Quickstart: your first agent

Put an AI agent to work safely: configure it, run it, and review what it proposes before anything changes. This is the orientation; the click-by-click guide lives in the agent-designer guides.

Note

This is an orientation to the flow. For exact screens and field names, follow the links to each component and to the full guides.

An agent acts as its own service account, not as you. Before it can write anything, it has to run in bounded or autonomous mode and hold a write-granting role. Start in propose mode so every write waits for your review.

What you’ll end up with

A configured agent you can run on demand, that hands you a reviewable proposal instead of changing your data on its own.

The path

1. Design the agent. Give it a charter (its instructions), grant it the tools it needs, and choose an autonomy mode. Start with propose. The designer warns you up front if the agent could not actually use a tool you granted. See Agents and Skills and knowledge.
2. Run it. Dispatch the agent and watch the run. It reads what it needs on its own; anything that would change data waits for approval. See the agentic loop.
3. Review the proposal. A propose-mode agent hands you a Plan: a proposal attached to a record that you approve or reject before anything happens. See Plans and governance.

You know it worked when

The run finishes as completed, awaiting approval, or failed, and a proposing agent leaves you a plan to review instead of changing data directly.

Now do it

• Read the full step-by-step in the agent-designer guides.
• Understand what governs an agent’s writes in the Safety model.