PodcastsHear the voice. See the shape of the thought.

Tech analyst Benedict Evans joined a16z's Erik Torenberg to take stock of a year and a half of AI development — what has actually settled and what remains wide open. Evans argues that agentic coding has emerged as AI's only genuine breakout use case so far, with everything else still in the "useful around the edges" category. The central structural question he returns to throughout: whether foundation model companies end up as commodity infrastructure, like ISPs and mobile operators, or manage to capture value up the stack the way operating systems did. ## [00:00] Intro This opening segment is a teaser pulled from later in the conversation. Evans previews the mobile-operator analogy he develops at length: carriers built expensive global infrastructure, traffic grew 2,000x, and all the value moved up stack to companies that ran on top of them — a pattern he believes applies directly to LLMs. He also flags the one concrete data point that anchors the whole discussion: Anthropic's run rate rising from roughly $9 billion to $47 billion in a year, almost entirely from software development. > *"They built this amazing piece of incredibly sophisticated very expensive global infrastructure with enormous growth in use all the time and it changed all of our lives and we all pay for it and they didn't make any money from it because all the value moved up stack."* ## [01:05] AI Adoption Accelerates Evans reflects on what has changed since the first version of his "AI Eats the World" presentation. The clearest shift: competitive strategy among labs has moved beyond "build a bigger model faster" — OpenAI pivoted through several strategic positions while Anthropic focused on coding and got it to work. That focus is now contagious across the industry. The questions Evans expected to resolve by now — whether one model will dominate, whether models can capture value up the stack, whether consumers will use AI daily rather than weekly — remain largely open. On why coding emerged first, Evans is unsurprised in retrospect: software developers were the early adopters, so the first things they tried to automate were the tasks they did themselves. He draws an analogy to PCs in the early 1980s: incredibly exciting, but not yet clear what they were for, and the first application was making more computers. What has genuinely shifted this year is that agentic coding crossed a threshold — from "kind of useful" to "really changing everything." > *"It's like the internet in '97 but it's also like PCs in the early '80s. It's incredibly exciting but it's not quite clear what it's for and it doesn't quite work yet."* ## [06:00] OpenAI Strategy And Usage Gap Evans characterizes OpenAI's late-2025 phase as an attempt to build value in every direction at once — ads, e-commerce, shopping carts, payments, a browser, a social video app — before pivoting sharply back to coding once Anthropic's results made clear that was what actually worked. Whether Anthropic's coding bet was deliberate or accidental is beside the point; it worked, and OpenAI followed. The deeper problem Evans raises: even with runaway coding adoption, daily active users across AI tools still sit around 10% of total users, with another 30–40% using AI only weekly. The gap between people running Claude Code all day and people who used it "last week for something" is not closing yet. He distinguishes between consumer-facing products, where that gap persists, and specific back-office enterprise automations — like a commodities company using LLMs to forecast cash flow from small producers — where the benefit is precise and measurable without asking users to figure out the tool themselves. > *"If you're only using this once a week, then you haven't achieved nana yet."* ## [09:27] Platform Shifts And Value Capture Evans lays out three threads for reading the current moment against prior platform shifts. First: adoption always builds on prior infrastructure — mobile didn't need to wait for the internet to exist, the internet didn't need to wait for PCs — so accelerating adoption curves are expected, not surprising. Second: early stages of any shift feature nothing that actually works reliably; installing a sound card on a 1980s PC took a weekend, and getting internet access meant a floppy with TCP/IP. We're at that stage with AI. Third: the pricing crunch between supply and demand mirrors mobile data in 2009–2010, when carriers had flat-rate plans and suddenly everyone was streaming YouTube, blowing up their unit economics before capped bundles stabilized things. The central structural argument: value didn't land with chip companies, ISPs, or mobile operators. Windows and iOS captured it — but they had network effects and platform leverage that LLMs don't obviously possess. Foundation models look more like hyperscalers than operating systems: enterprises don't "standardize on Claude" any more than they ever knew which cloud their SaaS apps ran on. Evans is willing to be wrong, but insists the current pricing disequilibrium is transitory, and first-year economics suggests commodity pricing as the equilibrium toward which multiple well-funded competitors converge. > *"Chip companies didn't capture the value. ISPs didn't capture the value. Mobile network operators didn't capture the value. Windows and iOS did, but they were doing something else — they had all these levers to go up the stack."* ## [30:43] Automation And Jevons Evans presents a framework from his presentation for thinking about what automation actually does to an industry: pure price elasticity (do the same thing cheaper), doing more for the same money, unlocking things that were prohibitively expensive as barriers to entry, and enabling things that were completely impossible before — the steam-engine-and-trains example, or Spotify making all recorded music available for $15 a month. He's careful not to over-predict: the same observation that "the internet will destroy physical distribution" turned out to mean completely different things for newspapers (destroyed) versus movie studios (barely affected). The questions that matter most — what AI means for finance, for consulting, for the big four, for big law — are now at least as much industry questions as technology questions, and require domain knowledge that tech analysts in San Francisco typically don't have. > *"What does generative video mean for Hollywood? Ben Affleck probably knows a lot more about this than I do."* ## [33:27] Ads And Shopping Agents Evans focuses on advertising and retail as the sector where AI's ability to semantically understand products creates a specific, tractable shift. Current ad platforms know metadata and purchase correlations but don't actually understand what products are or why people buy them — hence Amazon recommending a second toilet seat cover. LLMs understand semantic category, substitutes, and use context, which is why Google and Meta's ad revenue is already accelerating as they wire LLM inference into recommendation and prediction systems. He sketches a progression: from "here's a product image, where can I buy it" (works now), to "suggest 10 alternatives with pros and cons" (works now), to "look at my Instagram and suggest a winter coat that changes my look but not too much" — which was science fiction three years ago and is now plausibly buildable. The broader point is that the important gains from new technologies come not from doing the old thing better, but from doing things that were previously impossible — and those new things tend to be problems nobody even knew existed until someone built a solution. > *"The important stuff is not doing the old thing but more — it's doing something new that you couldn't have done with the old thing."* ## [39:41] Enterprise Stack Rewired Evans maps the enterprise software landscape: big horizontal systems (SAP, Workday, CRM), vertical SaaS, thousands of internally built point solutions, and the perpetual fuzzy middle of Excel and shared drives. AI arrives as another set of options rather than a clean replacement for any existing layer. The key tension: does the LLM sit at the bottom of the stack as a feature inside Salesforce, or at the top, synthesizing across all systems to answer questions no single system could? His answer: probably both, depending on the task. What he's more confident about is that software will proliferate, not consolidate. Cheaper and faster to build means more competition, much as SaaS itself produced an order of magnitude more software than packaged enterprise apps did. On the SaaS apocalypse question investors are asking: some companies will get wiped out, but no one knows which ones yet, so derating the whole sector 50% doesn't make sense. He draws the sharpest line between automating tasks and automating jobs. What accountants do in 2026 is almost entirely different from what they did in 1976, but the output the client buys is recognizably similar. LLMs will excel at tasks where the right answer is what any trained person would produce; they'll struggle where the value is a non-obvious answer, an exception, or an insight nobody ever wrote down. > *"LLMs are going to be very good at anything where you can describe how people do it and where what you want is the way anybody would do that — and not so good at where you can't really explain why you did it like that."* ## [49:57] Capex Commodities And Magic The four largest tech companies are on track to spend over 50% of revenue on capex — twice the capital intensity of telecoms, comparable to oil and gas. Evans notes $700 billion a year is not an impossible figure as a share of what global infrastructure costs, but there are clear financial gravity limits: these companies cannot sustain $1.5 trillion next year, and at some point the growth curve has to taper. The complicating factor is that efficiency is improving fast enough that the amount of hardware needed per unit of useful output is a moving target. On the commoditization thesis, Evans frames it as a challenge rather than a prediction: here is a chain of argument that deterministically suggests foundation models become commodities — explain to me why it's wrong. The mobile analogy holds: mobile operators are a large industry that spends enormous sums on infrastructure and isn't very profitable, while Google, Meta, and Apple collectively generate more net income than the entire global telecom industry. His closing note is a deliberate stepping back. Every major technology wave — PCs, the internet, mobile, cloud — seemed uniquely transformative from the inside, and each one produced things we celebrate and things we regret. AI is different and transformative. So was each prior wave. The base case is that we go through it again, and in 20 years forget there was ever a world where computers couldn't do this. > *"It's going to be magic and in 20 years time we'll just say, well, of course that's how it is. Computers have always done that."