Content
@
https://opensea.io/collection/dev-21
0 reply
0 recast
2 reactions
androidsixteen
@androidsixteen.eth
Does anyone else feel like Cursor Agent mode is a regression from Composer? I feel like I had a lot more control over Composer and the steps felt more “atomic”. Whereas Agent mode doesn’t know when to stop and keeps digging even after it does the intended ask (like the meme below) I could also be using Agent wrong 🤷
9 replies
3 recasts
27 reactions
Matt
@mattlee
I feel the exact opposite. I think thee kind of posts are people being mindless. It is pretty easy to keep track of what it is doing and tell it when it does something wrong. If you want to go crazy just do it on a new branch. Different workflows work for different people though
1 reply
0 recast
3 reactions
Antidote 🎩
@0xantidote.eth
I am not mindless. There has been multiple times when Cursor decided to change code not related to my prompt, sometimes even deleting parts without mentioning it. Yes, you can track in the codebase what the changes are, but you just don't expect it to just delete functionality without even mentioning it.
2 replies
0 recast
1 reaction
Matt
@mattlee
Didn't mean to call you out specifically. Different people's brains work differently and the agent works well for me but maybe not as much for you. I do think many people vibe code mindlessly and create problems and are posting stuff like that. Don't think you're in that group
1 reply
0 recast
0 reaction
Royal
@royalaid.eth
I think you can both be right about Cursor, the choice of model matters a lot. Specifically a lot of people are seeing perf. regressions around moving from Claude 3.5 to 3.7 because 3.7 is super eager to change a bit of shit without your direct input BUT if you get aggressive with your rule files it gets quite a bit better
1 reply
0 recast
1 reaction
