What Is an Agent?

An agent is a system that uses a language model to decide what to do next, takes an action, observes the result, and repeats.

The minimal formula

Agent = Model + Tools + Loop

• Model — reasons and plans.
• Tools — let it act (edit files, run commands, search).
• Loop — keeps it going until the goal is met.

Chat vs agent

A chatbot gives you one answer. An agent:

1. Reads the request.
2. Decides the first action.
3. Executes it.
4. Looks at the result.
5. Decides the next action.
6. Repeats until done — or until it runs out of steps or money.

That's why an agent can do a multi-step task (rename a function across 40 files, run tests, fix the fallout) while a chatbot can only describe how to do it.

The loop in code

A minimal agent loop is surprisingly small:

async function agentLoop(goal, tools, maxSteps = 10) {
  const history = [{ role: 'user', content: goal }];
  for (let i = 0; i < maxSteps; i++) {
    const reply = await model(history, tools);     // model decides
    if (reply.done) return reply.answer;           // finished?
    const result = await runTool(reply.toolCall);  // act
    history.push(reply.message, result);           // observe
  }
  return 'Stopped after ' + maxSteps + ' steps.';
}

That's the whole magic: decide → act → observe → repeat.

Why limits matter

Without a maxSteps cap and a budget, an agent loops forever. Real agents add:

• A step/budget limit.
• A "done" signal.
• Permission checks before risky actions.

If you remember one idea from this course: an agent is a loop, and the loop is where all the interesting engineering happens.