What Are AI Agents?

From Chatbot to Agent

A chatbot responds to messages. An agent pursues goals.

The key difference: agents take actions in the world — browsing the web, writing files, calling APIs, running code, sending messages. They don't just generate text; they do things.

An AI agent has three components:

1. Perception — it receives inputs (messages, data, tool results)
2. Reasoning — it decides what to do next
3. Action — it uses tools to affect the world

The Autonomy Spectrum

Agents exist on a spectrum from assistive to autonomous:

Most production agents today operate at L2–L4. L5 remains aspirational.

What Makes an Agent Powerful

Tool use — agents extend their capabilities through tools: web search, code execution, API calls, file operations, database queries.

Memory — agents that remember past interactions and context can build on prior work rather than starting fresh every session.

Planning — sophisticated agents break goals into steps, monitor progress, and adapt when plans fail.

When to Use Agents vs Simpler Approaches

Use an agent when: the task requires multiple steps, requires real-world actions, or benefits from persistent context.

Use a simple prompt when: a single LLM call with good prompting is sufficient. Agents add complexity and cost — only justify them when necessary.