All-In PodcastDOGE kills its first bill, Zuck vs OpenAI, Google's AI comeback
CHAPTERS
- 0:00 – 12:40
Banter, Bestie Intros, and Welcoming Box CEO Aaron Levie
The episode opens with the usual All-In banter about bar fights, personalities, and inside jokes before shifting to introduce returning guest Aaron Levie, CEO of Box, as the ‘fifth bestie.’ They joke about ski trips, Box’s all‑time‑high stock price, treasury Bitcoin experiments, and announce the new allin.com email list.
- •Lighthearted jokes about JCal and Sacks’ ‘codependent’ relationship and bar-fight archetypes set an informal tone.
- •Aaron Levie joins as guest, teased about Box’s strong stock performance and potential Bitcoin treasury moves.
- •The crew plugs allin.com, positioning it as an email list for pod drops and event tickets, tongue-in-cheek framed as ‘spam.’
- •They explain Sacks’s absence: he’s in Washington, D.C. doing meetings, setting up later policy discussion.
- 12:40 – 25:00
Sacks in Washington: AI and Crypto Regulation Priorities
The conversation turns serious as JCal asks Levie what he thinks of Sacks’s role shaping AI and crypto policy. Levie supports Sacks’s deregulatory instinct for AI and criticizes California’s SB 1047 and some Biden-era proposals as overcautious, while Chamath and JCal press on what practical crypto rules Sacks should push first.
- •Levie calls Sacks a strong pick for AI policy given his bias against early overregulation.
- •He argues SB 1047 and similar efforts frame AI progress as inherently risky, imposing heavy liability on model makers and threatening to chill releases like LLaMA.
- •He favors national over state-by-state AI rules to avoid a ‘world of hurt’ for developers and businesses.
- •Levie views the Biden EO on AI as relatively toothless but credits OSTP’s Arati Prabhakar as technically competent and not pro-overregulation.
- 25:00 – 45:00
Crypto 101, Stablecoins, and What Sacks Should Do First
Chamath lays out a two-step crypto agenda for Sacks: normalize and regulate stablecoins as core payment rails and use them to undercut legacy card networks on fees. He and Levie contrast ‘crypto as financial infrastructure’ with speculative tokenization and explain why stablecoins already power real businesses like SpaceX’s Starlink, arguing this path avoids political quagmires.
- •Chamath defines stablecoins and distinguishes Tether (offshore, historically opaque) vs USDC/Circle (US-based, more transparent).
- •At scale, stablecoin issuers earn enormous float yields from treasuries, making them powerful financial players.
- •Real-world use case: SpaceX/Starlink uses stablecoins in long-tail countries to avoid FX risk and wiring friction.
- •Regulated stablecoin rails could both discipline banks’ slow, fee-heavy infrastructure and challenge Visa/Mastercard/Amex economics.
- •Levie separates ‘Crypto 1.0’ (payments, Bitcoin as digital gold) from speculative tokens/NFTs that trigger thorny securities-law issues and are politically harder to fix.
- 45:00 – 53:20
Doge vs the Omnibus: Killing a $340B Bill in 12 Hours
The hosts unpack how Doge and Elon helped derail a 1,500‑page, $340B continuing resolution unveiled three days before a funding deadline. They detail the bill’s scope and pork, Friedberg’s historical perspective on runaway federal spending, and how AI-assisted citizen scrutiny turned Twitter into a real-time oversight platform, forcing Congress to back down.
- •The bill was enormous, rushed, and included controversial items (congressional pay raises, farm bill renewal, stadium and bridge funding, etc.).
- •Friedberg notes federal spending has grown to ~23–24% of GDP, with little structural check, contrary to Founders’ expectations.
- •He argues representatives are structurally incentivized to say ‘yes, but give me X for my district,’ swelling spending over time.
- •Chamath calls the bill’s defeat a ‘multi‑hundred billion dollar grift’ stopped in ~12 hours of tweets, enabled by AI summarization and mass online analysis.
- •They suggest future controversial bills could be rapidly dissected the same way, enhancing both pro- and anti-legislation movements.
- 53:20 – 1:01:40
The Case for Smaller Government, Austerity, and Transparency
Building on Doge’s win, the panel explores deeper structural critiques of U.S. governance and spending. They discuss misaligned incentives in disaster relief and farm subsidies, lack of constitutional fiscal limits, and the electorate’s shift from valuing rugged freedom to expecting government-provided lifestyle gains.
- •Friedberg explains how federal disaster relief and farm bill supports distort markets, encourage risky building, and prop up prices independent of risk.
- •The Founders assumed democratic processes would limit spending but did not codify hard caps; that assumption has failed over 250 years.
