Back to Changelog
New

AI Agent Scenarios for complex business logic

What's new

Your AI Agent now supports Scenarios -- a new way to define step-by-step instructions for specific business situations like returns, cancellations, booking changes, or complaint handling.

Until now, all AI Agent behavior was managed through a single set of instructions. That works for simple setups, but as you add more rules and edge cases, things get hard to maintain and unpredictable.

Scenarios change that. Each Scenario covers one specific process and tells the agent exactly what to do when that situation comes up.

How Scenarios work

A Scenario defines:

  • When the agent should activate -- based on customer intent, keywords, or context

  • What steps to follow -- in the order you define

  • What actions to take -- assign to a team or person, apply labels, add internal comments, use integrations

  • When to escalate -- hand over to a human with full context

When a Scenario is triggered, it takes full priority over your general Additional instructions and Handover criteria. If no Scenario matches, your default rules still apply as a fallback.

What the agent can do inside a Scenario

  • Assign conversations to a specific team or team member

  • Apply one or more labels for tracking and routing

  • Add internal comments (visible only to your team, not the customer)

  • Use connected integrations like Shopify Storefront or Notion

  • Use custom actions to call any external API (e.g. order status lookup, booking availability)

These actions are only available inside Scenarios.

How instructions now work together

Your AI Agent uses three layers of instruction:

  1. Behavior settings -- your agent's identity: job description, tone of voice, company context. Always active.

  2. Handover criteria + Additional instructions -- your default rules and escalation triggers. Active when no Scenario applies.

  3. Scenarios -- process-specific instructions. When triggered, they override Layer 2 for that conversation.

Example: handling a damaged item (e-commerce)

When a customer reports a damaged item:

  1. Acknowledge the issue and apologize

  2. Ask for a photo of the damage and the order number

  3. Confirm the damage back to the customer based on the photo

  4. Add an internal comment with order number, damage description, and whether a photo was provided

  5. Apply the label "damaged-item"

  6. Assign the conversation to the Fulfillment Team

Good to know

  • You can create up to 10 Scenarios per AI Agent

  • Each Scenario supports up to 5,000 characters of instructions

  • Use the @ button in the editor to reference teams, teammates, and labels directly

  • Scenarios don't chain -- each one covers a single, complete process

  • Test your Scenarios in Preview mode before going live