Agent Builder Monitor Tab

What it does

The Monitor tab in Agent Builder gives the owner of a declarative agent usage metrics and user feedback for that agent, scoped to a date range. It's how a maker sees whether an agent is being used and where its answers are landing badly, without needing a separate analytics product.

Key facts

You reach it from the agent's Edit menu in Microsoft 365 Copilot, then the Monitor tab. It's per-agent, for the maker — not the org-wide adoption reporting in the admin centre.
Date range is Last 7 / 14 / 30 days or a custom start/end. Everything on the tab respects that range.
Overview metrics: Total sessions, Avg. DAU, Avg. user messages per session, Avg. session duration.
The Engagement chart plots sessions and users per day; the Knowledge chart breaks down which knowledge sources the agent actually drew on, one line per source.
Reactions (thumbs up/down counts and split) and Comments (written feedback with date and reaction) only show up if an org admin has turned on agent feedback sharing and users have actually left feedback in the range.

When to use / skip

Use it after an agent has been shared and running for a week or two, when you want to know if adoption is real and whether the knowledge sources you wired up are the ones being used. Skip it for a brand-new agent with no traffic — there's nothing to read yet — and don't expect it to answer tenant-wide "how much Copilot are we using" questions; that's the admin centre's job.

Configuration decisions

Whether to enable agent feedback sharing at the tenant level — without it the Reactions and Comments sections stay empty, so if you want qualitative feedback you need the admin to switch it on first.
Who gets Can edit (or ownership) on the agent, since that access is what gates the Monitor tab.

Gotchas

Empty Reactions/Comments almost always means feedback sharing was never enabled, not that users are silent — check the tenant setting before concluding the agent is fine.
Metrics are bounded by the selected range, so a quiet-looking Overview can just be a short window over a slow week; widen it before drawing conclusions.
The Knowledge chart tells you what the agent used, not what it should have used — a source you expected to see missing is a signal the grounding isn't working, not a display bug.

Consultant notes

Clients scoping an agent rollout usually forget the feedback-sharing tenant setting, then wonder why the Monitor tab looks empty at go-live. Flag it in the environment prep, not after.
The Knowledge chart is the useful one in practice: it's the quickest way to catch an agent that's ignoring a knowledge source because the source was misconfigured or the instructions steer around it.
It's maker-facing telemetry, so don't sell it to a governance team as their monitoring surface — they'll want the admin centre usage reports and Copilot control system, which are a different thing.
Adoption reads over a 7-day window will look thin for anything but the busiest agents; set expectations that 30-day windows are where trends actually show.

Worth revisiting once Agent Builder monitoring picks up export or alerting, which it doesn't have yet.