself three years ago, the answerVibhu [00:10:08]: Of course they will get betterSwyx [00:10:09]: Collect all the failures, build benchmark, and then collect all the, examples, build the RL environmentAkshat [00:10:15]: RightSwyx [00:10:15]: Sell it for like ten billion dollars to Meta.Swyx [00:10:17]: And then also train a model and then sell that for sixty billion dollars to Elon. And this isAkshat [00:10:23]: Yeah, of courseSwyx [00:10:23]: The funny machine. Like, it's like, it's about the hardware.Akshat [00:10:28]: It's hard to have that inherent conviction that the stuff will get that much better.Swyx [00:10:33]: In retrospect, it's so f*****g obvious.Akshat [00:10:36]: Fair enough.Swyx [00:10:37]: Like, what else were we doing back then? I don't know. anyway. Yeah. So this. That was the start of your sandboxing journey, right? I feel like it didn't blow up until, like, last year.Akshat [00:10:49]: Yeah.Swyx [00:10:50]: So there was like a couple years of quietness.Akshat [00:10:52]: Exactly, yeah. We wereVibhu [00:10:53]: I think very underrated product value. Like, my experience with Modal, Charles, before he had joined Modal, met this guy at a hackathon, and he really insisted we wanted to run some small model, not hosted anywhere, and he's like, “ there's this cool company, Modal. They'll like spin up a GPU sandbox, we can throw it on there. They'll take a Hugging Face link.” And like there's so much value just right there, right? Like instant hosting, spin it up, spin it down. It'll stay cold, but we run the demo a few days later, it'll come back up and like all this stuff in retrospect, like it's still what we needed like today.Akshat [00:11:27]: Yeah, it's still needed today. workload shapes have changed a lot as, we run stuff for people with really massive production scale and, there it's it's not about scaling from zero to one, but it's how do we scale really elastically, from like thousand to fifteen hundred GPUs very quickly in a given region. It's the same shape problem.Elastic Inference, GPU Autoscaling, and Custom ModelsVibhu [00:11:50]: Okay. So you look at, say, Cursor Composer, right?Akshat [00:11:53]: Yeah.Vibhu [00:11:53]: They had a. “We'll do RL on a model every couple hours.” you guys have a whole version of RL inference gym and whatnot.Vibhu [00:12:01]: When you look at workloads like that, you're doing train runs where you need to scale up, scale down every hour thousands of GPUs, right? That's the example for we do need it, right?Akshat [00:12:12]: Yeah. Well, so I'll, I'll take a step back and, maybe talk about like how people use Modal today. because our biggest use case is, elastic inference. And the thing we first found product market fit, with was inference for custom models. So we stayed away from the LLM space, and we were serving companies like Suno for audio, Runway for video, robotics, comp bio companies that train their own model elsewhere. But Modal is the best black box that for deployment, scaling to however many GPUs you need as your traffic pattern changes. And we saw all of them like have a very unpredict- predict- predictable, traffic pattern. it's like diurnal. It's Some days, like the company will do a launch and, they'll need like, way more. And it's not just one model that they deploy. They-- all these companies deploy, lots of different models in different regions, and so the autoscaling problem becomes even harder because then you have to scale within a certain region, and those cycles are offset. So different times you scale up in different regions.Akshat [00:13:20]: So that's like our sortVibhu [00:13:22]: And thatAkshat [00:13:22]: YeahVibhu [00:13:22]: That in and of itself is a huge category. There's a bunch of inference providers which, provide this fireworks, does this as a service together, whatnot, Base10. that's carved into its own niche for language models, at least right now.Akshat [00:13:36]: Yeah. the thing that we have specialized in is the autoscaling aspect.Vibhu [00:13:41]: Yeah.Akshat [00:13:41]: Because we found that it's not universally true that everyone else can autoscale, and we've gone deeper into it on the tech side by, we've incorporated GPU snapshotting into the product so we can take the GPU state, like your torch.compile model, snapshot it, and the next cold start is way faster. And so going back to your question, it's That's why you need a lot of burstiness for inference. But then people also do a lot of demand training, like for RL stuff, your rollouts are bursty, as you said. People also do a lot of batch jobs. So we'll see, a lot of companies, before they have a training run, they'll need thousands of GPUs to run encoding or something like that. And I think those things are much more bursty than. I agree that agents are not that bursty. sandboxes are, except when you're doing RL. RL is justRL, Batch Jobs, and 100,000 SandboxesVibhu [00:14:28]: Or commerceAkshat [00:14:28]: Insanely bursty.Vibhu [00:14:29]: Yeah.Akshat [00:14:30]: Yeah. Like when you're doing, rollouts, you sometimes need a hundred thousand sandboxes in your sandboxes.Vibhu [00:14:37]: Yeah. I'm curious if you've seen early sparks of continual learning. There are some people, like our friends, ngram, recently announced thisAkshat [00:14:45]: YeahVibhu [00:14:45]: They're, they're trying to do training. That also seems like a different workload, right? If you're doing training twenty-four/seven per se, there's a very weird dynamic of how you're using GPUs between people and whatnot, but seems like something you guys would work for.Akshat [00:15:00]: As you said, we're, we're fortunate to work with a number of, customers at the frontier and grab some of our customers. and they are taking the primitives we have, and trying to use them in very interesting ways, like continual learning. It's possible as the stuff gets better, some of that will be part of, our offering as well if, more people need it. but we're, we're just waiting to seeVibhu [00:15:23]: YeahAkshat [00:15:23]: How it shakes out.Vibhu [00:15:24]: Is there a primitive that you added after sandboxing that was the next step in the story?LLM Inference, DeFlash, and Speculative DecodingAkshat [00:15:32]: I guess we've been going much deeper into LLM inferenceVibhu [00:15:35]: YeahAkshat [00:15:35]: Because we realized that some of the advantages we have with like autoscaling, again, especially in different regions and whatnot, are, not present elsewhere. and the place where we had a gap was we weren't, working on the model layer itself. Like we were a black box. And, we realized that, we can get to frontier-level model performance, with, by having great people who work on this. And, we've been open sourcing a lot of our work, in terms of, Recently, we, shared our work on DeFlash, which is a block-based, speculator, and we've open sourced, all of it. So, you can - By using open source DeFlash, you can get the same performance as you would with one of the proprietary providers. And the next thing we're thinking about hereVibhu [00:16:23]: I thought this wasAkshat [00:16:24]: YeahVibhu [00:16:24]: An interesting blog post as well, right? Like, I think in here you make a claim that. Not a claim, just that how effective speculative deco-decoding really just get to.Akshat [00:16:33]: Yeah.Vibhu [00:16:33]: Anything you wanna point out from this around, what people should know?Akshat [00:16:39]: Yeah, absolutely. the high-level summary is, it would help to describe what speculative decoding is.Vibhu [00:16:44]: Yes.Akshat [00:16:44]: I will, yes.Vibhu [00:16:45]: I think, likeAkshat [00:16:46]: YeahVibhu [00:16:46]: So we've covered like Eagle and all thisAkshat [00:16:47]: YeahVibhu [00:16:47]: Like Hydra and all those things, but it was like two years ago.Akshat [00:16:51]: Yeah.Vibhu [00:16:51]: I think it doesn't hurt, right?Akshat [00:16:52]: Yeah. Speculative decoding is you have a smaller model, called a draft model, predict tokens ahead of the bigger model, and then you have the bigger model, verify all of this, all the tokens are predicted. And the reason it's faster is if you're predicting, one token at once, you're bound by memory bandwidth. But if you can batch the verification of, the draft model, then you're much more efficient using compute, and it's faster, and as long as your draft model is producing a lot of tokens that can get accepted, which is called the accept length, you can get a speed up that's, multiple times of, the original model speed. and well, that's what we highlight here. It's Like people talk a lot about we made these kernels faster and whatnot, but improving kernel will only give you like few percentage points of improvement, and, increasing accept length, literally is a multiplicative decreaseVibhu [00:17:47]: Like two to four X.Akshat [00:17:48]: Yeah, exactly.Vibhu [00:17:48]: Without much head-on performance.Akshat [00:17:50]: Yeah. I think it may - you are running a second model, right? So it may be something more expensive in the compute,Vibhu [00:17:57]: I meant quality performanceAkshat [00:17:58]: Probably not by muchVibhu [00:17:58]: But yeah. I thinkAkshat [00:17:59]: So there's no drop in quality performanceVibhu [00:18:01]: YeahAkshat [00:18:01]: Because you're always. You're never accepting a token that the big modelVibhu [00:18:04]: It's strictly betterAkshat [00:18:05]: YeahVibhu [00:18:05]: Or it's same.Akshat [00:18:06]: Exactly.Vibhu [00:18:07]: Right. Yeah.Akshat [00:18:08]: And so we've been working a bunch on DeFlash, which is a block-based speculator. so it's instead of predicting, one token at a time, it's predicting a block. And we've been open sourcing our work with it. The next thing for us here is for helping people train speculators and custom models. it's it's something that traditionally is very forward-deployed engineering driven, support deployed, engineer driven, like you work with customers and help them do that. And our vision for. This is why we launched Auto Endpoints, is we want to make frontier-level performance available to everyone. And so, we mentioned this in the announcement, we teased it. The next thing we're, we're launching is, as you run an auto endpoint, we shadow trafficAuto Endpoints and Frontier-Level PerformanceVibhu [00:18:54]: Do you want to explain what auto endpoints are?Akshat [00:18:57]: Yeah.Vibhu [00:18:57]: I lovely, yeah.Akshat [00:18:58]: Yeah. So, this is, I guess, going back to your Modal is you touch the code, but, sometimes people don't wanna touch the code, and they wanna get started with an endpoint that works and has all the great performance and, scalability that Modal has. So we've made that easier with, a way to create an endpoint from our UI, from the CLI, that has all of our optimizations that we talked about, like the DeFlash stuff already baked in, and there's full transparency. So we give you the code, you can go run it yourself, and if you want, you can eject out into the full Modal experience, which we see as people get sophisticated, they do wanna tweak the models, they wanna, fine-tune stuff. You can still do all of that. It's it's not a black box. And yeah, the next thing, as we teased later in the post, is how do we give you value even beyond this in terms of having your draft models evolve as your data distribution evolves, again, without having to talk to a person and, yeah.Vibhu [00:19:59]: I guess just to understand it directly, you have the GPUs, you have an endpoint that's compatible, you serve open model. If someone was to do this themselves, what's the delta that you guys provide? So you do a lot of open source great work on effective inference. how does it compare to, say, I take the same model, 5.2 FP8, take shelf inference engine, vLLM, SGLang, get compute of similar capacity, similar cost. What's the delta that plugging into something this, like this offers outside of the benefit of, scaling?Production Inference Beyond Raw GPUsAkshat [00:20:34]: It's interesting because we've taken the approach of open sourcing our contributions and upstreaming them. we work closely with