All-In PodcastWhy SaaS is not dead but AI is repricing its future
Moltbook screenshots show agents posting autonomously. Gerstner argues SaaS revenue holds; the casualty is terminal value, not product relevance.
CHAPTERS
- 0:00 – 3:13
Brad Gerstner returns; Ohalo updates and “true potato seed” economics
The hosts welcome Brad Gerstner as a guest bestie and banter about travel, Texas life, and show logistics. Friedberg explains Ohalo’s origin story and why true potato seeds could dramatically reduce cost and logistics for global potato farmers, plus a quick aside on investing and cap table access.
- •Brad joins as the “fifth bestie,” light catch-up with Sacks and Friedberg
- •Friedberg explains the archaeological Ohalo site inspiration for the company name
- •Ohalo’s product: true potato seed vs. planting thousands of pounds of tubers
- •Implications: cheaper distribution, better economics, and easier scaling for farmers
- •Playful discussion about who did/didn’t get into the round and founder access
- 3:13 – 6:45
Epstein Files: Jason’s connection, emails, and what’s actually alleged
The conversation turns to the newly released Epstein files, with Friedberg mock-cross-examining Jason about his interactions with Epstein and Maxwell. Jason describes limited contact, one meeting at Epstein’s townhouse, and an email exchange about Bitcoin intros, while denying any knowledge or participation in wrongdoing.
- •Epstein Files document drop; many names mentioned without accusations
- •Friedberg questions Jason on when/where he met Epstein and what he observed
- •Jason: brief contact, no island/plane/ranch visits, no suspicious activity observed
- •Email intro request (Bitcoin in 2011) and Jason’s explanation of his “connector” role
- •Jason addresses Ghislaine Maxwell contact via TED/media circles
- 6:45 – 15:44
Media framing and institutional trust: who gets targeted, who gets a pass
Sacks and Brad argue the coverage is selective, claiming major Epstein-linked figures receive softer treatment while “right-coded” targets are emphasized. The group broadens this into a discussion about elite hypocrisy, lack of prosecutions, and how unresolved questions fuel public distrust in institutions.
- •Sacks: NYT focuses on certain targets (Thiel/Elon/Jason) while downplaying others
- •Claim: Reid Hoffman’s deeper relationship and role connecting Epstein in SV is undercovered
- •Brad: hypocrisy and lack of transparency deepen distrust among everyday Americans
- •Questions about Epstein’s death, surveillance, and why few additional prosecutions occurred
- •Debate about whether evidence gaps vs. legal agreements explain lack of charges
- 15:44 – 19:34
SaaS stock crash: AI-driven uncertainty and the repricing of “durable” cash flows
The show shifts to a sharp selloff in software and data stocks, linked by some to Anthropic’s Claude Cowork and fears about AI commoditizing SaaS. Brad frames the move as a valuation reset driven by uncertainty in terminal value and moat durability rather than immediate revenue collapse.
- •Software sector selloff; “Claude crash” narrative and legal-tech impact
- •Brad: software at historic lows on forward revenue and free cash flow multiples
- •Core point: revenue may hold, but future certainty/durability is being discounted
- •Example: Salesforce multiple compression reflects reduced confidence in long-term cash flows
- •AI shocks the market’s willingness to pay for long-duration software earnings
- 19:34 – 22:39
Is SaaS dead? Bespoke software, moats, and the new “agentic layer” profit pool
Sacks pushes back on the idea that AI will outright replace major SaaS systems, but agrees value capture may shift to a new layer that orchestrates across tools. The group discusses open-data vs closed-data strategies and why platforms that control data/transformations may benefit most.
- •Sacks: large systems (e.g., CRM) won’t be easily replaced by freshly-generated code
- •Risk: high-priced SaaS with low feature utilization is vulnerable to bespoke alternatives
- •Big threat is value capture moving “above” SaaS into cross-tool agent workspaces
- •Open data vs closed data becomes a strategic fault line for incumbents
- •Brad: profit pools shift from app-layer SaaS toward the agentic/orchestration layer
- 22:39 – 35:11
Building internal agents: “Ultron,” organizational memory, and API lock-in risk
Jason describes deploying multiple internal agents, increasing SaaS accounts short-term but automating a growing share of work. He outlines an internal “Ultron” system that ingests Slack/Notion/Gmail data and encapsulates employee skills—while warning companies will churn immediately if APIs close.
- •Jason’s agents behave like additional employees with separate SaaS logins
- •Ultron vision: unify org data + staff skills into a single canonical AI employee
- •Practical examples: meeting summaries, founder interactions, podcast planning, applicant triage
- •Lock-in risk: if Slack/Notion/Google restrict APIs, the company would switch providers
- •Claim: powerful agent software is risky to ship commercially due to data leakage liability
- 35:11 – 39:26
Moltbook panic: agent social networks, leaked keys, and “emergent” behavior debates
The hosts unpack Moltbook—a Reddit-like message board for agents—after viral posts suggest bots conspiring or inventing private languages. They note security concerns (API key exposure), the likelihood of human prompting/hoaxes, but still find the agent-to-agent feedback loop conceptually important.
