Overview

In this article, we cover how and when to split your workflows into multiple AI agents.

This article is part of a series covering setting up and deploying AI agents. See the other parts in the series:

1. Create an AI Agent - this article
2. Write a Behavior Description (Prompt)
3. Create Actions and Tools
4. Add Your Knowledge Center
5. Add Input Parameters
6. Check Your Agent
7. Test Your Agent
8. Integrate into a Flow
9. Use Multiple Agents - this article

Using Multiple Agents

👍

Tip:

Check out our Intro to AI Agents for a full walkthrough of the different types of agents and when you should use them.

AI agents should always be set up to do just a few specific things. For example, an agent which is configured to route conversations or an agent which handles sales-related tasks.

If you want your agents to be able to handle a wider range of tasks, we recommend creating an agent for each domain of responsibility. For example, if you want your agents to handle both sales and support tasks, you should create one agent for sales and another agent for support.

Assuming you want your agents to handle more than basic routing or basic Q&A tasks, we generally recommend a setup similar to the following, and in the following order:

1. One or more routing agents to handle introductions and to figure out what the customer wants to do. Routing agents, like the example above, should generally be able to decide where to send conversations and should be able to answer basic questions. They should also be provided tools to collect information from your customer you'll need in order to route them correctly, like customer identification and contact information.

   

2. A domain-specific agent for each area of responsibility. Domain-specific agents should operate in a specific area (Sales, Support, Finance, Logistic, etc.) and be able to handle most requests in that area. For example, a sales agent should be able to handle things like checking your product catalogue, transferring to a sales representative, and transferring to another agent that handles order placement.

   

3. Any number of specialist agents designed to perform specific, complex tasks. Specialist agents should be designed to use specialized tools or flow steps to complete the tasks they're responsible for. For example, an order placement agent should be designed to handle payment processing via an external payment processor, order recording in an external database, and receipting via email or through an in-ticket message.

   

Putting it all together, a fully agentic flow with multiple agents will look something like this (click to expand):