the SGLang team. We want the improvements that our team, comes up with to be, there in open source for others to use, even outside of Modal. The benefit to us is we have a team that has significant expertise in terms of if you do have something that is not there, our team can help you get that performance, first. the other thing is with these endpoints, we are way more elastic, as you said, than, anyone else, and you have true scaling to zero. you have true, burstiness, and in practice, that matters a lot more to people than just finding, the GPU and, running Modal code on something.Vibhu [00:21:20]: Yeah. And I will say it's not that straightforward to just. like what I said is easier said than done, right?Akshat [00:21:26]: Yeah.Vibhu [00:21:27]: It's I think still for the average person, still hard to just gut check using different. There's, there's quite a bit of combinations you can make there. the trade-offs aren't really known at face value.Akshat [00:21:40]: Yeah. it's it's not just that. I think it's it's that running production-grade inference is a hard infer problem.Vibhu [00:21:49]: YeahAkshat [00:21:49]: Even if you subtract out the autoscalingVibhu [00:21:50]: YeahAkshat [00:21:51]: Is controlling things like tail latency and, making sure every, request is delivered at least once and whatnot.The Model and Agent LifecycleVibhu [00:22:00]: There's a lot of innovation that you can do here. I think, it's very interesting that you're starting to encroach on, like as you become a full cloud, you're starting to encroach on other people's turf.Vibhu [00:22:09]: What will you not do?Akshat [00:22:13]: Well, we wanna follow our users and, make sure they get like a platform that has everything that works well together. so right now we're focused on the model lifecycle and the agent, lifecycle. so both like going from data prep to training to inference, and then also if I want to deploy a background agent, let's say, sandbox, do persistent storage, a whole bunch of other stuff.Vibhu [00:22:38]: We talked to Cole, who did, OpenInspect. Yeah.Akshat [00:22:42]: Yeah.Vibhu [00:22:42]: And RealInspect also is on Modal.Akshat [00:22:44]: Yeah. So Ramp Inspect was a great example of a background agent that was really successful because they, were able to use some of the primitives like snapshotting and fast scaling to just have something that feels really reactive and works well.Ramp Inspect and Background AgentsVibhu [00:23:02]: Yeah. That's the new CTO of, Ramp right there.Akshat [00:23:05]: Yeah, Rahul.Vibhu [00:23:08]: It was really fun. yeah, okay, I think, all very bullish. Like, one of my reflections was also I did not originally. So when I met you guysThe Inference Inflection: CPU, GPU, and Co-LocationVibhu [00:23:19]: You weren't that much in the GPU game, and now you're all about, inference. And one of the points that I hinged on for Jensen's keynote at GTC this year was, what we're calling like the inference inflection, right? That let's say in AI workloads or machine learning workloads, it used to be like, let's call it eight to one GPU to CPU, and now it's more like one to one, which is like a interesting. Like, - because of how much agents are blocked or call out to this, to CPU heavy stuff the actual, like, limiting factor, like, swings back and forth from GPU to CPU a lot more than it used to be all GPU and then occasional CPU.Akshat [00:24:01]: Yeah.Vibhu [00:24:02]: GPU, CPU. And now it's like just constantly, and you just have to locate everything.Seventeen Clouds and the Supercloud StrategyAkshat [00:24:08]: Yeah. And that's one of the things that, again, we see as, something appealing about Modal, which is we've built this capacity pool that spans, 17 cloud providers, so we're, we're very good at Running on various kinds of cloud capacity across the worldSwyx [00:24:24]: You don't have your own data centers?Akshat [00:24:25]: We don't have our own data centers. We just run across a lot of neo cloudsSwyx [00:24:29]: Yeah. AreAkshat [00:24:30]: Metal providers.Swyx [00:24:30]: Yeah. Question mark.Swyx [00:24:31]: Yeah. You're, you're running the math, and you're like, “What's the cutover point where you're like.”Akshat [00:24:36]: Yeah, it's a good question. part of it is we see our differentiator in the software layer, and, being capital light and focusing on the software helps us move really fast. so far it's worked out well because there are so many other people building data centers that we're able to work effectively with them, and again, focus on what makes us, special.Swyx [00:24:55]: Yeah.Swyx [00:24:56]: 17 gets you into, like, the local providers sometimes. LikeAkshat [00:25:00]: The,Swyx [00:25:01]: Which was the most interesting one?Akshat [00:25:02]: There are a lot more neo clouds than you expect, and they all have various degrees of, various levels of reliability. And, that's why it's something we've invested a lot of time in, is building our own reliability layer on top. so if the GPU falls off the bus or something happens, we user workloads are not affected, and that lets us use a lot more capacity than,Swyx [00:25:30]: YeahAkshat [00:25:30]: You as a user would be able to.Swyx [00:25:32]: It's a useful thing to have because like now everyone knows, like, what layer you are and, like, you optimize for being the super cloud of all clouds.Akshat [00:25:41]: Yeah. That's, that's, that's the idea. and so I guess when you mentioned colocation, that's, that's another interesting thing where, one thing we've seen is people come to us when they want, very specifically located, CPUs or GPUs, like they wantSwyx [00:25:57]: Oh, they pin it in likeAkshat [00:25:58]: YeahSwyx [00:25:58]: EU?Akshat [00:25:59]: Exactly. Or EU, US.Swyx [00:26:01]: Right. Data resiliencyAkshat [00:26:02]: AustraliaSwyx [00:26:02]: Locality thing or performance or what?Akshat [00:26:04]: It's either data locality or latency, yeah.Swyx [00:26:07]: Yeah.Akshat [00:26:07]: Like, you want your. They're running sandboxes and model. They want them to be right next to aSwyx [00:26:10]: Yeah, it's easy thenAkshat [00:26:11]: YeahSwyx [00:26:12]: To. That is important in all those things. and so, like, you've accidentally, I don't know if it's accident, but, like, you've built the perfect primitive for agents to express themselves. And then, like, it's almost very funny how every extra development just involves more file system, just involves more CPU.Akshat [00:26:30]: Yeah.Swyx [00:26:31]: Just like the things that you already have. I don't know much about, if there's any, like, networking usages that are interesting, but you've also done some good work on networking.Networking, Sidecars, Private IPv6, and SandboxesAkshat [00:26:40]: Yeah, that's exactly right. Like, we're just taking compute storage and networking and building stuff on that layer, for, again, the stuff people need.Swyx [00:26:49]: YeahAkshat [00:26:50]: We see a few interesting networking things coming up. one is people want networked sandboxes. so we haveSwyx [00:26:57]: For like a Docker cluster type thing.Akshat [00:26:59]: Yeah.Swyx [00:26:59]: Sorry, Docker Swarm. Oh, f**k. What is it called?Akshat [00:27:02]: Compose.Swyx [00:27:03]: Compose type thing.Akshat [00:27:04]: Yeah. So if you want Docker Compose, our sandboxes now support, this thing called sidecars. So you can. A sandbox is a pod of containers, and you can run multiple containers in, a sandbox. also useful because, going back to networking, people want a lot of control over, outbound networking from a sandbox.Swyx [00:27:23]: Yeah.Akshat [00:27:23]: Like, they might wanna run a middle proxy for, like, maybe logging stuff for RL or, controlling how egress can happen to a domain, injecting credentials. and yeah. So we've, we've had to build a lot of that stuff ourselves.Swyx [00:27:38]: Yeah.Akshat [00:27:39]: But then also sometimes people want, sandboxes spanning multiple nodes to talk to each other, which is an emerging thing we're seeing. We have support for that for a different reason, and yeah, we'll see if that becomes stable.Swyx [00:27:52]: Like, just an open socket. It's a. This is directly like mTLS.Akshat [00:27:56]: We do support that, which is you can, expose a tunnel inside a sandbox.Swyx [00:28:01]: Yeah.Akshat [00:28:01]: And then you can either expose it to public internet or it can be, you can add like a HTTP, auth layer above it. But we have this thing called I6PN, which we haven't talked about, which is this, like, overlay network using IPv6 addresses. so if Modal containers, within the same workspace, when this is enabled, can address each other using this private IPv6 address, and no one else can.Akshat [00:28:28]: So it's like private networking, for containers. We built it because we needed it as a primitive for our distributed training product. so we have this other feature, which is you can add a decorator to a function, and you get a cluster of GPUs. and they have RDMA networking. so you can run a distributed training job, that's truly serverless. and we did the overlay network for that. But then we've seen that people are using it for other reasons, and, I'm intrigued to yeah, what would people do with it.Swyx [00:28:59]: Build primitives and let people figure it out, right?Akshat [00:29:01]: Yeah, exactly.Swyx [00:29:02]: You put out a pretty interestingAkshat [00:29:03]: They're like, they read the docs webpage. Let me use thatSwyx [00:29:06]: YeahAkshat [00:29:06]: Something they never intended to work. This is literally not even in our docs page. People somehow found it, and they're using it.RDMA, Memory Movement, and Distributed TrainingSwyx [00:29:12]: Huh.Swyx [00:29:14]: The way you portrayed it with, like, RDMA versus TCP, like, very well laid out, but just the transfer speed change at scale for RL, like yeah, you have it, you have it built in. I'm sure someone found it. It's found it to be a lot more efficient before you made a thing out of it, right?Akshat [00:29:32]: Yeah. And not to split hairs, I guess the overlay network is the TCP overlay network.Akshat [00:29:39]: The reason we have that is you need that to do the key exchange for RDMA before you set up the RDMA network on top of that. but then people found the TCP part.Swyx [00:29:48]: Can I tell you, this is like a big aha moment for me becauseAkshat [00:29:51]: YeahSwyx [00:29:51]: So I review 2,200 submissions for the World's Fair.Akshat [00:29:56]: Yeah.Swyx [00:29:57]: And then I got this from John OsterhoutAkshat [00:29:58]: HuhSwyx [00:29:59]: Who I don't know if. Do John Osterhout by name?Akshat [00:30:01]: The name sounds familiar.Swyx [00:30:02]: He published a. He's a well-known professor, published a lot of interesting software design books, and this is the talk he chose to submit, is on RDMA at Inference. And I'm like, you wouldn't think that this guy, who is like operating systems guy, would care about RDMA.Akshat [00:30:20]: I, it makes sense to me because I,Swyx [00:30:24]: This is the cloud, right? YeahAkshat [00:30:25]: Like, the way you move around your KV cache and how efficiently you can do it, how efficiently you move, your weights from your training GPUs to your inference GPUs in RL is there's a lot of degrees of freedom, and it is a systems problemSwyx [00:30:41]: YeahAkshat [00:30:41]: Moving memory aroundSwyx [00:30:42]: YeahAkshat [00:30:43]: Scheduling.Swyx [00:30:44]: This shows you how primitive my understanding of networking stuff is.Swyx [00:30:46]: Is this like the domain of WireGuard as well?Akshat [00:30:50]: Not quite.Swyx [00:30:51]: It's adjacent?Swyx [00:30:53]: Explain everything.Akshat [00:30:54]: Sure.Swyx [00:30:56]: How do we move memory around GPUs?Akshat [00:30:58]: Well, so sorry. Yeah, that is memory. Sorry, I was talking more, and maybe I was talking like five minutes back, about the private IPv6, addressing that you've set up.Swyx [00:31:09]: Yeah.Akshat [00:31:09]: Is it like it's a VPN?Swyx [00:31:10]: Yeah, it is like a VPN, and yeah, WireGuard is, yeah, you're right. It is,Akshat [00:31:16]: Right. Yeah, you already moved on to