I have the same thing as @jrista constant “Your conversation is too long. Please try creating a new conversation or shortening your messages.” in 1.4 - I never had this before and I use a very strict workflow with what I do (same few prompts in a cycle to work through a task list). Now i can’t even get passed one prompt, from a new chat with my spec document attached (only 300 lines long). I have a whole team (enterprise edition) experiencing the same thing. This is starting to get unusable and has rendered it from being my daily driver, to a chore. Pushing me back into just doing it all myself. The last few updates to this app have completely ruined the experience.
- It rarely uses to-do lists and when it does it uses them wrong, I get better results with an attached markdown file with a task list in it and simple .cursorrules. What’s the point of this functionality if it barely works? It adds extra tokens to the context for very little results.
- It fails to use mcp tools anymore without explicit instruction, and when it does it uses them wrong a bunch of times before eventually figuring one of them out (usually executing the wrong one because it confused itself). It used to use them really well and I haven’t changed my servers.
- It gets caught in loops doing the same prompt over and over again.
- It keeps asking me to store memories of memories it should already have or almost identical to previous and things that are already in the .cursorrules so don’t need to be remembered again and fill up the context
- It randomly drops files off the context meaning I have to add them back
- It always re-adds the active tab to the context which I don’t want it to do.
- I use it in editor mode in zen mode but now it constantly opens all files in another side panel.
- Constant “conversation is tool long” warnings EVEN when on my first prompt in a new chat with barely any context
- Requests take ages to run, tools take ages to execute, responses are massively long and verbose, all of which mean it takes many times longer to do simple things compared to v1.2.x and prior performance
- Many more issues which I can’t remember off the top of my head but as a really heavy user of the application my list is getting much longer with each release.
I am assuming these are changes under the hood to reduce the token usage and improve the cost scaling of the product but it is severely gimping the real USP which should be improving code delivery. I would suggest as a business owner myself you survey your customer base more to get better feedback about user experience as it seems you are not aligned with the customer base.
Whatever decisions you are making with the product direction at the moment, they are the wrong direction. You have taken something which was an exciting tool to use to an over-bloated mess with useless additions and gimped main usp functionality. Unless the main agent actually goes back to producing at least passable quality results without all the friction introduced recently then I for sure will be looking at alternatives for my team. Looking around the forums it seems many people want the same things - consistency and results.