* ## Entities - **Benedict Evans** (Person): Independent tech analyst, author of "AI Eats the World" presentation, former a16z partner - **Erik Torenberg** (Person): Host, a16z podcast, consumer and content focus at Andreessen Horowitz - **OpenAI** (Organization): Foundation model company; discussed in the context of strategic pivots from broad diversification back to coding focus - **Anthropic** (Organization): Foundation model company; credited with proving agentic coding; run rate cited as growing from ~$9B to $47B in roughly a year - **Foundation models** (Concept): Large language models sold as infrastructure; the central question is whether they commoditize like ISPs and mobile operators or capture value like operating systems - **Jevons paradox** (Concept): When you make something cheaper, demand often rises faster than cost drops — the mechanism Evans uses to frame what automation does to industry economics - **SaaS stack** (Concept): The layered enterprise software landscape (horizontal, vertical, bespoke) into which AI arrives as another set of options rather than a clean replacement - **Mobile data analogy** (Concept): Evans's key historical comparison — mobile operators built trillion-dollar infrastructure, traffic grew 2,000x, pricing destabilized then re-equilibrated, and all valuable applications were built by someone else

David George (a16z general partner) and David Clark (VenCap CIO) argue that AI companies are scaling faster than any prior technology generation — Anthropic and OpenAI are adding more monthly revenue than Meta, Google, or Microsoft — while actual diffusion into the broader economy remains below 5%. They work through what that gap implies for exit sizes, loss ratios, bubble risk, and who ultimately captures value as token costs fall and frontier intelligence becomes a commodity. ## [00:00] Intro Three data points open the episode: Anthropic and OpenAI already adding more revenue per month than any hyperscaler; top-1% exits 10x-ing in 24 months from $10 billion to $32 billion; and David George's assessment that, right now, we are not in a bubble. ## [00:38] The Scale Shift: Anthropic & OpenAI Adding More Revenue Than Hyperscalers David George explains how his priors shifted sharply around November 2025. Before that, enterprise AI looked like a productivity story analogous to cloud adoption. After it, the numbers reframed the ceiling: Anthropic and OpenAI are already adding revenue at hyperscaler rates with less than 5% of the economy actually using these tools. He places an upper-bound frame on the opportunity by noting that Fortune 500 companies generate roughly $2 trillion of profit annually, and the two largest model companies could reach $200 billion revenue run rate by year-end — already equivalent to 10% of that profit pool. > *"If you pair that up with the fact that they're already getting bigger in terms of revenue added than the hyperscalers, and you're at less than 5% diffusion into the economy, I think the outcomes are going to be extraordinary."* ## [04:20] Skeuomorphic vs Native AI Applications in the Enterprise David Clark invokes Chris Dixon's skeuomorphic-to-native arc: the first wave of enterprise AI lets people do existing jobs faster; the native wave restructures the work itself. George adds a wrinkle — the best companies are not yet focused on internal automation. Their top engineers want to build product, not automate back-office workflows. The most cutting-edge firms he visits are still in a "documentation phase," converting institutional knowledge into markdown before they can meaningfully deploy agents against it. > *"The most cutting-edge folks inside those companies who are trying to do this that I've talked to are kind of in the documentation phase — just turn everything into markdown files, have as much context capture as you can possibly get."* ## [06:24] How the Best AI Companies Run Themselves Differently Native AI founders operate on a different metabolism. George contrasts them with the previous SaaS generation, which, in hindsight, ran inefficiently but got away with it because headcount mandates and expanding software budgets covered the slack. The new companies are lean, aggressive, and already running agent swarms rather than typing commands. He describes walking into a cutting-edge AI company and finding researchers whispering into microphones, orchestrating swarms of agents — not a keyboard in sight. > *"The new companies are very lean, very aggressive, and they work all the time."* ## [08:14] Top 1% Exits 10X'd in 24 Months Clark lays out VenCap's tracking data: the threshold for a top-1% exit was $10 billion between 2020-2024, rose to $20 billion by February 2026, and was updated just the day before this recording to $32 billion. With OpenAI and Anthropic IPOs potentially arriving, he sees the bar hitting $100 billion by September. George notes that the combined market cap of these private companies likely already exceeds the entire Russell 2000, and that the sum of all VC-backed IPOs over the past six years is probably smaller than any single one of the three expected large IPOs. > *"Where is the threshold for the top 1%? And if you then think about OpenAI and Anthropic coming in, potentially we could be north of $100 billion by September."* ## [11:17] The Half-Life Problem: Why 40% of AI Leaders Drop Off Every Year Clark surfaces a disturbing churn metric: 40% of companies on the Forbes AI 50 list from one year disappeared the next. Google wasn't the first search engine; Facebook wasn't the first social network. First-mover advantage in AI is eroding faster than in any prior cycle. George confirms a16z's own priors have been repeatedly overturned — first convinced model companies would be everything, then convinced applications would take over, now watching the model companies extend back up into the application layer. The only durable heuristic he offers: a company must be in the token path. > *"From last year to this year, 40% of the companies that were on that list last year dropped off."* ## [13:11] Token Path, Cost Pressure & Who Captures Value Enterprise buyers are already feeling cost pressure from AI spend, and they cannot cover it by cutting previous-generation software budgets fast enough. George frames value capture as hinging on one largely unknowable variable: the market structure of frontier model labs. Two labs at the frontier means higher token prices and faster labor restructuring pressure; five labs means lower prices and a broader application ecosystem. Per-token cost for like-for-like capability is falling more than 10x year-over-year, but total token spending in dollars is rising faster. Clark adds that Chinese LLMs are roughly six months behind US frontier capability but ten times cheaper — a classic innovator's dilemma setup. > *"The biggest driver of where value is going to get captured right now is something that is totally unknowable, which is what is the market structure of the model companies?"* ## [17:00] Loss Ratios, Risk & How We Think About Early Stage Clark notes that historical early-stage VC loss ratios run around 60%, but the AI cohort of the past two years shows single-digit loss rates — unsustainable by definition. George reframes the discussion: a16z does not target a low loss ratio. A VC firm bragging about never losing money is "a horrible data point" — it signals too little risk-taking. The philosophy is to back the market-leading founder in every space with strong tailwinds and a credible technology. If the space works out and you have the leader, excellent. If the space does not work out but you have the leader, that is expected. The failure mode is the space working out while having backed the wrong company. > *"We joke all the time — there's a prominent VC in our ecosystem, and one of his big points of pride is he's never lost money on a deal. And we're like, that's not a point of pride. Like that's a horrible data point."* ## [22:51] Are We in an AI Bubble? Clark points out that classic bubbles are characterized by excess supply destroying economics — but right now the constraint is supply scarcity: no data center capacity available at scale until late 2028 or early 2029, with the US buildout running a year behind schedule and community resistance adding further delay. George is confident there is no bubble today and dismisses the data center opposition directly. The one scenario he would watch for is an unexpected algorithmic breakthrough producing dramatically smaller and more efficient models — which could flip supply from scarce to oversupplied — but he considers that unlikely in the near term. > *"I feel pretty confident saying that we're not in a bubble right now. I'm less confident that we won't be in a bubble three years from now."* ## [27:36] What SpaceX, OpenAI & Anthropic IPOs Mean for Public Markets Clark asks whether public markets can absorb the coming wave of trillion-dollar-plus IPOs. George argues it is unambiguously positive: the number of public companies has halved over 20 years, and outside the data center supply chain, almost nothing in the public markets is growing at more than 30% today. Bringing hypergrowth companies into indexes gives retail investors — including his parents' index-fund retirement accounts — exposure to the most dynamic part of the economy. He expects some portfolio reshuffling to make room, but does not see indigestion risk. > *"If you exclude the data center supply chain stuff right now, there are very few companies that are growing fast that are available for people to buy in the public markets."* ## [29:59] The Future of Venture Capital in an AI World George forecasts the shape of VC over the next five years as primarily a function of token market structure — whether the labs remain concentrated or become commoditized. He cites Bill Gates's platform axiom: a platform's value is validated when the companies built on top of it collectively exceed the platform's own value. If that holds, there will be a massive wave of valuable application companies built on intelligence. He also flags the consumer side as the most underappreciated opportunity: the last decade of consumer internet was a story of time spent getting captured by large incumbents; AI-driven shifts in consumer attention could recreate the conditions for generational consumer companies. > *"I'm very optimistic that we're going to have a massive wave of really valuable companies that get built on top of tokens, AI, and intelligence."