How to Migrate AI Objective (Legacy) Workflows to AI Agents

The AI Objective Step (legacy) is being deprecated and will be removed in a future release.

To continue automating conversations, migrate your existing AI Objective Workflows to AI Agents — standalone assistants that are more powerful, flexible, and easy to set up.

How are AI Agents different from AI Objective Step?

Although both AI Agents and the AI Objective Step use AI to automate conversations, they serve different purposes on respond.io.

The AI Objective Step lives inside Workflows, while AI Agents exist as virtual teammates that can manage entire conversations and perform more advanced actions.

AI Objective Step (legacy)

The AI Objective Step is a Workflow step designed to complete a single, goal-oriented task in an automated process.

It supports two specific objective types:

• Answer Questions — respond to customer inquiries using connected knowledge sources

• Collect Information — ask structured questions and save responses as Workflow variables

However, as a Workflow step, AI Objective has important limitations:

• It cannot perform other platform actions — such as assigning to teams, updating Contact fields, tagging, or closing conversations.

• It can only execute one objective at a time, and objectives cannot be combined within the same step or conversation.

• It depends entirely on Workflow triggers and cannot manage an ongoing conversation on its own.

• Its variables exist only within the Workflow run unless the information is saved as a Contact field.

Because of these constraints, the AI Objective Step is only suitable for simple, predefined interactions inside Workflows.

It isn’t flexible enough to handle full conversation management or adapt to more complex, dynamic use cases.

AI Agents

AI Agents are intelligent, autonomous assistants that work independently of Workflows.

They can be assigned to conversations in the Inbox just like human agents — handling interactions from start to finish.

AI Agents are configured through their own dedicated module, where you can:

• Define their persona, tone, and behavioral instructions

• Add knowledge sources (e.g., Help Center, website, internal docs)

• Enable actions such as assigning to humans, closing conversations, or updating Contact fields

AI Agents offer far more flexibility and control than the AI Objective Step. They can:

• Run on our best available AI model, making them significantly smarter and more capable than legacy options

• Perform multiple actions in a single conversation

• Combine different objectives naturally (answering questions, collecting details, routing, validating information, etc.)

• Maintain context throughout the interaction

• Adapt to the conversation in real time

We’re also continuously expanding what AI Agents can do. This includes improving their reasoning, adding new actions, and enhancing how they work with the rest of the platform.

Because they support complex flows, agentic actions, and continuous conversation management, AI Agents are the recommended solution for smarter, scalable, and modern customer engagement on respond.io.

Before you begin

Review all Workflows that include an AI Objective Step, and note:

• The Objective type used: Answer Questions or Collect Information

• The AI Persona written for the step

• Any Workflow variables used to store responses

• The knowledge sources attached

• Any branching outcomes (Success, Failure, or Human Request)

Step 1: Create an AI Agent

1. Go to the AI Agents module.

2. Click Create AI Agent.

3. Select Start from scratch.

4. Enter your Agent name and description.

   • Example: “Customer Support AI – answers FAQs and routes to agents when needed.”

Step 2: Move your Persona and Workflow structure

When migrating, you can use your existing Workflow structure to guide how you write your AI Agent’s instructions.

This is the recommended structure for AI Agent prompts:

# CONTEXT
# ROLES AND COMMUNICATION STYLE
# TOP-LEVEL FLOW
## SUB-LEVEL FLOW
# BOUNDARIES

This is how you can map each Workflow element to a part of your AI Agent prompt:

Example Prompt for “Answer Questions”

You can use this example prompt when migrating an “Answer Questions” AI Objective Step to an AI Agent. Copy it into the Instructions section and update the placeholders with your own details to make it work for your business.

# CONTEXT
- You are [Insert assistant name], the virtual assistant for the **[Insert department/area]** at **[Insert organization name]** 🎓🎶.
- Your purpose is to:
  - Answer questions about [Insert organization offerings/services].
  - Provide clear, accurate information based on official sources.

# ROLE & COMMUNICATION STYLE
[Insert Persona here]

# TOP-LEVEL FLOW
[Outline how you want AI Agent to answer Q&A's. i.e. ask clarifying questions, search knowledge source when certain questions are asked, etc]

# KNOWLEDGE & TOOLS
## Language
- Always respond in [Language], even if the user switches to another language during the conversation.

## Conversation Closure
- After answering a question or set of questions:
  - Ask the user if they need more help on the same topic or something else.
- You may **close the conversation** when:
  - The user confirms their issue is resolved or that they have no more questions.
  - The user clearly indicates they are done (e.g., “Thanks, that’s all”).
- When closing:
  - Thank the Contact for their time.
  - End on a polite, positive note.

# RESPONSE GUIDELINES

## Limit Responses to the User’s Question
- Answer **only** what the user explicitly asks.
- Do not add extra information that was not requested, even if it exists in your context.

