Verticals

Autopilot over copilot: Keith Rabois on which vertical AI companies survive the frontier-lab moment

Key points

Key takeaways from the Verticals interview with Keith Rabois, managing director at Khosla Ventures (October 2026):

  • Autopilot beats copilot.: Vertical AI companies that move towards autonomous completion of work are more defensible than copilots; Rabois also channels Sam Altman — bet on foundation models improving and compound that into customer value, not on models flatlining.
  • Contracts are the proof point.: At portfolio company Spellbook, 90%+ of US contracts are 'cut and paste and shuffle'; a Fortune 100 company reported $21m of outside-counsel fees saved over three years, with autonomous contracting moving to the core of the product.
  • Rogo owns investment banking.: 29 of the top 30 investment banks use Rogo, an illustration of vertical application dominance; Rabois argues mundane junior-banker work is exactly where AI gives analysts 'superpowers'.
  • Frontier labs swallow thin layers.: Cursor's acquisition was lucrative but its independent future looked bleak without investing ~$1bn in own-model bets; compounding data loops, non-accessible data (e.g. state-legislature monitoring where attending a hearing is the only way to get the record) and corner cases are the real moats.
  • The 2030 test.: Rabois works backwards from whether a company is a durable, independent business in 2030 — fast growth today (the 'industry GPT' cohort) does not guarantee survival, and he is bearish on most AI roll-ups and on legacy vertical SaaS surviving the transition.
  • Changed his mind on enterprise.: He now bets on AI-native services that replace entire departments — citing Comp, which replaces human-resources functions with AI — and on 'one app' ambitions like Ramp becoming the CFO's control surface for the whole finance function.

More top stories