- •Chamath argues Americans still believe ‘more spending equals better outcomes,’ analogous to bloated tech companies that lose to lean startups.
- •They propose structural process reforms (e.g., minimum review days per 100 pages of legislation, breaking giant bills into smaller ones).
- •Levie emphasizes the scale of waste in government procurement—mandatory use of overpriced contractors with weak accountability.
- 1:01:40 – 1:10:00
Doge as a New Political Operating System and Biden’s Hidden Health
Aaron and Chamath frame Doge as a founder-like, startup-style approach to governance that the traditional political class doesn’t understand. They contrast Elon’s radical transparency on X with allegations that Biden’s staff hid his cognitive decline, raising ethical questions about who was actually making decisions and underscoring the need for structural transparency.
- •Levie notes DC misunderstood Elon’s willingness and capacity to ‘see this through’ using unprecedented digital tools and reach.
- •Doge’s brand may be polarizing, but its core principles—waste-cutting, efficiency, limiting pork—could draw cross-party support.
- •Chamath highlights ‘Today in Hypocrisy’: critics fear Elon as a ‘shadow cabinet’ while admitting Biden was effectively hidden from view.
- •They question whether hiding serious cognitive decline (if proven) is merely unethical or also criminal when nuclear codes and major policy decisions are at stake.
- •Chamath argues more direct communication from leaders (like Trump’s habit of saying exactly what he thinks) is a powerful check against shadow governance.
- 1:10:00 – 1:27:29
From Doge to Deregulation: Cutting Spend and Reducing Red Tape
The group connects fiscal reform with deregulatory opportunities, especially in heavily regulated states like California. Levie points out how overregulation produces perverse outcomes even in climate tech, and they argue Doge’s agenda must tackle not only spending but also regulatory sprawl to truly reboot U.S. competitiveness.
- •Levie cites climate-tech startups unable to build in California due to layered regulations, despite the state’s green rhetoric.
- •He links bloated spend and thick regulation: more rules require more overhead to interpret, enforce, and comply with.
- •The crew suggests sunset clauses or time limits on large swaths of regulations (e.g., 10–20 year clocks) as one way to force periodic pruning.
- •JCal references zero-based budgeting at Twitter/X as an example of ruthlessly questioning every role, SaaS subscription, and process.
- •They argue Doge should target regulations next, after spending, using the same transparency and public-pressure playbook.
- 1:27:29 – 1:33:20
Conspiracy Corner: New Jersey Drones, Dirty Bombs, or PSYOP?
In a lighter ‘Conspiracy Corner’ segment, the crew examines a rash of drone sightings over New Jersey and a temporary FAA drone ban. Friedberg offers three explanations—from classified U.S. activity to kids messing around to a foreign PSYOP designed to harden U.S. public opinion against drones and stall the emerging drone economy.
- •Thousands of drone sightings, an FAA ban citing ‘special security reasons,’ and rumors of missing radioactive material fuel speculation.
- •Friedberg notes the U.S. has heavy drone regulations (line-of-sight, etc.) that impede commercial innovation, while China’s Meituan is already doing large-scale drone delivery.
- •His PSYOP theory: a rival state (e.g., China) could orchestrate disruptive drone events to stoke fear, entrench regulators, and delay U.S. deregulation in drones and eVTOLs.
- •Levie is skeptical, arguing there are many more effective ways to undermine the U.S. economy than spooking people with drones.
- •They also discuss startups building radiation-sensing drones and poke fun at online conspiracy accounts and ‘expert’ pseudonyms.
- 1:33:20 – 1:40:00
OpenAI Under Fire: Zuck, Elon, and the Nonprofit-to-For‑Profit Flip
Attention shifts to AI industry politics as JCal outlines OpenAI’s $6.6B round, $157B valuation, and the poison-pill requirement to convert to a for-profit within two years. With Elon suing and Zuck/Meta petitioning California to block the conversion, Chamath analyzes market-share trends and likens OpenAI to a MySpace-like incumbent facing nimble challengers.
- •OpenAI’s new financing includes a clause forcing for-profit conversion or giving investors the right to pull capital.
- •Elon seeks a court order to pause conversion; Meta sends a letter to the California AG echoing concerns.
- •A Menlo Ventures chart shows OpenAI’s market share falling from ~50% to ~33%, while Anthropic, Google, and others gain.
- •Chamath draws parallels to Facebook overtaking MySpace: incumbents pioneer categories but can be overtaken as the field fills with upstarts.
- •He also flags xAI’s huge GPU procurement (100k now, 1M planned), which can reorder the hardware supply queue in Elon’s favor and pressure others into an arms race.
