Comment by cadamsdotcom

My hot take: Cursor is a bad tool for agentic coding. Had a subscription and canceled it in favor of Claude Code. I don’t want to spend 90% of my time approving every line the agent wants to write. With Claude Code I review whole diffs - 1-2 minutes of the agent’s work at a time. Then I work with the agent at a level of what its approach is, almost never asking about specific lines of code. I can look at 5 files at once in git diff and then ask “why’d you choose that way?” “Can we undo that and try to find a simpler way?”

Cursor’s workflow exposes how differently different people track context. The best ways to work with Cursor may simply not work for some of us.

If Cursor isn’t working for you, I strongly encourage you to try CLI agents like Claude Code.