All-In PodcastE168: Can Google save itself? Abolish HR, AI takes over Customer Support, Reddit IPO teardown
CHAPTERS
- 0:00 – 2:40
Cold open: Jason as a house guest, Kato Kaelin jokes, and bestie roll call
The episode opens with banter about Jason’s location, virtual backgrounds, and a riff on “Kato Kaelin-ing” through the friend group. The hosts then transition into the standard All-In intro and set the stage for the week’s topics.
- •Jokes about Jason’s background/location and being a “great house guest”
- •Kato Kaelin reference and quick digression on who he is today
- •Show intro stinger and episode framing
- •Besties introductions and light personal updates
- 2:40 – 5:54
Groq momentum and the North Star metric for developer platforms
Sacks shares a Groq update, highlighting a surge in developer interest and why developer demand is the most important early indicator for platform success. The group discusses the flywheel of developers building “while you sleep” and how AI tools may expand the global pool of people who can code.
- •Groq waitlist growth (nearing ~10k developers) as a signal of platform pull
- •Why developer adoption is a better leading indicator than near-term revenue
- •Platform dynamics: third parties create unexpected breakout apps
- •AI-assisted coding expanding who counts as a “developer”
- 5:54 – 15:26
Google’s Gemini image-generation backlash: what it signals about culture and competitiveness
Jason frames Google’s ongoing Gemini controversy as more than a PR issue—an indicator of deeper organizational dysfunction. The discussion centers on whether Google can ship great consumer AI products quickly, and whether leadership and incentives are aligned to compete in an AI-disrupted search market.
- •Gemini controversy as a symptom of product governance and incentive failures
- •Investor pressure, employee frustration, and comparisons to Meta’s 2022 reset
- •Core Google economics: Search profits subsidize everything else
- •Question of founder intervention vs. bureaucratic drift
- 15:26 – 30:38
Responsible AI, DEI ‘veto power,’ and why ‘structural changes’ could mean big cuts
The besties debate how internal review groups (Responsible AI/DEI) can accumulate decision-making power, chilling dissent and slowing product iteration. They argue this dynamic can turn product decisions into risk-avoidance theater, and discuss what meaningful reform would look like (empower product leaders vs. expand oversight).
- •Claimed ‘asymmetric’ internal power: reviewers can veto; dissenters fear reputational risk
- •Memo language (“structural changes”) interpreted as potential reorg/downsizing
- •Culture as monoculture: difficulty detecting bias when everyone shares assumptions
- •Fix proposals: restore product-owner accountability; reduce policy-by-committee
- 30:38 – 39:58
Abolish HR? Sacks’ model: outside counsel for investigations, teams own benefits and hiring
The conversation pivots to how large HR organizations can become internal ‘policing’ bodies and slow execution. Sacks outlines a model with no traditional HR department: use third-party employment counsel for serious issues, let teams design benefits, keep hiring with functional leaders, and enforce performance management rigorously.
- •Sacks’ claim: in companies he controls, he doesn’t keep a traditional HR department
- •Use external law firm as an ‘escape valve’ for serious complaints/investigations
- •Benefits set by employee committees; hiring owned by department heads
- •Annual performance discipline: manage up/out bottom 5–10%
- 39:58 – 46:13
AI data licensing wave (Reddit, Stack Overflow, media): is this ‘TAC 2.0’?
Google’s content/data licensing deals are framed as a new kind of platform payment: not for traffic, but for training data. The besties compare these agreements to Google’s historic Traffic Acquisition Costs and debate whether the new market will be recurring like ad syndication or “chunky” like content licensing.
- •TAC 1.0 recap: Google paying partners (e.g., Apple) for default search placement
- •TAC 2.0 idea: paying for proprietary datasets to improve model training
- •Entrepreneur opportunity: unique corpuses become high-margin licensing revenue
- •Key unknown: attribution and pricing for smaller sites vs. major platforms
- 46:13 – 53:41
How valuable is content to LLMs over time? Staleness, ‘dark data,’ and supplier vs buyer power
Chamath challenges the TAC analogy, arguing these deals resemble Netflix-style content licensing more than ongoing traffic purchases. They discuss how fast new data is generated, why older content may depreciate quickly, and how fragmented suppliers could weaken negotiating power versus a small set of model buyers.
