Claude Code has made me more careful about what I consider “Done”.

One thing that changed for me after using Claude Code regularly is how I define a finished task. A clean diff and passing tests are useful, but they don't always tell you whether the original problem is actually solved. I've had changes that looked correct in code review but still missed the real behavior I was trying to fix. So now I try to verify the outcome separately from the implementation whenever I can. For UI work that might mean checking the rendered result. For backend changes it might mean reproducing the original failure and confirming it no longer happens. Claude has made implementation faster, but it has also made me less willing to treat “the code changed” as proof that the work is complete.