AI research agents move from literature search toward closed-loop discovery

Sakana AI’s Agentic Scientist and related systems use parallel coding agents, sandboxed experiments, evidence retrieval, baselines, ablations, and rubrics to automate parts of the research loop. Anthropic’s funding and compute commitment supports community activity in mathematics and science, but is not yet evidence of research impact; provenance, reproducibility, reliable evaluation, and human or laboratory validation remain necessary.