So I really gave it a try. Not a huge try because I was honestly disappointed from the very beginning. But I did try it. The experience is terrible, to be honest.
I was expecting Claude Projects to be superior to Cursor Projects. Maybe the wait time after they announced it made me believe that they really nailed it and they were just fine-tuning it for general usage.
It’s a very poor DX, IMHO. There is the Agent Coordinator that does not coordinate anything. It just opens threads and passes over my same message. I was expecting a bit of prompt engineering fine tuning my bare request. I have to baby sit the freaking threads.
I am working on a knowledge graph that has a visual representation and I tried to implement a new feature that extends existing graph nodes. I have to keep my eyes on the threads and I barely moved the needle trying to implement the feature. I’m guilty of not sitting for 2-3 hours to get it done. I stepped away several times a day, but every time I came back, my input was expected. I needed to steer individual agents. As I said, the coordinator agent just copied/pasted my bad-written prompt. Yeah, I know models are good at figuring it out, but come on.
I did not know if I was on a branch, on which branch, or in a worktree either. What’s the point of a coordinator if I cannot do parallel work because I don’t know if I’m going to be messing stuff up.
Orca set the bar high, I know. Or maybe I just liked that DX in particular. But I feel Claude let me down here. It was a bad moment for letting me down because I was already considering downgrading my Claude Max plan. I just did it. I did not do it because of Claude Projects. I did it because SpaceX AI unified the subscription of X + Grok + Cursor and I found it particularly interesting to try out. So yeah, the disappointment kind of pushed me to try different things. Cursor Projects is way better. Even Grok Bot spinning Cursor Agents is better.