I have been noticing recently that Cursor introduced their own model, which is pretty limited; it is very fast, but it is not smart. They might have to improve that, along with the agent selection methodology. When you select auto, you expect it to use the best agent possible and think of the money savings, but it does not always work as expected, sometimes providing bad code or code that has bugs. I think the auto mode selection can improve. Another concern is the pricing; we have been paying a lot for Cursor recently, which I feel has increased within the last few months, possibly due to our usage going up. We are paying an amount similar to a developer's salary for Cursor now. I believe all the needed improvements are primarily around the auto agent selection mode and pricing. If they could be more transparent about it, that would be appreciated. We only see the bill at the end of the month, and it is often a high amount, so transparency in the pricing would be very helpful for us as developers.
