New Worktree workflow seems like a downgrade

I’ve been using the worktree feature for running parallel agents on my app development and I really liked it.

I treat them like seperate employees - give each one a feature that they can develop, once they are done I can test the feature in isolation by applying it to the worktree, if it’s not how I expected, I undo the apply and let the agent improve it while I test the other agents features.

That way, each feature gets a seperate commit and I can test them in isolation. That helps me move forward way quicker.

With the new worktree update, each time I try a workspace command, it takes an agent atleast 10-20 seconds to do what I’ve done before with a single click. This makes that feature unusable to me, as I cant even be sure the command is executed correctly.

Maybe I’m missing something, but where is the benefit of having it that way/what was the idea behind it? It just consumes unnecessary time and tokens and I dont see any benefit over having just a git command executed.

Hey, thanks for the feedback. You’re definitely not alone, in the Cursor 3 Worktrees announcement thread Cursor 3: Worktrees & Best-of-N a few users describe the same issue. Switching to the agent workflow feels like a downgrade in speed, token usage, and reliability compared to the previous implementation.

The team is aware of this feedback. Colin has already confirmed known issues, including /worktree plus Plan Mode, and mentioned that improvements for worktrees in the Agents Window are coming. There’s no specific timeline yet, but your report helps with prioritization.

Let me know if you notice anything else on this.