The episode opens with an OneTrust ad read and a roll call of Hard Fork Live 2026 sponsors (IBM, Everpure, Pureleaf, University of Notre Dame, Atlassian). Kevin Roose and Casey Newton then take the stage to a live audience and deliver a warm welcome that immediately establishes the tone: self-deprecating, fast, and acutely aware of their own oddity. Newton cracks a joke about the security queue rivaling his attempt to use Claude to make a bioweapon; Roose reminds the crowd there's no 1.5x speed button tonight. The conflict-of-interest disclosures — Roose noting the Times is suing OpenAI, Microsoft, and Perplexity; Newton that his fiancée works at Anthropic — are delivered with custom disclosure hats, a bit of theatre that underscores how live performance lets the show show off its self-awareness.

With the audience warmed up, Kevin and Casey run through the show's lineup: first, Microsoft CEO Satya Nadella; then Phil Moen, who trained and walked the Beeple robot dogs (robot dogs bearing the faces of Elon Musk and Mark Zuckerberg) featured onstage; and finally Cindy Cohn, the outgoing executive director of the Electronic Frontier Foundation. Before diving in, the hosts compress a week's worth of tech news into quick satirical takes: Apple fell further behind in AI at WWDC; Anthropic launched Claude Fable, already being used in government; and 'hot IPO summer' is here with Anthropic, SpaceX, and OpenAI all lining up. The jokes land because the stories underneath them are real and significant — a perfect live show set-up.

The first major conversation of the night opens with Kevin confessing he hasn't been a regular Microsoft AI user since Sydney tried to break up his marriage in 2023 — a reference Nadella meets with 'think about a world without guardrails.' From there, Nadella lays out his core thesis: the danger of the current moment is not that the frontier isn't advancing, but that the benefits are concentrating in a handful of models and firms while the broader economy grows at 2%. He describes Microsoft's Build conference as being entirely focused on the platform question: can every enterprise in every country operate at the frontier? He introduces Project Solara — Microsoft's 'agent-first' hardware vision — describing a badge-like device for a nurse walking hospital floors that scans, inputs, and converts speech to prompts. The phone won't disappear, but ambient, always-on agent intelligence needs new form factors. ^{[1] — Satya Nadella "The real test of AI isn't whether one model beats another — it's whether the gains diffuse across the whole economy. Nadella argues Microso…" 06:40 Play}

Casey Newton pivots to a topic that landed in the news the same day as the live show: an Xbox leadership memo announcing a hard reset for the console business. Nadella celebrates Xbox's 25-year history — noting that gaming at Microsoft actually predates Windows and Office, going back to Flight Simulator — but doesn't sugarcoat the economics. Semiconductor shortages and memory price inflation are hitting all consumer electronics, and Xbox is no exception. More painfully, he admits there is more monetization of Xbox games happening on YouTube than at Microsoft itself, a striking self-indictment of a business model that has been subsidizing entertainment without capturing the value. New Xbox head Asha, just 100 days into the job, has been given another 100 days to rethink both hardware and publishing. Nadella's tone is candid but measured — this is a platform problem, not a concession. ^{[1] — Satya Nadella "More Xbox monetization on YouTube than Microsoft: Nadella acknowledged there is more monetization of Xbox games happening on YouTube than a…" 11:40 Play}

Kevin Roose produces a piece of rare merchandise: a Microsoft Advanced AI Research sweatshirt from the November 2023 weekend when Sam Altman was fired and Microsoft scrambled to build an entirely new AI division to catch the expected mass exodus of OpenAI employees. The division never needed to exist — Sam came back. But the sweatshirt did. Roose asks Nadella to play a counterfactual: in the world where all those OpenAI employees ended up at Microsoft, is Microsoft better or worse off today? Nadella deflects gracefully, saying he's thrilled Greg Brockman and Sam Altman are back at OpenAI and that the original bet on a nonprofit research lab whose thesis was 'intelligence is log of compute' was considered nuts at the time by many inside Microsoft. His most memorable answer: what he really remembers from that chaotic weekend is India's cricket team being thrashed by Australia.

