> ## Documentation Index
> Fetch the complete documentation index at: https://docs.forge.shieldtp.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Operate AI Engineer

> Choose an execution mode, review eligible plans, monitor runs, and handle human handoffs.

AI Engineer performs eligible ticket work in the background. It can begin as a reviewable assistant in front of your existing helpdesk and expand as your runbooks, connected systems, and operational confidence improve.

## Choose the automation boundary

Administrators configure which tickets are eligible and how far AI Engineer may proceed.

| Setting             | Behavior                                                                                    |
| ------------------- | ------------------------------------------------------------------------------------------- |
| **Runbooks only**   | Background execution is limited to work governed by an applicable runbook.                  |
| **Full autonomous** | Additional eligible work can use the autonomous path configured by the organization.        |
| **Pilot**           | The autonomous path stops for review before execution.                                      |
| **Auto**            | The autonomous path can continue without that review step when all other controls allow it. |

Availability, role and system policy, provider support, and ticket-specific conditions still apply.

## Review a plan

The review view can show the proposed plan, ticket activity, ticket events, quality judges, and the runbooks or skills used.

Before approving:

* confirm the account and ticket;
* check that the plan addresses the current request;
* verify the tool and system assumptions;
* look for customer communication or change-approval requirements;
* identify a clear success condition and handoff path.

**Approve** allows the run to proceed. **Reject** returns the work to a person; include a reason that helps improve the next plan.

## Monitor runs

AI Engineer can show states such as **Claimed**, **Awaiting approval**, **Ready to execute**, **In progress**, **Scheduled**, **Handoff requested**, **Completed**, **Needs human**, and **Failed**.

Treat **Needs human** as an intentional safety outcome, not automatically as a failure. Read the reason, the evidence gathered, and any partial work before taking over.

## Respond to outcomes

<Steps>
  <Step title="Verify the source record">
    Confirm that the ticket status, notes, communication, and other expected writebacks appear in the PSA.
  </Step>

  <Step title="Resolve the immediate issue">
    Complete any requested human step or recover the run.
  </Step>

  <Step title="Improve the repeatable path">
    If the same gap will recur, update the runbook, skill, mapping, Availability, or connected-system setup.
  </Step>
</Steps>

<Warning>
  Interactive approval settings do not gate AI Engineer. Configure tool Availability for the Helpdesk surface to control unattended access.
</Warning>