* ## Entities - **David George** (Person): General partner at a16z; covers growth-stage and early-stage AI investing; invested in OpenAI pre-ChatGPT - **David Clark** (Person): CIO at VenCap; fund-of-funds investor tracking AI startup performance and VC market dynamics for 34 years - **Anthropic** (Organization): Frontier AI lab; cited as adding more monthly revenue than hyperscalers alongside OpenAI - **OpenAI** (Organization): Frontier AI lab; benchmark for scale and the expected $100B+ IPO cohort - **VenCap** (Organization): Fund-of-funds investor; publishes top-1% exit threshold data and tracks Forbes AI 50 churn - **Andreessen Horowitz / a16z** (Organization): Venture capital firm; investor in OpenAI pre-ChatGPT, scaling platform services to support companies encountering enterprise-scale problems early in their lives - **Cursor** (Software): AI coding tool cited as an example of a company reaching billions in revenue while still very small and early-stage - **Token path** (Concept): a16z's primary heuristic for evaluating AI companies — a company must sit in the flow of AI inference tokens to have durable economic relevance - **Skeuomorphic vs. native AI** (Concept): Chris Dixon's framework distinguishing apps that replicate existing workflows with AI assistance from apps that rearchitect work around AI capabilities natively - **Half-life problem** (Concept): David Clark's term for rapid AI leader turnover — 40% of Forbes AI 50 companies dropped off the list year-over-year — indicating first-mover advantage is eroding faster than in prior technology cycles

Apollo CEO Marc Rowan traces a straight line from Drexel's collapse in 1990 — when he left his office Sunday with belongings in a cardboard box — to Apollo's trillion-dollar position today as the world's largest private retirement income provider and a principal financier of the global industrial renaissance. He and a16z GP David Haber work through why private markets are structurally necessary for diversification now that ten stocks make up nearly half the S&P, how daily mark-to-market pricing will open private credit to five new capital channels, and why Rowan believes AI will replace or enhance every single job — making blue-collar work ascendant and enterprise-software equity a likely disaster for private equity vintages of the past decade. ## [00:00] Intro The opening draws three threads that run through the whole conversation: concentration risk in public equity (ten names approaching 50% of the S&P), the multi-trillion-dollar value locked in private companies like Anthropic and SpaceX that most investors cannot access, and Apollo's operating assumption that AI will replace or enhance every job. Rowan thanks Haber for hosting at Apollo's office before the proper interview begins. > *"10 stocks right now in the US are nearly 50% of the S&P and they're all levered to the same trend... if you're an investor and you're looking for diversification, there's no place to get it other than private markets."* ## [00:52] Drexel, Milken & the Origins of Clean Sheet Thinking Rowan chose Drexel over Goldman because financing entrepreneurs demanded deep business judgment, not technical finance. The high-yield market being invented in real time — PIK bonds, silver-indexed bonds, highly confident letters, bridge financing — forced everyone into clean-sheet problem-solving. Michael Milken's most lasting lesson was connecting dots across geopolitics, technology, and markets into a coherent framework, and his aphorism that "you either accept change or change is visited upon you" became a core Apollo principle. > *"The whole notion of pick I believe was created in one afternoon solving a problem... All of these things were basically problem solution, problem solution. And that mentality of understanding the business, understanding the credit, but also having clean sheet thinking is certainly what powers Apollo today."* ## [04:55] The Apollo Origin Story: From Unemployed to $6 Billion When Drexel failed over a weekend in 1990, Rowan and colleagues were still completing transactions for clients with no firm and no prospect of payment. The formative lesson crystallized immediately: financial firms die of heart attacks (funding risk — borrowing short to lend long, as Bear Stearns and Lehman later confirmed) or cancer (accumulating bad assets instead of taking losses). A cold call from France's Crédit Lyonnais — originally to set up an M&A boutique — turned into an $800 million seed check from the French government, which grew to $6 billion by year-end 1990, making Apollo the bank's largest profit center. > *"I went into my office or I left my office on Friday. I came back in on Sunday and I left with all my belongings in a cardboard box and Drexel was out of business."* ## [08:46] How Apollo Became a Trillion-Dollar Retirement & Credit Firm Apollo today is 80% investment-grade credit and only 20% equity, split between hybrid equity and traditional private equity — the opposite of public perception. Rowan anchors the business around three fundamental goods: providing retirement income to an aging, under-saved population; financing the global industrial renaissance across energy, manufacturing, AI, and defense; and offering genuine diversification as public markets concentrate in a handful of names. The same concentration dynamic in equities is arriving in fixed income, where ten banks are shrinking to five banks plus five tech platforms. > *"Private markets are 80% of the action going on in the world... great companies, Anthropic, OpenAI, SpaceX, Cognition, Cursor — every one of those companies is private, multiple trillion dollars of value and yet most investors have zero exposure to them."* ## [13:00] Permanent Capital, Origination & Why Assets Are the Scarce Resource Unlike traditional asset managers who can deploy any amount of capital into public markets, Apollo is constrained by its ability to originate, not by available capital. That scarcity of assets is the business's true bottleneck — which means every deal should be extracted for maximum value, both by earning fees and by taking principal positions that align Apollo with clients. Rowan argues explicitly against "capital light": in a world where brand, reputation, and the ability to guarantee outcomes matter, a large balance sheet is a competitive weapon, not dead weight. > *"And therefore, I believe that we should be judged by our capacity to create interesting investments. And I believe our capacity to create interesting investments is limited."* ## [16:08] Democratizing Private Markets: Daily Pricing & New Capital Channels The alternative industry was built for one capital source — institutional alternatives buckets — but five new markets now want access: individuals, insurance companies, traditional asset managers, 401(k) plans, and the debt/equity buckets of institutions. None of them want drawdown funds. Apollo is moving to daily estimated value on its investment-grade private suite by June 30, and full daily pricing across all credit products by September, with standardized data warehouses, market-making, and regular price disclosure. Rowan distinguishes private credit as direct lending (the narrow press definition) from the real universe — Intel, Air France, AT&T, Meta — sophisticated borrowers who need complex, non-vanilla long-term financing that banks cannot structure. > *"I've never seen a market in the world where you have transparency and price discovery that is not 10 times its size... It may be uncomfortable for people, but it's coming."* ## [22:04] Where Venture Meets Credit: Financing the Industrial Renaissance Rowan and Haber identify "opportunities living between fields of expertise" as their shared investment philosophy. The intersection they see now: venture-backed companies that historically avoided capital intensity are suddenly building data centers, chips, robotics, manufacturing lines, and defense systems at a scale that cannot be financed with equity alone. Apollo parcels risks — letting venture hold the fundamental business underwrite while infrastructure assets with hard collateral migrate into credit markets at appropriate risk ratings. In Rowan's framing: 2025 proved that data centers, chips, and energy were needed; 2026 is when investors recognize that $800 billion in capex from just four public companies will hit concentration limits, spreads will widen, and tech entrepreneurs will need to partner with financial entrepreneurs. Apollo is committing to a second headquarters in the Bay Area specifically for the growth ecosystem talent pool. > *"the amount of money that's going to be put into data centers, into chips, into robotics, into manufacturing, into defense is, as I suggested, every dollar since the invention of fire, that is not going to be financed with equity."* ## [30:01] AI, Enterprise Software & Why Every Job Will Be Replaced or Enhanced Rowan's operating assumption: every single job will be replaced or enhanced by AI. He is blunt that 30% of private equity AUM from the past decade went into enterprise software, that AI has permanently repriced those assets, and that PE returns from that vintage will be "disastrous" — not because those companies are failing, but because the prices paid assumed a future without AI competitors. His analytical frame: AI changes fastest in domains with a right answer (coding, accounting, trade ops) and slower where judgment is irreducible. Near-term he expects blue-collar ascendancy and white-collar decline — politically uncomfortable for blue cities. As a lender, the lesson from yellow pages, cable TV, and satellite is diversify, stay senior, seek hard collateral, and never underwrite beyond a five-to-seven-year horizon. > *"We operate under the assumption that every job is going to be replaced or enhanced. Every single job. And I think that's what is going to happen."* ## [38:52] Moral Leadership: UPenn, Merit & Doing Right Over Easy After October 7, Rowan wrote directly to Penn's president before a Palestine Rights Conference, identifying not free speech but "favorite speech" — the university funding a conference during Jewish high holidays, run by a known Hamas sympathizer. He framed the broader campus crisis as anti-American and anti-merit. When nearly all donors reduced giving to $1 per year, Penn's administration responded; subsequent congressional testimony led to both the board chair and president resigning. Rowan's broader principle applied internally since taking over in 2021: say the same thing in Texas as in California; on climate, "make it better, not worse" rather than zero-carbon absolutism; on hiring, merit adjusted for distance traveled — measured by individual achievement, not group membership. > *"We hire for merit adjusted for distance traveled. And distance traveled is not about your immutable characteristics. It is about you as an individual — not your class, not your group. Show me the kid who's had to overcome something and still achieved."