Agent patterns
Agent patterns are reusable, composable building blocks that can be tailored to specific domains, use cases, and levels of complexity. Agentic systems differ, however, from traditional applications. At the heart of all AI agent designs is a conceptual model anchored in the following three foundational principles:
-
Asynchronous – Agents operate in loosely coupled, event-rich environments
-
Autonomy – Agents act independently, without human or external control
-
Agency – Agents act with purpose, on behalf of a user or system, toward specific goals
The triangle in the following diagram represents the core building blocks of a software agent: perception, reason, and action. This enables an agentic system to observe, make decisions, and act within its environment.
By design, agentic patterns provide a modular design language for building AI systems, which means they're accessible, operational, extensible, and production ready. Designing these systems requires careful attention to the following three interrelated dimensions, which are further discussed later in this guide.
In this section