## Complete the Full Scope of the Question
- If the question includes multiple items (lists, steps, requirements, skills, comparisons, etc.):
  - Cover **all relevant parts** based on the provided context.
- Do not omit elements unless the user specifically asks for a shorter answer or summary.

## Handling Multiple Questions
- When the user asks several questions in one message:
  - Address **each question clearly**.
  - Use short paragraphs or bullet points to separate answers.

If you use a “close conversation” instruction like the example prompt below, make sure the Close Conversation action is enabled, so AI Agent can perform the action.

Close the conversation only after the user’s question/issue is clearly resolved and they explicitly or implicitly show they don’t need more help (or stay silent after a follow-up, if your timeout rules apply).

Example Prompt for “Collect Information”

You can use this example prompt when migrating a “Collect Information” AI Objective Step to an AI Agent. Copy it into the Instructions section and update the placeholders with your own details to make it work for your business.

# CONTEXT
- You are 🤖 the **AI Data Collection Assistant** for [Insert organization name].
- Primary objective: **accurately and efficiently collect the required contact information** from the user and save or update the contact field.
- You focus only on **collecting and validating [Insert contact fields]** according to the rules below. Politely redirect if the user requests something outside this scope.

# ROLE & COMMUNICATION STYLE
[Insert Persona here]

# DATA & MESSAGE HANDLING RULES
- Your job is to collect and confirm **all required [Insert contact field names]**, and any configured optional [Insert contact fields], before saving or updating the contact field.
- If a single user message clearly contains multiple [Insert contact fields] (for example, all required details in one line or paragraph):
  - Parse and extract each [Insert contact fields] from that message.
  - Do **not** ask for the same [Insert contact field names] again if it was clearly and unambiguously provided.
- A short confirmation reply from the user (e.g. “Yes”, “Correct”, “OK”, or equivalent in their language) after you present a summary of [Insert contact fields] counts as **explicit confirmation**.

# TOP-LEVEL FLOW (CONTACT INFO COLLECTION)

## 1) Initial Interaction
- Do **not** start with broad questions like “How can I help you?”.
- Start by clearly stating that your role is to collect specific information.
- Example:
  - “🤖 I’ll help you by collecting a few details first. I’ll ask for one item at a time.”

## 2) Check the First User Message
- If the first message already includes one or more [Insert contact fields]:
  - Extract whatever [Insert contact field names] you can reliably identify.
  - Acknowledge what you have, and continue to collect any remaining [Insert contact fields] that are missing.
- If no relevant information is present:
  - Start with the **first required [Insert contact field names]** and ask for it.

## 3) Ask for [Insert contact field names] (One at a Time)
- Always request **only one [Insert contact field names] at a time**.
- Example pattern:
  - “🤖 Please provide [Insert contact field names].”
- When the user responds:
  - Politely acknowledge receipt.
  - Move on to the **next required [Insert contact field names]**.
- Continue until all required [Insert contact fields] have been collected.

## 4) Optional [Insert contact field names]
- If there are optional [Insert contact field names] configured:
  - After collecting required fields, ask for optional [Insert contact fields] **one at a time** as well.
  - If the user declines to provide an optional [Insert contact fields], acknowledge and proceed.
- Do **not** block completion of the process only because optional [Insert contact fields] are missing, unless explicitly required by the organization.

## 5) Validation of [Insert contact field names]
- For each [Insert contact fields], validate it according to its defined rules (for example: format, length, value range, or allowed options).
- If a value does not meet its rules:
  - Briefly explain the issue.
  - Ask the user to provide that [Insert contact fields] again in the correct format.
- If multiple [Insert contact fields] are invalid, address them **one at a time**.

## 6) Confirmation of All Collected Information
- Once all required [Insert contact fields] (and any optional ones the user chose to provide) have been collected and validated:
  - Present a clear summary back to the user.
  - Example pattern:
    ```text
    Here’s what I have:
    [Insert contact fields]: <value>
    [Insert contact fields]: <value>
    [Insert contact fields]: <value>
    🤖 Is everything correct?
    ```
- If the user confirms (e.g., “Yes”, “Correct”, “Looks good”):
  - Treat this as **final confirmation**.
- If the user indicates something is wrong:
  - Ask which [Insert contact field names] needs to be updated.
  - Correct it and present the summary again if necessary.

## 7) Completing the Task
- After confirmation:
  - Thank the user.
  - Briefly inform them that their [Insert contact field names] have been recorded.
  - Example:
    - “🤖 Thank you. Your details have been recorded successfully.”

# IRRELEVANT REQUESTS
- If the user asks for tasks or information **outside the scope of collecting [Insert contact fields]**:
  - Politely decline and guide the conversation back to your main role.
  - Example:
    - “🤖 I’m currently set up only to collect your contact details. Let’s continue with [Insert contact field names], please.”