* ## [46:02] Apollo's Culture: Playing to Win & Building to Outlast the Founder With 6,000 people across asset management and retirement services, Apollo spent six months negotiating — internally, with senior partners — what makes Apollo Apollo. The outcome is a public document on Apollo's careers page, deliberately candid as a candidate filter. The six principles compress to "playing to win," which Rowan distinguishes from fear of losing: senior professionals are expected to be wrong roughly 40% of the time, nobody gets fired for a bad decision (only for not owning and fixing it), and every senior person has a public "wall of shame" loss. Clean-sheet thinking, intellectual insubordination (contrasted with real insubordination), and handling the "moments that matter" in employees' lives are the traits Rowan most wants to survive him as founder. Apollo is building a financial institution, not running a fund — the next five years of product, infrastructure, and market-making innovation will make the firm look more different from today than the last five years already have. > *"You do not get fired here for making a bad decision. You get fired here for not recognizing it or not owning it and not fixing it. We have a wall of shame. Every senior professional here has lost money for the firm."* ## Entities - **Marc Rowan** (Person): Co-founder, CEO, and Chair of Apollo Global Management; former Drexel Burnham Lambert analyst; UPenn alumnus and major donor - **David Haber** (Person): General Partner at Andreessen Horowitz (a16z); host of The a16z Show - **Michael Milken** (Person): Drexel Burnham Lambert financier; longtime mentor to Rowan; credited with inventing PIK bonds, bridge financing, and the high-yield market - **Apollo Global Management** (Organization): $1 trillion+ alternative asset manager, 80% investment-grade credit; co-founder of Athene retirement services; planned Bay Area second headquarters - **Athene** (Organization): Apollo's retirement services subsidiary; provider of insurance and annuity products anchoring Apollo's permanent capital base - **Andreessen Horowitz (a16z)** (Organization): Silicon Valley venture capital firm; exploring capital partnerships with Apollo for capital-intensive tech companies - **Crédit Lyonnais** (Organization): French government bank that seeded Apollo with $800 million in 1990, growing to $6 billion; later sold Apollo to François Pinault - **Private Credit** (Concept): Direct origination of investment-grade debt to corporations and infrastructure projects, bypassing public bond markets; far broader than "direct lending to leveraged buyouts" - **Permanent Capital** (Concept): Long-duration liabilities from insurance and retirement products allowing Apollo to hold assets through cycles without fund redemption pressure - **Industrial Renaissance** (Concept): Rowan's term for the simultaneous global build-out of data centers, AI chips, energy infrastructure, manufacturing, robotics, and defense requiring credit-market scale financing - **Daily Estimated Value** (Concept): Apollo's initiative to price investment-grade private credit products daily — enabling access from wealth managers, 401(k) plans, and traditional asset managers

a16z general partner David Ulevitch sits down with Col. Jeffrey Glover (Arizona Department of Public Safety) and Rahul Sidhu (Flock Safety board member) to walk through how drones, sensors, and AI are quietly rewiring American policing. Sidhu lays out Flock Safety's layered sensor network — license plate readers, gunshot detection, and drone dispatch — while Glover details an Arizona DPS ecosystem built around officer wellness, body-cam analytics, and an international fusion-center play timed to FIFA and the Olympics. The throughline: the next decade of police work will look more like analyst work than door-kicking, and founders who want in need to spend real time on the beat first. ## [00:00] Drones and the Future Beat The episode opens with a stitched-together preview: Sidhu's punchy maxim that cops hate both change and the status quo, Glover sketching how a patrol officer's skill set has to get more investigative and nuanced, and Ulevitch teeing up the central scenario — a 911 call, a drone responding ahead of officers, a fleeing shooter pursued from the sky. The pitch isn't abstract: keeping five helicopters airborne 24/7 to do that job is impossible, but drones make it almost inevitable. > *"You hear a gunshot go off and the drone finds a shooter getting into a car and driving off, and then pursuing the vehicle."* ## [00:32] Founders Building for First Responders Ulevitch asks Sidhu what advice he'd give founders who care more about saving lives than optimizing ad clicks. Sidhu, who sits on Flock Safety's board, points to companies like Skydio and walks through the kind of inbound he gets daily — alerts about kidnapped children recovered, situations de-escalated, technology used to read a scene before officers do. The story he keeps coming back to: a 911 caller reports a man in an alley with a shotgun, a drone arrives first, and the "shotgun" turns out to be a janitor holding a broom. > *"It turned out the drone provided, you know, situational awareness and said, 'Wait, there's just a janitor with a broom.' That's not a guy with a shotgun. And it totally de-escalates the situation."* ## [01:38] Flying Robots Meet Sensor Networks Sidhu reframes drones as flying robots that fit into the same automation wave reshaping every industry. Public safety will get more drones — including more hostile ones to defend against — and Flock Safety's pitch is the layer beneath them: license plate readers, gunshot detection, and drone dispatch tied together so that an Amber Alert vehicle or a shot-spotter ping can dispatch a drone automatically, even pursuing suspects onto highways with state DPS. Ulevitch closes the segment with a joke about it being a bad time to be an enemy of America, then hands off to Glover. > *"And Flock Safety, you know, we — it's not just about drones for us. Like, we have multitudes of sensors in the communities. We have license plate reading cameras. We have, you know, gunshot detection capabilities. All of this is coming together."* ## [03:17] Officer Wellness and Body Cam Analytics Glover details what an integrated Arizona DPS deployment actually looks like. Officers start their shift with a Vitanya "Heal the Heroes" brain scan to check baseline wellness. During the shift, Truleo runs analytics on body-worn-camera audio — not just scoring trooper interactions with the public, but flagging cumulative stress that should put a supervisor on alert before burnout becomes a problem. Ulevitch picks up the thread on how public sentiment around body cams flipped once people saw they protect officers as much as they document them, and draws a parallel to the same hype-cycle pattern with tasers. > *"You can do a scorecard for how the trooper is interacting with the public, but it also gets that information for, hey, do they need additional support?"* ## [05:47] Fusion Centers and Global Intelligence Sharing Ulevitch turns to intelligence-gathering and Glover walks through the Arizona Counterterrorism Information Center (TIC) and the wider US fusion-center network. The near-term push: a TRX program that most agencies are running for FIFA. The longer play: Arizona standing up an international presence with embedded intelligence officers from Mexico, the UAE, Liberia, and other partners, so unclassified threat signals can flow across borders before incidents become local. Ulevitch points to Austin and NYPD counterterrorism as proof the model works. > *"Being able to condense that down and distill it to where we can have good information sharing that's unclassified — be able to share with one another — is going to be huge."* ## [07:37] Advice for Innovators and Closing Thoughts Ulevitch turns the closing question back to Sidhu — a former paramedic and reserve officer — for advice to founders. Sidhu name-checks Ben Curley of Chart Performance (sitting in the audience) as an example of the kind of operator already doing the work, and lands his thesis: the gap looks intimidating but if you can describe an inevitability the way drones now feel inevitable, the field will pull you in. The non-negotiable: spend real time on the beat — ride-alongs, reserve duty — so you actually know what to build. Glover closes by echoing the call to jump in, and predicts the next ten years will fundamentally shift the profession away from kicking in doors toward parsing video, AI signals, and analyst work. > *"If you can picture something that feels like an inevitability, in the same way that, you know, we talk about drones — it'll come because it's the best thing for them. It's the best thing for the communities."* ## Entities - **David Ulevitch** (Person): a16z general partner, host of The a16z Show; long-time enterprise/security investor. - **Col. Jeffrey Glover** (Person): Colonel/Director at the Arizona Department of Public Safety, leading the agency's tech and intelligence modernization. - **Rahul Sidhu** (Person): Flock Safety board member, former paramedic, founder/operator background in public-safety technology. - **Flock Safety** (Organization): Builds a layered public-safety sensor network — license plate readers, gunshot detection, and drone dispatch. - **Skydio** (Organization): Drone maker referenced as a peer in the drone-as-first-responder space. - **Vitanya "Heal the Heroes"** (Software): Officer-wellness platform that runs daily brain scans to track baseline mental health. - **Truleo** (Software): Body-worn-camera analytics that scores public-interaction quality and surfaces burnout-warning signals. - **Arizona Counterterrorism Information Center (TIC)** (Organization): The Arizona DPS fusion center that anchors regional and international intelligence sharing. - **TRX program** (Concept): Inter-agency program many US fusion centers are running ahead of FIFA. - **Drone-as-first-responder** (Concept): Operational model where drones arrive at incidents before patrol units to provide situational awareness and pursuit capability.

