Long context may help AI agents trace messy real code bugs
was tested with about 550,000 tokens from a real inherited software project to see whether helps with practical debugging. The input combined a Djan, React frontend code, outdated , logs, and notes.
The bug was a login loop that appeared after token expiration and frontend retries. The backend logs did not show one clear failure point.
The cause was split across AuthContext.tsx and .py, which are hard to bring together with normal chunking unless their connection is already known. This was not a local run; it used a hosted run because the was not ready for more than 500,000 tokens.
Key points
- handled roughly 550,000 tokens from a real project test.
- The input included backend code, frontend code, stale docs, logs, and notes.
- The login loop involved both AuthContext.tsx and .py.
- Normal chunking may miss links between files unless the relationship is already known.
- The test used a hosted run, not a .