A Nature news feature published April 13 reports that human scientists substantially outperform today's best AI agents on complex, open-ended scientific tasks, even as researchers have embraced AI tools at scale — the share of natural-sciences papers mentioning AI grew roughly 30-fold from 2010 to 2025. The finding complicates AI-as-researcher narratives: despite AI-augmented scientists publishing more and receiving more citations, there is still little evidence AI improves core scientific productivity on hard problems. The report arrives as labs including Sakana AI and several university groups continue pursuing autonomous AI research systems, and as DOE's Genesis Mission funds 24 organizations to apply AI to scientific discovery.