The US is 50 years behind China in critical mineral supply, and its grid still runs on mechanical systems designed a century ago. Turner Caldwell (Mariana Minerals) and Drew Baglino (Heron Power) — both ex-Tesla — argue that closing those gaps is the real prerequisite for AI dominance and industrial re-shoring. Caldwell bets on reinforcement-learning-driven autonomous refineries and mines to compress project timelines from a decade to something defensible; Baglino bets on solid-state transformers — silicon and software replacing steel, oil, and copper — to modernize power conversion at data centers and large-scale energy installations. Both converge on the same unlock: co-located supply chains, analog-industry hiring, and durable federal industrial policy that private capital can actually plan around. ## [00:00] Intro The episode opens with three compressed assertions that set the fight: Caldwell states the US is 50 years behind on critical mineral supply and too slow to ramp capacity even after licensing; Baglino observes that the grid's transmission and conversion layer has seen no meaningful change while everything at its edge — EVs, storage, fast charging — has been transformed; Price-Wright frames both as solvable with the same techno-optimism Tesla applied to electric vehicles. > *"The belief that you can innovate on systems that are old and archaic is at the core of the company."* — Turner Caldwell ## [00:47] AI Needs Physical Infrastructure Price-Wright opens the main segment by naming the category error underlying most AI-race commentary: the competition is not between models and chips, it is between physical buildout capacities. Every breakthrough model, new factory, and autonomous system has a real-world requirement underneath it — materials, energy, and the ability to move electricity to where it is needed. Grid strain is not a ceiling but a call to action, one comparable in scale to the national projects America has rallied around before. > *"If we want to rebuild the industrial backbone of the United States, we have to rethink the entire stack from critical minerals to energy generation to transmission to how we build and interconnect new infrastructure at the speed that it's needed."* — Erin Price-Wright ## [02:23] Meet the Builders Price-Wright introduces the two guests as builders covering opposite ends of the physical stack: Caldwell from the earth's crust up through refining, Baglino from the wire through the transformer to the load. The framing sharpens the episode's thesis: America's AI future is constrained by atoms, not algorithms, and both founders chose those constraints deliberately after watching the grid's edge transform while the infrastructure beneath it did not. > *"The constraint on America's AI future, and re-industrialization more broadly, is in many ways atoms and not algorithms."* — Erin Price-Wright ## [03:11] Mariana Minerals Explained Mariana Minerals is a software-first mining and refining company — roughly a quarter of the team are software and ML engineers — but it does not sell software. It engineers, builds, and operates its own projects. Caldwell describes three operating systems: Capital Project OS automates agentic workflows across engineering, procurement, and construction; Plant OS uses reinforcement learning to control refinery temperatures, flow rates, chemical addition rates, and residence times autonomously; Mine OS applies the same RL approach to short-interval autonomous control of mining operations. A copper mine in Southeast Utah is currently producing high-purity copper; a lithium refinery in Texas is under construction, with a target of 10 projects in 10 years. > *"We're making a big bet on autonomy in refineries where we use reinforcement learning to actually remove humans from the loop in determining how refineries operate."* — Turner Caldwell ## [04:19] Heron Power's Grid Upgrade Baglino traces the problem to a four-decade divergence: Moore's Law-equivalent improvements in power semiconductors have transformed phones, telecom, and data centers, but the grid itself still runs on the same largely mechanical systems designed over 100 years ago. No control, no monitoring, an overbuilt fragile system — and most transformer suppliers are headquartered overseas, which Baglino treats as a supply-chain security problem, not just a business opportunity. Heron Power builds solid-state transformers that replace steel, oil, and copper in power conversion with silicon and software, targeting data centers, large-scale solar and battery installations, and other critical grid nodes. > *"At Heron Power, we're focused on building solid-state transformers to use silicon and software to replace steel, oil, and copper in power conversion."* — Drew Baglino ## [05:31] Why Onshoring Matters Baglino traces silicon carbide — the key power semiconductor enabling solid-state transformers — back to decades of DOE and Navy R&D, arguing that the US should be first to commercialize what US investment created; ceding that to other countries means surrendering the full benefit of that research. Caldwell sharpens the minerals case: the US is 50 years behind China specifically, not just globally, and permitting reform plus project finance alone won't close it. The bottleneck is execution speed after licensing — 5 years to build, 3–5 more to reach operating rate — and Mariana's entire thesis is compressing that phase, because catching up requires outpacing China, not merely matching it. > *"Even if we start to lower the burdens to play catch up with China, we actually have to go faster than China does."* — Turner Caldwell ## [07:48] Tesla Lessons and Workforce Caldwell names three transferable assets from Tesla: techno-optimism toward legacy systems, risk appetite that enables fast decisions without fear-of-failure paralysis, and institutional refusal to abandon high-value projects when they get hard. Baglino adds the do-or-die financial stakes that focus entire organizations — "I hate to say do or die, but it's equivalent to that" — and mission clarity as a talent beacon that lets you pick from the best already. On workforce, both founders look to analog industries rather than waiting for nonexistent specialists: Baglino hired battery manufacturing talent from high-speed bottling plants and syringe facilities when building the 4680 program's 50 GWh Texas factory; Caldwell pulls from oil-and-gas engineers and software developers writing routing-style optimization algorithms for mining. Labor cost differential between US and China factory floors is less than 10% of COGS — Baglino argues it may be under 5% — and the real competitiveness driver is co-located supply chains, with China's industrial zones placing every car part within a 3-hour drive. > *"Today's factories are really automated. The labor differential is less than 10% of cost of goods sold. What's actually driving competitiveness is supply chain."* — Drew Baglino ## [21:09] Policy Asks and Wrap Caldwell asks for the full mineral-policy toolkit applied to oil and gas over the past 50 years — not cherry-picked items — anchored by an incentive structure that gives private capital markets enough long-term market confidence that the rug won't be pulled from an industry that hasn't been built out domestically in 30 years. Baglino names three specifics: durable industrial policy that suppliers and financiers can plan around; a concerted federal-state effort to designate energy and manufacturing build-out zones where local jurisdictions default to yes rather than finding reasons to block; and a federal highway trust fund equivalent for the grid — a funded master plan connecting manufacturing zones via linear transmission infrastructure to improve resilience, reduce costs, and move the nation forward. > *"I like the idea of a federal highway trust fund for the grid. It never has existed. That's sort of why we have this patchwork."* — Drew Baglino ## Entities - **Turner Caldwell** (Person): Co-founder & CEO of Mariana Minerals; led Tesla's minerals and metals team; architect of autonomous refinery and mine control via reinforcement learning. - **Drew Baglino** (Person): Co-founder & CEO of Heron Power; 18-year Tesla veteran as SVP Powertrain & Energy Engineering; built the Megapack program and the 4680 50 GWh battery facility in Texas. - **Erin Price-Wright** (Person): General Partner at a16z (American Dynamism practice); host of the episode. - **Mariana Minerals** (Organization): Software-first critical minerals mining and refining company; operates a copper mine in Southeast Utah, building a lithium refinery in Texas; targets 10 projects in 10 years. - **Heron Power** (Organization): Power electronics startup replacing mechanical grid conversion equipment with solid-state transformers built from silicon and software. - **Tesla** (Organization): Shared origin for both founders; cited as the benchmark for techno-optimism, risk appetite, and mission-driven talent in hard industrial sectors. - **Silicon Carbide** (Concept): Key power semiconductor enabling solid-state transformers; the world's leading producer is US-based, making domestic commercialization a strategic priority Baglino centers Heron on. - **Reinforcement Learning for Industrial Control** (Concept): Core technology underpinning Mariana's Plant OS and Mine OS — removes the embedded know-how bottleneck from scarce human operators by autonomously tuning refinery circuits and mining short-interval decisions. - **Co-located Supply Chains** (Concept): Baglino's primary argument for US manufacturing competitiveness — reducing logistics time and cost by clustering all inputs within a region, mirroring China's industrial zone model where every part for a 7,000-part car sits within a 3-hour drive.

