Introducing /multitask in the new Cursor 3 interface. Cursor can now run async subagents to parallelize your requests instead of adding them to the queue. For already queued messages, you can ask Cursor to multitask on them instead of waiting for the current run to finish.

Another way to parallelize work is with new and improved worktrees in the agents window. Run isolated tasks in the background across different branches. When you're ready to test changes, move any branch into your local foreground with one click.

We've also added multi-root workspaces for cross-repo changes. A single agent session can now target a reusable workspace made of multiple folders.

Download Cursor 3.2 to try these new features in the agents window:

sleep on cursor at your own peril.

nahhh this is crazy, might just have to circle back to cursor just for this!

that’s insane. Daily cursor update smacked today.

the part that always bites me running parallel agents: two siblings make contradictory architectural choices and you don't notice until merge. async UX only helps if there's a synthesis pass before the work lands. is /multitask going to surface conflicts, or just queue them?

The cross repo feature is sooo good Finally you can work on back and front without any friction Cursor is getting more and more interesting everyday

idk why people don't realise how good cursor is, how much better it is getting over the past few months! the UI is great familiar, you can granularly edit the codebase, you can choose models, you can plan first and then execute and it works great! it has never disappointed me after from Opus 4.5 launch

This will come super handy, thanks! <3

Everyone will now understand why @xai made the deal. @cursor_ai is leading the charge while @claudeai and @OpenAI are catching up. Cursor offers everything they do and is always a step ahead. We’re living crazy times.

I use a structured process to keep multitasking with AI agents from turning into chaos. Wrote some notes on vibe coding a game with Cursor for #vibejam:

The quota is too small, and it will be gone after a few uses. The self-developed model is really not good. Changing the code changes a lot of bugs. Hey, money is not earned in this way. Extra money to buy an API and use the model can remove a few dollars.

🤌

@asahiXXXXXXXXX Multiple branches at once? Codex can’t do that yet.

Multi repo, Multi task, Multi chat... I like where it is going haha

moving from sequential to parallel workflows is a huge unlock especially for multi-step builds

this is pretty cool. hope you guys lean heavy into the Grok thing. could be the perfect model for you

This is the parallel agent workflow I’ve been begging for. No more serial queue pain in big projects!

great change

I really love your new Cursor Agents, the IDE seems less and less important the more we progress. You're doing great work - keep on.

cooking

Now shipping faster than ever before!!! 💪 Just a genuine question, what happens if a task interferes/depends on another one still running?

This multitasking feature is huge. Finally, agents can actually run in parallel instead of waiting in line. Cursor is cooking.

Total amazed by this multitask feature... used cursor after months got curious to see what multitask does.. open one repo and asked to run 20+ agents and write complete test suite for frontend and backend.. and man it literally did ran 20 agents and composer 2.5 did really well

thank you for consistently leading the way with novel agent UX. trying to use this but it just makes the window hang when starting agents

Cursor just gave its AI agents ADHD and a Red Bull IV drip

很棒 不过我还没有开始使用

Multitasking + subagents = game changer for dev workflows

🥰 love it

Truly agentic is here, great job team

so cool-

I will @edwinarbus @cursor_ai

We run 12 parallel agents daily. Parallelism is the easy part. The real engineering is coordination: agent A writes a file agent B needs, two agents hit the same config, shared state drifts. Excited to see how Cursor handles that layer.

Async subagents work for me on 4 separate branches because the branches keep them from touching each other's files. Same workspace, two agents editing the same component, that's an hour of debugging where neither one understands what the other did.

bro my cursor is literally spawning subagents rn while i wait for the other one THIS IS INSANE

i want 1000刀 cursor