AI coding agents can modernize research software but can't judge if the science is right

A joint project between OpenAI and academic researchers demonstrates that coding agents can modernize outdated scientific software, delivering performance speedups of up to 60 times. While these tools significantly accelerate development, researchers caution that the agents frequently produce convincing but inaccurate code. This shifts the expert's primary responsibility from manual coding to the rigorous verification of AI-generated results, highlighting a persistent limitation in the technology's ability to ensure scientific validity.