Skip to main content
In What Is an Agent?, we defined an agent as a runtime built around four things: a role definition, the inputs it observes, the SOPs it executes, and the outbound messages it sends. The Agent Structure section unpacks what that looks like in practice — the concrete components that make an agent work, the workflows that govern its decisions, and the adapter patterns that connect it to the systems around it. Every department agent — whether it runs as Department A, VM-1, or any other designated runtime — is assembled from the same set of parts. The shape of those parts differs from one deployment to the next, but the structure does not.

​
What this section covers

Components

The five building blocks of every agent: role definition, input sources, SOP logic, escalation rules, and outbound interfaces. This is where the agent’s scope and procedures become concrete configuration.

Workflows

How an agent sequences its decisions — the patterns it follows when processing a trigger, evaluating conditions, and producing an outbound message or action.

Integrations

How agents connect to the systems around them — including adapter patterns for your ERP system, your CRM, and any other platform the agent needs to read from or write to.
You do not need to understand every detail here before engaging on a build. This section exists so that when you review a design document or request a change, you know which part of the agent you are talking about.