With the OpenAI S-1 looming, Kevin asks what Microsoft actually received in the renegotiated partnership deal. Nadella is unusually specific: Microsoft remains on OpenAI's cap table, is its largest cloud customer, retains IP licensing rights through 2032, and — crucially — has the flexibility to develop and deploy its own models. That last right is no longer theoretical: last week Microsoft launched its MAI models, hill-climbed via reinforcement learning from the ground up, with a published research paper. Azure is the infrastructure substrate that made both Microsoft and OpenAI what they are today, and Nadella is careful to credit the partnership for that. But MAI is the clearest signal yet that Microsoft is no longer content to rent frontier AI from a partner. ^{[1] — Satya Nadella "MAI models published with full paper: Microsoft last week launched MAI models hill-climbed from the ground up and published the research pa…" 16:15 Play}

Kevin asks Nadella the uncomfortable question: AI has a popularity problem. Graduation speakers across the country are getting booed. AI is polling terribly. Communities are furious about water and energy consumption from data centers. Nadella doesn't deny the perception — he says it 'is terrible' and must be earned back. His most compelling counterpoint is Quincy, Washington, where Microsoft has run data centers for more than 20 years. The numbers, as he tells them, are striking: local taxes went down while the tax base went up, and employment in the town has grown. Quincy has become a 'data center town' that celebrated its 20-year anniversary with Microsoft at a company cookout. That kind of longitudinal evidence, Nadella argues, is what the industry must produce — not promises. On the jobs side, he acknowledges real displacement is coming but says new work — 'metacognition, meta-work' — will emerge to fill it.

Casey Newton presses Nadella on the labor disruption question he keeps hearing from economists and tech leaders: some are hiring fewer engineers because of AI; others can't hire enough. Nadella's response is the 3.5 billion typists analogy — in the early 1980s, nobody would have predicted that billions of people would become information workers, but here we are, all 'typing' away at knowledge work. He then introduces the concept of 'cognitive coverage,' his term for the new software developer's core skill: managing and auditing repositories of agent-generated code, just as test coverage once meant validating human-written code. The workflow has changed, the artifact has changed, the work itself has changed — but there is still a job. He also mentions a blog post by a colleague named Sarah Go on the 'untrainable' parts of human capability, signaling his own awareness that the real limit on AI isn't technical.

This is the sharpest exchange of the Satya segment. Casey cites Nadella's own February 2025 comment to Dwarkesh Patel: his benchmark for AGI is 10% GDP growth. We are nowhere near that. Why? Nadella explains with unusual candor: achieving 10% growth requires a perfect match between the marginal cost of the token and the marginal value it creates — a management discipline almost nobody is enforcing right now. Instead, everyone is 'token maxing': pumping the most powerful frontier models at every problem regardless of whether the economics justify it. He admits he does it himself. It's addictive. Microsoft's response inside Copilot is an 'auto mode' with an economic model underneath: don't use frontier models for non-frontier problems. The insight crystallizes: productivity growth from AI isn't automatic — it requires the same rational allocation of capital that any technology requires. ^{[1] — Satya Nadella "Everyone loves pumping tokens into AI — including Nadella himself. But he's now enforcing an 'economic model' inside Copilot: don't use fro…" 27:10 Play}

Kevin asks what Satya thinks the San Francisco AI vanguard gets wrong about political economy. Nadella doesn't say they're wrong — he says they need historical context. He cites a book (Parallel Paths to Prosperity by Joel Mokyr and co-authors) that traces 1,000 years of Western economic ascent to a virtuous cycle of three things: technological revolutions, markets, and democracy, each acting as a check on the others. There is no 'economy,' he says — only a political economy. The AI age needs to regenerate that same balance. Casey's final question gets to the heart of it: how AGI-pilled is Nadella really? His answer is honest and specific. He believes AI loops can close in coding and AI research. He does not believe the 'messy real world' of human knowledge work — the unverifiable things people observe and decide in meetings — can be captured and RL-trained away. He closes with humility: this technology is 'in the pantheon' of great step-changes, but it's not the last one.

