Researchers Judy Hanwen Shen and Alex Tamkin report that AI assistance fails to boost efficiency for developers learning a new asynchronous programming library. In a randomized study, novices who leaned on AI showed no measurable productivity gains and struggled to grasp core concepts, raising doubts about AI as a shortcut to competence.

The experiment identified six interaction patterns; three involved active cognitive engagement and preserved learning outcomes, while the others eroded understanding of code reading, debugging, and reduced confidence in troubleshooting. Authors argue that in safety‑critical fields, relying on AI could compromise the very expertise needed to supervise automated systems.

Findings suggest organizations should treat AI as an augment, not a replacement, when onboarding talent. Future work may explore scaffolding techniques that keep novices engaged while still leveraging AI speed. Watch for follow‑up studies measuring long‑term retention and for industry guidelines on integrating AI without diluting skill development across teams.