Popular posts

Great points. from my experience:

Hallucinations: Claude Code is the most "cautious"—it often runs terminal commands to verify files exist before editing. Cursor is faster but more prone to inventing imports that aren't there.

Limits: Claude hits token/rate limits much harder and is pricier. Copilot is the "unlimited" budget option, but the logic is shallower.

Sessions: Claude uses CLAUDE.md for project-wide memory, which is great for persistence. Cursor handles the immediate UI context better but gets "lost" if the chat session grows too long.

Errors: Claude handles terminal/build errors natively (it tries to fix its own crashes), whereas Cursor still mostly relies on you to feed the error back into the chat.

I definitely find Claude more reliable for complex logic, even if it's more restrictive. What’s been your experience with the "loops" or session timeouts?

I'm using Claude code web for 2 months - many errors, many times it guessing instead of check Claude.md. From the last "innovations", you can't send image to it image without context, you'll get a continuous api error and probably will start a new session. Fun thing, after a huge session, it started to make mistakes one after another. When I asked why, it answered, sorry, I honestly tired, you maybe too, just rest and start a new session)