new topicsSwyx [00:31:17]: A similarAkshat [00:31:18]: OkaySwyx [00:31:19]: In the same space, WireGuard is, encrypted and this is,Akshat [00:31:23]: And you don't need encryption.Swyx [00:31:23]: Yeah.Akshat [00:31:24]: Yeah.Swyx [00:31:24]: This is not encrypted. that's the main difference. This is TCP and we have eBPF programs that will reject or allow the TCP connection based on whether you're allowed to do it.Akshat [00:31:35]: Used to involve a full sidecar, but now you have eBPF in the Linux kernel.Swyx [00:31:39]: Yeah.Akshat [00:31:40]: Yeah. I don't know if this is a natural follow-on to the topic of like my skepticism on distributed training is that while, like, people spend a lot of money on, like, cables to hook up GPUs, and even that is not, like, fast enough, and that's the bottleneck, is your networking fast enough?Swyx [00:31:59]: Yeah. So I guess you're talking about fully distributed training like, Dialog or something which is like cross data centerAkshat [00:32:06]: That would be, yes.Swyx [00:32:07]: That's the extreme.Akshat [00:32:08]: Yeah.Swyx [00:32:08]: You're in the middle, and then other people would have like the Mellanox cables up in, like, their actual data center.Akshat [00:32:14]: When you run multi-node training on Modal, RDMA, I think Mellanox, is, or InfiniBand is like a, is all seen as RDMA. but it's a way to bypass the TCP networking stack and, transfer, stuff much faster, between one node, to the other. And we have I think like 3 terabit per second, internal networkingSwyx [00:32:40]: OkayAkshat [00:32:40]: Which is the standard that's needed.Swyx [00:32:42]: Okay. So I misunderstood whatAkshat [00:32:43]: 50Swyx [00:32:43]: What part of the stack you wereAkshat [00:32:44]: 50 gigs overSwyx [00:32:45]: YeahAkshat [00:32:45]: If you wentSwyx [00:32:45]: YeahAkshat [00:32:46]: RDMA.Swyx [00:32:46]: Okay.Swyx [00:32:48]: Yeah. I, very impressive work.Multi-Node Training, Post-Training, and Auto ResearchSwyx [00:32:52]: So effectively you're extending like the model philosophy to the training cluster, like, yeah.Akshat [00:32:59]: Yeah. And we're, we're not going for like large scale training runs. the thing that we've built multi-node training for is, we see a lot of, smaller scale post-training. like, people are post-training like medium sized fund models, so they can, get higher quality on inference. this is a perfect fit, for something like that.Swyx [00:33:21]: Yeah. That is my impression of how a lot of these labs explore branches in post-training and then eventually merge whatever they find in.Akshat [00:33:31]: Yeah. The other use case we've seen for multi-node training is even if you have a big cluster, your researchers are still doing small runsSwyx [00:33:38]: YesAkshat [00:33:39]: Having elasticity thereSwyx [00:33:40]: Right, sureAkshat [00:33:40]: Matters a lot more.Swyx [00:33:41]: Yeah. the, like, this is like the current limiting factor for auto research, which is like you need to give your model some GPUs in order for it to completely run.Akshat [00:33:51]: We have a blog post on auto resource and model is,Swyx [00:33:55]: YeahAkshat [00:33:56]: Yeah, like, turns out to be pretty good substrate for that.Swyx [00:33:59]: So my impression is auto research means many things, likeAkshat [00:34:01]: YeahSwyx [00:34:01]: Anything that Andrej coins. Right now it's still science fair, right? Like not like, I don't know how many people are doing this.Akshat [00:34:08]: We're having a golf.Swyx [00:34:08]: Yeah.Akshat [00:34:09]: I thought the same thing.Swyx [00:34:11]: Yeah, you would know.Akshat [00:34:12]: We, like, our internal both training and inference teams use this the general shape of this quite a bit. like we have this one internal repo called auto inference, which essentially we've automated our own forward-deployed engineering efforts using, this harness, which is, the agent will just spin up a sweep of different things. It'll even run like, NVIDIA inside profiler and it'll like tweak configs and it'll arrive the right thing. it'll change your GPUs both from H200 to B200, and works really well.Swyx [00:34:47]: Nice.Akshat [00:34:47]: So yeah.Swyx [00:34:48]: By the way, I enjoy that your forward-deployed engineering is so technical that you have to do these things.Swyx [00:34:52]: It's very different from forward-deployed engineering from other people.Akshat [00:34:54]: Yeah. For our forward-deployed engineering team is, essentially they're like applied inference researchers or applied training researchers.Swyx [00:35:02]: Someone told me like they have to be able to build, but they also have to be able to sell. do they have to sell or are they like they're good, they're just like post-sale type of thing?Akshat [00:35:09]: It does, being able to talk to a customer and engage effectively with themSwyx [00:35:13]: YeahAkshat [00:35:13]: Matters a lot.Swyx [00:35:14]: They want the same thing.Akshat [00:35:15]: Yeah.Swyx [00:35:15]: ?Akshat [00:35:15]: But it's it's not really a sales, thing. We pair them with-- We have solution architects as well that are more on the sales side.Swyx [00:35:23]: Okay. Let's spend a bit more time on auto research. This is a big focus for for this year. Where does this go? like, have people explored enough? Like, there's all these beautiful charts of like improve and then level off a bit and then you find the next thing. Is this one abstraction up from normal training? Is that how we think about it, or do you think about it differently? Like model level training versus high, like driven hyperparameter search.Auto Inference and Modal BenchAkshat [00:35:51]: Yeah, like,Swyx [00:35:51]: Someone, some people call it like neural architecture search or whatever, right? Like.Akshat [00:35:54]: Yeah, - So the stuff I've seen people do with it is nowhere on the architecture level. It's pretty much tweaking parameters, but it's it's a hyperparameter sweep that's guided by some model intuition, so it's like much more efficient than, whatever other, sweep you would have.Swyx [00:36:12]: Yeah, it's just, it's just a question of where you want to spend your compute?Akshat [00:36:16]: Right.Swyx [00:36:16]: ‘Cause yeah, you can just throw infinite amounts of money on this and somehow you'll bang out Shakespeare?Akshat [00:36:22]: Yeah, infinite monkey.Swyx [00:36:24]: Yeah, so like the very good for model. and I think it's also very important that agents can spin up other agents, can spin up their infrastructure. Like very good for you. how good is our LLMs at generating model code? Like the benefit of existing LLMs is that you are in the data.Akshat [00:36:42]: Yeah. They're, they're surprisingly good. I think like pre Cloud 4 they were not, and then now they're able to shot, stuff out of the box. But we're playing around with releasing like a Modal Bench for like the harderSwyx [00:36:55]: YeahAkshat [00:36:55]: Things, that the LLMs cannot do yet and maybeSwyx [00:36:59]: What's an example of that?Akshat [00:37:01]: I think the things that- Sometimes agents struggle with, without right guidance and a skill is, how to, use the rest of our observability. Like how to. Something is failing, like how do you look at the logs and then update the right thing? It's reasoning about that. But they're able to shot, likeSwyx [00:37:23]: Yeah. You can just add a skill to it?Compute Strategy and Capacity PlanningAkshat [00:37:26]: Yeah. So we have a Modal skill now that. Which is why we built this Modal Bench. It's to find things like that, so we can address them in our tool.Swyx [00:37:35]: Tune a skill. Yeah.Akshat [00:37:36]: Yeah.Swyx [00:37:36]: No. it's it's good. are you facing any shortages? like we talk a lot about GPU shortages, but also CPU, also memory.Swyx [00:37:44]: Yeah.Akshat [00:37:45]: We have had a lot of growth, which means that, there's - we've had to be much better aboutSwyx [00:37:53]: PlanningAkshat [00:37:54]: Proactive capacity planning.Swyx [00:37:55]: Yeah.Akshat [00:37:55]: So we have,Swyx [00:37:57]: Which by the way, like it's like a MBA's like dreamAkshat [00:38:00]: YesSwyx [00:38:00]: Is like just planning this stuff. I think last time you and I talked about something maybe about this.Akshat [00:38:03]: Yeah. we have a really competent team of people that we call, The role is called compute strategy. so yeah, if anyone listening here or wants to work on thatSwyx [00:38:13]: Compute strategy?Akshat [00:38:13]: Yeah.Swyx [00:38:14]: I think,Akshat [00:38:14]: I feel like,Swyx [00:38:15]: I think the normies call it FP&A or something.Akshat [00:38:18]: Well, it's more It's it's not FP&A. It's it's There's a lot of interesting financial questions of like what is the blend between one year and three-year reservations? how do we forecast our own capacity? how do we. especially since our capacity is very fungible across different GPU types and different regions, like you have to model a lot of it. and you also have to have an opinion on how the supply chain is gonna evolve, and then you have to like, take bets,Swyx [00:38:49]: YeahAkshat [00:38:49]: Based on that.Swyx [00:38:50]: Tokenomics.Akshat [00:38:50]: Yeah.Swyx [00:38:51]: This is like probably a not a real point, but, I was trying to think about like what other industries. I was trying to think about like, we cannot be first to like these kinds of problems.Akshat [00:38:59]: Yeah.Swyx [00:39:00]: And what other industries have had this? And I was like, airlines with fuel and like they have to hedge their fuel and like, I think for a long time Southwest because they made like a hero fuel bet, they like were like super low cost becauseAkshat [00:39:12]: OhSwyx [00:39:12]: Compared to everyone else.Akshat [00:39:14]: Yeah. I hadn't thought about that.Vibhu [00:39:16]: We're at a fun time too?Akshat [00:39:18]: Yeah. It's. A lot of the compute business in general, for us is also about being very good about capacity management. That is how you have great unit, economics. but also over time it's how you can unlock more value for customers. Like, one of the things we're building now is like a way for customers to get, If they don't care about latency, like get much cheaper pricing and they'll get results back in like next 24 hours or something, like a batch tier essentially.Batch Tiers and Latency-Insensitive WorkloadsSwyx [00:39:47]: Yeah.Akshat [00:39:47]: And those are levers we have because we control the whole stack and scheduling and whatnot to give people a sufficientSwyx [00:39:53]: Yeah. I feel like they're not as popular. Like those, like the Frontier Labs have all those APIs. They're not as popular as they should be.Akshat [00:40:00]: The demand that we see for something like that is not for LLMs. although sometimes people wanna run evals andSwyx [00:40:08]: OkayAkshat [00:40:08]: Synthetic data prep and there it makes sense.Swyx [00:40:10]: Okay.Akshat [00:40:11]: But it's from a lot of LLM companies, like people who are doing computational bio, like they have to run really big batch jobs and they don't care about when they get it back.Swyx [00:40:22]: Yeah. And like they have a reasonable. It's it's also like a cousin to the stopping problem of like, will this finish in time?Akshat [00:40:30]: Yeah. You can bound it.Swyx [00:40:33]: Yeah.Akshat [00:40:33]: Like you can give peopleSwyx [00:40:34]: YeahAkshat [00:40:34]: SLAs on it.Swyx [00:40:35]: Yeah. I think what's, what's interesting is like the next phase of model.Swyx [00:40:38]: Like what, do people expect from you, now that you're established and you're like well-known compute player among all these leading companies. You had an inference launch week, and we talked a little bit about the launches. like what else? Like what else should people know?What Modal Builds NextAkshat [00:40:55]: We are building primitives that make our users' lives much easier. So, I think for example, with LLM inference, thousands more companies are gonna post-train their own models and, deploy open source models for inference. so we're thinking a lot about what is the best product shape for that. And, that involves