Open Agents to turn WorkMonitor’s built-in AI agents on, run any of them on demand, and see the findings they raise for a person to review.
Who this is for
Owners, by default. Opening Agents needs the permission that reads agents’ data, and the default Manager role doesn’t include it — only Owner’s full access does. If your organisation has granted that permission to another role, that role can open it too, but out of the box this is an owner-only screen.
What you’ll see
The header reads Agents, with the context line “Named, persistent agents · advisory & human-in-the-loop,” and, once any agent has run, a chip showing how many are switched on — for example, “3 of 7 on.”
Below that, a notice: “Advisory by design — a person always decides.” — “No agent moves money or auto-decides. Each one cites its evidence, degrades honestly without an AI key, and every finding is challengeable with a dispute.”
Next, four running counters: Findings (“raised for a human to review”), Time saved (“soft value · never dollarized”), Anomalies caught (“priced at detection time”), and Compliance blocks (“non-compliant writes stopped”).
The agent crew section lists every agent as a card. Each card shows the agent’s name, its rollout phase, and — for Burnout Guardian only — a Consent-gated badge, plus a short description of what it does:
Runtime probe — “Deterministic smoke agent proving the harness, scheduler, and console end-to-end.”
Manager Coach — “Three grounded things to act on this week — the insight layer, always advisory.”
Standup Scribe — “Turns your own day into a grounded standup — worker-first, self-serve, advisory.”
Timesheet Detective — “Flags and prices pre-approval timesheet anomalies for a human to weigh — never auto-rejects.”
Burnout Guardian — “Surfaces a consented, aggregate workload signal for a human 1:1 — never affect, never a verdict.”
Compliance Sentinel — “Flags monitoring configs that are likely non-compliant for the org’s jurisdiction (not legal advice). Preventive: blocks a likely-illegal policy write unless a human overrides on the record.”
Invoice Prover — “Assembles a client evidence pack (JSON/HTML/PDF) that substantiates a finalized invoice from stamped time records — advisory, unsigned, never moves money.”
Every agent ships switched off. Each card has a toggle, captioned On or Off, to turn that agent on or off, and a Run now button that runs it immediately regardless of the toggle. A card also shows when the agent last ran, or “Never run — scheduler may be off; run on demand.” if it hasn’t yet.
Beneath the crew, a Findings section lists everything the agents have raised, with severity, cited evidence, and controls to decide or dispute each one — see “Act on a finding.”
What happens next
Switching an agent on lets the scheduler run it automatically from then on; switching it off stops future scheduled runs (a run already in progress isn’t affected). Selecting Run now runs the agent right away, and a toast reports the result — a run with AI synthesis, or “Ran in basic mode (no AI key).”
If it doesn’t work
If you don’t have access to Agents at all, you’ll see “Agents are unavailable” — “You don’t have access to this workspace’s agents.”
If switching an agent on or off fails, you’ll see a toast: “Couldn’t update the agent”.
If running an agent fails, you’ll see a toast: “Couldn’t run the agent”.
Related
Act on a finding