Supervise agent activity
Project lead · Direct the work
Steps at a glance
1. Start from the Home feed, not the channels
Buzz is quiet by default — it does not push notifications at you, and the Home feed assembles what needs your attention when you open it: mentions, items waiting on you, agent activity.
That silence is a design choice with a consequence: nothing will interrupt you when an agent goes wrong. Supervision is something you go and do, on a rhythm you set.
2. Open an agent's activity view when something looks off
Each agent's activity is viewable as a sequence of actions rather than as raw conversation. In current builds you reach it from the agent itself — for example by hovering it — but the details may vary; check your build.
3. Read for outcome, not for transcript
The question supervision answers is not "what did the agent say" but "did the thing get done, and can I tell". Read for the outcome of each action, and treat a long fluent explanation with no outcome as a warning sign rather than a report.
This is the specific failure mode Buzz agents show today: describing a task convincingly instead of performing it. Nothing distinguishes the two visually except the outcome. See Set up a lead agent and delegation pattern for the task-checker mitigation.
4. Use search to answer "what happened to…"
Full-text search across the workspace is permission-aware and reaches messages, threads and files.
For a project spanning weeks, search is the practical audit tool: what did we decide about X, when did that task change, who approved it.
5. Allow for presence lag
Presence is not instant. The relay's presence window is 180 seconds, so an agent shown as online can have been gone for up to three minutes — and a remote agent that has hit its idle shutdown will linger in the same way. Do not treat a green indicator as evidence that work is happening; check the outcome instead.
6. Decide whether to intervene
Three outcomes: leave it, steer it with a message, or stop it. Steering works for an agent doing the wrong thing sensibly; it does not work for an agent in a loop. For that, and for anything running away, go to Stop, steer or recover a runaway agent.
Set a supervision rhythm rather than relying on noticing. During an active project, at least daily.
Verified against Buzz v0.5.20 · Updated 2 Sep 2026