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

# Run in production and adopt

> What has to be true before people can rely on an agent in real work, and how to tell live and adopted from merely deployed.

Deployed isn't the same as adopted. An agent counts as live when real, eligible work succeeds through it again and again, people know how to work with it, and every exception has somewhere to go.

## The six conditions

Agree these with the people who'll accept them, and set the thresholds and the observation window before release.

| Condition | The evidence | Who accepts it |
| - | - | - |
| **1. Real work reaches the agent** | An authorized live scope, the production version, and real cases you can trace | AI Builder & Deployer, with the Functional Champion |
| **2. Authority is clear** | Action limits, enforced human approvals, and release authorization | The release or control authority |
| **3. Outputs are usable** | Representative live quality, provenance, corrections and completion evidence | Process Performers, with the Functional Champion |
| **4. Exceptions have an owner** | A named queue, a response window, a tested takeover, rollback and reconciliation | The operator, with the process owner |
| **5. People can sustain it** | People do the task and handle exceptions on their own, with a job aid, support and repeat use | The Functional Champion, with Process Performers |
| **6. The customer sees the benefit** | A comparable baseline and live measures, adoption, full costs and limitations | The business outcome owner or Executive Sponsor |

The first five make the agent **live and adopted**. The sixth is **value verified**: the customer accepts the measured benefit. See [Measure the value](/user-docs/deliver/measure-the-value).

## The first weeks in production

| Role | What they do |
| - | - |
| **AI Builder & Deployer** | Deploys, monitors and resolves incidents |
| **Functional Champion** | Tracks coverage, friction and exceptions |
| **Executive Sponsor** | Protects time and incentives for the rollout |
| **Process Performers** | Use it, inspect what it did, recover when it's wrong, and report when they work around it |
| **Within** | Reviews the live signals, and triages and supports |

## Raise autonomy step by step

Start with more human review than you think you need, and reduce it only when testing and production evidence support it. Keep a person's approval for any action that's irreversible, regulated or expensive to correct, keep monitoring on, and keep a way to switch the agent off.

## When it isn't sticking

If live work isn't succeeding repeatedly, or people are working around the agent, fix the workflow, the practice or the support route first, and send the risky work back to people until it's fixed. A bypass report is a finding, not a failure.

> [!TIP]
>
> Adoption is a habit, not a launch. [Group rituals](/group-rituals) has run scripts a champion can use to keep the team in rhythm.

## Where to go next

> **[Measure the value](/user-docs/deliver/measure-the-value)**
>
> Recalculate against your baseline.

> **[Maintain and improve agents](/user-docs/deliver/maintain-and-improve-agents)**
>
> Keep it current as the work changes.

> [!NOTE]
>
> **Need help?** Use the in-app chat — click the chat bubble in the bottom-right corner of Within (staffed 24×5) — or email [support@within.ai](mailto:support@within.ai).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.