I’m using Agent largely in Auto mode but I’m a bit annoyed by its behavior. I’m working on code changes and appreciate the agent explaining changes and giving me the option to review, accept (or drop) changes and with the development server running these changes are available immediately for testing.
However I might be in a session for a while and the behavior suddenly changes. Code changes are no longer applied but rather are highlighted in the chat and then require me to apply myself slowing the progress as a result.
Is this due to Cursor changing the underlying model for whatever reasons? If not what is it? Do I have any control over this?
That is exactly what it is going on. It’s one of the downsides of Auto. The best way to get around this is to use a consistent model that is very cheap or free. I think GPT-5-mini is free as of now. But if you have to use Auto. Create some cursor rules like Create files and change any files to implement the suggested changes. Do not just show the code in chat, unless User requested to not change files in chat. A cursor rule is basically text that is added behind the scenes with every request.
Ok, not ideal but thanks for the clarification. I’ll try that Cursor rule and see if Auto improves. I basically stumbled upon Auto as I was on Claude-3.7 as opposed to Auto but ran out of credits (and bonus credits) and didn’t want to use 2 weeks so found Auto.
I have an API key for Gemini 2.5 Pro. So can I run the agent on Gemini (as opposed to Auto). How does that compare with the Pro license? Hypothetically speaking could I run Cursor via Gemini?
I think when you use API keys, you lose the tab autocomplete. I am not 100% sure on that. There may be other features that don’t work with API keys. Did you ever give it a try?
Hey, hypothetically you can, but you’d also lose other features like the Cursor tab, agent mode, background agents, and many other useful functions. If you have a Pro subscription and are using your API key, you won’t lose any features included in your plan, though there could be issues on the provider’s side of your API key.
I tried the rules path but still ran into situations where the agent wouldn’t update files for me.
I’m using the Gemini API key and use the Agent in Auto but strangely I’m still at 0% of your usage limit even though I have used the agent a couple of hours a day since the reset (28/8). How that possible?
so found Auto.