Both the GitHub Copilot CLI and Visual Studio Code now support calling custom subagents from the main thread. Which means you can start to get into some pretty interesting orchestration scenarios.

@github @code Well done! Exactly what I was looking for.

@github @code Ty!

@github @code If I issue a single premium request that invokes subagents, am I billed for a single premium request or 1+n for each subagent?

@github @code You guys absolutely cooked with the subagents. It's a game changer!

@github @code Excellent video Burke! Thanks for sharing!!

@github @code ty franz!

@github @code Waiting for sub-agent beast mode lol

Subagent orchestration is where things get really interesting for distributed teams. Imagine each team member having specialized agents that coordinate async, handling code reviews, docs updates, and deployments in parallel. This is the kind of workflow we're exploring at Remoty for remote teams. The future of dev tooling is multi-agent.

@github @code Loved it!!! I just finished uploading my brand new GitHub Copilot 101 to YouTube, and I covered custom agents really quickly - LOVED the orchestrator use case!!! Here's my new copilot 101 if you'd like to check it out:

@github @code I haven't installed your agents yet, but tried your example 'Use Opus, Gemini, and Codes to review this code...blah blah'. Once it ignored and just reviewed with current model other time it replied with image. I see you went to terminal before you did it. Only CLI?

@github @code CLI only atm if you are just doing it in the prompt. In @Code you need to specify custom agents with the model in the front matter...

@JamesMontemagno @github @code /fleet does feel like magic.

@github @code Thanks to your work on the ACP we can support this in Clairvoyance. The agents will specialize out the work.

@github @code Couldn't find the link in the comments for some reason so there goes:

Installing your agents Burke, I noticed that I don't have the 'memory' tool available. I remember seeing it in the built-in section at some point but it's not there atm. Using the latest VSCode Insiders. Is your memory reference to an installed MCP Server outside of the built-in ones?

@github @code No it should be there. Just enable in settings ?

@github @code That was it! Thank you. Should've checked ...

@github @code This was a pretty good tutorial but shouldn't the product just do this vs. needing to teach people how to do it.

@github @code Hey Burke, this is an awesome video! As an SDET, I would like to adapt these Orchestrators to Agents specialized in Testing. Could you please share the repository with the Architect.agent. md documentation? Thank you for sharing!

Oh this multi agent is a dream, I can already see asking for a design improvements using both claude and gemini3 and then use the best of them all, but this kind of use will be very expensive so out of reach, but super powerful. Even multiple agents for code review and having them check each other is amazing, but still very expensive these runs.

@github @code Does each subagent count as additional premium request ?

@github @code I am testing your custom agents and for some reason the orchestrator never delegate tasks to subagents. It ends saying it doesn't have permission to edit anything and stops. I have activated in the settings the memory, subagents, etc.

@github @code This is so close. But my orchestrator seems to fail or give up a lot. It will ask me questions for the coder model or say “going on to next phase” then just stop for no reason. But this is a really nice framewok

@github @code You could try a bigger model as the orchestrator or try enabling more tools.

@github @code Trying with opus 4.5 right now to see if it will keep it running. Just so risky to test at a 3x lol Also is your design agent allowed to code or just build structure to hand off? When I asked codex what it thought, it wanted design to just write up docs an the structure, no code

@github @code this custom subagent thing is a game changer, can't wait to see what kinda automation workflows people come up with now. Good for enterprise customers and bad for people complaining about copilot.

@github @code The implication that we might want to look at things like gastown I also think is a bit incorrect; we should be using Process Coordination Theories like Pi Calculus, Petri Nets, Fork-Join Parallelism, etc. to discuss optimal ways to coordinate agents. This is a solved problem.

@github @code which transcription tool are you using in the cli? @burkeholland

@github @code The main reason I've been running multiple subagents is just to save the context window. I use claude-opus a lot and it fills the context window pretty fast.

@github @code It would be great to be able to assign models to subagents in @code just with prompting, as you do in your example in @github Copilot CLI

@github @code Superb! Thanks for sharing.

@github @code Is there a way to get these .md files for reference?

@github @code Excellent video