Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More

A modern-day AI agent consists of, at least, a large language model (LLM) that has been enabled to call some tools. Given the right set of tools for coding, it would start by generating the code, be able to run it in a container, observe the results, modify the code and therefore have a better chance of producing useful code.

By contrast, a generative AI model takes some input and, through the process of predicting expectations, produces an output. For example, we give it a coding task, it produces some code, and, depending on the complexity of the task, the code may be usable as is.

As they take on different tasks, agents should be allowed to talk to each other. For example, imagine your company intranet with its useful search box directing you to the apps and resources you need. If you are a large enough company, these apps owned by different departments each have their own search boxes. It makes a lot of sense to create agents, maybe by using techniques like retrieval augmented generation (RAG), to augment the search boxes. What does not make sense is to force the user to repeat their query once the search box has identified it as useful given the initial query. Rather, we would prefer the top agent to coordinate with other agents representing various apps and present a consolidated and unified chat interface to you, the user.

A multi-agent system representing software or an organization's various workflows can have several interesting advantages, including improved productivity and robustness, operational resilience and the ability ability to perform faster upgrades of different modules. Hopefully, this article will help you see how this is achieved.

But first, how should we go about building these multi-agent systems?

Capturing the organization and roles

First we should capture the processes, roles, responsible nodes and connections of various actors in the organization. By actors, I mean individuals and/or software apps that act as knowledge workers within the organization.

An organizational chart might be a good place to start, but I would suggest starting with workflows, as the same people within an organization tend to act with different processes and people depending on workflows.

There are available tools that use AI to help identify workflows, or you can build your own gen AI model. I've built one as a GPT which takes the description of a domain or a company name and produces an agent network definition. Because I'm utilizing a multi-agent framework built in-house at my company, the GPT produces the network as a Hocon file, but it should be clear from the generated files what the roles and responsibilities of each agent are and what other agents it is connected to.

Note that we want to make sure that the agent networ ...