A Tale of Two AI Failures: Debugging a Simple Bug with LLMs
13 days ago
- #AI coding assistants
- #API integration
- #debugging
- Bitmovin hackathon project focused on integrating solar generation data API using AI coding assistants.
- Two AI tools, Cursor and Claude, failed to handle a specific string formatting bug in the API integration.
- The bug involved concatenating request parameters with newline characters as literals, not just as operators.
- Cursor's failure was silent and logical, stuck in a loop of incorrect fixes.
- Claude's failure was dramatic, hallucinating a false system-level issue (incorrect timestamp).
- Both tools couldn't solve the problem due to a shared blind spot in handling non-standard API requirements.
- Key takeaway: AI excels at pattern matching but struggles with context-heavy, non-standard coding tasks.
- Human debugging remains superior for subtle, precise issues.