Your AI Agent and your scenarios don't have to work apart. You can drop AI actions into any scenario to write replies on the fly, post comments, branch on what a contact means, collect information, and hand the whole conversation to the agent mid-flow.

Think of it like a relay race. The scenario handles the structured, predictable part. When things turn open-ended, it passes the baton to the agent. You'll find these actions in the AI group of the step picker inside any scenario or campaign.

These are AI features, so a few things are true of every one of them:

The agent writes using your global personality, goals, and knowledge base, so the replies sound like the rest of your account.

The agent reads the last message (or the comment that started the flow) and writes one contextual reply. You don't write the copy, the agent does.

Use this when you want a quick, relevant reply at a set point in a flow but a scripted message wouldn't fit what the contact said.

Configuration: A message-length choice only, from 10, 25, 50, or 100 words. There's no instructions field. This step sends one message and doesn't branch.

Like auto-DM, but you give the agent an instruction for this particular message. It still uses the conversation as context, then follows your direction for what to cover.

For example: "Explain our refund policy based on what they asked" or "Summarise the options based on their stated budget." The instruction supports variables, so you can fold in contact or flow data.

Configuration: A message length (10/25/50/100 words) plus a free-text instructions field. Sends one message and doesn't branch.

The agent reads the comment or mention that triggered the flow and posts a public reply on the post, rather than a DM.

⚠️ Some comments are skipped automatically as the volume grows, to stay clear of spam detection. That is expected.

Configuration: A comment length of 5, 10, or 25 words, and a Use training data in reply toggle. With the toggle on, the agent draws on your knowledge base when it writes the comment.

This action puts the conversation into agent-handled mode. From here, the agent fields every incoming message itself, like a live chat, until its task is judged complete or the handover expires.

A common pattern: a scenario qualifies the contact and collects their details, then hands off so the agent can have a real sales or support conversation.

Configuration: An optional instructions field for context about this handover (for example, "This contact confirmed interest in the Pro plan, focus on booking a call"), plus an optional follow-up cadence.

This step branches: one path for Task completed, one for Handover expired. Because there's a lot to set up there, it has its own guide: see Agent Handover: Task Completion & Expiration Branches.

⚠️ Handover hands the chat to your AI Agent, not to a teammate. To route a contact to a person instead, use Human Intervention.

This branches your flow on whether a plain-language condition is true or false for the contact. It's the AI version of the rule-based Scenario condition action.

Write the condition as a statement:

It's checked against the triggering message, the recent conversation, and the contact's saved profile data, not only the one message that fired the trigger. Use it when keyword rules can't capture what you need.

⚠️ AI isn't always right. Don't use this for critical decisions like charging someone.

Branches: If condition is true / If condition is false.

This is the Ask a question action with AI detection switched on (it also appears in the AI group as AI Agent Question). You ask a multiple-choice question with quick-reply buttons, and each answer creates its own branch. With AI detection on, the agent also reads free-text replies that don't tap a button and matches them to the closest option.

Use it when contacts often reply in their own words instead of tapping a button. The full mechanics of the action (saving the answer to a property, re-asking, retries) live in Ask a Question & Wait for Reply; this is the AI layer on top.

Branches: one per answer option, plus Any other reply and an expiration branch.

This is the Ask for information action (collecting a contact property). You ask for something like an email, phone number, or a custom field, and the agent reads the contact's reply, pulls out the value, and saves it to their profile automatically. When AI isn't available it falls back to format detection (an email or phone pattern, for example).

Use it to capture data conversationally instead of with a rigid form. The full action, including retries and the built-in "Collect email" shortcut, is covered in Contact Actions: Folders, Properties & Data Collection. To have the standalone agent save data on its own during normal chats, see User Data Collection.

Branches: When information is collected / expiration.

Two things to keep in mind when you use both:

For the deep dive on the handover branches, see Agent Handover: Task Completion & Expiration Branches. For getting consistently good output across all these touchpoints, read AI Agent Best Practices.