Sholto Douglas

“I think we're near guaranteed at this point to have effectively models that are capable of automating any white collar job by 2027, 2028 — or near guaranteed end of decade.”

Sholto DouglasResearcher

AI researcher, Anthropic (previously Google DeepMind)

Douglas caveated that white-collar tasks are unusually susceptible to current algorithms because they can be retried on computers and have abundant internet data, unlike robotics or biology.

“There is this whole spectrum of crazy futures. But the one that I feel we're almost guaranteed to get — this is a strong statement to make — is one where, at the very least, you get a drop in white collar workers at some point in the next five years. I think it's very likely in two, but it seems almost overdetermined in five.”

Sholto DouglasResearcher

AI researcher, Anthropic (previously Google DeepMind)

Douglas argued the decline would happen even if algorithmic progress stalled, because existing models are already good enough for many white-collar tasks given the right data; Trenton Bricken agreed with the five-year timeline.

“We haven't yet demonstrated long-running agentic performance. You're seeing the first stumbling steps of that now, and should see much more conclusive evidence of that basically by the end of the year, with real software engineering agents doing real work.”

Sholto DouglasResearcher

Researcher (RL scaling), Anthropic

By the end of 2025 agentic coding tools such as Claude Code and OpenAI's Codex agents were in broad production use doing genuine multi-step software engineering work, which meets the bar Douglas set. The related forecast he and Bricken made about reliable general computer-use agents ('go do my taxes') remained clearly unmet, so only the software-engineering half of their outlook landed.