Cursor Composer 2.5 feels better at adding code than fixing bugs
works well when the job is clear and focused on building something new. It can follow a direct goal and keep producing code. The weakness appears during and work.
Once it forms an idea about what caused a bug, it may keep following that idea instead of stepping back and checking other possible causes. The resulting fix can look logical and sound convincing, but only because the starting assumption was wrong. The real bug can remain.
These wrong fixes also add cost: more code, more complexity, more tokens, and more context. After several rounds, a lot of time can go into building solutions for a problem that was never properly understood.
Key points
- is useful for clear new-feature tasks.
- It can struggle when the task is or .
- It may stick with its first theory about a bug instead of reconsidering.
- A fix can sound right while leaving the actual bug untouched.
- Wrong attempts can add code, complexity, tokens, and context.