- •Content freshness problem: some corpuses age poorly (news, ephemeral discussions)
- •Scale of data generation and ‘dark data’ outside the public web
- •Non-exclusivity vs exclusivity: why buyers might want exclusivity and sellers resist it
- •Potential industry coordination (music-style licensing) vs highly competitive supply
- 53:41 – 1:00:33
Klarna’s AI customer support claims: 700 agents’ worth of work and the real economic impact
The besties dissect Klarna’s announcement that AI agents handle the majority of support chats with faster resolution times and comparable satisfaction. They debate whether the savings become pure profit or are reinvested into higher-value work, and what this implies for call-center and support software industries.
- •Klarna metrics: faster resolution, fewer repeat inquiries, large profit impact claim
- •AI as first major “in production” workflow with measurable value (beyond toy apps)
- •Market shock example: call-center outsourcer valuation hit on the news
- •Likely trajectory: Level 1 support automated first, then gradual move up complexity ladder
- 1:00:33 – 1:13:02
Open source as an AI strategy: why Klarna (and Meta) might share the playbook
Sacks and Chamath argue Klarna should open source its support-agent workflow because it’s not a core product and would benefit from community improvement. They connect this to Meta’s logic for open sourcing models: if AI isn’t the direct product, open sourcing accelerates innovation and reduces internal engineering burden.
- •Open source reduces duplicated effort and lets the community advance the tooling
- •No major competitive loss if the real moat is proprietary data and integration
- •Meta’s stated rationale: open source because AI improves products rather than being sold directly
- •Implication: point-solution SaaS (e.g., support tooling) faces major pressure from agent workflows
- 1:13:02 – 1:17:18
Reddit IPO S-1 teardown: growth spike, low ARPU, and the risks of the ‘directed share’ plan
They walk through Reddit’s financials and user metrics, focusing on recent usage acceleration, strong gross margins, and persistent losses. The group debates valuation, the sustainability of the recent growth bump, and whether Reddit can ever monetize like Meta given user anonymity and ad skepticism.
- •Key S-1 stats: revenue growth, losses narrowing, high gross margin
- •DAU growth acceleration: why the last two quarters differ from prior periods
- •ARPU weakness vs Meta and the challenge of ad targeting on anonymous platforms
- •Directed share program to mods/users: potential volatility and “what could go wrong?”
- 1:17:18 – 1:22:49
Reddit’s ‘logged-out’ growth theory and the long-game Condé Nast spin-out narrative
Sacks proposes that a product decision around logged-out access could explain the sudden growth and may be tied to Google search visibility. They also recap an anecdotal ‘scheme’ narrative (shared by a former Reddit CEO) about how Reddit was structured and financed to enable a later spin-out from Condé Nast, culminating in today’s IPO.
- •Logged-out user growth: mobile app gating vs open web access tradeoffs
- •Search rank authority as a growth lever (and how toggles change top-line metrics)
- •Valuation framing: depends heavily on whether recent growth is repeatable
- •Spin-out/backstory: early sale price, cap table creation, outside investors, eventual independence
- 1:22:49 – 1:26:51
Apple cancels Project Titan: why the Apple Car never fit (and why AI does)
The episode closes with Apple’s reported shutdown of its decade-long car effort and reassignment of some staff to AI initiatives. The besties debate whether a car was ever strategically coherent for Apple, what might have caused the cancellation, and how Apple’s core strengths align better with AI than automotive manufacturing.
- •Project Titan shutdown and staff redeployment to generative AI
- •Skepticism: cars as a fundamentally different product category for Apple
- •Speculation: opportunity cost vs AI push and strategic focus
- •Counterfactual: Apple acquiring Tesla during the Model 3 era
