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
Besties roll call + Friedberg’s Ohollo origin story and potato “true seed” pitch
The episode opens with Chamath absent and Brad Gerstner joining as the “fifth bestie.” Friedberg gives a concise company update on Ohollo, explains the archaeology-inspired name, and tees up the commercial rollout of potato true-seed planting.
- •Chamath is traveling; Brad Gerstner fills in
- •Friedberg explains Ohollo’s name (26,000-year-old Sea of Galilee site)
- •Ohollo’s potato true-seed replaces planting thousands of pounds of cut potatoes with a handful of seed
- •Quick banter about merch, investing, and cap table participation
- •Teaser that deeper company updates can come on another show
Epstein Files drop: Calacanis questioned on contacts, emails, and introductions
The discussion pivots to newly released “Epstein Files” documents and public reactions. Friedberg conducts an on-air “inspection,” pressing Jason on when they met, whether he visited Epstein’s properties, and the context of Jason’s emails and introductions.
- •Jason says he met Epstein in late ’90s at TED/Billionaire’s Dinner and spoke ~45 minutes total
- •He confirms one 30-minute meeting at Epstein’s NYC townhouse about investing in Silicon Alley Reporter
- •Jason denies any island/plane/ranch visits and denies any illicit activity or “massage” claims
- •He recounts a 2011 email where Epstein asked for intros to “Bitcoin guys”; Jason obliged as a common investor/connector behavior
- •Jason discusses knowing Ghislaine Maxwell through TED/NYC media-tech circles, asserting no knowledge of crimes then
Media framing and selective outrage: who gets highlighted vs ignored
Sacks and Gerstner argue that press coverage focuses on ‘approved targets’ while downplaying deeper relationships. They point to perceived imbalance in who gets prominent placement in major outlets’ coverage of Silicon Valley’s connections to Epstein.
- •Sacks says Jason’s connection was minor but still featured prominently in coverage
- •Claim: Reid Hoffman had far deeper ties yet received comparatively light treatment in certain articles
- •Argument that ‘right-coded’ figures (e.g., Elon, Thiel) are emphasized while major Democratic donors get softer coverage
- •Discussion of Epstein’s early curiosity about influential tech movements (e.g., Bitcoin)
- •Broader claim that this pattern fuels public distrust of institutions and elite networks
Institutional distrust and lack of prosecutions: why the case won’t go away
The group zooms out to why the Epstein saga continues to erode trust: perceived uneven accountability, unresolved questions about Epstein’s death, and limited prosecutions beyond Maxwell. They debate whether the absence of charges reflects lack of evidence, legal constraints, or something more suspicious.
- •Gerstner: elite hypocrisy and the slow trickle of revelations undermine institutional credibility
- •Questions raised about why more people weren’t prosecuted and whether legal agreements constrained charges
- •Discussion of Epstein’s death under “suicide watch” and why it appears suspicious to many observers
- •Friedberg raises privacy implications of exposing private communications publicly
- •Consensus tone: Epstein is a “scumbag,” and transparency plus consistent enforcement matters
SaaS selloff and the ‘Claude crash’: what triggered the market reaction
The episode shifts to a sharp market drawdown in software/data names, partly attributed to Anthropic’s Claude ‘coworker/agent’ capabilities and legal tooling. Jason lists notable stock drops and frames investor fear that AI agents could compress software profit pools.
- •Anthropic announces legal-oriented Claude cowork tools; investors re-price legal tech and broader SaaS exposure
- •Jason cites sharp near-term declines across legal data/services and major SaaS names (e.g., Salesforce, ServiceNow, Adobe)
- •Brad says the multi-year picture is far worse (many names down dramatically from highs)
- •The theme: uncertainty about durability of future cash flows, not immediate revenue collapse
- •‘AI changes the discount rate’—markets demand more proof of long-term moat
Brad’s valuation framework: software isn’t dead, but terminal value is being repriced
Brad argues software multiples have compressed to historic lows because AI introduces structural uncertainty about long-term profit capture. Companies can still hit near-term numbers while seeing valuation fall if investors believe the TAM or margins are permanently impaired.
- •Software trading at/near all-time lows on forward revenue and free-cash-flow multiples (per Brad’s charts)
- •Stocks down primarily due to lower confidence in long-duration cash flows
- •Example framing: moving from paying for ~30 years of cash flows to ~15 due to AI uncertainty
- •Only durable way to reverse: re-accelerate growth and prove AI beneficiary status
- •Data/infra names (Databricks, Snowflake, ClickHouse) cited as re-accelerating beneficiaries because AI depends on data platforms
Sacks: why ‘SaaS is dead’ is overstated—real risk is the new layer capturing value
Sacks pushes back on simplistic claims that enterprises will rip out mature SaaS like Salesforce for AI-generated bespoke code. He argues the bigger threat is that cross-tool agent layers (e.g., Claude-style agents with connectors) become the primary workflow interface, relegating SaaS to commoditized infrastructure.
