Team patterns for AI work

A team of agents needs explicit ownership. A shared chat transcript is not enough: it can hide which worker executed a tool, which step failed, and whether two workers repeated the same action.

Use a job ledger

{"job":"customer-research","session_id":"team-job-7","owners":{"research":"agent-a","verification":"agent-b"},"steps":[{"name":"research","status":"pending"},{"name":"verification","status":"pending"}]}

Keep the ledger in the host system and use Zambo receipts as execution evidence. The session trail records the calls made through the MCP connection; it does not replace your team's authorization or review policy.

Assign read and write boundaries

Let one worker gather read-only facts, a second worker verify sources, and a human or explicitly approved worker handle consequential actions. Zambo's result status and approval fields should be preserved rather than flattened into a success label.

Handoff format

• Stable session ID.
• Last completed receipt URL.
• Exact next step.
• Known missing evidence.
• Owner responsible for the next call.

When the final worker finishes, run verify mode and return the receipt trail with the answer. Connect Zambo →