Are we automating science? No. We aim to reduce the effort scientists spend working together.
The move. This is not a CI pipeline where resumes drop in at one end and papers roll off the other. It is a human-AI partnership with mandatory stopping points. Every computed step ends in a boxed call-out with two parts:
The frame: a 14-step loop, each step small and checkable. Define the area → define “recent” → list best papers → extract strengths → assemble critics → search the literature → pick the open problem → write paper[0] (title, abstract, elevator — if these cannot be written, go back) → write paper[i+1] → offline human read → apply critics → discuss with humans → apply fixes + zero-flip self-test → ship or loop. Humans receive succinct conclusions they can inspect, dispute, and change before moving onwards.
One process rule for team sessions: while the team is together, do not polish. Joint sessions seek the big picture; mark polish spots with a to-do and divide them at day's end. Polish applied while the structure still moves is polish thrown away.
Newell's FAQ. Doesn't strengths-first drive a career into a rut? No: personal strengths choose the tools, but the field's open problems choose the targets. The hammer is aimed, not wandering — and the literature step forces contact with material outside prior experience.
Combines with: every other trick — the pause boxes are what make 1, 2, 3, and 6 safe to automate.