- 1:40:00 – 1:46:40
Enterprise AI: Model Promiscuity, Open Source Pressure, and Pricing Gravity
Levie describes how Box and other enterprises are becoming ‘model agnostic’—routing workloads among multiple LLMs based on cost, latency, and quality. He explains there are ‘no secrets in AI’ because research advances diffuse quickly, so open source plus hyperscaler competition will push inference pricing down to something close to compute cost.
- •Large buyers already use several AI vendors; they will likely end up with dozens, using routers or orchestration layers to optimize per-task model choice.
- •Open-source models from Meta (LLaMA, etc.) act as a price umbrella; if Zuck open-sources top benchmarks, closed vendors can’t sustain high API margins.
- •Levie compares this to storage economics: Box’s underlying storage costs collapsed over time, they now offer ‘unlimited storage’ yet enjoy 82% gross margins because value moved to the software layer.
- •He expects similar dynamics: infrastructure (tokens/inference) commoditized; differentiation moves to application logic, UX, and integration.
- •JCal and Chamath speculate that OpenAI could drift to #3–5 in market standing as Google, Meta, and xAI leverage larger data and hardware advantages.
- 1:46:40 – 1:55:00
Will AI Shrink or Expand the Software TAM?
Chamath posits that AI may compress the current $5T ‘software industrial complex’ by making code production nearly free, eroding the pricing power of incumbents. Levie and Friedberg partly disagree, arguing that while legacy enterprise systems and bloated SaaS look vulnerable, AI will expand the addressable market by automating offline services and enabling bespoke internal tools.
- •Chamath estimates roughly $5T spent annually on software, consulting, and IT staff, and predicts that figure could shrink by 10x as AI slashes development costs and commoditizes functionality.
- •He argues incumbents like ERP/CRM vendors charge tens of millions for a handful of workflows; AI will let firms roll bespoke replacements cheaply, undermining traditional pricing.
- •Levie counters that many organizations don’t want to own core systems engineering and prefer to buy turnkey solutions; history shows low-cost clones (e.g., Zoho) haven’t displaced incumbents solely on price.
- •Friedberg describes running hackathons where non-developers built useful internal tools with AI coding assistants, anticipating a future where employees can verbally specify software and have it designed, tested, and deployed.
- •They agree AI will massively expand software into service-like jobs (accountants, VAs, social media managers) via agents, even as some legacy categories face deflation.
- 1:55:00 – 2:02:30
Regulation, Compliance, and AI-Built Systems in Heavily Regulated Industries
The group digs into when and where AI can fully own application development, especially in regulated sectors like banking and life sciences. Levie and Chamath stress that QA, audits, and human sign-off are the real bottlenecks, while Friedberg argues regulators and tools will eventually adapt to AI-generated software, though likely on a decade-long horizon.
- •Levie notes clinical and financial systems require exhaustive, auditable QA and change management, making end-to-end AI-generated code hard to accept under current rules.
- •Chamath emphasizes human ‘unit tests’ and sign-offs remain necessary for regulators; bots producing probabilistic code won’t be trusted by a human inspector after a breach.
- •Friedberg imagines a future where you can instruct an AI to design systems that automatically meet regulatory and security standards, with AI also generating the required test harness and documentation.
- •Levie suggests AI-built tools will proliferate first in the long tail of internal apps (today’s ‘Access/Retool’ universe) before taking on hardened, regulated systems.
- •They converge that regulators will eventually adapt, but for the next decade, human oversight will keep AI from fully replacing core systems in critical industries.
- 2:02:30
Google’s AI Comeback: Gemini, Veo, Genesis, and Video/3D Breakthroughs
The final segment celebrates Google’s recent AI surge. Friedberg outlines how Gemini 2.0, Veo, and open-source models like Genesis tap into vast video datasets to build models that understand physics and 3D space, potentially transforming games, film, and interactive experiences. The Besties argue Google has shifted from cautious to aggressive and now rivals or surpasses OpenAI on multiple fronts.
- •Friedberg claims Gemini has not only caught up but sometimes exceeds OpenAI for his use cases; he sees usage as fungible and users willing to switch to the best engine.
- •He highlights Veo’s apparent ability to ‘render physics’ and Genesis’s 3D object modeling, enabling camera control, dynamic scene generation, and real-time virtual worlds from text prompts.
- •Levie points to Project Mariner/browser control and webcam-enabled Gemini demos that let AI observe and manipulate real interfaces, leveraging YouTube’s massive trove of screen-share content.
- •They note video data is estimated to be a billion times larger in volume than text data, giving Google a unique edge for multimodal and embodied AI.
- •The panel praises Sundar and Sergey’s renewed focus and aggressiveness, suggesting Google was late but now ‘in it to win it,’ with JCal predicting OpenAI may fall behind Google, Meta, and xAI in the medium term.
