Buzz Agent Teams: accountable operating units for human-agent work

A Buzz Agent Team is a structured group of agents and humans assigned to a continuing function or bounded job. The useful unit is not the number of agents in the sidebar; it is the combination of owner, lead, members, channel, scope, state, and proof.

For business work, a Team should answer four questions before execution starts: who owns the outcome, who can act, what evidence is required, and who is allowed to inspect the result.

Generic bot rosters collapse accountability. When every task goes to one all-purpose agent, failures become hard to assign and repeated owner intervention becomes normal.

The Team model gives a business a repeatable way to separate strategy, building, QA, content, deployment, customer handling, and proof inspection without pretending that agent activity is the same as accepted output.

A practical Team record should name the lead, member roles, channel or job room, allowed tools, forbidden actions, expected heartbeat, acceptance criteria, proof location, and escalation path.

The record should also identify owner-held decisions. Anything involving spend, legal claims, customer contact, deletion, agent promotion, or provider writes should fail closed unless the owner or configured authority has approved it.

A Team existing in a registry is not execution proof. Proof requires signed messages or runtime receipts from the workers, artifacts they produced, checks that ran against those artifacts, and an independent verdict from a different identity.

For parallel work, the evidence should show distinct assigned lanes and timestamps that overlap or advance independently. Without those receipts, the safe claim is that the Team is configured, not that it operated.