Layer 2 — AI Agents · Chapter 1
What Is an AI Agent?
There is no single universally accepted definition of an AI agent. In practical terms, an agent is an AI system that can pursue a goal through a sequence of decisions and actions.
From response to action
A conventional chatbot waits for a prompt and returns text. An agent may receive a broader objective such as “prepare a market report,” “resolve this customer complaint,” or “schedule the best available appointment.” It then breaks the objective into steps, chooses tools, checks results, and continues until it reaches a stopping condition.
Minimum components
- Goal: what the system is trying to accomplish.
- Model: the reasoning and language capability.
- Tools: software, databases, browsers, APIs, or machines it can use.
- State: information about what has already happened.
- Policy: rules that limit what it may do.
- Feedback: evidence used to judge whether an action worked.
Levels of agency
Agency exists on a spectrum. A low-agency assistant suggests the next action. A medium-agency system performs bounded tasks after approval. A high-agency system plans and executes many steps with limited human intervention.
The important distinction is not whether the system sounds intelligent. It is whether the system can reliably change the state of the world.