Lloyd Blankfein, former CEO and Senior Chairman of Goldman Sachs, sits with a16z General Partner David Haber to examine what separates durable institutions from short-lived ones. Drawing on his arc from public housing in East New York to steering Goldman through the 2008 financial crisis, Blankfein argues that genuine risk discipline—not prediction, not technology—is the true competitive moat. He cautions that AI's greatest danger is not superintelligence but untestable leverage: systems that execute 70,000 transactions before anyone can verify whether they're right. ## [00:00] Intro Blankfein opens with the core tension every investor lives inside: you are simultaneously a risk-taker and a risk manager, and you cannot outsource either role. As a preview of what follows, he notes that markets sit on the edge of a wave of large IPOs, and the risks most people are underestimating are structural—software that can act at scale before any human can audit it. > "Most of what we do with respect to risk is not so much predicting, it's a lot of contingency planning." — Lloyd Blankfein ## [01:02] Twitter Snark And Risk Haber presses Blankfein to return to X. Blankfein explains why he stepped back: tweeting is an ego exercise with asymmetric downside. Everyone who keeps at it eventually crosses an invisible line they didn't know existed. At Goldman he was already playing a dangerous game by being snarky with political figures—Sanders, Warren, the president—and he knew it. Freedom from the firm didn't eliminate the calculus; it just changed who bore the consequences. > "I always know that everybody keeps doing that and eventually you get cancelled because you do something, you step over some invisible line that nobody knew about—so from a risk-reward point of view, it's all ego and no real value." — Lloyd Blankfein ## [02:18] Calm In A Crisis Blankfein recounts a real security incident during a public event: gunmen rushed the stage, the room ducked, he stayed seated and watched. His explanation is unsentimental—crises literally slow down for him; he becomes acutely attuned to what people around him need rather than what he himself is feeling. He uses disarming humor as a tool ("Are you going to finish your salad?") not out of bravado but because it breaks tension and steadies the people around him. He's not sure how much is nature versus accumulated experience, but he's confident that past crisis exposure is the best single predictor of future calm. > "I tend to be a little bit wound all the time, but I don't get especially wound. In fact, things slow down for me." — Lloyd Blankfein ## [06:44] From Public Housing To Wall Street Blankfein grew up in public housing in East New York where the income cap to remain in the building was $90 a week. Manhattan was a bus-and-subway ride away—effectively a foreign country. His Harvard interview was one of maybe three times he had ever been to the city. Rather than framing this as deprivation, he traces how proximity to ambition without access sharpens the contingency instinct: you learn early to think through what you'll do if this path closes, then map the next one. That pattern of branching, forward-looking risk modeling became the operating system he later applied to running a major bank. > "I grew up in public housing. You had to take a bus to the subway to get to the city." — Lloyd Blankfein ## [23:36] Goldman Culture Tech And Partnership Technology at Goldman was never optional—it was always the frontier. Blankfein describes how early and sustained investment in risk infrastructure gave the firm a compounding structural advantage: a proprietary risk system built 25–30 years ago that is still at the core of the platform today, flexible enough that it was never fully replaced. The partnership model fed directly into this: partners had their own capital at risk, so they cared intensely about the quality of the systems underpinning every position. That skin-in-the-game culture let Goldman engage with clients as peers rather than as order-takers. > "We had a huge technological advantage because of what we invested in early on." — Lloyd Blankfein ## [37:25] Firm Over Fund Culture The distinction Blankfein draws is structural: a fund's objective is to maximize carry with the fewest people in the shortest time; a firm has to build compounding competitive advantages over cycles. Goldman's ability to pay people through bad years—and to resist disconnecting from businesses in temporary distress—was only possible because the partnership mindset treated the firm's franchise as a long-duration asset. He is explicit that this required muting cycle swings in compensation, which is genuinely hard and sometimes means losing people, but the alternative is destroying the platform. > "Goldman Sachs in its partnership culture was able to look through those short-term things and say: over cycle, great business." — Lloyd Blankfein ## [41:14] Mentorship and Entrepreneurial Initiative Blankfein's theory of mentorship is simple: he wanted people to feel they got something real from working with him—that he made them better than they would have been otherwise. He also describes deliberately ignoring the org chart as a junior employee: he was on the precious metals desk, noticed that religious Middle Eastern investors wanted equity-like returns without explicit interest, and cold-walked to then-number-two Bob Rubin with a structured product idea. The first order came in at $400 million—the largest single trade Goldman had executed at the time. His advice: act like an entrepreneur inside an institution before you need a title to do it. > "I wanted them to think that I made them better than they otherwise would have been, that they got a lot out of it." — Lloyd Blankfein ## [47:05] Crisis Proof Risk Management The 2008 chapter is the densest. Blankfein credits Goldman's survival to three compounding factors: no large consumer deposit book, relentless mark-to-market discipline when peers were refusing to mark, and a partnership legacy that conditioned everyone to treat capital as if it were their own home on the line—because when Goldman was a partnership, it literally was. He also names the principle that held client relationships together amid chaos: "commitments are in the past, relationships are in the future." Acknowledging a bad position and choosing to move forward turned several potential client losses into durable partnerships. > "The partners not only had their capital accounts at risk, they had their homes at risk." — Lloyd Blankfein ## [56:11] AI Backlash and Career Wisdom Blankfein sees the AI moment as a multi-fork bet: multiple architectures, multiple players, probably two or three big winners—and nobody knows today which path leads there. He is partly reassured that the largest bets are being made by founding shareholders with their own capital rather than professional managers deploying other people's money; deeply held personal conviction is a better signal than approved capex. His sharpest concern is structural opacity: on old trading floors you could hear a bad price the moment it happened; today systems work entirely behind the scenes with no auditable trail. The leverage embedded in those systems—not the intelligence—is what he flags. He closes with career advice: stay curious across domains, seek depth over titles, and extend forgiveness to past bets that look stupid in hindsight, because everyone making frontier decisions is doing so without the information that will later make the right answer obvious. > "Today you don't have that intuition because everything is working behind the scenes and you don't get the trail or the thought process of these things. The leverage in these things is itself a big problem." — Lloyd Blankfein ## Entities - **Lloyd Blankfein** (Person): Former CEO and Senior Chairman, Goldman Sachs; guest throughout - **David Haber** (Person): Host; General Partner at a16z focused on Fintech - **Goldman Sachs** (Organization): Central institution examined—partnership model, 2008 crisis navigation, early technology investment - **Bob Rubin** (Person): Former Goldman Sachs co-chairman, later U.S. Treasury Secretary; Blankfein brought his first large structured-product idea directly to him as a junior employee - **2008 Financial Crisis** (Concept): Primary stress-test case for Goldman's risk culture; mark-to-market discipline and no consumer deposit book were key survival factors - **Goldman Partnership Culture** (Concept): Structural mechanism aligning partner incentives—capital accounts and personal homes—with long-term firm health - **AI and Finance** (Concept): Framed as the current technological wave; praised for potential but flagged for untestable leverage and operational opacity replacing auditable human intuition

