When configuring your AI Agent, it’s important to understand how it behaves.

You don’t need to know the technical details but there are a few key rules that influence how you should write your instructions and Scenarios.

The AI Agent only does what it can actually do

Your AI Agent will never claim to perform an action it cannot execute.

If a customer asks for something that requires a tool or integration that isn’t connected, the AI Agent will clearly state that it doesn’t have access instead of guessing or providing incorrect information.

For example, if you instruct the AI Agent to check an order in Shopify but the integration isn’t set up, it won’t fabricate a response.

Practical implication: Only reference actions that are supported by your connected integrations. Make sure everything is properly set up before using it in Scenarios.

The AI Agent always responds in the customer’s language

The AI Agent automatically detects the language used by the customer and replies in that same language.

If it cannot determine the language, it will fall back to your configured default language.

Practical implication: You don’t need to create multiple versions of your Scenarios in different languages. Write them in the language that works best for you.

Scenarios always take priority

When a Scenario is triggered, it takes full control of the conversation.

The AI Agent will follow the Scenario step by step, including:

  • What information to collect
  • Which actions to perform
  • When to hand over to a human

General Handover criteria and Additional instructions remain active, but only as a fallback when no Scenario is applied.

Practical implication: If a Scenario defines escalation rules, those will override your general Handover settings. Make sure your Scenarios fully cover the intended behaviour.

Escalation is a one-way door

Once a conversation is escalated, the AI Agent hands over completely and cannot take further actions.

Before escalating, the AI Agent completes all required steps, such as:

  • Adding internal comments
  • Applying labels
  • Collecting necessary information

Practical implication: Always place the escalation step at the end of your Scenario. Any actions you want the AI Agent to perform must come before the handover.

The AI Agent won’t ask for information it already has

If the customer has already provided certain information, the AI Agent will not ask for it again.

When data collection is configured, it only requests missing details.

Practical implication: You don’t need to add conditional logic like “ask for the order number if not provided.” The AI Agent handles this automatically.

What the AI Agent cannot see

There are a few limitations to keep in mind:

  • The AI Agent can only see the current ticket
  • It has no memory of previous conversations with the same customer
  • It does not know which AI Journey it is part of
  • It cannot access systems or data that are not connected

Summary

Understanding how your AI Agent works helps you configure it more effectively:

  • It only performs actions it can actually execute
  • It always responds in the customer’s language
  • Scenarios take priority over other instructions
  • Escalation ends the AI Agent’s involvement
  • It avoids asking for duplicate information
  • Its visibility is limited to the current conversation and connected systems

Use these principles to write clearer Scenarios and create a more reliable AI Agent.