All-In PodcastBig Fed rate cuts, AI killing call centers, $50B govt boondoggle, VC's rough years, Trump/Kamala
CHAPTERS
- 0:00 – 6:50
Summit Afterglow, Roles, and Learning to Delegate
The hosts open by celebrating the success and viral reach of the All-In Summit, joking about Friedberg’s ‘afterglow’ absence and highlighting how the team divided responsibilities. Chamath and Sacks praise Jason’s improved moderation, tying it directly to him delegating logistics and focusing on his ‘unique value add.’
- •Summit clips have already hit ~20M views with more to come.
- •Friedberg is credited as de facto ‘conference CEO’ for content and speakers.
- •Jason emphasizes that trusting a professional operations team let him focus on moderation.
- •Chamath argues Jason is “exceptional as a moderator” but “mostly average as a conference producer,” illustrating the power of focusing on strengths.
- 6:50 – 15:30
Fed Cuts 50 bps: Soft Landing or Recession Signal?
They dissect the Fed’s surprise 50 basis point cut off a 23-year rate high, comparing it to past cycles where similar moves preceded recessions. The group debates whether markets should treat this as a bullish catalyst or a warning sign of underlying economic weakness.
- •First rate cut since March 2020; CPI has recently printed with a ‘2 handle.’
- •Historically, 50 bps kickoff cuts in 2001 and 2007 were followed by 26–31% market declines.
- •Chamath says the likely path is cuts to ~2–3% by 2026 due to rising economic tension.
- •Sacks notes the yield curve has de-inverted, historically the phase when recessions hit.
- •They flag likely downward GDP revisions and softening labor markets as key watchpoints.
- 15:30 – 17:35
Labor Market Shifts and the Hollowing of Mid-Tier Jobs
Jason describes a sharp reversal in hiring difficulty, with suddenly abundant qualified candidates for roles that used to be hard to fill. He worries about pressure on $150k-ish ‘upper-middle-class’ jobs amid immigration, offshoring, and impending AI automation.
- •Hiring narratives flipped from ‘can’t find talent’ to surplus qualified candidates.
- •Early-stage startups are seeing easier recruitment compared to the ZIRP era.
- •Jason connects immigration, offshoring, and AI to potential hollowing of mid-paying jobs.
- •Sacks counters that in tech, AI is creating a strong tailwind, especially for AI-focused startups, while non-AI areas normalize.
- 17:35 – 33:41
AI Targets Call Centers: Error Tolerance, Data, and Jobs
The conversation turns to AI’s most imminent disruption: call centers and customer support. Sacks outlines why support is a perfect first target—ample training data, tiered escalation, and acceptable error rates—while Jason notes consumers’ growing preference for fast, automated solutions over human agents.
- •LLMs plus voice (e.g., OpenAI’s audio API) can already handle level-one support.
- •Customer support has built-in failover (L1 → L2 → L3), enabling safe AI deployment.
- •Error tolerance in support is higher than in domains like legal or healthcare.
- •Call centers are a major employer in specific geographies (Denver, Salt Lake, Arizona, Florida), making disruption economically sensitive.
- •They predict millions of jobs will be transformed or eliminated in 2–3 years.
- 33:41
Beyond Support: High-Stakes AI, 100% Accuracy, and Reasoning Models
Chamath previews a use case from his startup 80/90, claiming 100% accuracy over 10 days in a highly regulated system-of-record workflow, after iterating up from mid-80s accuracy. The hosts discuss OpenAI’s new reasoning model (o1), chains-of-thought, and how stitching multiple models together is still a hard, human-intensive engineering art.
- •80/90 reached sustained 100% accuracy versus the legacy deterministic system in a public, highly regulated company.
- •Initial AI versions were mid-80s, then mid-90s, then 97–98% before reaching 100%.
- •Chamath argues this proves AI can match deterministic software in zero-tolerance environments, not just toy use cases.
- •o1 ‘preview’ is described as a powerful reasoning engine that anticipates multi-step prompts but is compute-heavy and rate-limited.
- •Real-world reliability still depends on how teams orchestrate multiple models and control error ‘blast radius.’
- 33:41 – 39:10
Government Waste Exhibit B: $50B for Rural Broadband and EV Chargers
They dig into the $42B rural broadband and $7.5B EV charging programs that have delivered almost nothing years after passage. The group frames this as a mix of incompetence, political retaliation against Elon, and structurally broken incentives in public spending.
- •After ~1,000 days, zero rural households connected and only eight EV chargers built.
- •Private sector has already rolled out >1,000 charging stations in six months and Starlink has thousands of planes and rural customers online.
- •FCC canceled Starlink’s $885M subsidy claiming it couldn’t deliver high-speed internet, then later argued Starlink was so strong it risked monopoly—contradictory positions.
- •Sacks labels this ‘pure, naked retaliation’ for Elon’s political independence and non-union operations.
- •Chamath laments normalization of tens-of-billions waste and warns of an ‘administrative state’ that punishes disfavored actors at national cost.
