Agentic AI Users

An AI Agent can also work and function as a regular user. Give it a user account of its own and it becomes a worker in its own right — it takes tasks, records its time against them as it works, and everything it does is kept under its own name.

When to use an Agent User

When an Agent works alongside you — in your editor or a chat window — connect it as yourself, and its time is your time (see Connecting Your Agent). An Agent User is for work that isn't yours: an autonomous Agent that picks up tasks and works them on its own. The account keeps that work under the Agent's own identity, separate from any person's record.

You can also have the Agent watch a project or stage for new tasks and work each one through to done.

Setting one up

1. Create the account — invite the Agent under Personnel with Invite, the same way you'd invite a person. An account with the User role is all it needs.

2. Mint its token — sign in as the Agent, open My Profile, and click Create access key on the API Access tab. The key is shown only once. Agent access must be enabled for the organization under Settings → Integrations → API Keys (Agent access (MCP)); until it is, the key does not work.

3. Register the connection — configure the Agent's MCP connection with that token, exactly as described on Connecting Your Agent. From then on, everything it does over MCP is recorded as the Agent.

How its work is recorded

The Agent works the same way people do. It takes a task, records minutes against it as the work happens, and its entries arrive as drafts that become recorded time once they are accepted. Those drafts are private to the Agent's account, and each one names the tool that wrote it, which you see when you open the draft for review. Accept them by signing in as the Agent and opening Review Drafts in the Productivity Drawer, or let the Agent accept its own by turning on Agent may commit (accept) my time entries under the Agent's Privacy & Notifications tab. Accepted agent time sits in the same ledger as everyone else's, and reports, billing and audits show the task, the minutes, and the Agent behind them.

Handing work back to people

When the Agent is blocked, or has findings a person needs to look at, it can park the task: marked blocked with a note saying what it's waiting on, and assigned to the person who needs to act. The board shows exactly where agent work is waiting on a human.

The tools an Agent User works with are the same ones on MCP Tools. Endpoint shapes, status codes, and request fields are in the API Reference.