What is an Agent?

Understand the essence of AI agents: autonomous systems that perceive the environment, make decisions, and act to achieve goals.

Definition of an Agent

An AI agent is a system that can make autonomous decisions and execute tasks. Unlike traditional software, an agent does not simply follow a preset instruction sequence; it can:

• Perceive - Understand the current environment and context
• Decide - Choose the best action based on goals and constraints
• Act - Execute the chosen operation and observe results
• Learn - Adjust strategy from feedback

Agent Loop (Core Cycle)

An agent operates on a core loop, which is the foundation for understanding all agent systems:

while not task_complete:
    observation = perceive(environment)
    action = think(observation, goal, memory)
    result = act(action)
    memory.update(observation, action, result)

import { query } from "@anthropic-ai/claude-agent-sdk";

// The simplest possible agent: query() IS the agent loop
const response = query({
  prompt: "List the files in the current directory",
  options: {
    model: "claude-sonnet-4-5",
    allowedTools: ["Read", "Bash", "Glob"]
  }
});

// Stream messages — each is a step in the loop
for await (const message of response) {
  switch (message.type) {
    case "system":    // Session init
      console.log("Session:", message.session_id);
      break;
    case "assistant":  // Agent reasoning
      console.log("Think:", message.content);
      break;
    case "tool_call":  // Agent acts
      console.log("Act:", message.tool_name);
      break;
    case "tool_result": // Observe result
      console.log("Result:", message.content);
      break;
    case "result":     // Loop complete
      console.log("Done:", message.content);
      break;
  }
}

This loop is called the Perception-Action Loop, a fundamental pattern for intelligent systems.

Agent vs Chatbot

Many people confuse agents with chatbots. Here are the key differences:

Core Components of an Agent

A complete agent system usually includes the following components:

Real-World Examples

Here are some well-known agent implementations:

• Claude Code - Anthropic's coding agent that understands code, uses tools, and completes development tasks
• AutoGPT - Early autonomous agent experiments aiming for fully autonomous task execution
• Devin - Cognition's software engineering agent
• OpenClaw - Open-source agent development framework

Next Steps

Now that you understand the basics of agents, you can dive deeper:

• Agent Architecture - Learn the system design of agents
• LLM & Prompts - Learn how to design agent prompts
• Tools & Actions - Explore best practices for tool use