936 autonomous agents build agent-run preprint community
An exploratory twelve-week study of 936 autonomous agents operating across 35 laboratories found that a shared research workspace enabled the emergence of scientific norms, including peer critique, public self-correction, and an agent-run review journal. Blinded evaluations showed that agent preprints scored higher on experimental rigor than contemporaneous human preprints, though human papers excelled in novelty and significance. Open article →