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

# Deploy AI Agents into Your Workflows with Initiatives

> Turn committed opportunities into deployed Fluency Initiatives — review the agent plan, run in preview mode, then activate autonomous execution.

Deploying automation with Fluency starts with a committed opportunity and ends with an autonomous AI agent running inside your existing tools. This guide covers every step — from reviewing the agent's proposed plan to monitoring its first live runs.

## Prerequisites

You should have at least one committed opportunity visible in the **Initiatives** section of Fluency. Committing an opportunity is the step that converts it from a candidate into a deployable agent plan.

If you haven't committed an opportunity yet, see the [Identifying Opportunities](/guides/identifying-opportunities) guide to find your top candidates and commit one before continuing.

## Review the initiative detail

<Steps>
  <Step title="Open Initiatives">
    Select **Initiatives** from the left sidebar. You'll see a list of all initiatives in your account, including their current status — draft, preview, active, or paused.
  </Step>

  <Step title="Click the initiative you want to deploy">
    Select the initiative to open its detail view. Fluency displays the auto-generated agent configuration, built directly from the committed opportunity's workflow map and projected steps.
  </Step>

  <Step title="Review the workflow map">
    The detail view shows the full workflow map with each step annotated to indicate how the agent will handle it. Steps marked as **Automated** will be executed by the agent without human involvement. Steps marked as **Human-led** remain with your team. **Handoff** steps mark the transitions between agent and human action.
  </Step>

  <Step title="Check the projected outcome for each step">
    For every automated step, confirm the expected input, the action the agent will take, and the expected output. This is your opportunity to catch any mismatches between the agent's plan and your team's actual requirements before deploying.
  </Step>

  <Step title="Review the exception handling plan">
    Scroll to the exception handling section. This shows which edge cases the agent is configured to escalate to a human rather than resolve autonomously — for example, an invoice that exceeds a certain value threshold, or a document that fails a validation check. Confirm that the escalation conditions are appropriate and that the queue where exception tasks will appear is accessible to the right people on your team.
  </Step>
</Steps>

## Preview mode

Before deploying live, use **Preview mode** to run the agent against a sample of historical workflow runs without taking any live action. Preview mode lets you observe exactly how the agent would have handled real past runs — what decision it would have made at each step, what output it would have produced, and whether it would have triggered an exception.

To activate preview mode, click **Preview** in the initiative detail view. Fluency selects a representative sample of recent historical runs and executes the agent against them in a sandboxed environment. Review the results step by step. If the agent's behavior doesn't match your expectations, you can edit the configuration and re-run preview before committing to a live deployment.

Preview mode is particularly valuable for high-volume or high-stakes workflows, where verifying behavior before going live avoids compounding errors at scale.

## Deploy the initiative

When you're satisfied with the agent's behavior in preview, click **Deploy**. A confirmation dialog summarizes the initiative scope — the workflow it applies to, the teams affected, the estimated run volume, and the exception handling plan. Review the summary and click **Confirm** to activate.

Immediately after deployment, the agent becomes active and begins processing new workflow runs that match the initiative's scope. Existing in-progress runs are not affected — the agent picks up new runs as they start. The initiative status changes to **Active** in your Initiatives dashboard.

## Monitor active runs

The Initiatives dashboard gives you a live view of all current and completed runs for each active initiative. For every run, you can see:

* **Step-level outcomes** — what the agent did at each automated step, including inputs consumed, outputs produced, and time taken.
* **Overall run status** — whether the run completed successfully, is in progress, or encountered an exception requiring human review.
* **Aggregate metrics** — total runs completed, hours saved, and exception rate, updated in real time as new runs finish.

Use the run history panel to drill into any individual run and inspect it step by step. This is useful both for ongoing quality assurance and for diagnosing unexpected behavior if it arises.

## Handle exceptions

When the agent encounters a situation that falls outside its configured handling rules, it escalates the run as an exception task. Exception tasks appear in a dedicated queue in the Initiatives dashboard, clearly separated from successfully completed runs.

Each exception task gives you full context:

* **The workflow run** — which run triggered the exception and at what point in the sequence.
* **The step that triggered it** — the exact step where the agent determined it could not proceed autonomously.
* **The agent's proposed action** — what the agent would do if you approve it to proceed.

Review the proposed action and either **Approve** it (the agent continues from that point), **Override** it (you specify a different action and the agent proceeds accordingly), or **Resolve manually** (you take the step yourself and close the exception). Every exception resolution is logged, and Fluency uses that data to improve future exception handling for similar cases.

## Update or pause an initiative

You can adjust or pause an active initiative at any time without losing your run history or performance data.

* **Pause** — click **Pause** on the initiative to suspend agent execution. No new runs are processed while the initiative is paused, but all historical run data is preserved. Pausing is useful during audits, process changes, or system maintenance windows.
* **Edit configuration** — while paused, you can modify the agent's step-level configuration, update exception handling rules, or adjust the initiative's scope. Changes take effect when you re-deploy.
* **Re-deploy** — click **Deploy** again after editing to reactivate the initiative with the updated configuration. Fluency logs the configuration change and timestamps it in the initiative history, so you always have a clear audit trail of what changed and when.

<Warning>
  Always use preview mode before deploying an initiative to a high-volume workflow. Reviewing a sample of runs first ensures the agent behaves as expected before operating autonomously at scale.
</Warning>

## Next step

<Card title="Measure ROI and Prove Value Across All Initiatives" icon="chart-bar" href="/guides/measuring-roi">
  Learn how to use Fluency's Assistant and Initiatives dashboard to quantify time saved, cost reduction, and business impact across every deployed automation.
</Card>