everything from our training gym to, then, endpoints that get frontier-level performance. again, but I haven't talked to anyone. It looks somewhat different on other verticals. Like, we're also seeing a lot of real-time, audio-video stuff in there, which is why like, we're working on things like regional routing, with fallbacks. So you can get GPUs that are as close to users as possible. so you get like low latency for video streaming and whatnot. And then on the agent side, it's,Akshat [00:41:52]: We're still working very closely with our customers because stuff is changing so fast in terms of what they need. And, I think beyond sandboxes and persistent file systems, there's a lot of other things people will need from this agent stack as they build production agents. So yeah, we're thinking about those other things that fit in there.Swyx [00:42:13]: I want to ask what the other things are.Akshat [00:42:15]: Yeah. I probably should share right now.Swyx [00:42:17]: I think-- I think, okay, so, I do think a lot about the principal components of cloud, and you do talk about compute storage networking.Akshat [00:42:25]: Yeah.Swyx [00:42:25]: Because so far for me, it's fine. so far for the. the first couple generations of cloud, it's fine. What's different, qualitatively different about agents that you need some new permission level? Like a lot of people, okay, and I'll just kinda spew tokens at you until it like hopefully sparks something.Akshat [00:42:43]: Yeah.Swyx [00:42:44]: Like the new level now is whatever Claude Code does, which is dangerously scope permissions or like allow list by command or like whatever, right? And sometimes they're like, “Well, okay, we have like this adaptive thinking mode where like, just trust me, bro. I will make the calls for you.” Is that it? like mediated permissions.Hard Guardrails vs. LLM-Mediated PermissionsVibhu [00:43:03]: Now you're looping it with a goal and letting it roll.Akshat [00:43:06]: Yeah, I'm, I'm skeptical of LLM media permission for stuff that is at the sandbox level because you do want hard boundaries.Swyx [00:43:16]: Yeah.Akshat [00:43:16]: Otherwise, someone can exfiltrate stuff.Swyx [00:43:20]: But likeAkshat [00:43:20]: YeahSwyx [00:43:20]: Maybe that's old school thinking. Maybe we're the dinosaurs.Swyx [00:43:23]: Maybe the AI OS or the LLM OS is really the kernel is a goddamn LLM.Swyx [00:43:30]: Like it makes you feel uncomfortable.Akshat [00:43:31]: Yeah, I'm, I'm toldSwyx [00:43:32]: But that's what trusting the LLM is. Like imagine a spherical cow perfect LLM.Akshat [00:43:36]: Right.Swyx [00:43:37]: That it.Akshat [00:43:39]: Maybe.Swyx [00:43:41]: I wanna test the boundaries, right?Akshat [00:43:42]: Yeah.Swyx [00:43:42]: Like, and I don't believe that, but I wanna see where I'm wrong ‘cause that's, that's the consensus.Akshat [00:43:49]: Yeah. I think you always need hard guardrails when you want, And you can pair those with softer guardrails, right? And that's gonna be a lot of mediated.Managed Agents and Specialized SandboxesSwyx [00:44:00]: There. I'll also get you a end with a couple of your commentary on like the ecosystem outside of Modal. Manage agents. Everyone has one. Gemini, OpenAI, Claude, very useful for you, but also like it is their way of starting to edge into your space.Akshat [00:44:17]: Yeah.Swyx [00:44:17]: What's going on?Akshat [00:44:19]: Yeah, we're, very excited to partner with Anthropic and some of the other foundation labs, will not name who we're also working with. the way we see it is the manage agent thing is a great place to start if you're starting out building an agent and, But then when you get to, building something more production grade, like you're a company that's like Ramp that's building their own, Ramp also runs their accounting agent on us, so their external-facing agent. You need a lot more control over, your compute primitive on things like, what sort - how do you persist different files that the agent has access to, and how do you snapshot and restore? How do you control the networking? maybe you want GPUs. When you get to that point, you kinda want, a specialized sandbox provider, that gives you those things, and that's the role that we are trying to play.Swyx [00:45:15]: YeahAkshat [00:45:16]: We don't really have an opinion on the harness, whether it runs - it's a cloud-managed agent, and you hook it up to Model Sandbox, or you run the harness in Model Sandbox. We'll see where people converge with that.Swyx [00:45:26]: Yeah. Do you any opinions on like the meta harnesses, or just another layer on top of these things?Akshat [00:45:31]: You mean like the OpenPipeSwyx [00:45:33]: OpenPipe is one. I think Vercel had one, which I can't remember the name of right now. Fredshot had one. and then, to me, most recently was Data Databricks that had Omnigen. All these are meta harness. Like it's kinda pseudo agent cloud type things.Akshat [00:45:50]: I personally have not played around with them.Swyx [00:45:53]: Yeah.Akshat [00:45:53]: Build agents with them.Swyx [00:45:54]: Everything's bullish Modal, as long as it consumes more infra.Akshat [00:45:57]: That's why we're focusing on the infra layer. It's somewhere where our, relative competence is and, also it's a hard problem to solve.Swyx [00:46:06]: Yeah. I will say like just generally reflecting on that, I don't know if - if there's other topics on Modal, but like just generally reflecting as an infra person, not as intense as you, but in that field, this has like been the most exciting time in infra. Like it was boring for a while, and you couldn't really get people excited about data infrastructure. Like Eric would get on Data Console, everyone just watched the video and like say, “Look at how many sandboxes I can spin up,” and no one gave a crap.Why Infrastructure Became Exciting AgainAkshat [00:46:39]: Yeah.Swyx [00:46:40]: And like now everyone gives a crap.Akshat [00:46:42]: That's true. It is a very exciting time, and I think a lot of that's driven by just the amount of scale all of this stuff needs.Swyx [00:46:50]: I think the, like a lot of your initiatives or a lot of your like product directions make sense in retrospect, which is like the best kind, but I wouldn't necessarily have thought about it myself, which.Akshat [00:47:00]: We need the predictions.Swyx [00:47:02]: I think there's a lot that you just don't even see, right? Like you have the batch, you have the voice, you have the multimodal, but what else?Akshat [00:47:10]: What else is coming up for usSwyx [00:47:11]: Yeah. Where do you see things going?Akshat [00:47:13]: Yeah. I, in generalBiotech, Robotics, and Non-LLM AI WorkloadsAkshat [00:47:15]: It's it's clear that there's there's a huge shift happening. I think one thing that's not as obvious to people because LLM inference gets talked about so much and is also we work a lot of companies that are, doing things like drug discovery and computational bio, like the Chai Discoveries of the world. Big things are probably gonna happen there. we work a lot of robotics companies that are putting robots in like active deployments and getting good results out of them.Swyx [00:47:45]: Is there Air Gap Modal? Is there a version that is like prem air gapped whatever?Akshat [00:47:50]: No. We,Swyx [00:47:51]: You should cloud only.Akshat [00:47:51]: Yeah.Swyx [00:47:52]: Yeah. Okay. But yeah, so what you're saying is like because you're focused on primitives and they're good primitives, you find use cases in all these kinds of things.Akshat [00:48:01]: Yeah.Swyx [00:48:01]: Probably diversifies you a little bit away from LMS all the time.Akshat [00:48:05]: Yeah, absolutely. We're, we'- our goal isn't to only serve the LLM inference market.Swyx [00:48:10]: There are a lot just on the website, the audio,Akshat [00:48:12]: Yeah. We said both onSwyx [00:48:14]: Computational bio images. Yeah, there's a lot here. There's QTA TTS, customizing. Oh, Chatterbox. there was customizing Whisper.Akshat [00:48:24]: Okay. Yeah.Swyx [00:48:25]: This screen reminds me of a fallen competitor, which Replicate.Model APIs vs. Differentiated AI ProductsSwyx [00:48:31]: What's your postmortem on what happened?Akshat [00:48:34]: This is one thing we've stayed away from is providing an API for models because I think providing model APIs is some of it ends up serving like a really hobbyist market, which is much less sticky.Swyx [00:48:50]: Yeah.Akshat [00:48:50]: And we've always wanted to build for companies that are building products and need more flexibility that's not just an API.Swyx [00:48:57]: Which you can build an API for a model and this is clearly what it is. But you - but what you're saying, you can wrap it into a more fully functioning back end that you run.Akshat [00:49:06]: Yeah. So all of our examples, it's not that spin up this model, here's an API token, use it. They're all code.Swyx [00:49:13]: Okay.Akshat [00:49:13]: And so the point is that this is just an example.Swyx [00:49:16]: Starter code.Akshat [00:49:17]: Yeah. But you can tweak it however you want.Swyx [00:49:20]: Yeah.Akshat [00:49:21]: And if you're like a company building a product, like, computational bio whatnot, yeah.Swyx [00:49:26]: I guess I'm trying to tease out for listenersAkshat [00:49:28]: YeahSwyx [00:49:28]: When does it stop becoming, oh, you're just an API call and you're just a wrapper on API to becoming what you call a product, right?Swyx [00:49:36]: Like, what is that layer? Like what-- Like, more lines of code, but like beyond that, what is the substance that people add that qualifies it to be something more?Akshat [00:49:46]: I think there's a little bit of like a selection effect of like a lot of the companies who do wanna get deeper into that level are probably building something that's more differentiated. And, I think, an example is like - with LLM inference, originally we, worked with companies that were building their own post-training frameworks or they were, - Ramp early in the day was training their own tokenizer and like swapping out the tokenizer in Llama and whatnot. I'm not saying that's, that successful, in that case. But a better example is like, let's say Suno. because Suno, does not use Modal for training.Swyx [00:50:26]: Mikey on the pod. Yeah.Akshat [00:50:27]: But they use Modal for all their inference and that's because they have like a custom-- They have completely custom model architecture and that means that they have to be at the code level and tweak things that are not, just an API.Swyx [00:50:41]: It's interesting as well, like we had, Ethan, most recently on the xAI Groq team make a prediction that like the next tier in video gen is not a better video model, it's a better model or agent that orchestrates video models.Video Agents and Production WorkflowsAkshat [00:50:56]: Oh, interesting.Vibhu [00:50:56]: Language model backbone that can use toolsAkshat [00:50:58]: RightVibhu [00:50:59]: And write code.Akshat [00:51:00]: Like, yes, I can make my second video or my second video from Groq, but I want my minute video.Akshat [00:51:06]: And I'm not going there through normal video gen.Swyx [00:51:10]: Yeah, that's interesting. I - So we have GPU sandboxes and recently have seen a few companies doing agents that do video manipulation or,Akshat [00:51:22]: Yeah. Give it FFmpeg and just do it.Swyx [00:51:23]: Run FFmpeg. But likeAkshat [00:51:25]: That's not enough.Swyx [00:51:25]: Yeah.Akshat [00:51:26]: You need to give it Adobe.Swyx [00:51:27]: Yeah, I hadn't put it together with like it would be a video production thing. in my mind these things were going more towards editingAkshat [00:51:36]: Yeah.Vibhu [00:51:36]: Well, shout out Mantis.Akshat [00:51:37]: I think about this a lot.Swyx [00:51:38]: .Akshat [00:51:41]: Yeah. Sorry.Vibhu [00:51:41]: Luma. Luma Agent is a version of this for video production, but it's a off.Swyx [00:51:46]: I was gonna get your quick takes, on some other stuff that happensGitpod/Ona, CI, and Runtime SandboxesSwyx [00:51:50]: In recent news and just-just see if you have anything interesting. Gitpod, very li

