Cursor was genuinely a good product in the beginning. It was easy to recommend because it combined a strong IDE experience with a practical AI workflow that felt predictable, powerful, and worth the subscription.
That has changed.
My biggest issue is that Cursor keeps walking back the things that originally made it appealing. The Composer agent, in particular, is not something I want forced into my workflow or near my codebase by default. If I want to avoid it, the answer i get is “switch to Max,” but that means spending dramatically more per request just to preserve the workflow I already had. I should not have to pay far more simply to avoid a model or agent behavior I did not ask for. (see previous feedback)
The same applies to newer models gated to Max usage. A request that used to feel predictable under the old credit/request system can now burn through many times more usage, often with less clarity around what I am actually getting in return. From a user perspective, the incentive feels obvious: push people toward workflows that consume credits faster.
The billing changes are also a major step backward. Moving more usage toward API-style billing, while leaving users to rely on Auto when they want to preserve credits, makes the product feel less transparent and less controllable. Auto may be fine for small edits or basic UI changes, but for serious engineering work I often want direct control over the model and workflow. I should not have to restructure a workflow I have spent over a year building just to accommodate new pricing mechanics.
I am not canceling my legacy $20 / 500-credit plan because it is still not worth giving up. But that is not the same as being happy with the product. It feels like Cursor tried to change the value proposition after users had already committed their workflows to it.
The feedback starts here:
I used to recommend Cursor without hesitation. I cannot and wont do that anymore.
Cursor as an IDE is not meaningfully better than JetBrains or Visual Studio on its own. The value was the AI workflow.
Cursor’s moat was the combination of strong models, predictable usage, and a smooth agentic development experience. That moat is shrinking quickly. Claude Code, Codex-style workflows, and other tools are getting close enough that the switching cost is starting to feel smaller than the frustration of staying.
The forced or default Composer behavior is especially irritating. It feels like a product decision that benefits Cursor’s model strategy more than the user’s actual workflow. The fact that disabling or avoiding it requires digging through settings, Googling forum posts, or changing plan behavior makes the whole thing feel even worse.
Right now the product feels less transparent, less predictable, and more focused on extracting usage than helping developers do meaningful work. Please give users clear model control, clear billing, clear limits, and the ability to disable unwanted agent behavior without forcing them into a much more expensive mode.