The core loop is: observe the current state, decide on an action, execute it, and feed the result back as a new observation. Everything else in an agent framework, from memory to planning to guardrails, is an elaboration of one of those four steps.

The Agent Loop

An observation is whatever the program hands back to the model after an action: a tool result, an error string, a file's contents, a user reply. Observations are the agent's only channel of truth about the world, so their fidelity bounds how well the agent can possibly behave.

Observation

An agent is a system that pursues a goal by choosing its own actions in a loop, rather than executing a fixed script written by a developer. The defining property is autonomy over control flow: the model decides what happens next, and the surrounding program supplies the means to act and the results of acting.

Agent

A topic of the Building AI Agents course. Planned for 2 hours at Medium difficulty.

University of Michigan - Ann Arbor

A category of the Building AI Agents course, covering 4 topics.

Foundations of Agentic Systems

A topic of the Building AI Agents course. Planned for 1.5 hours at Easy difficulty.

What Makes a System an Agent

The Perception, Reasoning and Action Loop

Agent Architectures: Reactive, Deliberative, Hybrid

Choosing a Model for an Agent

For agents, the deciding capability is rarely raw knowledge; it is whether the model reliably respects tool schemas, stop conditions and formatting contracts across dozens of turns. A model that drifts from its contract at turn fifteen cannot be deployed unsupervised.

Learn Before

Related

Learn After