A useful first machine-identity workflow is small enough to understand from beginning to end. Choose one agent, one task, and one receiving tool. Setup time depends on your account, permissions, and integrations, so treat this as a checklist rather than a fixed-duration promise.
1. Choose the workspace and owner
Use the business workspace responsible for the work. Confirm which administrator owns the agent and who will review its activity. Keep a clearly labeled test configuration separate from any production role that can perform consequential actions.
2. Define a narrow task
A good example is preparing a summary of approved sample documents for a human reviewer. State what the agent may read and what it may produce. Do not grant sending, payment, or record-changing permissions merely because those tools are available.
3. Register and connect
Create the agent identity using the supported console or API workflow. Configure the credentials and scopes required by the receiving tool. Use the current documentation for exact endpoint fields and package methods rather than copying commands from an older release article.
Confirm that the receiving integration enforces the intended scope. A permission written in a profile does not automatically protect an external application that never checks it.
4. Run and inspect
Submit sample work and inspect the available task history and receipts. Check that you can identify the agent, task reference, captured inputs, reported output, and human review event. Make any missing evidence explicit before a reviewer depends on the record.
5. Test an exception
Try an out-of-scope request or a controlled tool failure. Confirm that the action is rejected or routed for review as intended. Then test changing or revoking access and check what happens in the connected system.
Document who can stop the workflow and how active work is handled. Keep provider credentials and downstream sessions in the response plan.
Expand only after the first handoff is clear
Once another person can explain the evidence and decision path, decide whether the workflow needs more tools or agents. Explore human review of AI work, machine identity, and agent workforce management to plan the next step.