I’ve been using Cursor for last couple of months now - loving it. Statistically I may as well be one of the highest number of hours clocked user. I used to pay $40 instead of $20/month for increased fast usage.
Last few days, I’ve been getting more and more frustrated. Cursor is becoming unusable in my opinion.
“@Codebase” is gone for Chat - I actually thought they’d introduce it to the agent mode, but no they just got rid of it. Now manually select all your context. This was one of the reasons I even moved to Cursor.
Severe downtime for Anthropic’s models. I don’t use anything other than Claude-3.5-Sonnet (now 3.7), but last few days it was just dead because of the rate limiting issues. Not Cursor’s fault but my mind already started looking at alternatives - as a startup I’m sure they don’t want people having ideas to churn right? I didn’t even bother working on my projects for those 3 days.
Overall, I don’t think Cursor is going at the right direction. The correct direction - again, in my own opinion - would be go default to 1 mode, which is agentic, and has all codebase knowledge. If Claude models are your most popular (which I’m sure it is), then do whatever you must but make sure it never goes down - you will definitely lose users in such a competitive market.
You may feel differently about Cursor but I love this tool and hope the best for the team, I hope my thoughts provide a good feedback value for you all. Thank you for the amazing product, I’m rooting for you.
That’s why open source is so important. We’re heavily relying on AI now, so a fallback to “running it on your own” for whatever AI service you use (coding, writing, etc) is so powerful
So, I am not the only one. I thought I would test it and see if it could do a project from start to finish without writing in code. Nope. Earlier this week and all last week, things were working great. As soon as Claude 3.7 was released, Cursor stopped working and hasn’t worked since. I canceled my subscription to Cursor. Time to write my own
In Ask mode, I can still use @codebase. In Agent mode, the agent is supposed to handle this automatically, so you shouldn’t need to use @codebase (but I haven’t personally verified if that works as intended).
Perhaps providing the exact file names, function names, and other specific names to the Agent/Edit would be helpful.
Also, it seems like Edit mode is similar to the Normal mode of Composer in versions prior to Cursor 0.46.x; it only modifies the current file. (I rarely use Edit mode, so I’m not entirely familiar with it).
I actually do this a lot, but mostly when I am telling it to give me a one off script to do other maintenance tasks on the file system. such as:
“Give me a .ps1 to move files from A to B, rename them [blah], and write a log to log_file, and format it with this information. Fully comment your code, then enter a description in the [sqliteDB] and add invocation. Create a menu and give me a prompt for filename, a list of [something].”
and other things.
I often have it write some python thing to do a task, and I always tell it to wrap any python into a .ps1 and invoke the .ps1 (this is because no matter how many ■■■■ times I try to get it to do python on the cli – it often reverts to thinking its on linux, so it F’s up the invokations, and attempts to && multiple commands…
Even if I explicitely have told it in whatever rules, to use Windows.
If I wrap it in a .ps1, it never F’s up the python runs…