Subagents just got a big upgrade. Previously, all subagents ran synchronously (the parent agent had to wait for each one to finish before moving on). Now subagents can run asynchronously, so the parent keeps working while subagents handle tasks in the background.
Subagents can also spawn their own subagents, creating a tree of coordinated work. This opens up bigger tasks like multi-file features, large refactors, and complex debugging where research and fixes can happen in parallel.
We’ve also improved subagent performance since 2.4, with lower latency, better streaming feedback, and more responsive parallel execution.
We’d love your feedback!
How are async subagents handling your larger, more complex tasks compared to before?
Have you noticed the performance improvements in subagent latency and streaming?
What kinds of tasks would benefit most from deeper subagent trees?
If you’ve found a bug, please post it in Bug Reports instead, so we can track and address it properly, but also feel free to drop a link to it in this thread for visibility.
I must have missed something, I’m using Cursor 2.5.17 and I still can’t get a subagent to run in the background, or have the coordinator/subagent pattern working. Do you have any tips to get this going?
I find it much more repeatable to tag every agent in the starting prompt and say “Orchestrate among these subagents to complete the plan/task/etc.”
Recently, I’ve seen significant improvement in the rules helping with subagent orchestration. Because of that, I’ve put together this orchestration rule that gives all the context the agent needs to start orchestrating subagents.
You don’t have to use this, I just opened it up to make it helpful to others. But you can also copy this repository as a template by instructing your agent in Cursor:
Copy the subagent/rule setup in this repo https://github.com/tmcfarlane/oh-my-cursor but set it up like.... <insert any preference>
It would be really handy if we could use subagents with composer 1.5, it’s my go to model and it can’t even spawn subagents, only models charged on API can, I’d rather pay API costs on composer 1.5 and have access to the features I’m paying for.
is it supposed to just work if you tell the agent to parallelize, or do you need to structure the prompt a specific way? the docs don’t really explain the handoff between parent and child
Yeah nothing I do can seem to trigger async mode unfortunately. It would be especially helpful if I asked a question if it didnt cancel all the subagents so the main agent could answer it
@charles I’m checking on this with the team. Adding notes in the description seems to do a good job of telling the subagent to run in the background, but is_background: true does not.
I ran it to build a feature I explicitly mentioned to create subagents and make it run parallel they are good. But I see the parent agent is not verifying the work it is missing the proper verification.
I created feature adding both my UI and API in same workspace and task to plan and add test cases also sub agent worked well but that is the only thing verification is what I didn’t satisfy
Closing this thread since it has been a while since this was announced. Any new questions, or loose ends from this thread should be filed as new reports in the appropriate forum category!
