AI Agents — When AI Takes Action
- Understand the difference between a chatbot and an AI agent
- Know how agents work through a perceive-plan-act-check loop
- Recognise where you may already be encountering agents
- Know the practical safety principles for using agents
From Answering to Acting
Every AI interaction in this track so far has followed the same pattern: you type something, the AI responds with text. That is useful, but it is only one mode of AI. There is a second mode that is becoming increasingly common: AI that does not just answer questions — it takes actions.
This kind of AI is called an agent. The difference is clearest with an analogy.
A chatbot is like calling a travel agent and asking them to describe what a trip to Tokyo would involve. An agent is like calling a travel agent, saying "book me a trip to Tokyo for next month" and coming back to find the flights selected, the hotel reserved, and a packed itinerary in your inbox.
One gives you information. The other gets things done.
How Agents Work
Agents follow a loop rather than a single exchange:
- Perceive — the agent reads the goal you have given it and any relevant context (files, emails, a web page, a codebase)
- Plan — it breaks the goal into steps
- Act — it uses tools to take each step: searching the web, reading a file, writing code, calling an API, sending a message
- Check — it evaluates the result of each step and adjusts
- Repeat — it continues until the goal is complete or it needs your input
What makes agents possible in 2026 is two things arriving together: models became reliable enough to follow multi-step plans without going off course, and they got connected to real tools — your calendar, your files, the web, your code editor.
Where You Already Encounter Agents
Agents are no longer experimental. You may already be using them:
- Gemini on Android — when Gemini completes a cross-app task like placing a food delivery order or scheduling an appointment, that is agentic behaviour. It is taking actions on your behalf across multiple apps.
- GitHub Copilot Agent Mode — open a task, and Copilot plans and executes the code changes across multiple files autonomously, rather than suggesting one line at a time.
- Claude Code — give it a software task and it reads the codebase, plans the changes, edits files, runs tests, and iterates — without you directing each step.
- ChatGPT with Operator — OpenAI's agent that can navigate websites and fill out forms on your behalf.
- n8n AI Agents — automated workflow agents that combine LLMs with triggers, data sources, and external services to run entire business processes.
What This Means for You as a Beginner
You do not need to build or configure agents right now. But understanding the concept matters for two reasons.
First, you will encounter them in products. When Gemini offers to complete a multi-step task on your phone, or a customer support widget says "I can take care of that for you," you are dealing with an agent. Understanding what that means helps you use it deliberately rather than accidentally.
Second, agents have a different risk profile from chatbots. A chatbot that gets something wrong gives you a wrong answer. An agent that gets something wrong takes a wrong action — sends the wrong email, books the wrong thing, deletes the wrong file. That distinction matters when you decide what to let an agent do on your behalf.
Using Agents Safely
Two practical principles for anyone starting to use AI agents:
- Start with reversible tasks. Drafting a document is reversible. Sending an email to 500 people is not. Build trust with an agent on low-stakes tasks before letting it take actions that are hard to undo.
- Review before high-stakes actions. Good agents pause and ask for your confirmation before taking irreversible actions. If an agent does not ask before sending, posting, purchasing, or deleting, that is worth noticing.
Agents are one of the most powerful directions AI is heading in 2026 and beyond. This lesson gives you enough of the concept to understand what you are dealing with — later tracks on this site go deep into building and working with agents if you want to go further.
- A chatbot answers questions; an agent takes actions — a fundamentally different mode of AI
- Agents work in a loop: perceive the goal, plan steps, act with tools, check results, repeat
- You already encounter agents in Gemini on Android, GitHub Copilot Agent Mode, Claude Code, and ChatGPT Operator
- Agents have a different risk profile from chatbots — wrong actions, not just wrong answers
- Start with reversible tasks and look for confirmation prompts before irreversible actions
Progress is saved in this browser only — clear your cache or switch device and it's gone. Create a free account and it follows you everywhere.