Address a coding agent by name in a live session
Coding agents now sit in the session roster alongside the people on the call, and you can hand one a task from a to-do, the chat bar, a highlighted passage or the session brief. Anything aimed at someone else's machine waits for their approval.
Reaching a coding agent from a live session used to mean clicking through and pushing a slice of the meeting over to Claude Code, with no record of what came back. Agents are now participants you can talk to by name.
Agents in the session roster
Available agents appear beside the people on the call, each attributed to the machine offering it. You can see who brought which agent before you hand it anything.
Four ways to hand off a task
You can address an agent by name from a to-do row, the session chat bar, a passage you have highlighted in the transcript, or the session brief. Pick the surface where the work came up instead of switching context to start it.
Approval for someone else's machine
An ask aimed at a teammate's agent lands on their machine as a request and runs nothing until they allow it. Nobody's laptop starts working because someone else typed a name.
Agent work view and end-of-call digest
Agent activity gets its own view next to the transcript and the summary, so you can follow what each ask turned into while the call is still running. When the session ends, Chief writes a digest of the agent work, and the web app shows it alongside the rest of the session.
Offers and findings during the call
Chief offers to help when it spots something it can take on, and researches on its own during the session. Offers and findings collect against the session itself, so two people's devices reconcile into one account of what happened.
Start your next session and check the roster for the agents your machine is offering, then send one a to-do and watch the result land in the agent view.