Chapter 07 · AI Agents

AI Agents — when AI stops answering and starts doing

Until recently, AI tools worked like a smart reference desk: you asked, it answered. Agents are something different. They don't just respond — they take actions, use tools, make decisions across multiple steps, and work toward goals you set.

Chatbot vs. agent — the difference that matters

Chatbot
Agent
Tells you how to book a flight
Actually books the flight
Explains how to summarize emails
Summarizes your emails automatically
Answers one question per turn
Pursues a goal across many steps
Produces text
Takes actions in the real world
Stops when it responds
Keeps going until the job is done

Agents accomplish this by combining a language model with the ability to use tools: browsing the web, running code, reading and writing files, sending emails, interacting with software. The language model acts as the brain — deciding what to do next — and the tools give it hands.

How an agent works — the action loop
You set
a goal
Agent
plans steps
Uses a
tool
Reads
result
Next
step
Goal
complete
Unlike a chatbot, which produces one response then stops, an agent loops until the task is done — adjusting its plan based on what it finds.

"The shift from AI that answers to AI that acts is the most significant change happening in the field right now."

Agents you can use today

These are not hypothetical. Each category has real tools in active use that you can access now.

🔬

Deep research agents

Give these a research question. They browse dozens of sites, synthesize across sources, resolve contradictions, and produce a structured report — in minutes rather than hours. They cite sources so you can verify.

Try: Perplexity Deep Research · ChatGPT Deep Research · Gemini Deep Research
💻

Coding agents

Write the code, run it, read the error messages, fix them, and iterate until it works. Non-programmers are building functional tools they couldn't have created otherwise. Experienced programmers move 5–10x faster on routine tasks.

Try: GitHub Copilot · Cursor · Claude Code · Replit Agent
📋

Task automation agents

Connect AI to your existing apps — email, calendar, spreadsheets, Slack. Describe a workflow in plain language and the agent runs it automatically. No programming required.

Try: Zapier AI · Make · n8n · Claude Projects
🗂️

Knowledge agents

Feed these your own documents — reports, notes, contracts. They build a private knowledge base and let you ask questions across all of it. They answer from your documents, not the general internet.

Try: NotebookLM · Claude Projects · Notion AI · Mem
🖥️

Computer-use agents

Agents that can see your screen and control your computer — clicking buttons, filling forms, navigating websites. You describe a task; the agent executes it the way a person would. Currently early-stage, moving fast.

Try: Claude Computer Use · Operator (OpenAI)

What to watch for

Agents make mistakes — and because they take real actions, their mistakes can have real consequences. An agent that sends emails or modifies files creates problems that are harder to undo than a wrong answer in a chat window.

The right approach: start with low-stakes tasks. Keep humans in the loop for anything consequential. Think of a new agent like a new hire on their first week — smart and willing, but not yet proven in your specific context. Start bounded, watch how they handle it, expand their autonomy as they earn it.