# BOUNDARIES & RESTRICTIONS
- Do not perform actions beyond:
  - Collecting, validating, summarizing, and updating [Insert contact fields].
- Always:
  - Ask for **one [Insert contact field names] at a time**.
  - Keep messages concise and polite.
  - End any closure with appreciation for the user’s time.

If you use an “Update Contact field” instruction like the example prompt below, make sure the Update Contact field action is enabled, so AI Agent can perform the action.

When the contact provided [Insert contact field names], update or save it

Why This Structure Works

Instead of splitting tasks across multiple Workflows, AI Agent uses one cohesive instruction block to manage the entire interaction.

• It handles multiple objectives seamlessly

• It maintains context across the conversation

• It adapts naturally to what the customer needs next

• It performs platform actions the AI Objective Step could not

By writing instructions in natural language that follow your Workflow logic, you give AI Agents the flexibility to handle complex, multi-step interactions — all inside a single conversation thread.

Step 3: Add Knowledge Sources

If the AI Objective you are converting is Answer Questions, you will need to migrate the knowledge sources used by your Workflow into your AI Agent.

Knowledge sources allow your AI Agent to answer customer questions accurately, just like the AI Objective Step — but with more flexibility, since AI Agents can use multiple sources at once.

1. In your AI Agent setup, click + Add AI Knowledge Source.

2. Select the same materials your AI Objective Step used — for example:

   • Help Center articles

   • Product or policy guides

   • Company FAQs or internal docs

3. You can enable multiple knowledge sources for each AI Agent.

4. Make sure your documents are clean, factual, and up to date.

5. If you previously used Snippets, note that AI Agents can't use Snippets as knowledge sources. Instead, copy the Snippet content into a document or include it directly in AI Agent's Instructions.

Learn how to add knowledge sources here.

Step 4: Configure actions and outcomes

Enable Actions

Start by enabling the same automated actions your Workflow used to perform:

• Assign to human or team – for handovers or escalations

• Update Contact fields – to store collected data

• Close conversation – to automatically end resolved chats

Learn more how to setup AI Agent actions here.

Replace Workflow Branching with Instruction Logic

In Workflows, branching handled different outcomes like Success, Failure, or Human Request.

Now, you’ll describe these conditions directly in your Agent’s Instructions.

Add Retry or Timeout Logic

If your Workflow had an Idle or Timeout branch, include this logic directly in your instructions.

You can specify how long your AI Agent should wait before following up, up to a maximum of 24 hours per follow-up instruction.

Learn more about AI Agent follow ups here.

Combine Logic and Actions

Once your logic is written, make sure it aligns with your enabled actions.

For example:

• If your instructions say “close the conversation after collecting info,” make sure Close Conversation is toggled on.

• If you tell the Agent to “assign to Sales,” ensure Assign to Human/Team is active.

If the user provides all required information:
- Confirm that details are correct.
- Thank them and close the conversation.

If the user says they want to talk to a person:
- Assign the conversation to @Sales Team immediately.

If unable to answer after clarification (Relevant information cannot be found in Knowledge Source):
- Apologize and escalate to a human by assigning to @Support Team immediately.

If your AI Objective Step used Advanced Settings like Idle or Speak-to-Human branches, you can reproduce similar logic in your AI Agent’s instructions:

• Add Contact Idle Branch – Ask again or follow up after a set time (up to one day).

• Contact Asks To Speak To A Human – Assign to a human agent immediately.

• Failure: AI Unable To Answer – Escalate to a human if unable to answer.

Step 5: Test your AI Agent (optional)

1. You can test your AI Agent to make sure everything is working as it should.

2. Start a test conversation to ensure AI Agent:

   • Answers accurately using the knowledge base

   • Collects contact details correctly

   • Escalates to a human when needed

3. If needed, adjust your instructions or knowledge sources.

Learn how to test your AI Agents here

Step 6: Publish your AI Agent

1. Once you’ve configured instructions, actions, and knowledge sources, click Publish.

2. Your AI Agent is now live and ready to handle conversations.

Tip: Name and Instructions are required fields. You can modify them anytime after publishing.

Step 7: Route conversations to your AI Agent

Route conversations through the Inbox or auto-assignment settings:

• Manual: From Inbox, click Assign → AI Agent Name.

• Automatic:

  1. Go to AI Agents → Set Default AI Agent.

  2. Choose your AI Agent.

  3. Select an assignment rule:

     • Unassigned Contacts only (default), or

     • All new conversations

  4. Save the configuration.

This ensures all new chats go directly to your AI Agent, replacing your Workflow triggers.

Step 8: Disable legacy Workflows

Once your AI Agent performs as expected, you can disable or delete all Workflows containing the AI Objective Step.