Cursor: I've been using Cursor and had to reinstall recently. Almost accidentally agreed to data sharing, which now is a compulsory must to agree on to use (or must be manually disabled later). Almost missed it, as I'm happy to share usage but not my code (A lot of the code I write have bits I would like to remain private, and anything I'm happy to make public is already public). Not super happy with this.
Their agent mode is also not that great, it either overshoots or misses because it doesn't get the context. Worse, it will refactor and delete other parts, such as comments that I put there as guides for myself.
Was happy to use it in chat though - really sped up my troubleshooting, and that itself made it worth the price. Also, the latest models are actually quite a lot better than the older one, especially for novel code.
Copilot (on VS Code): Stopped using it after I found cursor. The earlier autocomplete models were nice for boilerplate but now that I have used the newer models (deepseek, gemini 2.5 models, sonnet 4) those feel really bad, and not useful for novel code.