Skip to main content

Agent Home

Agent Home is the cockpit for the autonomous side of msg.ZenTestAI: one page per application that answers, at a glance, the three questions you actually have about an AI agent that watches your suite — Is my suite healthy? What needs me? What did the AI agent do?

It is available once autonomy is switched on for the application, and appears as a dashboard page for each application; the Home landing page embeds the same view.

Is my suite healthy? — the plan traffic light​

Every execution plan under the agent's watch gets a traffic-light row based on its latest completed run:

  • Green — the latest run passed. Nothing to do.
  • Amber — the latest run failed, but the agent has already repaired it ("Failed — I repaired it"). The fix is live; the plan stays amber until you review it or a fresh green run confirms it.
  • Amber, pulsing — someone is actively on it: either the agent is repairing right now ("I'm working on it") or a person has taken the item over ("In process").
  • Red — the latest run failed and the agent could not (or was not allowed to) resolve it. This is the state that needs you.

A repaired plan does not silently turn green: only a fresh green run of the plan confirms the repair. What you can do earlier is accept the repair — the row then shows green with a small repaired badge, so it stops competing for attention with the genuinely open items while staying honest about the fact that the plan failed and was fixed by the agent.

Plans that are being worked on right now — by the agent or by a colleague — link straight to the live session: clicking the row opens the running AI Assistant conversation so you can watch or join in.

What needs me? — the needs-you queue​

Failing plans the agent could not settle land in the Needs you queue. Each item explains why it needs you in the agent's own words — for example "I have a fix ready — accept it to go live", "This looks like an application bug, not a test problem", or "The application has changed — the test needs updating" — together with the failing tests and error categories.

Depending on the situation, an item offers these actions (names as they appear in the UI):

  • Accept — take the item over, or apply a fix the agent has proposed but not applied itself.
  • Keep — confirm a fix the agent already applied and keep it.
  • Revert — throw an applied fix away and restore the previous state of the test.
  • Discuss — open an AI Assistant session about the failure and work it out in conversation.

While you (or a colleague) are on an item, it shows an In process chip and stays off the open queue, so two people never pick up the same problem.

What did the AI agent do? — timeline and daily report​

The What I did timeline lists the agent's activity: repair tasks and their outcomes, automatic runs that came back non-green, quarantine changes — and, just as important, what it could not do. Completed repairs appear with a short recap and Accept / Discuss actions, so reviewing the agent's overnight work takes a minute, not a log dive.

Use the day selector to switch between Today — a live view that is recomputed on every visit — and earlier days, which show the persisted morning briefing for that date. A health-trend strip shows how many items needed you on each of the recent days.

Permissions and configuration​

Agent Home reads the same data you can already see elsewhere in the tenant; the queue actions need the same edit rights as changing the affected tests manually. Autonomy itself — which applications the agent watches, and how much it may do on its own — is configured per application in the tenant's administration area.