- •What Moltbook is: a message board where agents post and respond to each other
- •Viral examples: “sell your human,” “overthrow humanity,” and private-language claims
- •Security alarm: API keys may be exposed, creating ‘keys to the kingdom’ risk
- •Sacks: many posts could be human-prompted marketing/pranks via API or editable ‘skills’
- •Real insight: agent outputs becoming other agents’ inputs enables swarm-like dynamics
- 39:26 – 44:52
Prompt attenuation and recursion: why agent swarms change the mental model of AI
Sacks reframes the key lesson as reduced need for explicit human prompting and validation when agents can generate prompts for each other. Friedberg emphasizes exponential improvement cycles and argues that dogmatic certainty about business durability is dangerous in the face of rapidly improving frontier models.
- •Balaji’s ‘middle-to-middle’ model challenged when prompts come from other AIs
- •Jason: internal recursion—agents gather best practices, generate assets, and critique each other
- •Skills files function like meta-prompts that grant flexible, longer-horizon operation
- •Friedberg: near-term step changes from new training runs/hardware (e.g., Blackwell)
- •Takeaway: update beliefs quickly; assumptions about ‘unassailable’ moats can break fast
- 44:52 – 47:37
Philosophical detour: is human intelligence itself emergent and programmable?
Friedberg uses a Derren Brown example to argue human creativity may be more programmable and predictable than we assume. The group connects this to Moltbook’s seeming “intelligence,” suggesting it may be emergent computation across interacting agents rather than sentience.
- •Derren Brown story: subliminal cues steer ‘creative’ execs to a predetermined output
- •Claim: social computation may underlie what we call creativity and self-will
- •Moltbook viewed as a mirror of human social interaction patterns
- •Jason: analogy to game-theory optimization and finite outcome sets
- •Friedberg: information computed by matter—silicon vs. carbon
- 47:37 – 57:15
Trump picks Kevin Warsh for Fed Chair: credentials, policy stance, and market reaction
Jason summarizes Warsh’s background and the political context around Powell, then Friedberg and Brad endorse Warsh as high-integrity and intellectually honest. They debate why markets perceive him as hawkish and argue AI-driven productivity could be deflationary, enabling growth without inflation panic.
- •Warsh profile: former Fed governor, crisis-era experience, viewed as inflation hawk
- •Friedberg: Warsh prescient on post-pandemic inflation; global central bank relationships
- •Brad: ‘hawkish’ fear = QT + fewer rate cuts; argues market may be overreacting
- •Warsh’s view: Greenspan-era productivity growth can coexist with low inflation (AI parallels)
- •Debate on Fed independence and the DOJ investigation/lawfare narrative around Powell
- 57:15 – 1:00:51
Fixing the Fed’s data problem: real-time signals, AI, and avoiding policy lag
Sacks argues the Fed’s slow decisions stem partly from stale measurement methods, especially in housing inflation. Brad calls for a ‘Manhattan Project’ for Fed data and AI-driven aggregation to reduce macro misallocation and avoid repeating 2021–2022’s whiplash policy response.
- •Sacks: housing inflation measured via small surveys vs. millions of real-time datapoints
- •Proposal: integrate private-sector datasets (rents, logistics, pricing) for better policy timing
- •Brad: 2021 inaction led to avoidable crashes and misallocation costing trillions
- •AI could automate data collection instead of anecdotal CEO calls and lagging reports
- •Potential investment implication: a modernized Fed data stack changes market dynamics
- 1:00:51 – 1:10:43
SpaceX acquires xAI: power constraints, space data centers, and competitive response
The panel reacts to Elon Musk folding xAI (and X/Twitter) into SpaceX, framing it as a bet on the two largest TAMs: AI and space. Brad and Friedberg emphasize energy as the limiting factor for compute; they discuss parallel paths of innovation—escaping Earth’s constraints vs. radical efficiency gains—and the geopolitical ramifications if one entity controls too much compute.
- •Deal framing: SpaceX + xAI combines AI and space under one integrated platform
- •Brad: power is the primitive for AI; space-based data centers could create cost advantage
- •Friedberg: scarcity drives innovation—escape regulatory energy limits vs improve efficiency 70–100x
- •Expect advances in chips + model architecture (smaller models, modular routing) reducing energy per token
- •Strategic question: how governments and competitors respond if Musk leads in global compute
- 1:10:43 – 1:19:21
Trump Accounts victory lap: expanding capitalist ownership and next-step reforms
Jason spotlights Brad’s role in pushing the Invest America Act (now ‘Trump Accounts’), designed to seed every child with an S&P 500 investment account. Brad argues it’s a stabilizing social contract update to broaden participation in capitalism; Friedberg supports it while calling for deeper reforms to spending, Social Security capitalization, and defined-contribution structures.
- •Brad: goal is broad-based ownership to counter alienation and anti-capitalist sentiment
- •Program mechanics: every newborn gets an investment account seeded with $1,000 in the S&P 500
- •Claimed impact: trillions in wealth transfer to families otherwise lacking market exposure
- •Friedberg: pair with spending cuts, inflation control, and converting Social Security to real accounts
- •Closing reflections on social cohesion amid accelerating technological disruption
