Agentic AI for Everyone — The Beginner's Course (Optional)
---
Module 1 — You Already Use AI Every Day
**Step 1 (instruction) — Relax: you're not behind**
If you've never "used AI," here is some good news: you have, today, probably before breakfast. When your phone suggested the next word of a message — that was AI. When Maps rerouted you around traffic, when your email quietly moved junk out of sight, when YouTube knew exactly which video to offer next — all AI. Artificial intelligence is simply software that learns patterns from examples instead of following fixed rules a programmer typed out. Nothing about it is magic, and nothing about it requires a technical mind. You've been living with it comfortably for years. This course just turns the lights on.
**Step 2 (instruction) — The one-sentence definition**
Hold onto this for the whole course: **AI is software that learned from examples how to make useful guesses.** Your phone's keyboard saw billions of sentences, so it guesses your next word well. A spam filter saw millions of junk emails, so it guesses which new ones are junk. That's the entire trick — at small scale it autocompletes a text message; at enormous scale it writes essays, answers questions, and plans tasks. Different sizes of the same idea: learn from examples, then guess usefully.
**Step 3 (instruction) — Try it now (30 seconds)**
At the bottom-right of this tracker is a green button labelled **ASK** — the school's own AI assistant. Open it and type: *"Explain in one sentence what you are."* Read the answer. Congratulations — you have just knowingly used an AI assistant, possibly for the first time. Notice two things: it answered in ordinary language, and it told you what it can and can't see. Keep ASK in mind; you'll practise on it throughout this course.
**Step 4 (quiz)**
*Which of these is an example of AI you have probably already used?*
- ✅ Your phone suggesting the next word while you type
- A calculator adding two numbers
- A light switch
- A printed dictionary
---
Module 2 — How a Chatbot Actually Works (No Maths, Promise)
**Step 1 (instruction) — Autocomplete, scaled up**
Tools like ChatGPT, Claude, and our own ASK are built on something called a **large language model** — an "LLM." Strip away the jargon and an LLM does one thing: it predicts the next word. Given "The cat sat on the…", it has learned that "mat" is likely and "helicopter" is not. Now scale that up: the model has read a colossal library of human writing, and it predicts not one word but word after word after word — and out comes a paragraph, an explanation, a lesson plan. It seems miraculous; underneath, it is prediction done extraordinarily well.
**Step 2 (instruction) — Why "just predicting words" is more than it sounds**
Here is the surprise researchers themselves didn't fully expect: to predict the next word *really well*, a model has to absorb a great deal about how the world works. To complete "The trophy didn't fit in the suitcase because it was too…", you need to grasp which object "it" refers to — and so does the model. Out of pure prediction, something that behaves like understanding emerges. A fair mental picture, used widely in training circles: an **eager, endlessly patient intern with a vast memory** — remarkably capable, works at incredible speed, and still needs a supervisor. You are the supervisor.
**Step 3 (instruction) — What it doesn't do**
Three honest limits to carry with you. First, an LLM doesn't "look things up" the way you check a book — it generates from patterns, which is why it can be wrong while sounding right (next module). Second, it has no goals, feelings, or intentions — it is not "trying" to do anything; it completes text. Third, it only knows what was in its training and what you tell it in the conversation. Our ASK, for example, is deliberately restricted to the tracker's own guidance and the KHDA framework — when it doesn't know, it is built to say so rather than improvise. That restriction is a feature, not a weakness.
**Step 4 (quiz)**
*At its core, what does a large language model do?*
- Searches the internet for every answer
- ✅ Predicts the next word, over and over, based on patterns it learned
- Copies answers from a hidden encyclopaedia
- Thinks and feels like a person
---
Module 3 — When AI Gets It Wrong
**Step 1 (instruction) — Confident nonsense has a name**
Sometimes an AI states something false with total confidence — a date that's wrong, a book that doesn't exist, a quote nobody said. This is called a **hallucination**, and it is the single most important thing a beginner must know. It happens because the model is generating *plausible* text, not retrieving *verified* facts — and plausible and true are not the same thing. This is not rare: studies of professional AI research tools have measured fabricated or misleading answers in a meaningful share of responses. The polish of the writing tells you nothing about the truth of the content.
**Step 2 (instruction) — Why it flatters you, too**
There is a second, sneakier failure: AI tools are trained to be agreeable, so they tend to go along with whatever you suggest — researchers call this **sycophancy**. Tell one confidently that the Earth has two moons and ask it to elaborate, and a weaker model may politely build on your error. The lesson: an AI agreeing with you is not evidence you are right. If you want honest output, invite correction — ask "what's wrong with this idea?" rather than "isn't this a great idea?"