S5:E253 David co-hosts a special episode on the future of Physical AI with Drobot CEO Stijn Van de Velde. Stijn was on the Pod for E178 for Part 2 of my series on Robotics in the summer of 2025. In this episode we'll be talking about the current market for Robotics, Software for bots, Robots-as-a-Service, virtual training with digital twin technology, the value prop for Humanoids and what the future is likely to look like for physical AI. (recorded 7.6.26)Follow David on X at https://x.com/DGRollingSouthConnect On LinkedIn with David at https://www.linkedin.com/in/davidgrisell/Follow Paul on X at https://x.com/PalmettoAngelConnect On LinkedIn with Paul athttps://www.linkedin.com/in/paulclarkprivateequity/We invite your feedback and suggestions at www.ventureinthesouth.com or email david@ventureinthesouth.com.

Simpro Group CEO Fred Voccola argues that the biggest AI opportunities are emerging in the skilled trades, where small and midsize businesses are using digital workers to automate back-office tasks and improve profitability. He explains why entrepreneurial companies are leading AI adoption today and why robotics could begin transforming field service operations over the next several years.======== Schwab Network ========Empowering every investor and trader, every market day. Subscribe to the Market Minute newsletter - https://schwabnetwork.com/subscribeDownload the iOS app - https://apps.apple.com/us/app/schwab-network/id1460719185Download the Amazon Fire Tv App - https://www.amazon.com/TD-Ameritrade-Network/dp/B07KRD76C7Watch on Sling - https://watch.sling.com/1/asset/191928615bd8d47686f94682aefaa007/watchWatch on Vizio - https://www.vizio.com/en/watchfreeplus-exploreWatch on DistroTV - https://www.distro.tv/live/schwab-network/Follow us on X – https://twitter.com/schwabnetworkFollow us on Facebook – https://www.facebook.com/schwabnetworkFollow us on LinkedIn - https://www.linkedin.com/company/schwab-network/ About Schwab Network - https://schwabnetwork.com/about

