WHAT WORKS
- code editing
- long context
I use it mainly for writing and editing code across a whole project, not single files. It holds enough context to know how the pieces connect, so I spend less time re-explaining the codebase before getting anything useful back.
Two things bother me. Usage limits run out fast on a heavy day. And it tends to over-solve: ask for a small fix and you can get a rewrite you did not ask for. Both are manageable once you learn to scope requests tightly.
Still the first tool I open.
What did you use before Claude? ChatGPT for general work, and Cursor with Copilot inside the editor. I still keep both around, but most of the actual building moved here.
What surprised you after a week with Claude? How much project context it actually holds. I stopped pasting files in one by one and started pointing it at the repo, and the answers got better instead of vaguer.
Who is Claude not for? If you want a tool that changes the exact line you pointed at and nothing else, this will annoy you. Same if you work in long unbroken sessions and cannot afford to hit a usage cap in the middle.
WHAT NEEDS WORK
- aggressive defaults
- usage limits
Usage limits are the main friction. On a long build day I hit the cap mid task and have to stop. The other one is scope: a request for one small change can come back as a refactor of files I never asked it to touch. Tighter prompts help, but the default should be more conservative.
- Ease of use
- Reliability
- Value for money
- Customisation
