Agentic coding tools such as Claude Code, Cursor and Codex are now part of the daily workflow of research groups willing to use them seriously. Used well, they can migrate legacy build systems, write numerically validated simulation code and maintain CI across multi-language stacks. Used badly, they produce plausible nonsense at scale. This talk distils 12 months of hands-on experience, including more than 2,000 Claude co-authored commits across 26 repositories, into practical patterns and anti-patterns: how to structure tasks, what context agents need, and how to verify their output when correctness matters.
[more]