OpenAI Field Report: AI agents rewrite fragile research software, but researchers must rigorously verify
OpenAI and academic partners found that AI coding agents can accelerate and maintain aging research software, but the burden shifts from programming to verification. In eight case studies, agents modernized tools, some running over 60 times faster, yet they often behaved confidently while producing incorrect code. Researchers emphasize the need for independent validation and scientific oversight.
OpenAI
Anthropic

