Agent mode quality is down bad

Hi guys,

I’ve been a Cursor user for over a year now, and the past few months for the agent mode have been extremely deceiving.

It constantly loses sight of the goal, tries to implement things you haven’t even discussed, edit irrelevant files, stopped doing proper research of the codebase before proposing code changes, etc. That’s on top of certain features being purely gone (custom modes, @ a specific function, etc) and the plethora of bugs being introduced all the time (changes not applied after clicking on Apply in Ask mode, Agent continuing even after clicking the stop button, etc).

I’ve slowly been using the Agent mode less and less and started going back to the Ask mode, but even it is becoming more and more irrelevant… although I’m using models that are supposedly more and more performant.

I used to love Cursor, and I’m now using the AI features less and less and strongly considering ditching it. Only the Tab completion has remained amazing.

I believe it might be time to tackle the tech debt and put more emphasis on ensuring the good quality of the features that are being delivered, rather than focusing on the velocity alone.

Please, I don’t want to stop using Cursor.

1 Like

yeah, the agent mode is driving me crazy, the auto model is soooo incredible terrible that i’m seriously considering switching the tools, to either the new neovim thingy or windsurf (which is just cheaper and does the same)
but seriously its hard to not start hating the cursor model. even the worst junior i’ve ever had managed to follow easy tasks which the auto model just refuses to. it also just ignores rules and just implements stuff i didnt ask for in the worst bad practice. And its kinda sad because it used to work pretty okayish a few month ago.

i absolulty feel you.