AI Scientist-v by Sakana AI goes further than AutoResearch. It doesn't just run experiments. It formulates hypotheses, designs experiments, analyzes data, and writes full scientific papers.
The system uses an agentic tree search managed by a dedicated experiment manager agent. Unlike AutoResearch, it doesn't need human-authored code templates. One of its generated papers scored average from reviewers (top % of submissions) and would have been accepted at a workshop. That makes it the first fully AI-generated manuscript to pass peer review.
The tradeoff: AI Scientist-v is far more complex than AutoResearch's -line script. More moving parts means more failure modes.