20VC: Why "Pacing the Frontier" is BS | Instinct Raising $1BN at $10BN & Meta Launches Muse | Miro Sells for $1.35BN After a $17.5BN Valuation | Mistral Raises €3BN & Could Sam Bankman-Fried Win His Freedom?

Summary

The conversation opens with Dario Amodei’s “pace the frontier” framing and quickly becomes a debate over which AI risks are real versus politically effective. The speakers say cyber risk is genuine, open-weight models make guardrails porous, and the most important danger is losing control of recursive self-improvement. They also think the rhetoric will trigger hearings, regulatory scrutiny, and attention from the public, even if markets barely react. From there, they assess Meta’s Muse as technically strong but still lacking a clear killer app, and they are skeptical that standalone AI assistant startups can win long term against platform incumbents with superior infrastructure and model access. The episode closes on valuation compression in software, arguing that once growth slows, companies like Adobe and Canva get judged on cash flow or EBITDA rather than narrative, while Mistral’s giant raise is interpreted as more about European AI sovereignty than true frontier-model parity.

Notes

• Dario Amodei’s “pace the frontier” warning is treated as a debate about infrastructure, energy, and frontier-model providers.
• The episode frames the next battleground as AI assistants: startup-side Instinct versus Meta’s Muse.
• Dario’s stated risk buckets are cyber attacks, economic disruption, and loss of control of AI.
• The speakers dismiss unemployment as a weak reason to slow invention.
• Cyber risk is treated as real, but no longer containable in principle.
• The core technical concern is loss of control over recursive self-improvement.
• Jason argues that Anthropic’s rhetoric will catalyze repeated congressional hearings.
• The public reaction is framed as existential fear rather than a niche technical debate.
• Open weights are described as making guardrails ineffective because dark versions can be recreated.
• Market reaction to the AI safety debate is described as mild: semis dipped a bit and cyber stocks rallied.
• Jason mocks the tiny market move relative to the headline extinction risk.
• Meta’s Muse is praised as software that works immediately and feels unusually polished.
• Muse is shown doing practical tasks like reservations, email, story collection, and rebuilding a website.
• Jason estimates comparable AI site-builder delivery costs at roughly $3 to $4 per user.
• Meta’s infrastructure and its own LLM are presented as the main reasons Muse feels fast.
• Jason says there is still no obvious killer app for Muse or the broader assistant category.
• Harry flags latency as a genuine issue for Instinct, with responses sometimes taking minutes.
• The assistant category may exist, but the UI remains awkward and the winner is unclear.
• Instinct is rumored to be raising $1 billion at a $10 billion valuation.
• Jason says he would not invest because Meta has the infra and LLM advantages.
• He argues the broader threat is competing against an incumbent that can subsidize the app with enormous infrastructure.
• Speaker B says Meta should make Muse broadly accessible across Messenger and iMessage.
• Instinct is framed as a venture bet that may exit via acquisition rather than IPO.
• Jason warns against venture bets that absolutely depend on M&A.
• The speakers think the risk-return math gets worse fast as Instinct’s valuation steps up.
• Harry says the consumer pull is unusually strong, citing more than 1,000 DMs from a single Instagram reel.
• Menlo Ventures is said to be modeling 100 investments aimed at $25B+ exits.
• Jason cites a world model with 8,125 tech companies above $25B, up from 2,310 a decade earlier.
• Miro is described as a 2021 high-flyer that sold to Bending Spoons for about $1.3B after a $17.5B valuation.
• The stale 2021 valuation is presented as making a sale almost inevitable.
• Bending Spoons is portrayed as a tougher operating owner that can force cash-flow improvement.
• Harry says Bending Spoons’ acquisitions have brutal churn but pair that with price increases and aggressive cost cuts.
• Jason says Miro still had a strong business with roughly $600M ARR, high-single-digit growth, and positive cash flow.
• The discussion turns to SBF’s appeal, with an Eighth Amendment argument against an enormous fine.
• Speaker B agrees white-collar crime should be punished but says 30 years may be too much.
• WordPress/Automattic is criticized as using its leverage against the open-source ecosystem.
• Jason says Automattic would have been better off without venture capital.
• He later argues venture-backed open-source businesses must be ruthless to avoid losing control of monetization.
• The conversation contrasts late-cycle businesses with frontier bets like Instinct and Jeff Dean’s spin-out.
• Harry says a major CIO told him the current cycle is more extreme than the last 30 years.
• Speaker B compares AI to Y2K and dot-com mania, but with much larger capital intensity.
• Mistral’s €3B round is framed as Europe’s largest-ever tech round.
• The round is described as more about AI sovereignty than frontier-model competition.
• Jason questions whether Samsung-led capital should be treated as a clean market-price signal.
• The valuation is portrayed as strategically driven rather than purely market-comped.
• The Adobe segment starts with a complaint that AI ARR can rise while total ARR still falls.
• The host and guest peg Adobe’s market cap at roughly $100B to $105B on about $25B of revenue.
• Jason thinks Adobe could trade roughly flat in three years.
• Adobe is treated as a durable cash-flow compounder unless ARR actually evaporates.
• Jason says Adobe can either keep running its long cash-flow playbook or find an unexpected AI breakthrough.
• Canva is described as still growing, but slowing from 30% to 20%.
• A key valuation rule is stated: above 30% growth, revenue multiples work; below 30%, EBITDA multiples take over.
• The valuation transition can take years and requires substantial cash flow growth just to keep the stock flat.
• Stripe is cited as unusually able to weather slower-growth valuation pressure because of its cash generation.
• Vanta is pitched as an agentic trust platform for compliance and continuous control monitoring.
• Deel is described as managing hardware, software, and IT access across 130 countries from one system.
• Framer is pitched as an AI website platform with CMS, SEO, custom code, enterprise hosting, security, and 99.99% uptime SLAs.