agents · Level 2
What is an AI agent? Agentic AI explained
How agents differ from chatbots and fixed workflows, how the loop works, what can go wrong, and when not to use one.
Start with the essentials
The short answer
An AI agent is a system that pursues a goal by choosing its own steps: it uses a model to decide what to do, calls tools such as search or file access, looks at the results and repeats until it decides the goal is met. A chatbot answers once. A workflow follows steps a person fixed in advance.
What you will learn
- You will be able to tell a chatbot, a fixed workflow and an agent apart, and say who decides the next step in each.
- You will be able to name the four parts of an agent and walk through its loop.
- You will understand levels of autonomy and be able to place a product claim on them.
- You will be able to describe the main risks: compounding errors, prompt injection, over-broad permissions and runaway cost.
- You will have a paper design for an agent, with limits and human checks built in, or a reasoned decision not to use one.
Who it is for
Anyone who has used a chatbot and keeps hearing about agents, at work or in the news. No coding is needed.
Before you start
- Some experience of using an AI chatbot. What is AI and Your first prompts are ideal preparation.
Keep learning
The complete workbook
This workbook explains AI agents without hype. You will learn how an agent differs from a chatbot and a fixed workflow, what its parts are, how its loop runs, what agentic AI really means, which risks matter most, and how to design one safely on paper.
- 01Chatbot, workflow or agent?In the workbook · 1 exercise
These three words get mixed up constantly. One question separates them: who decides the next step?
- 02The parts of an agent, and the loopIn the workbook · 1 exercise
Strip away the marketing and most agents have the same four parts and run the same loop.
- 03Levels of autonomy, examples and the word 'agentic'In the workbook · 1 exercise
Autonomy is a dial, not a switch. Knowing where a tool sits on the dial cuts through the hype.
- 04One agent or several, and when not to use oneIn the workbook · Reading
More agents is not automatically better. Often the best agent is no agent at all.
- 05Risks, and keeping a human in the loopIn the workbook · 1 exercise
Agents are useful because they act. That is also why their failures matter more than a chatbot's.
- 06Design an agent on paperIn the workbook · 1 exercise
The best way to understand agents is to design one carefully, then decide whether to build it at all.
Also inside: a 9-point checklist, a glossary of 12 terms and 10 questions and answers to test yourself. 5 hands-on exercises, each with a worked answer at the back where the workbook gives one.
No login, no card, no account. Before the download we ask you to follow Mickai (two quick links). Free to download and use for personal learning, study groups and inside your own team. Please do not resell the workbooks or republish them as your own. Link people to trust-agent.ai instead.
Test yourself
Questions and answers
What is an AI agent?
It is a system that works towards a goal by choosing its own steps. A model decides what to do, tools carry out actions, and the results feed back in. The loop repeats until the goal is met or a limit stops it.
How is an agent different from a chatbot?
A chatbot replies once and you decide what happens next. An agent takes a goal, then plans, acts with tools, reads the results and continues by itself. That freedom makes it more useful and also riskier.
What does agentic AI mean?
It describes AI that pursues goals over several steps and actions instead of answering once. It is a spectrum and the term is used loosely, so ask who decides the next step, what it can change, and who checks.
Is an agent the same as a workflow or automation?
No. In a workflow a person fixes the steps in advance and an AI may perform some of them. In an agent the model chooses the steps as it goes. Workflows are more predictable, and often the better choice.
What can go wrong with AI agents?
Errors can compound across steps, hidden instructions in web pages or emails can hijack the agent, broad permissions can let a mistake do real damage, and loops can waste money. Limits, approvals and checks reduce each risk.
What is prompt injection?
It is when text an AI reads, such as a web page or email, contains instructions that try to redirect it. Models cannot reliably tell your instructions from content they read, so treat all such content as untrusted and limit what the agent can do.
What does human in the loop mean?
A person stays involved: approving risky actions before they happen, reviewing results afterwards, and able to stop the system at any time. The more costly or irreversible the action, the more important the human check.
Do I need multiple agents?
Usually not. One agent is simpler, cheaper and easier to follow. Add more only when you can name a specific problem they solve, such as separate roles or an independent check, and accept the extra cost and complexity.
When should I not use an agent?
Avoid an agent for a single question, for repeatable steps a workflow can handle, for tasks where errors are costly or cannot be undone, and when you cannot limit its access or define what done means.
Are AI agents safe to use at work?
They can be, with care. Give minimum permissions, require approval for risky actions, work on copies, keep logs and follow your organisation's rules. Start with low-stakes tasks, and remember that you remain accountable for the results.
When you have finished
Get your certificate of completion
Type your name and download a certificate for this workbook as a PDF, ready to print or to add to LinkedIn. It is made on your own device, so your name is never sent to us. It is a self-declared certificate, not an accredited qualification.
Learn the language
Key terms
- Agent
- A system that pursues a goal by choosing its own steps, using a model to decide, tools to act and results to adjust.
- Agentic AI
- A loose label for AI that works towards goals through several steps and actions rather than answering once.
- Workflow
- A sequence of steps designed in advance by a person. An AI may perform one or more steps, but the order is fixed.
- Tool
- An action an agent can take, such as searching the web, reading a file or sending a message. Each tool is a permission.
- Memory
- What an agent can draw on: the current context (short-term) and information saved outside the model (long-term).
- Agent loop
- The repeating cycle of plan, act, observe and decide that an agent runs until the goal is met or a limit is reached.
6 of the workbook's 12 terms. The complete glossary is in the workbook.
Follow the evidence
Sources and checks
Facts last checked: .
These workbooks use AI assistance. See how the workbooks are made.
- Prompt injection is not SQL injection (it may be worse)National Cyber Security Centre
- LLM01:2025 Prompt InjectionOWASP Gen AI Security Project
Created by Mickarle Wagstaff-Irons - Micky Irons with the Mickai team. Published by Mickai LTD. Last updated 25 September 2026.
NextKeep going
Where to go next
Recommended for you
The loop harness: how AI agents actually run
How an agent harness runs a model in a loop: context, tools, permissions, budgets, stop conditions, logging and testing, with a minimal Python loop and a paper design exercise.
Recommended for you
AGI explained: what it means and what it does not
A balanced guide to artificial general intelligence: the main definitions, one published levels framework, how to read headlines and timelines, and what matters in practice.
Recommended for you
Build your first chatbot
Build a chatbot that behaves: model access, system prompt, history and context limits, guardrails, tone, a test set, logging and cost, on a no-code path or a code path.