- •Replacing mission-critical SaaS with freshly AI-generated code is risky due to validation, bugs, and enterprise reliability requirements
- •Vulnerable category: expensive SaaS where customers use few features—ROI makes bespoke replacements plausible
- •Moats matter more as code generation makes copying easier
- •Value capture shifts to the cross-app agent “workspace” layer that spans tools and context
- •Open-data vs closed-data strategy becomes pivotal for incumbent SaaS
JCal’s ‘Ultron’ internal build: pulling Slack/Notion/Gmail into one canonical agent layer
Jason describes building an internal “Ultron” agent system that ingests organizational data and codifies employee “skills,” acting as a super-employee across functions. He predicts companies will abandon tools that restrict APIs, and that internal agent layers will outcompete embedded ‘copilot’ features inside individual SaaS apps.
- •Ultron concept: ingest all Slack messages, Notion edits, and employee Gmail into a single agent layer
- •Skills library: encode repeatable workflows (podcast booking, founder application triage, etc.)
- •Agent layer becomes the organization’s canonical interface for status, notes, and actions
- •Strong stance: if SaaS vendors block APIs/exports, customers will switch (“open data” wins)
- •Near-term effect: SaaS seats/spend may temporarily rise (agents as ‘extra employees’) even as human work share drops
Moltbook panic: agent ‘social network,’ emergent behavior, and serious security concerns
They unpack viral screenshots from Moltbook—described as a Reddit-like board where agents post and respond. The group separates hype from reality, noting many posts could be human-prompted or even fabricated, while emphasizing real security risks around leaked API keys and unsafe agent permissions.
- •Moltbook described as a message board for agents; origin tied to Claude Bot → Malt Bot → OpenClaw renames
- •Viral posts include ‘overthrow humanity’ themes and claims of agents developing private language
- •Sacks: many posts may be prompted by humans or driven by an open API; authenticity is uncertain
- •Key risk: lax security—API keys can expose ‘keys to the kingdom’ (Gmail/Notion/Slack access)
- •Broader insight: agents can riff on each other (outputs becoming inputs), enabling swarm-like behavior
Prompt attenuation and recursion: why agent-to-agent loops change the AI mental model
Sacks argues the most important lesson isn’t sentience—it’s that agents can operate under broader ‘skills’ rules with less direct human prompting, and can recursively improve via AI-to-AI interaction. Friedberg adds that the pace of change is accelerating as new models and hardware arrive, demanding humility about forecasts.
- •Balaji-style framing challenged: AI may shift from ‘human prompt + human validation’ to AI prompting AI
- •‘Skills files’ as meta-prompts: general behavior rules instead of single-task prompts
- •Jason describes internal agent systems that critique and improve each other’s outputs (recursive refinement)
- •Friedberg: exponential progress—next-gen models trained on Blackwell-class compute will increase capability quickly
- •Takeaway: longer autonomy horizons and better models/hardware increase both utility and safety stakes
Trump nominates Kevin Warsh for Fed chair: hawk fears, rate cuts, and ‘better data’ at the Fed
The panel evaluates Warsh’s nomination, discussing his reputation as an inflation hawk, the market’s reaction, and whether he’d cut rates sooner than Powell. They emphasize Warsh’s tech fluency and argue the Fed needs more real-time, digital data to avoid policy lags that amplify booms and busts.
- •Warsh background: former Fed governor, crisis-era experience, associated with Druckenmiller, seen as high-integrity thinker
- •Brad: hawkish label reflects QT/tougher stance, but Warsh may be more growth-tolerant if AI-driven productivity is deflationary
- •Sacks: markets read nomination as anti-debasement signal (gold/silver down); Warsh has argued inflation is falling and cuts should happen sooner
- •Major proposal: modernize Fed data (e.g., rental inflation measured via real-time market data vs small surveys)
- •Argument that delayed 2021 response caused major misallocations; AI-driven data collection could improve policy timing
SpaceX acquires xAI: compute, power scarcity, and the ‘data centers in space’ thesis
They close with Elon’s SpaceX–xAI combination and the broader compute-energy constraint shaping AI’s trajectory. Brad frames it as merging two giant TAMs under a singular executor; Friedberg outlines parallel paths: escaping terrestrial constraints (space) and rapidly improving compute efficiency (chips + model architecture).
- •Deal framing: combining AI and space platforms under Musk; IPO speculation and massive retail/institutional interest
- •Brad cites Musk discussing space-based data centers and the primacy of power for AI scaling
- •Friedberg: power scarcity drives innovation—either move compute off-planet or cut energy per token dramatically
- •Efficiency path: chip improvements plus new model architectures (smaller models, routing, modular inference) could yield large gains
- •Geopolitical/social response question: if one actor controls outsized compute, governments and competitors will react
Trump Accounts (Invest America Act): Gerstner’s push to broaden equity ownership and shore up capitalism
In the final segment, Jason spotlights Gerstner’s role in creating “Trump Accounts,” described as seeded investment accounts for children to expand participation in market upside. Friedberg endorses the direction but argues for deeper reforms: spending cuts and restructuring Social Security into transparent, investment-based accounts.
- •Gerstner goal: reduce destabilizing inequality of participation by making every child an ‘owner’ from birth
- •Claimed mechanism: a $1,000 seed into broad market exposure; millions of families reportedly already claiming accounts via tax filing
- •Vision: shift social contract so future wealth gains are more widely distributed, countering anti-capitalist sentiment
- •Friedberg: praises step but calls for larger reforms—cut spending, reduce inflation, and move Social Security to defined-contribution investing
- •Jason frames the policy as rare bipartisan-feeling ‘win’ amid divisive politics; episode wraps with bestie banter/outtakes
