Where is this setting?
Maybe. Can’t you force it to tell you which model it is using? Usually its Sonnet 3.7 or 4 for me, but maybe its lying. When it does use gpt its obvious, but it also tells me now, so I am not surprised.
Yes this is what i was thinking too. They have a usage algo that basically will make the agents dumber for a specific period of time.
No - Specifically, I notice extremely laziness within the last 48 hours. This is in auto mode and specifically in agent mode as well.
Basically, I’m in a situation where its not actually doing anything anymore and defaulting or reverting itself behind the scnes (i think) to ask mode. So, I’ll say something like “yes perform the actions” liek an agent should.
Then it gives me back some sort of hypotehtical “template” (that seems new) consisting of a Summary, a very clear “how to fix” section and then a chart on the possiblities. It spends more time on re-summarizing what im telling it to do, as opposed to doing it.
use this in user rules
#Core Rules
MANDATORY: At the beginning of every task or step, you MUST clearly state the model name, model size, model type, and its revision (updated date). This requirement is non-negotiable and must be followed without exception.
whenever u see claude models u use that session. or else just stop and open new chat. its working for me very well
1 Like
Yes, the auto mode prompts too much for user approval.
I wonder if this is related to Cursor doing too much, spiking my requests & lines of code to to a daily all time high a couple of days ago. My Anthropic usage ran out faster than normal, and I’ve been managing this less intelligent model.
Wondering if this is related to the new Claude sonnet model’s tendency to go overboard..I wonder if Cursor is able to control the agent’s behavior to not do that..
Has anyone had luck with Cursor Rules actually improving the agent’s response?
I noticed this when using o3, it just does not want to follow instructions and tends to do not even the bare minimum
yes, could be.
I usually now step by step it now and only agent once im happy with the plan.
rather than watch it painfully spitting out a bunch of stuff that is not required or low quality.
That’s not strictly true…
https://platform.openai.com/docs/models/chatgpt-4o-latest
chatgpt-4o-latest is a surprisingly useful model in Cursor! (needs to be added manually)
1 Like
Thanks @mwjt42, yes true about the model.
If that example is how you normally write your prompts, then it’s no wonder why you’re getting bad results from the AI. Detailed prompts with correct spelling and grammar will get you much better results.
Also, in Cursor Settings, you can turn on Auto-Run Mode which may help with what you’re trying to do.