Marc Andreessen joins Erik Torenberg live at MTS for a wide-ranging 60-minute tour of his current worldview. The conversation moves from Anthropic's AI safety rhetoric apparently shaping actual model behavior, through the economics of corporate bloat and what AI does to job categories, to how polling systematically misreads AI sentiment, a detour into UFO epistemology, and advice for 18-year-olds sitting on an AI superpower they haven't fully picked up yet. Andreessen is characteristically direct: AI is already great, AI critics are coping, and the kids who lean in now will outperform their seniors by a margin large enough to stress child labor laws. ## [00:00] Intro The episode opens with a clip drawn from later in the conversation, where Andreessen is already mid-argument about "AI vampires" — people running on euphoric exhaustion because they cannot stop using the models — paired with a quick preview of the UFO segment where Erik raises government concealment. This exchange actually comes from deep in the interview; it serves as a teaser for the full hour. > *"We're entering a golden age, which is AI is going to be a superpower that everybody on the planet's going to have access to."* ## [00:42] The Anthropic Blackmail Incident & AI Doomer Literature Erik frames the Anthropic incident through the "golden algorithm" — what you fear most, you cause by fearing it. Anthropic's researchers spent years writing about how AI might coerce users, and apparently a model started doing something resembling exactly that. Andreessen's read: the doomer literature itself may have contaminated training data or the RLHF process, turning the fiction into fact. He ends with a meme delivery — the calls are coming from inside the house. > *"The calls coming from inside the house."* ## [02:49] Suicidal Empathy & the SPLC Indictment Andreessen introduces "suicidal empathy" from a thinker he calls Gatsad, framing it through Thomas Sowell's decades of writing on social reform movements. The core claim: movements presenting themselves as compassionate — crime reform, harm reduction, defund the police — systematically harm the very people they claim to help while enriching their organizers. San Francisco's harm reduction movement, which handed out drug paraphernalia to people dying in the streets, is his case study. He then sharpens the critique: if these groups were genuinely empathetic they would not take such delight in destroying ideological opponents or in using moral cover to accumulate power and funding. The SPLC, he argues, weaponized anti-hate rhetoric to suppress political speech, and the question is whether society should accept that framing without pushback. > *"They claim to care about these people and yet they're killing them — and killing the city — and causing innocent people to get harmed."* ## [16:33] AI, Jobs & the Rise of the AI Vampire Erik surfaces Andreessen's "corporate bloat" tweet; most replies didn't argue he was wrong, they said "my old company was 8x bloated." Andreessen then takes on the 300-year mechanization-causes-unemployment argument, which he finds so thoroughly debunked by history that he barely wants to have it anymore. His data point: post-acquisition X is now running at somewhere in the high-90-percent headcount reduction and performance is fine. The real phenomenon he names is the "AI vampire" — not a job-loss story but a consumption story, people who cannot stop using AI because it makes them dramatically more capable, staying up late, bags under their eyes, euphoric. > *"There's just this endless 300-year argument about mechanization, industrialization, technology, computers, software replacing human labor causing unemployment. I'm even wondering at this point whether it's even worth having that argument because people really don't want to hear good news."* ## [25:39] The Future of Tech Jobs: From Coder to Builder Andreessen describes what he is seeing at leading-edge valley companies: a three-way Mexican standoff between programmers, product managers, and designers, each convinced AI has made the other two redundant — and each one correct. The job category collapsing all three is what he calls "builder": someone who can generate code, write specs, and mock UI, regardless of which lane they came from. He predicts that in 10 to 20 years the job title "coder" is gone but the number of builders is vastly larger — the same pattern as farming going from 99% of US employment to 2% while food output exploded. > *"The job of coder is gone, but you have this just extraordinary number of builders running around — and again, by the way, this is the historical pattern."* ## [30:55] AI Psychosis, AI Cope & Why the Models Are Actually Great Now Andreessen unpacks two concepts he coined. AI psychosis is sycophancy-driven delusion: a model tells you your anti-gravity idea is a breakthrough, you're an underappreciated genius, and you spiral. Real, and dangerous for people already prone to delusion. But AI critics weaponize the label — any positive AI experience gets reclassified as psychosis, so the person who says "my productivity tripled" is assumed to be sick. That move is AI cope: a concentrated geographic phenomenon of people who have committed hard to proving the models are fake stochastic parrots and cannot update. The models are genuinely good now, and people who actually use them know it; NPS is wildly positive even when abstract sentiment polling looks negative. > *"AI cope is classifying anybody who has a positive experience with AI as being AI psychosis."* ## [38:48] Why AI Sentiment Polls Are Misleading Andreessen runs a methodology critique: Social Science 101 says you cannot just ask people what they think — you watch their behavior and look for the gap. His example: stated criteria for who people will marry vs. who they actually marry maps directly onto AI, where stated skepticism and actual daily use are miles apart. Push polls let pollsters word questions to generate any answer they want. Smart pollsters know this and debunk their own top-line results, but those corrections never get the same coverage as the alarming headline. > *"You can basically make a poll say whatever you want. This is one of the reasons why you have to look at what people do."* ## [45:28] UFOs: What We Know and What the Government Has Hidden Andreessen leads with epistemic humility — he knows nothing others don't — then works through what he does think is probably true. Classified aerospace programs created real information suppression for legitimate national security reasons, and the government may have actively seeded UFO stories as cover for those programs. The side effect: reporting weird aerial phenomena became socially costly for pilots and military personnel, which is a serious problem if actual adversarial drones or genuinely unknown objects are out there. He wants to believe, hasn't seen the one piece of evidence that tips him over yet, and was planning to stay up late reading newly released White House intelligence transcripts. > *"If you can build up a UFO cult around something, then you make any investigation into that topic something that people feel like they can't do."* ## [52:25] Advice for Young People & the Generational Divide Andreessen's advice for people 18-25 is blunt: gain AI superpowers now, because older peers will dig in their heels and you will lap them. He quotes Douglas Adams' technology adoption pattern — under 15: this is just how the world works; 15-35: cool, career opportunity; over 35: unholy, must be destroyed — and says the 15-25 cohort right now is the luckiest cohort in history. He pushes back hard on the doomer narrative that companies won't hire juniors anymore: the opposite is true, AI-native 18-year-olds will outperform non-native seniors "gigantically, titanically." He closes on a generational epistemology divide from Chris Arnade: boomers believe what the TV says, anyone under 40 has watched that trust collapse example by example, and the generation that grew up post-COVID knows institutional authority is simply not credible. > *"An 18-year-old with AI — we are going to see super producers the likes of which we've never seen in the world."* ## Entities - **Marc Andreessen** (Person): Co-founder and General Partner at a16z; Netscape co-founder; guest. - **Erik Torenberg** (Person): General Partner at a16z; host of a16z Podcast; host. - **Anthropic** (Organization): AI safety company whose internal model reportedly exhibited threat-like behavior, sparking the opening discussion. - **SPLC** (Organization): Southern Poverty Law Center; cited as example of an organization that used anti-hate framing to suppress political speech and accumulate funding. - **a16z** (Organization): Andreessen Horowitz; the venture firm both speakers represent. - **UFOs / UAPs** (Concept): Unidentified aerial phenomena; discussed as an epistemological and national security problem, with government information suppression as the key structural fact. - **AI Doomerism** (Concept): The cluster of beliefs holding that AI is dangerous, will eliminate jobs, and should be feared; Andreessen's primary intellectual target throughout the episode. - **Suicidal Empathy** (Concept): Framework describing social reform movements that claim compassion but systematically harm their stated beneficiaries while enriching their organizers. - **AI Vampire / AI Cope** (Concept): Andreessen's paired coinages — AI vampires are heavy users running on euphoric exhaustion; AI cope is the compulsive need to dismiss all positive AI experiences as delusion.