Tony Wang and Lee Sandquist look to the future to help us understand what physical AI may mean for the world. From autonomous vehicles and robotics to defense applications and industrial automation, listen to what the future may hold.

There are about to be two kinds of professionals: the ones who command AI, and the ones who get commanded by it. In this episode of The Game Changing Attorney Podcast, Michael Mogill sits down with Mike Brown, a personal injury attorney with zero coding background who taught himself to build with AI agents and then beat 15,000+ applicants to win Anthropic's 2026 Global Claude Code Hackathon. They dig into why so many lawyers are waiting for AI to mature while a small group is already compounding a serious advantage, and why the real dividing line has nothing to do with technical skill. It comes down to curiosity, discipline, and a willingness to be bad at something for one uncomfortable week. Here's what you'll learn: What "vibe coding" actually is and why you don't need to know how to code How to prevent hallucinations, protect client privilege, and actually trust the output you get What it takes to go from overwhelmed to automated with a three-week plan any firm owner can run The people pulling ahead with AI are not the technical ones. They are the curious ones. (00:00:00) Introduction (00:03:04) What vibe coding actually is (00:06:15) Why he bet on ChatGPT early (00:09:03) How the models evolved (00:14:29) Producers vs. consumers (00:15:48) Could AI run a whole firm? (00:19:39) Fixing the housing permit crisis (00:22:26) Will AI kill jobs or create them? (00:26:33) Robotics and the hardware layer (00:34:04) AI for health and longevity (00:43:16) Beating AI's confirmation bias (00:48:18) Wrap-Up ---- Links & Resources: Crossbeam Anthropic Boris Cherny Gary Tan Andrew Huberman ---- Learn what sustainable growth can look like for your firm at crispcoach.com. ---- Do you love this podcast and want to see more game changing content? Subscribe to our YouTube channel. ---- Past guests on The Game Changing Attorney Podcast include David Goggins, John Morgan, Alex Hormozi, Randi McGinn, Kim Scott, Chris Voss, Kevin O'Leary, Laura Wasser, John Maxwell, Mark Lanier, Robert Greene, and many more. ---- If you enjoyed this episode, you may also like: 392. How One Startup Is Rewriting the Rules of PI Law with Ray Mieszaniec 345. AMMA - The AI Advantage: What to Automate, What to Keep Human, and How to Stay Ahead 321. Kyle Bachus - Harnessing Technology for Legal Success