- 39:10 – 41:00
Will AI Commoditize Customer-Support Startups and SaaS Giants?
Sacks warns that many high-flying AI customer support startups may see their value eroded as foundation models improve rapidly. Chamath agrees, saying he deliberately avoided customer service because it will be ‘run over’ by foundational models, and instead targets complex, regulated domains.
- •Customer-support AI startups are reaching unicorn valuations despite thin defensibility.
- •If models get much better, a small team could use next year’s LLMs to undercut them.
- •Chamath’s strategy: focus on hard, regulated areas where customization and compliance are core moats.
- •They predict that many support-focused apps will be commoditized, potentially run locally on powerful consumer hardware (e.g., M3 MacBooks) with internal datasets.
- 41:00 – 42:30
Media Silence, Partisanship, and a Proposed ‘Waste, Fraud, and Abuse’ Watchlist
They argue that modern media’s tribalism has killed the old watchdog function that once exposed Pentagon and federal waste. The hosts propose using their own platform and site to build a running public list of such scandals and encourage whistleblowers to leak examples to them.
- •Chamath notes he’s become numb to $50B figures because everything is now in ‘hundreds of billions and trillions.’
- •Sacks recalls when $40B-type scandals would have been major 60 Minutes segments; now they’re barely covered if they embarrass Team Blue.
- •They stress that this behavior will be copied by future administrations of both parties unless it’s checked now.
- •Jason proposes celebrating officials who save money and instituting a recurring ‘waste, fraud, and abuse’ segment on the podcast.
- 42:30 – 46:40
Reverse-Engineering Enterprise Systems: Klarna vs. Salesforce and Workday
Using Klarna’s statement about ‘deprecating’ Salesforce and Workday, they explore how AI agents can watch user interactions and recreate core functionality, effectively building a digital twin of large systems of record. Sacks is skeptical of full generalization but concedes that narrow usage patterns can be replaced more easily than previously thought.
- •Chamath describes agents that observe inputs/outputs and iteratively infer internal code paths.
- •Once parity is achieved with the legacy system, companies can shut it off and save millions.
- •He frames this as a ‘machine that builds a machine’—a massive technical challenge and leadership risk, but high reward.
- •Sacks questions why we haven’t yet seen a wave of startups doing this generically, suggesting projects are highly customized and not easily productized.
- •They agree that while not trivial, this approach challenges the long-held dogma that core systems of record are untouchable.
- 46:40 – 47:10
Government Waste Exhibit A: Oracle’s $600M NYC Portal
The hosts roast New York City’s billion-dollar course management portal built on Oracle PeopleSoft that looks like a 1990s intranet. They argue that such egregious waste and low-quality output should be impossible in an AI-enabled world and see it as symptomatic of deeper procurement and incentive problems.
- •NYC paid $600M (potentially $1B with more customization) for an outdated-looking portal.
- •Jason notes similar functionality could be built on modern platforms for ~1% of the cost.
- •They see this as classic regulatory capture plus old-boys-network relationships, not a technology problem.
- •Chamath predicts AI tooling will make it harder to justify such massive price tags for simple software.
- 47:10 – 53:00
Venture Capital’s Rough Decade: YC, DPI, and Vintage Distortion
Shifting to venture, they use a thread critiquing Y Combinator’s recent cohorts and Carta data on undelivered DPI to illustrate how hard it has become to generate real returns. Chamath shares his own fund performance and the unglamorous tactics required to turn paper gains into cash for LPs.
- •YC’s top companies list skews heavily to 2009–2016 vintages; few recent breakouts by revenue.
- •Over 40% of 2018 vintage VC funds have not made a single distribution.
- •Chamath’s funds required aggressive use of secondary markets to deliver DPI for institutions like Mayo Clinic and Sloan Kettering.
- •Gestation for big exits is often 11–13 years now versus the old 5–7 year expectation.
- •Board and founder pushback against secondary sales is common, but LPs need liquidity to keep funding the ecosystem.
- 53:00 – 57:30
ZIRP Hangover: Too Much Money, Too Little Ownership
Sacks and Jason diagnose the structural damage from 2020–2021’s liquidity flood—overcapitalized rounds, inflated entry valuations, and the ‘peanut butter’ spreading of talent, customers, and cap tables. They argue that average VC returns will be structurally lower for a decade, especially for managers who chased size and velocity.
- •COVID-era stimulus injected ~$10T into the economy; VC annual deployment spiked from ~60–100B to ~200B.
- •Entry valuations roughly doubled, turning a typical 2x fund into a 1x if outcomes didn’t scale proportionally.
- •Overfunding drove top operators to leave strong companies, fragmenting talent across too many startups.
- •Excess capital created 20 lookalike companies per category, diluting revenue, customers, and eventual outcomes.
- •Ownership in seed/Series A rounds collapsed (e.g., 20% → 10%, 5% → 1%), making fund-returning outcomes much rarer.
