Anthropic internal poll: staff expect AI to do 80% of model-building by end of 2027
Afinetheorem · x · 2026-10-17
Anthropic's internal employee survey and blog report "The AI expectations gap" (Oct 16, 2026; authors include Szymon Sacher, Charles I. Jones, Anton Korinek):
- Share of model-building: median staff expectation is AI does 80% of frontier-model building by end of 2027 (median 51% today, 95% by 2030).
- Expectations gap: 91% of staff (vs. 39% of US adults) expect AI to make a Nobel-level discovery by 2030; among US adults using AI at work, 26% say a task takes ≤25% of the time with AI vs. 59% of staff.
- Own jobs: over half of staff expect AI to do ≥80% of their current work by 2030, vs. 5% of US workers.
- Unemployment perception: those expecting AI to take more of their own job anticipate higher unemployment — and this link is strikingly similar for staff and the public.
Context: Anthropic's Michael Sellitto told the IMF that Claude could lead 80-90% of frontier model training by next year, up from 3% in April.
More from AGI Musings
- Walter Isaacson: Steve Jobs would be appalled by today's out-of-control agentic AI — johncoogan · 2026-10-17
- Ex-Meta researcher rbhar90: anthropomorphizing Claude is dangerous, it's AlphaGo-like RL not sorcery — rbhar90 · 2026-10-17
- Recommended read: The economics inside AI matters as much as AI's economic impact — SuB8u · 2026-10-17
- Anthropic exec: Claude now writes 80-90% of the company's internal code — GregCook2011 · 2026-10-17
- AlphaFold already won a Nobel; why do only 58% of Anthropic staff expect Claude to match it by 2027 — tszzl · 2026-10-17
- IBM's 1979 rule resurfaces: a computer must never make a management decision — sanderssays · 2026-10-17