After more than 6 months of continued fighting against persistently higher error rates, performance issues, downtime and rising costs I will be leaving Cursor at the end of August.
I have made heavy use of Cursor in this time, building complicated systems with decreasing levels of accuracy and support across all models in that time. There have been a few recent months that topped $1250 in usage fees despite less valuable output.
My experience with Cursor used to be great, it used to reduce friction and speed things up, now it causes friction and slows me down. For me a threshold has now been reached. The tipping point this morning was being billed for unused inference during a major outage that the team failed to report on the status page.
You’ve wasted my time and money with poor comms and buggy systems for the last time.
I will be going to Windsurf, where the bills are more reasonable and the team are at least somewhat more accountable for problems with the tool they provide.
For a long time I didn’t want to make this decision, it’s disruptive and annoying, but that’s what companies like Cursor rely on as they raise prices, they expect it to be sticky enough that you won’t leave.
There is a limit to all things, and billing me for something I didn’t use because of poor systems design is unacceptable.
You guys have a lot of lessons to learn.
I hope other people follow suit, as regular feedback and user complaints do not seem to be enough for you to get your act together. It doesn’t have to be to Windsurf as long as it isn’t Cursor.
Cats would be envious of the number of lives you’ve managed to use up without dying, maybe that luck will run out soon, as it should.