- 57:30 – 1:05:20
Fund Math, Time Diversity, and the Future of Liquidity
They drill into internal fund math and portfolio construction: how follow-on checks often don’t pencil out, why time diversification across 3–4 years is critical, and why many managers who deployed funds in 18–24 months were effectively running fee machines. They foresee more secondaries, smaller rounds, and a needed reinvention of public exit pathways.
- •Jason realized small follow-ons into later-stage rounds often couldn’t return the fund, and redirected capital to more seed bets instead.
- •Chamath contrasts ‘return-optimized’ funds (small, time-diversified, grand-slam-seeking) with ‘fee-optimized’ funds (large, rapid deployment).
- •Typical 10-year fund lives are misaligned with 15-year company exit timelines; extensions and cross-fund conflicts become common.
- •The percentage of first-time managers who can raise a second fund dropped from >50% to <15%.
- •They call for new liquidity mechanisms beyond traditional IPOs—citing SPACs, direct listings, and more robust secondary markets as partial experiments.
- 1:05:20 – 1:07:18
Rate Cuts, AI Tailwinds, and the Possibility of a New ‘Golden Era’
Before Chamath drops off, Sacks notes that if rate cuts continue and inflation truly subsides, AI could fuel a strong, non-bubble ‘golden era’ for tech and venture. They see the current painful shakeout as the tail end of a cycle that might set healthier foundations for the next one.
- •Further rate cuts (another potential 50 bps) could meaningfully lower financing costs.
- •AI is described as the most exciting wave since the commercial internet.
- •If discipline around valuations and capital intensity holds, AI could drive strong, sustainable returns rather than another 2021-style bubble.
- •They encourage entrepreneurs to start companies now, suggesting conditions may soon be favorable again.
- 1:07:18 – 1:13:00
Trump–Kamala Debate: Performance, Fact-Checking, and Media Bias
With Chamath gone, Jason and Sacks dissect the Trump–Kamala debate. They agree Harris overperformed expectations thanks to polished, canned answers, but Sacks argues that lopsided real-time fact-checking and her sorority ties to a moderator made the contest fundamentally uneven.
- •Harris delivered memorized talking points effectively, rarely knocked off script.
- •Subsequent interviews show near-identical ‘jukebox’ answers, exposing their scripted nature.
- •ABC moderators fact-checked Trump repeatedly, sometimes incorrectly (e.g., Springfield pet-eating complaints), while offering Harris a free pass on debunked hoaxes.
- •Sacks believes many observers fairly awarded Harris the debate-night ‘win’ on style, but sees her post-debate polling bounce fading back into a tight race.
- 1:13:00 – 1:17:00
Working-Class Realignment, Cultural Issues, and Union Voters
They explore why Trump is now leading among Teamsters despite Biden once having an eight-point advantage, attributing it to the Democrats’ shift from ‘beer track’ to ‘wine track’ priorities and Harris’s cultural emphasis on DEI over lunch-pail economics.
- •Teamsters reportedly swung from Biden +8 to Trump +26 after Harris became nominee.
- •Sacks frames the Democratic Party split as ‘beer track’ (working class) vs. ‘wine track’ (professional class, boutique cultural issues).
- •He sees Harris and Newsom as emblematic of the California-style ‘wine track’ that alienates blue-collar voters.
- •Jason agrees Harris struggles in unscripted, high-IQ exchanges compared to JD Vance or Trump, but notes she has successfully reassured some moderates she’s not ‘crazy.’
- 1:17:00 – 1:25:00
Paths to Victory: Chaos, Abortion, and the ‘Killer Issues’
They game out why each candidate might win or lose. Jason argues that fears of chaos and abortion rights could cost Trump moderates and women, while Sacks insists Trump still owns the core ‘wrong track’ issues of inflation, border, and cultural overreach, and would be favored in a fair media environment.
- •Jason: If Trump loses, it will likely be due to fears of renewed chaos, Roe v. Wade backlash, and women/suburban moderates defecting.
- •Sacks: If Harris wins, media’s overwhelmingly positive coverage of her and negative coverage of Trump will be the decisive edge.
- •Both acknowledge hidden Trump support in polls due to social pressure and stigma around admitting a Trump vote.
- •They agree the race is extremely close, flipping frequently in battleground polls.
- 1:25:00
Assassination Attempts, Dehumanizing Rhetoric, and Mental Illness Online
The show ends on a sober note discussing a second assassination attempt on Trump. They connect extreme rhetoric—e.g., framing him as an existential threat to democracy—with the presence of mentally ill followers who might act violently when they take such language literally.
- •Jason estimates 0.1–1% of any large following may be severely mentally ill and interpret charged language as literal instructions.
- •Sacks notes the attacker echoed mainstream phrases about Trump being a threat to democracy, heavily used by Biden, Harris, and media.
- •They argue that years of ‘Orange Hitler’ framing and hyperbolic narratives are ‘asking for trouble.’
- •Jason urges all sides to de-escalate rhetoric and be mindful of how extreme language can land with unstable individuals.
