What's this about?
Foundation models have enabled AI scientists to automate increasingly complete research workflows — hypothesis generation, code execution, manuscript preparation. This paper (83 upvotes) argues workflow coverage alone is not enough, and presents an omni-modal, omni-discipline system that works across data types and fields rather than within one.
What exactly is the "workflow coverage isn't enough" argument?
It's easy to build a system that touches every stage of research and does each one badly. Covering the workflow demonstrates plumbing, not competence — the same trap as the AutoResearch diagnostic paper identified this week, where an agent can complete every stage while compounding errors through all of them. The omni-modal, omni-discipline framing is a claim about what real research requires: papers involve images, tables, equations and code together, and the most interesting findings often come from moving a method between fields. A system confined to text in one discipline can't do either.
Why is this worth noticing?
- It's part of a visible cluster this month — Apodex, Large Discovery Models, the AutoResearch diagnostic, today's Brain Researcher — all pushing past workflow automation toward whether the science is actually good.
- "Omni-discipline" is the ambitious half, and cross-field transfer is exactly what Ethan Mollick argued would be revolutionary if LLMs did nothing else.