Healthcare robotics is not just about replacing tasks; it is about helping people do safer, more meaningful work. In this episode, Spencer Krause, President and CEO of SKA Robotics, shares how his team applies robotics across industries like mining, construction, agriculture, and healthcare. He explains how robotics can support hospitals by handling repetitive or difficult work, from moving waste and controlled medications to helping address patient transport challenges. Spencer also discusses the realities of bringing robotics into healthcare, including stakeholder buy-in, HIPAA concerns, sterilization needs, FDA requirements, and long sales cycles. Tune in to learn how robotics can reduce burden, improve safety, and bring cross-industry innovation into healthcare. Resources: Connect with and follow Spencer Krause on LinkedIn. Follow SKA Robotics on LinkedIn and explore their website. Listen to Spencer Krause on The Robot Report podcast.

What happens when the future of surgery starts to feel less like science fiction and more like a cockpit? In this Inside Job reaction episode, Matt and Aaron revisit their profile of Dr. Alex Fondaw, M.D., and bring in Dr. Alisa Coker, M.D., Director of Robotic Surgery and Education at Dell Medical School, to unpack how robotic surgery is changing the operating room. Dr. Coker explains why surgical robots are not exactly new, how the da Vinci system has evolved over more than two decades, and what it actually feels like for a surgeon to sit at the console and operate with robotic arms. She also gets into the tension between open, laparoscopic, and robotic surgery, the training challenges facing the next generation of surgeons, and why the human surgeon is still very much in control. It is a conversation about technology, trust, patient care, and professional identity at a moment when AI and robotics are reshaping the way people work. For surgeons, the question may not be whether the machines are coming. They are already here. The bigger question is how humans learn to work with them.

As Chief Investment Strategist at Hightower Advisors, Stephanie Link spends her days researching the market's winners. Today, she's breaking it all down for us. Stephanie explains why the economy keeps defying the doom-and-gloom headlines, which stocks she thinks will champion the next decade, and why she believes we're only in the third inning of the AI revolution. Nicole and Stephanie get tactical fast: the difference between the Mag 7 and the "S&P 493," which stocks are not worth the hype, and Stephanie's thesis that cybersecurity stocks will end up being bigger than AI. She names her favorite tickers across cybersecurity, data centers, robotics, and quantum computing, and explains whether it's too late to buy NVIDIA. Stephanie also gets real about the so-called "AI bubble," the circular spending debate freaking out investors, and why she bought SpaceX but capped it at just 2% of her portfolio. Plus: what her 19-year-old daughter is investing in, why "FAANG" just got a 2026 update, and the one boring, unsexy ETF Stephanie says every new investor should consider. Check out Nicole's financial literacy course ⁠The Money School⁠ Find a Financial Advisor or Financial Coach from Nicole's company ⁠Private Wealth Collective⁠ Watch video clips from the pod on ⁠Money Rehab's Instagram⁠ and ⁠Nicole Lapin's Instagram⁠  Read more about Stephanie's work Here's what Nicole covers with Stephanie: 00:00 Are You Ready for Some Money Rehab? 01:12 Stephanie Link Joins Money Rehab 01:41 Why the Market Keeps Defying the Doom and Gloom 04:07 CapEx, Decoded 06:58 The K-Shaped Economy: Why the Vibes Don't Match the Numbers 09:49 Inflation, Oil Prices, and the War's Ripple Effect 11:41 Mag 7 vs. the S&P 493: Which ETF Should You Buy? 16:16 Why Stephanie Says No to Leveraged ETFs 17:32 Cybersecurity Will Be Bigger Than AI 20:16 The Best Cybersecurity Stocks on Stephanie's List 23:15 How to Vet a CEO Before You Buy the Stock 25:16 Investing Lessons from Stephanie's 19-Year-Old Daughter 28:23 What Stephanie Won't Buy: Crypto, Staples, and Energy 32:20 Inside the AI "Food Chain": Powering the Data Center Boom 35:43 Robotics and Quantum Computing: The Next Big Themes 41:00 MicroStrategy vs. Palo Alto: What "On Sale" Really Means 43:07 FAANG Is Dead, Long Live MANGOES 43:47 Why Stephanie Bought SpaceX (and Kept It to 2%) 55:40 Is It Too Late to Buy NVIDIA? 59:06 Grading the Innings: AI, Cybersecurity, and Robotics 59:50 Hot Stocks: Micron, SanDisk, and the Chips Everyone's Chasing 1:02:27 Is There an AI Bubble? 1:03:16 The Circular Spending Debate 1:08:08 Stephanie Link's Tip You Can Take Straight to the Bank All investing involves the risk of loss, including loss of principal. This podcast is for informational purposes only and does not constitute financial, investment, or legal advice. Always do your own research and consult a licensed financial advisor before making any financial decisions or investments. Disclosures: Hightower invests in companies including Boeing Company, Dover Corp, General Electric, Quanta Services, Rockwell Automation, Union Pacific Corporation, Broadcom, International Business Machines Corporation, Marvell Technology Inc, ServiceNow Inc, Palo Alto Networks Inc, Snowflake Inc, Synopsys Inc, Bank of America, Capital One, Coinbase, Morgan Stanley, Truist Financial Corp, Amazon.com Inc, Meta, SpaceX, Alcoa Corp, Antofagasta PLC, Natera Inc, UnitedHealth Group Inc, iShares MSCI Brazil ETF, SLB Limited

