What happened
A research paper surveys how language models could autonomously design and execute scientific experiments, from forming hypotheses to analyzing results. This is still mostly theoretical — the paper is arguing what's possible, not showing it working at scale in real labs.
Why it matters
The paper is asking whether the next generation of AI can move from answering questions to actually doing science: designing experiments, running code, interpreting results, deciding what to test next. Right now, this exists mostly in vision form, not in working systems. If it works, labs could run experiments continuously without researcher intervention. If it doesn't, we'll learn which parts of science require human judgment in ways AI can't yet replicate.
The signal
Look for the first published papers describing autonomous agents actually running multi-step experiments in a real lab (not simulation), with results that a human researcher had to verify but didn't have to design.