The AI agent is an autonomous worker for your documentation. Where the [Writing Copilot](/ai/copilot) helps with the page in front of you and the [Workspace Assi...
The AI agent is an autonomous worker for your documentation. Where the [Writing Copilot](/ai/copilot) helps with the page in front of you and the [Workspace Assi...
The AI agent is an autonomous worker for your documentation. Where the Writing Copilot helps with the page in front of you and the Workspace Assistant answers questions interactively, the agent takes on recurring jobs unattended — sweeping a project for issues, keeping content fresh, and drafting pages — on a schedule and within guardrails you define.
You manage it in the Agent Control Center (/agent).
The control center at a glance
The Agent page shows live metrics across the top — Assistant sessions, Answer rate, Pending review (agent suggestions awaiting a human), and AI tokens — followed by the configuration sections below.
Task templates and scheduled runs
The heart of the agent is Task Templates and Scheduled Runs. Each task picks a type, a mode, a schedule, and a scope:
| Task type | What it does |
|---|---|
| **Freshness** | Find pages past their review cadence and flag or refresh them. |
| **Gap detect** | Surface topics that appear under-documented. |
| **Terminology** | Enforce preferred/banned terms across pages. |
| **Link check** | Detect broken links and references. |
| **SEO audit** | Check titles, descriptions, and metadata. |
| **QA audit** | Run a quality sweep against the five [scan dimensions](/ai/scan-and-content-quality). |
| **Review** | Move qualifying drafts into the review flow. |
| **Generate** | Draft new pages from a prompt or source. |
For each task you also set:
- Mode — Suggest (propose only) or Autonomous (act within scope).
- Toggles — Enabled, Email notifications, and Include drafts.
- Task locales — which languages the task covers.
Autonomy
How much the agent may do on its own is capped by the Maximum autonomy in your Workspace AI policy:
| Level | Behavior |
|---|---|
| **Suggest only** | AI proposes, humans decide — nothing is written. |
| **Draft for review** | AI drafts; humans approve through the [review workflow](/collaborate/review-workflow). |
| **Autonomous** | AI acts within scope without review. |
Content scope and automation safety
The Content Scope and Automation Safety section bounds what the agent can touch:
- Indexed locales — which languages are in scope.
- Included statuses — whether the agent considers Draft (and other) pages.
This keeps automated runs from straying outside the content you've approved for them.
Content quality reports
Configure a recurring Content Quality Report at a chosen frequency — Daily, Weekly, Monthly, or Manual — to get a regular read on project quality without opening the dashboard.
Recent runs and the review loop
Two sections close the loop on everything the agent does:
- Recent Runs — the live queue and execution history for the selected project, so you can see what ran and what it changed.
- Human Review Loop — pending agent suggestions that still need a reviewer's action (approve or dismiss). This is where Suggest and Draft for review output lands for sign-off.
A typical setup
A common, safe configuration for a growing docs team:
A QA audit task in Suggest mode runs weekly and posts findings to the Human Review Loop.
A Link check task catches broken links before readers do.
A Freshness task lists pages past their review cadence so owners can act.
A Weekly content quality report keeps the team aware of trends.
Governance
The agent is bounded by your Workspace AI policy: enable or disable the AI agent, set the Agent tier override, cap Maximum autonomy, and apply the workspace approval policy and escalation rule. Token use counts against your budget.