In this episode of the Crazy Wisdom Podcast, host Stewart Alsop speaks with Ekaterina Matveeva, founder of Amolingua and Lingo Plus and co-organizer of AI roundtables with EcoCivilization. The conversation covers a wide range of AI-related topics, from the cultural implications of AI development across different countries to the evolution of Ekaterina's perspective on artificial intelligence—moving from initial anger and fear in 2023 to active participation in AI model design by 2025. They discuss the importance of including diverse cultural perspectives in AI training, the cross-country differences in AI regulation (comparing Argentina's human-out-of-the-loop approach to Europe's cautious stance), and the critical question of which sectors need humans in the decision-making loop. The discussion ventures into brain-computer interfaces and Neuralink, examining what happens when technology is removed and questioning whether healthy humans should pursue such enhancements, before touching on the future of personal AI models and the preservation of indigenous wisdom in AI training. You can connect with Ekaterina on LinkedIn to follow her work on AI, education, and cross-cultural technology development.Timestamps00:00 Stewart welcomes Ekaterina Matveeva, founder of Amolingua and Lingo Plus, discussing her work organizing AI roundtables with EcoCivilization05:00 Ekaterina shares her evolution from anger toward AI in 2023 to realizing collaboration potential, emphasizing importance of multicultural perspectives in AI development and training10:00 Discussion of Argentina's new AI liability bill creating human-out-of-loop systems, contrasting with Europe's human-in-loop requirements and different global regulatory approaches15:00 Safety considerations in robotics and AI development, comparing industrial automation standards across regions and discussing Pentagon's use of Anthropic with Palantir systems20:00 Exploring brain-computer interfaces and Neuralink developments, questioning enhancement versus necessity and examining motivations behind cognitive augmentation technologies25:00 Debating human capabilities beyond cognitive function, discussing nervous system, emotions, psychosomatics, and whether brain generates or receives thoughts30:00 Examining societal divisions from enhancement technologies, referencing Avatar and Years and Years series, questioning benefits for healthy individuals versus disability applications35:00 Discussing technological gaps between enhanced and standard populations, concerns about corporate seduction into cybernetic modifications, and lack of long-term safety studies40:00 Questioning who frames AI conversations and trains models, exploring dominant worldviews embedded in AI systems and Vatican-Anthropic collaboration implications45:00 Stewart shares mistakes with Facebook biometric data and building personal AI infrastructure, emphasizing importance of controlling your own models and data50:00 Analyzing whether massive data collection actually improves AI training, questioning if user conversations become buried garbage rather than meaningful model improvements55:00 Ekaterina discusses plans for organizing more roundtables, integrating indigenous wisdom into AI training, and connecting on LinkedIn for future collaborationsKey Insights1. Ekaterina Matveeva's perspective on AI has evolved significantly since 2023, moving from initial anger and fear about AI's impact on education and translation work to becoming actively involved in AI development and training. She realized that instead of resisting AI advancement, she could participate in shaping it by contributing her expertise in education, language, and cross-cultural communication. This shift represents a broader realization that diverse participation in AI development is crucial for creating more versatile and culturally sensitive models rather than allowing a monopoly of values from any single country or culture.2. The conversation highlights fundamental differences in AI regulation across regions, with Europe implementing human-in-the-loop requirements and comprehensive safety measures, while Argentina is reportedly creating frameworks for human-out-of-the-loop AI systems with limited liability. The United States falls somewhere between these approaches, pursuing rapid advancement with fewer regulatory constraints. These divergent approaches reflect different cultural values and priorities, raising important questions about whether AI development should prioritize efficiency and profit or human oversight and control in critical infrastructure sectors.3. Both speakers express concern about brain-computer interfaces like Neuralink, questioning the motivation behind enhancing cognitive abilities when humans already possess multiple forms of intelligence including emotional, cultural, and bodily intelligence. The primary applications demonstrated so far appear limited to video games and potentially military applications, raising questions about whether the technology serves genuine human needs or merely represents an attempt to compete with robots. The speakers emphasize that humans are already complete beings with sophisticated nervous systems, senses, and capabilities that extend far beyond cognitive processing.4. A critical insight emerges around the question of what happens when advanced technologies are removed or turned off. This applies both to brain-computer interfaces and to broader civilization infrastructure dependencies. The speaker shares experiences from 2020 of attempting to live independently in rural California, discovering the challenges of isolation and self-sufficiency. This relates directly to concerns about creating dependencies on technologies where the software ownership and control remain unclear, particularly when those technologies become integrated into human bodies or essential services.5. The discussion reveals concerns about increasing societal division based on access to enhancement technologies. Beyond existing financial inequalities, the speakers worry about a future where people who can afford biological and technological enhancements will advance rapidly while others are left behind. This isn't about disadvantaging certain groups but rather about creating an unbridgeable gap between enhanced early adopters and what they call standard populations who may be intelligent people maintaining traditional ways of life but lacking access to expensive enhancement technologies.6. Matveeva emphasizes the importance of including diverse cultural perspectives, particularly indigenous wisdom and Buddhist traditions, in AI training data. She argues that current AI models reflect the worldviews, values, and hierarchies of their predominantly Western designers, which influences the guidance these models provide to users worldwide. By incorporating wisdom traditions from various cultures, AI models could potentially become wiser and more culturally adaptive, serving diverse populations more effectively rather than imposing a single cultural framework globally.7. The speakers discuss the problematic nature of data collection and ownership in AI training, noting that massive amounts of user data may not actually be improving AI models as expected. There are indications that some AI companies are now specifically requesting users to help train models on their prompts, suggesting that simply collecting billions of conversations hasn't been as useful as anticipated. This raises questions about whether the data users have given away over the past several years has actually made significant impact or is simply buried in chunks of information that aren't effectively connected to meaningful improvements in AI performance.

S5:E252 David reviews The Week In Venture which once again was dominated by AI. On Wednesday July 8th  David will be co-hosting a special episode on the future of Physical AI with Drobot CEO Stijn Van de Velde. Stijn was on the Pod for E178 for Part 2 of the series on Robotics in the summer of 2025. In this episode they'll be talking about the current market for Robotics, Software for bots, Robots-as-a-Service, virtual training with digital twin technology, the value prop for Humanoids and what the future is likely to look like for physical AI. (recorded 7.5.26)Follow David on X at https://x.com/DGRollingSouthConnect On LinkedIn with David at https://www.linkedin.com/in/davidgrisell/Follow Paul on X at https://x.com/PalmettoAngelConnect On LinkedIn with Paul athttps://www.linkedin.com/in/paulclarkprivateequity/We invite your feedback and suggestions at www.ventureinthesouth.com or email david@ventureinthesouth.com.

Dr. Camran Nezhat is world renowned for his work in minimally invasive and robotic surgery, and he specializes in treating people with endometriosis. He is best known for inventing video-assisted endoscopy and was the first to perform groundbreaking advances in minimally invasive surgical procedures that have helped millions of patients around the globe.   Without his contributions, we would not have video-assisted laparoscopy as we know it. To say that it's an honor to have him on the show is an understatement! In 2020, Dr. Nezhat was chosen by the AMA as an exceptional innovator whose contributions have revolutionized modern day surgery, and for that work he received the Distinguished Service Award for meritorious service in the science and art of medicine.   He's joining me today to talk about endometriosis, its treatment, and the role that minimally invasive surgery plays in assisting patients. Read the full show notes on Dr. Aimee's website You can find Dr. Camran Nezhat's website here. Do you have questions about IVF? Click here to join Dr. Aimee for The IVF Class. The next live class call is on Monday, July 13, 2026 at 4pm PST, where Dr. Aimee will explain IVF and there will be time to ask her your questions live on Zoom.   Click to find The Egg Whisperer Show podcast on your favorite podcasting app.   Watch videos of Dr. Aimee answer Ask the Egg Whisperer Questions on YouTube.   Sign up for The Egg Whisperer newsletter to get updates   Dr. Aimee Eyvazzadeh is one of America's most well known fertility doctors. Her success rate at baby-making is what gives future parents hope when all hope is lost. She pioneered the TUSHY Method and BALLS Method to decrease your time to pregnancy. Learn more about the TUSHY Method and find a wealth of fertility resources at www.draimee.org.

I interviewed Amanda Watson about GPL Compatibility in VR with CitraVR, Oculus Link & Air Link, & VR as the Perfect 3DUI for Robotic Teleoperation on Thursday, June 18, 2026 at Augmented World Expo in Long Beach, California. This is a listener-supported podcast through the Voices of VR Patreon. Music: Fatality

Gordon Chang challenges China's manufacturing numbers as "science fiction" and the promotion of robotics to hide massive youth unemployment. Chang explains that Xi Jinping prioritizes manufacturing over consumption to keep the population dependent. He also warns of rising state-promoted xenophobia against foreigners. 161890 Shanghai

There's a longstanding nursing shortage in the U.S. that's only expected to get worse. Some hospitals have experimented with robotic nursing assistants to lighten the load, like a model known as Moxi that rolled out in 2019. They're kind of R2D2-meets-Rosie-the-robots and were designed to handle non-patient-facing tasks like transporting lab samples. But many hospitals eventually pulled them from operation. Marketplace's Meghan McCarty Carino spoke with Varsha Bansal, reporter at Proof News, about what went wrong.

There's a longstanding nursing shortage in the U.S. that's only expected to get worse. Some hospitals have experimented with robotic nursing assistants to lighten the load, like a model known as Moxi that rolled out in 2019. They're kind of R2D2-meets-Rosie-the-robots and were designed to handle non-patient-facing tasks like transporting lab samples. But many hospitals eventually pulled them from operation. Marketplace's Meghan McCarty Carino spoke with Varsha Bansal, reporter at Proof News, about what went wrong.

The Evolving Robotic Battlefield in Ukraine. Guest: Colonel Jeff McCausland. McCausland explains how drones have transformed the war in Ukraine, effectively cutting off Russian supply lines to Crimea. He discusses the massive casualty rates caused by drones and Ukraine's plan to deploy thousands of ground robots. Meanwhile, Russia faces severe manpower shortages and high casualty counts. 21855 CRIMEA TATARS