Agents using debugging tools drastically outperformed those that didn’t, but their success rate still wasn’t high enough.
Credit:
Microsoft Research
This approach is much more successful than relying on the models as they’re usually used, but…
Article Source
https://arstechnica.com/ai/2025/04/researchers-find-ai-is-pretty-bad-at-debugging-but-theyre-working-on-it/