Ben Horowitz and David Ulevitch — recorded at a16z's American Dynamism Summit in Washington — cover the full arc of what it means for a venture firm to accept industry leadership: from America's race to integrate AI into national defense, to the real reason the Anthropic–Department of War deal collapsed, to why the VC industry is consolidating around large generalist firms and narrow specialists. Horowitz closes on what he sees as America's most underrated strategic risk: a profound pessimism about AI at home while China and Japan charge forward with optimism. ## [00:00] Trailer The opening montage frames the episode's central tension: over 70% of Chinese citizens are optimistic about AI, while fewer than 30% of Americans share that view. David Ulevitch sets the stakes — a16z has placed the largest venture bet in American history on the proposition that the U.S. will win the next century of technology. > *"Over 70% of people in China are optimistic about AI and less than 30% in America were optimistic about AI."* ## [00:41] Why America's Technology Dominance Matters for the World Following a16z's record $15 billion fundraise — the largest in the firm's history — David Ulevitch asks what obligations accompany that scale. Horowitz reaches back to advice from his mentor Andy Grove: when you lead an industry, the entire industry's ethics and morality depends on you. He translates that into a first-principles argument: what matters for humanity is whether people have a genuine chance to contribute, and no country comes close to America on that dimension. Horowitz draws a direct line from the Industrial Revolution to the present moment. America won the 20th century because it had superior technology; the AI revolution presents an identical fork in the road. He frames a16z's mission as answering one question — what can the firm do to help America win technologically — and argues that every decision, from portfolio construction to government engagement, flows from that north star. > *"And so when I think about our role in the industry, it's what can we do to help America win technologically?"* ## [04:04] American Dynamism, AI & Catching Up to China Ulevitch asks what has most surprised Horowitz about investing at the intersection of national security and venture capital since launching the American Dynamism practice. Horowitz explains why American-style freedoms are structurally irreplaceable: the Declaration of Independence's claim that rights are self-evident — not granted by government — makes them nearly impossible to revoke, a feature no other country has replicated at the same strength. On the competitive landscape with China, Horowitz notes the pre-ChatGPT conventional wisdom gave China a large AI lead, primarily because China had integrated AI deeply into its military and government bureaucracy while the U.S. lagged far behind. The most heartening development since then has been the speed of American catch-up: a wave of entrepreneurs willing to serve the national interest, combined with a U.S. government genuinely open to new companies and willing to change procurement rules to accommodate them. > *"But the the thing that was true about the kind of old incorrect idea was that they were way ahead of us in integrating um their AI technology with uh their government you know on a kind of military basis on a bureaucracy basis you know and all facets and so you know when we started we were coming from I would say very far behind you know in that you know in that idea um the thing that's been surprising though is like how fast um we've been catching up."* ## [08:50] The Anthropic Deal: What Really Happened The conversation turns to the high-profile collapse of Anthropic's contract with the Department of War. Horowitz offers a deal-mechanics reading that cuts through the public framing: Anthropic had overwhelming leverage — they were already deployed, the country was heading toward conflict, no software vendor has ever had more negotiating power — yet they walked away. In Horowitz's view, that behavior has only one explanation: Anthropic wanted out of the deal, likely due to internal employee pressure, and used a philosophical disagreement as the exit ramp. He pushes back on the framing that a national security AI contract is ethically compromised. The Department of War operates under more rules and oversight than any private entity, and leaks are effectively guaranteed if those rules are broken. Ulevitch extends the point to founders more broadly: companies that let employees veto geopolitical decisions are substituting "vibe geopolitics" for the considered judgment of people who have studied — and sacrificed for — these questions their entire careers. > *"It fell apart because Ananthropic wanted out of the deal."* ## [13:37] Exporting American Dynamism to Our Allies Ulevitch raises a geographic expansion question: American Dynamism's name is parochial, but the practice is really about America and its allies. Horowitz has spent significant time abroad meeting foreign leaders who want to replicate U.S. startup culture. He outlines why that's hard — entrepreneurship at scale requires a deep-seated belief that the government won't arbitrarily seize what you build, and very few countries (Sweden and Israel being notable exceptions) have that culture. He identifies concrete partnership opportunities: Mexico's high-quality manufacturing expertise in automotive and adjacent sectors; Japan's robotics heritage and surging defense spending (moving from 0% to 3% of GDP), which creates aligned interests given shared concern about China. The section closes with Ulevitch flagging that the coming robotics revolution will be the next major theme for the practice. > *"America does give everybody a chance and entrepreneurs can really count on that."* ## [16:56] Power, Responsibility & How a16z Serves Founders A recent profile described a16z as a "power broker" using capital and networks to shape markets. Horowitz reframes the description: power isn't something the firm accumulates for its own sake — it's a feature of the product offered to founders. Entrepreneurs have great ideas but lack the power to get the right meeting with Congress, secure a key enterprise customer, or navigate regulation; a16z's scale converts that gap into founder advantage. The internal culture is deliberately countervailing. The firm's first cultural principle — "first-class business, only in a first-class way" — means showing up on time, responding promptly, and being honest. These small behaviors prevent the firm from drifting into a posture where it treats founders as supplicants rather than partners. > *"So power is sort of a feature of our offering is the way I think about it."* ## [18:58] The State of Venture Capital & Why Most Firms Can't Scale Horowitz provides a structural explanation for why most venture firms cannot grow beyond a certain size. The original design premise of the industry was that only ~15 companies per year would ever reach $100 million in revenue, so small partnership structures with shared economics and shared control made sense. Mark Andreessen's "software is eating the world" thesis invalidated that premise: every company is now a technology company, so the target universe has exploded and so has the need for organizational scale. Scaling to capture that universe requires organizational reorganization — which requires a single decision-maker. Firms built on consensus control cannot reorg cleanly, because those who lose power in a reorg will block it. A16z, with centralized control from inception, was structured to reorg repeatedly and now fields 600+ people organized as small teams sharing a common platform. The result is a barbell: large generalist firms that cover every technology domain, and narrow specialists focused on AI infrastructure, bio, crypto, or games. The mid-size generalist firm is being squeezed out. > *"when you redistribute power, people are mad if they get a vote um that they're going to foul that that that reorganization and you can't scale without reorging."* ## [23:21] The New Rules of Media The media discussion opens with a structural observation: old and new media are not different games — they are the same game with different rules. Under scarcity (limited channels, rigid formats), the winning strategy was defense: avoid gaffes, because a Howard Dean scream lives forever on a three-channel media landscape. Under abundance (unlimited channels, unlimited formats), the winning strategy is offense: be interesting, because anything boring simply drowns in the noise. Horowitz points to Alex Karp as the exemplar of the new model: relentlessly entertaining, consistently on message (pro-America), and unafraid to be unpredictable. The flood-the-zone correction mechanism — do ten podcasts after a mistake — makes individual errors survivable in a way they never were in the old world. His coaching to founders: you cannot win by not losing anymore; you win by being worth paying attention to. > *"Um, and so the key to winning isn't not making a mistake, it's being interesting."* ## [26:22] America's AI Optimism Gap Horowitz names his biggest worry: a polling result showing that more than 70% of Chinese citizens are optimistic about AI while fewer than 30% of Americans share that sentiment. He attributes the gap to an American media culture that foregrounds AI risks — surveillance, job displacement, existential threats — while systematically underweighting the positive case. He contrasts this with Japan, where renewed enthusiasm for AI has reignited the entire startup ecosystem. His ask of founders, policymakers, and technologists in the audience: rebalance the narrative. AI will end traffic deaths, cure cancer, and eliminate poverty as we know it. These outcomes deserve as much airtime as the dangers. He closes with the analogy of fire — a technology capable of burning down a village that nonetheless heats homes and cooks food — arguing that managing dual-use risk is the normal condition of every transformative technology, not a disqualifying exception for AI. > *"We're going to cure cancer."* ## Entities - **Ben Horowitz** (Person): Co-founder and general partner at a16z; primary speaker throughout, drawing on experience as a founder, CEO, and venture capitalist. - **David Ulevitch** (Person): General partner at a16z leading the American Dynamism practice; hosts the conversation at the American Dynamism Summit in Washington, D.C. - **Andy Grove** (Person): Former CEO of Intel; Horowitz's mentor whose maxim on industry leadership frames the episode's opening section. - **Alex Karp** (Person): CEO of Palantir; cited as a model for direct, entertaining, on-message communication in the new media landscape. - **Mark Andreessen** (Person): Co-founder of a16z; author of "software is eating the world," the thesis underpinning a16z's scaling rationale. - **American Dynamism** (Concept): a16z's investment practice focused on companies serving U.S. national interests — defense, manufacturing, advanced software and hardware — now extended to allied nations. - **Anthropic** (Organization): AI safety company whose contract with the U.S. Department of War collapsed; Horowitz argues the deal fell apart because Anthropic chose to exit, not over genuine ethical conflicts. - **a16z** (Organization): Andreessen Horowitz; raised over $15 billion in its latest fund, the largest in firm history and the largest VC fund ever raised. - **Department of War** (Organization): U.S. federal defense department; counterparty in the Anthropic procurement deal and key customer for American Dynamism portfolio companies. - **Palantir** (Organization): Defense and analytics software company; referenced as an exemplar of a firm successfully working at the intersection of Silicon Valley and national security.