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

# Migrate from Classic AI

> Map Classic AI and workflow concepts to AI Agents.

Move your existing Classic AI and workflow automation to AI Agents over time. You do not need a risky all-at-once change: migrate one use case, prove it in a controlled rollout, turn off the overlapping legacy setup, and then expand.

## What to migrate first

Prioritize migration when:

* a Classic AI setup needs frequent manual tuning
* a workflow tree is becoming hard to maintain
* a workflow trigger prompt is trying to match broad customer intent
* the automation needs judgment instead of fixed branches
* the agent should use multiple actions or integrations
* you need better tests and run history

Stable legacy setups can stay active while you build and prove their replacements. The goal is a safe, complete migration rather than running both systems indefinitely.

## Concept mapping

| Classic setup             | AI Agents equivalent                                                                                                                                                                                                                                    |
| ------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Classic AI sources        | Agent instructions, skills, and connected knowledge                                                                                                                                                                                                     |
| Guidance                  | Skills or instructions for a topic                                                                                                                                                                                                                      |
| Handover topics           | Escalation instructions and routing actions                                                                                                                                                                                                             |
| Auto-tagging              | Native tag auto-tagging settings, or agent classification behavior when the tag depends on broader support context                                                                                                                                      |
| AI workflow trigger       | Agent trigger conditions and semantic filters                                                                                                                                                                                                           |
| Workflow action node      | Agent action                                                                                                                                                                                                                                            |
| Workflow approval mode    | Action approval policy                                                                                                                                                                                                                                  |
| Flow test                 | Agent test run                                                                                                                                                                                                                                          |
| Flow execution logs       | Agent run history                                                                                                                                                                                                                                       |
| Auto-resolve workflow     | Substring-based [native Auto-resolve settings](/inbox/auto-solve) only for narrow phrases, senders, or subjects that always need no action, or an [AI Agent example](/ai-agents/examples) whenever the surrounding conversation can change the decision |
| CSAT survey workflow      | [CSAT AI Agent example](/ai-agents/examples)                                                                                                                                                                                                            |
| Classic Takeover behavior | AI Agent ownership, human handoff, and fallback behavior in [Routing and ownership](/ai-agents/routing-and-ownership)                                                                                                                                   |

## Migration path

<Steps>
  <Step title="Pick one legacy use case">
    Start with one high-volume case, such as order status, returns, invoices, or B2B escalation.
  </Step>

  <Step title="Write the agent job">
    Describe what the AI Agent owns, what it should avoid, and when it should escalate.
  </Step>

  <Step title="Move trigger logic">
    Replace workflow trigger prompts with exact filters and semantic filters.
  </Step>

  <Step title="Move reusable knowledge">
    Convert useful guidance, policies, and examples into instructions or skills.
  </Step>

  <Step title="Replace workflow nodes with actions">
    Give the agent the smallest set of read or write actions needed for the job.
  </Step>

  <Step title="Add approval rules">
    Require approval for refunds, address changes, cancellations, and other high-impact actions.
  </Step>

  <Step title="Review ownership and handoffs">
    Decide which customer requests the AI Agent owns, when it hands the ticket to a person or team, and what happens after that handoff. Remove overlapping trigger conditions so the AI Agent and legacy automation do not both match the same customer message. See [Routing and ownership](/ai-agents/routing-and-ownership).
  </Step>

  <Step title="Test against old cases">
    Use tickets that the Classic AI or workflow handled before. Test requests that should match, requests that should not match, and cases that should hand off to your support team.
  </Step>

  <Step title="Run a controlled rollout">
    Start with a narrow scope, such as one channel, language, team, or ticket type. Keep the legacy setup active only outside that scope while you review agent runs and customer outcomes.
  </Step>

  <Step title="Turn off the overlap">
    Once the AI Agent is proven, disable the Classic AI trigger or workflow that handles the same tickets. Confirm that customer messages no longer match both systems before you widen the rollout.
  </Step>

  <Step title="Expand one use case at a time">
    Increase the AI Agent's scope, repeat the same checks, and retire each remaining legacy path after its replacement is proven.
  </Step>
</Steps>

## Best practices

* Migrate one use case at a time.
* Keep the new agent narrower than the legacy setup at first.
* Preserve escalation behavior before adding autonomy.
* Test both matching and non-matching trigger examples.
* Use run history to decide what to improve next.
* Keep stable legacy automation only while its replacement is being tested.
* Turn off overlapping legacy triggers before expanding the AI Agent to more tickets.

<Note>
  AI Agents are the product direction for both new and existing automation. Use Classic AI and workflows only during the staged migration or to maintain a legacy setup until its replacement is proven.
</Note>