How the Agent step works

With the Agent step, you can combine hard business logic with Agent networks layered on top, allowing for both risk mitigation and Agent control, as well as conversational flexibility and capability.

This component allows developers to designate specific parts of a conversation where the AI can operate independently while maintaining structured flows elsewhere in the experience.

When a user interaction reaches an Agent step, AI takes over with the ability to:

• Plan and execute appropriate responses based on user needs
• Access relevant knowledge bases when necessary
• Connect with external tools and APIs to retrieve information
• Determine when to exit the Agent step or hand off to another Agent step

Voiceflow gives businesses a complete toolkit from Agentic networks to human-written responses and business logic to ensure you always have the right tool for the job.

‍

Use Case: Customer Support Agent

Consider a customer support scenario: When a user initiates a return request, the deterministic flow handles initial validation and account verification. Once basic information is collected, the Agent step activates, allowing the AI to autonomously:

1. Assess the return reason and determine eligibility based on policies
2. Access order history to verify purchase details
3. Generate a return label if appropriate or escalate to a specialized Agent step for exceptions
4. Exit back to the deterministic flow for confirmation messaging

This hybrid approach ensures compliance with return policies while providing intelligent handling of complex cases. You can also create specific exit scenarios that automatically exit the Agent back to the safety of deterministic logic. These exit scenarios include situations such as an upset customer, requesting to speak with a human, or the agent detecting malicious usage outside its scope.

‍

Tying it all together

The addition of an Agent step eliminates the trade-off between reliability and flexibility. Deterministic flows provide consistency for critical processes, while Agent steps enable adaptability for complex scenarios—all within a unified development environment.

Voiceflow provides a practical solution for building AI Agents that balance structure with intelligence, giving you control over exactly where and when autonomous capabilities are deployed.

Next up: Agent Frameworks!

The Agent step's modular design supports advanced implementation patterns. Developers can connect multiple Agent steps to create sophisticated frameworks, including the widely-used Supervisor pattern where specialized agents handle different conversation aspects.

We are excited to see the novel Agent frameworks people develop using the Agent step!