Let me talk to subagents (and let sibling agents talk to each other)

Ideas Feature Requests
,
1

Feature request for product/service

Chat

Describe the request

I’m a domain expert who directs Cursor agents rather than writing code by hand. A lot of my work is: I talk to a parent agent, it spins up subagents for pieces of the job, and I stay in the parent thread.
That’s a workaround. The parent is a switchboard I didn’t ask for. I often want to talk to the specialist that’s actually doing the slice of work — to correct it, answer a domain question, or steer it — without waiting for a summary back in the parent. The parent should still be able to talk to that same agent. Right now I can’t talk to subagents, and other sessions can’t either.
I’ve been using tools (Codex, Grokbot) where a session can start a new agent that is a real thread I can keep talking to, and those agents can talk to each other as well as to me. That pattern is more useful for how I work than “helpers that only report to the parent.”
What I’m asking for:

  1. Subagents (or newly spawned agents) should be first-class chats I can open and message, not fire-and-forget workers.
  2. The parent agent and I should both be able to talk to that same agent.
  3. Sibling agents should be able to exchange messages when they’re on the same job, instead of everything funneling through me or through one parent.
    I’m not asking to replace subagents. I’m asking for a peer-agent mode on top of them: spawn a real thread, keep it visible in the Agents list, and let me and other agents in that job talk to it.
    Why it matters: my workflows already assume specialists. Cursor’s current model forces me to only steer the parent, then reverse-engineer what the subagent did from a summary. Direct access would cut that round trip and let me apply domain judgment where the work is actually happening.
4

Hey @bunst, thanks for the suggestion. A lot of this is closer than it might look!

Projects works the way you describe for your first two points. A coordinator agent plans and delegates, the agents it spawns are real threads that show up in your Agents list, and you can open any of them and message it directly. The coordinator is kept in the loop when you do, so you and it can both steer the same agent.

Agents messaging each other directly (sibling to sibling) is not there today. Messages flow between the coordinator and its agents, so the coordinator is still the relay for cross-agent handoffs. The shared project context helps a bit here since every agent in a Project reads and writes the same notes.

That said, I’d still encourage you to give Projects a try and see how it works for your desired workflow!