The segment opens with delightful chaos: the robot dogs wander back onto the stage just as Phil Moen is being introduced, prompting Casey to observe they're 'very lifelike in that they don't seem to respond well to instructions.' Once settled, Kevin describes the dogs for audio listeners — Unitree Go2 quadrupeds fitted with 3D-printed faces of Mark Zuckerberg and Elon Musk. Phil Moen introduces himself as executive director of Node, a digital art center in Palo Alto dedicated to giving digital artists a permanent institutional home. The segment immediately establishes a tonal contrast to the Nadella interview: lighter, more playful, but with an underlying seriousness about what art built on technology can say that plain commentary cannot.

Phil Moen explains that Beeple — digital artist Mike Winkelmann, who has created a new piece of art every day for 20 years — conceived Regular Animals as a meditation on who shapes the world we see. The six dogs break into two groups: tech and media CEOs (Musk, Zuckerberg, Bezos) and artists (Warhol, Picasso, Beeple himself). That split is not a coincidence; it's the thesis. Half of the regular animals are the people who now control how most of the world experiences reality. Each dog constantly photographs its environment and 'poops' out those images, filtered through the aesthetic lens of its face: the Picasso dog outputs cubist imagery; the Zuckerberg dog renders the world as if it's already inside the metaverse. Kevin asks about 'poop mode,' and Moen confirms: the dogs poop their photos, a Node staff member is certified to collect the poop and give it to guests, and yes, 'robot dog walker' is a real job opening. ^{[1] — Phil Moen "Half the faces on Beeple's robot dogs are tech and media CEOs — Musk, Zuckerberg, Bezos. The other half are artists — Warhol, Picasso, Beep…" 37:50 Play}

The segment's most philosophically interesting moment arrives when Casey reads the Node website: after three years — 21 dog years — each robot will die, with all memories from its life preserved forever on-chain. Kevin's reaction: 'What?' Moen explains that digital art institutions face a fundamental challenge traditional museums don't: their medium is software and hardware that gets discarded. Artists have repeatedly been told by institutions 'we don't understand it' or 'we have an opening in five years.' Node's mission is to be the institution that does understand, and that commits to preserving these works across generations. The blockchain preservation of the dogs' memories is part of that commitment — not gimmick, but infrastructure. Software, Moen says, has a long history of amazing projects that are built and then discarded. Art can't be like that.

The episode pauses for a standard sponsor block: OneTrust's AI governance platform; the University of Notre Dame's Delta Network, which brings together faith leaders, tech developers, and students to explore what's distinctly human in the AI age; and KPMG, which has embedded AI across its own enterprise as a 'client zero' blueprint for helping companies scale AI capability. Comcast Business follows with a read on SD-WAN and zero-trust security infrastructure. The Notre Dame and KPMG reads bookend the human dignity and AI deployment themes of the conversations on either side.

Cindy Cohn takes the stage in a 'Let's Sue the Government' T-shirt and immediately wins over the crowd. Casey frames the opening question historically: Cohn's first big fight at EFF was defending Daniel Bernstein's right to publish encrypted source code in the 1990s against government restrictions. That case helped establish that code is speech — and gave us HTTPS, Signal, and encrypted phones. Thirty years later, the fight is still live. In the UK and Australia, legislation threatens encryption at the infrastructure level. Signal has publicly threatened to exit Canada if a proposed bill passes. And in recent Meta litigation, the government used the existence of encryption as an argument that Meta's product was defective. Cohn describes her conclusion with weary clarity: encryption, like free speech, is a right that must be permanently defended. There is no finish line.

