Skip to main content

Agent

An AI that can take actions, not just respond to questions.


Definition

An agent (智能体) is an AI model that doesn't just generate text — it can also:

  • Read files on your computer
  • Edit documents
  • Run commands
  • Browse the web
  • Use external tools

The key difference from a regular chatbot: an agent acts on your environment, not just inside a chat window.


How It Works

An agent is not a special kind of AI. It's the same language model (like Claude or GPT), but with:

  1. Tools — Capabilities to interact with the outside world (file system, web, databases)
  2. Autonomy — The ability to decide which tools to use and in what order
  3. Feedback loops — It can observe the results of its actions and adjust

When you ask an agent to "proofread my document," it:

  1. Decides it needs to read the file (uses the Read tool)
  2. Analyzes the content
  3. Decides to edit the file (uses the Edit tool)
  4. Reports what it did

Examples

Not an AgentAgent
ChatGPT web interfaceClaude Code
Copy-paste workflowDirect file editing
You run the commandsAI runs the commands

Why It Matters

The shift from chatbots to agents is the paradigm shift this guide is built around. Instead of AI being a passive advisor, it becomes an active collaborator that can do work on your behalf.


  • Prompt — Instructions that guide agent behavior
  • MCP — A protocol for giving agents new tools

See this term in context: Task 1: Grammar Check