SF's highest-margin startups are data generation companies selling RL environments to AI labs
davidyin44 · x · 2026-11-15
A thread argues that some of the highest gross margin startups in SF right now aren't SaaS companies — they're data generation companies selling to AI labs.
Key points:
- Verifiable domains are getting crushed by AI: mathematics (formally provable with Lean) and software engineering (run the code, run the tests) advance insanely fast because they have a reliable reward signal.
- The next frontier is harder: GDPval-style knowledge work and long-running workflows that take humans days or weeks, with no clean verification mechanism.
- RL environments are the bottleneck: models are only as good as their data and training environments, and labs need ever more high-quality RL environments to keep scaling — physics, biology, healthcare are potential markets.
More from AGI Musings
- Checkout upsells are 100%+ of Amazon's retail profit — why merchants fear agentic commerce — davidyin44 · 2026-11-15
- "AI Writing Makes Everyone Sound Like the Same Intern": Founders Weigh In — MatthewChang · 2026-11-15
- "AI Can't Lie": Dev Explains Anthropomorphism Bias in Heated Thread — gerardsans · 2026-11-15
- Debate: AI self-reflection, not raw capability, may be the real danger — ZeroStateReflex · 2026-11-15
- Delip Rao asks: who is mirroring all of Hugging Face before it gets locked up? — deliprao · 2026-11-15
- Enterprise AI Needs Tools, Context and Evals, Not Omniscient Models — hwbhatti · 2026-11-15