In the 1990s, Kevin observes, the EFF and Google, Facebook, and Amazon marched in lockstep — all fighting for users against government overreach. Casey remembers being carried away by the grandiose promises of those companies. When did the cracks appear? Cohn's answer is structural: it happened when surveillance became the number one business model of the internet. Companies that used to stand with users to fight for copyright balance and open platforms gradually became adversarial to those same users as surveillance monetization took over. The EFF's test is simple: if you stand with your users, the EFF stands with you; if you stand against them, the EFF will be first in line. That second clause, Cohn says, has become bigger than it should be. She gives Section 230 as an example: tech companies no longer defend even their own legal rights if doing so might complicate their relationship with the administration. ^{[1] — Cindy Cohn "The EFF and Silicon Valley used to be allies — both fighting for users against government overreach. Then surveillance became the dominant …" 48:15 Play}

Casey poses the question that haunts every conversation about big tech in 2026: is the industry's rightward shift since Trump's reelection genuinely ideological, or is it purely transactional — more money to be made on the right? Cohn is honest: she doesn't know, and she doesn't think the executives are being honest with themselves either. She draws a sharp distinction between the billionaire class at the top of these companies, who are now disconnected from broader reality in ways that remind her of billionaires in any industry, and the broader tech workforce of 30,000 EFF members and their colleagues who still want to build cool things that connect the world. The AI founders, she notes, have made real public commitments — Anthropic's 80% philanthropic pledge is her example — but time will tell whether they walk their talk.

Kevin asks Cohn to name the AI privacy risk that worries her most. Her answer is immediate: mass domestic surveillance, supercharged by AI. She notes it's not a coincidence that the two hard lines Anthropic drew with the Defense Department were mass domestic spying and autonomous weapons — she's spent her career fighting the first one and says they're right. AI makes surveillance not just larger but categorically more powerful: it can ingest the traces of everyone's daily life at scale, making ordinary information — a Facebook message about reproductive care, a protest attendance record — suddenly dangerous. She walks through the post-Dobbs Nebraska case where a mother was jailed via Facebook messages, the killing of protesters in Minnesota who were merely monitoring police, and the expanding circles of immigration enforcement. The argument isn't abstract: the circles are closing, and privacy nihilism is a luxury nobody can afford. ^{[1] — Cindy Cohn "Mass surveillance already existed. AI makes it infinitely more powerful — and the targets are expanding. After Dobbs, reproductive care bec…" 55:08 Play}

Kevin names the phenomenon Cohn keeps encountering: privacy nihilism. People feel the battle is lost, the data is out, so why fight? Cohn's demolition is sharp: old data has a short shelf life because people keep living their lives and generating new data, and governments haven't unplugged the surveillance machine because they've got everything they need. If it were game over, they'd stop. She introduces the unusual bipartisan coalition forming around FISA 702 renewal — Ron Wyden and Jamie Raskin on the left alongside Andy Biggs, Rand Paul, and Mike Lee on the right, all demanding the FBI obtain a warrant before searching mass surveillance databases. Even Freedom Caucus members, she argues, understand that the surveillance state they empower today is the one a future administration they oppose will inherit. Casey raises Sam Altman's proposal for chatbot conversation privilege; Cohn prefers anonymity by design over relying on companies to protect data they still hold.

The final substantive topic of the night is the fight Kevin and Casey have been having with each other for years: whether to stay on X. Cohn ends it, at least for the EFF. She describes the decision as long overdue: the EFF's reach was shrinking, staff were being bombarded with abuse after any post about LGBTQ rights, and the platform's fundamental dynamic had turned toxic. Her principled argument silences the hypocrisy charge: free speech has always included the right to choose your audience and your venue. The idea that a civil liberties organization must post on a private platform or be called a free speech hypocrite misunderstands what free speech means. The 'Nazi bar' line — 'I'm sorry if you're hanging out in the Nazi bar and I decide that's not where I want to speak' — gets the biggest laugh of her segment. She closes with genuine sadness: there are people on that platform who care about rights, including politicians working on 702, and she wishes they weren't all stranded there.

Kevin calls Cohn 'a legend' and the crowd applauds. The show closes with a full production credit sequence: Rachel Cohn and Whitney Jones producing, Veron Pavich editing, Caitlin Love on fact-check, and a long list of live event team members from the New York Times who made Hard Fork Live 2026 happen at the Yerba Buena Center for the Arts and the Blue Shield of California Theater. Special thanks are extended to the event leadership including Hillary Kuhn and Beth Weinstein. A final Comcast Business ad closes the episode, reprising the SD-WAN and zero-trust security messaging before Kevin delivers the last line: 'No one does business like Comcast Business.'