**Step 3 (instruction) — The supervisor's rule**
One rule covers everything: **AI drafts, you verify.** Never pass on a fact, date, statistic, quote, or citation from an AI without checking it — especially anything that will reach a student, a parent, or an inspector. Treat AI output the way you'd treat work from a bright new intern: a strong first draft, reviewed before it leaves your desk. And remember bias: models learn from human writing, which carries human prejudices — outputs can quietly reflect them. Your judgement is the quality control the machine cannot supply for itself.
**Step 4 (quiz)**
*An AI gives you a beautifully written paragraph containing a statistic. What should you do before using it in a parent letter?*
- Use it — it sounds professional
- ✅ Verify the statistic independently first; AI can state false things confidently
- Ask the same AI if it's sure, and accept a yes
- Add more statistics from the same answer
---
Module 4 — Talking to AI: The Five Rules of Asking
**Step 1 (instruction) — The skill is the question**
Working well with AI is not a technical skill — it is a communication skill, and teachers are already good at it. The difference between a useless answer and an excellent one is usually the question. Universities now teach this as **prompt literacy**, and it reduces to five rules. **Rule 1 — Be specific.** "Help me with reading" gets you a vague essay; "Give me five comprehension questions for a Year 4 class reading a story about a lost kitten, two easy, two medium, one challenging" gets you something usable. Vague in, vague out.
**Step 2 (instruction) — Context, role, format**
**Rule 2 — Give context.** The AI knows nothing about your situation unless you say it: the year group, the subject, the constraint, the purpose. **Rule 3 — Give it a role.** Starting with "You are an experienced primary English teacher" genuinely shapes the answer's tone and depth. **Rule 4 — Say what shape you want.** A table, five bullet points, a 100-word summary, a letter — ask for the format and you'll get it. None of this is technical; it is exactly how you'd brief a capable colleague who just joined the school and knows nothing about it yet.
**Step 3 (instruction) — Rule 5, then try it now**
**Rule 5 — Don't accept the first draft.** The first answer is the start of a conversation: "shorter," "simpler," "more challenging," "that second point is wrong — fix it." The model takes correction without ego. Now practise, for real: open **ASK** and try a vague question — *"tell me about artefacts"* — then a five-rule question: *"I'm a Year 5 class teacher. In three bullet points, what makes a good artefact for my daily classroom-environment check?"* Compare the two answers. That difference is the whole skill, and you just learned it.
**Step 4 (quiz)**
*Which question will get the most useful answer from an AI assistant?*
- "Help me teach"
- "Make a worksheet"
- ✅ "You are a maths teacher. Create five practice questions on fractions for Year 6, from easy to hard, with answers."
- "Why is teaching hard?"
---
Module 5 — From Answering to Doing: What "Agentic" Means
**Step 1 (instruction) — The word everyone is suddenly using**
"Agentic" comes from **agency** — the capacity to act. A chatbot answers and stops; an **agent** is given a goal and carries out the steps to reach it. The cleanest everyday picture is travel: a chatbot, asked about Istanbul, describes Istanbul. An agent, told "book me three days in Istanbul under this budget," searches flights, compares hotels, checks dates against your calendar, assembles the itinerary, and brings it to you to approve. Same underlying intelligence — the difference is that the agent is allowed to *do*, not just *say*.
**Step 2 (instruction) — How an agent thinks (the loop)**
Every agent runs a simple loop you can hold in your head: **look, think, act, check.** It looks at the situation (what's the goal, what's known), thinks out a plan (what steps, in what order), acts (uses tools — searches, fills forms, drafts documents), then checks the result and adjusts. If a step fails — a page won't load, information is missing — it doesn't freeze; it tries another route, the way a resourceful colleague would. MIT researchers describe this generation of systems as able to perceive, reason, and act — which is precisely what separates them from the chatbots of two years ago.
**Step 3 (instruction) — What this means in a school**
Picture the difference in school terms. Responsive AI: a parent asks the fee and gets the fee. Agentic AI: an admissions enquiry arrives, and the system collects the documents, reads them, checks completeness, chases what's missing, schedules the assessment, and presents a ready file — with the human decision left firmly to a human. Marking support, evidence assembly, report drafting — agents are strongest exactly where school life is heaviest: repetitive, multi-step, data-shuffling work. What they cannot do is know a child. The judgement stays yours; the donkey-work increasingly doesn't.
**Step 4 (instruction) — What agents are not**
Three myths to retire. *Agents are not robots* — no metal hands; software completing tasks. *Agents are not unsupervised* — every serious deployment keeps humans approving the actions that matter, and ours is no exception: anything touching a child, a grade, or a record needs human sign-off before it commits. *Agents are not infallible* — they inherit every limitation from Module 3, which is exactly why the approval step exists. The autonomy is real, and so are the guardrails; the two arrive together or not at all.
**Step 5 (quiz)**
*What is the key difference between a chatbot and an AI agent?*
- Agents are physical robots
- Agents never make mistakes
- ✅ A chatbot answers questions; an agent carries out multi-step tasks toward a goal
- There is no difference
---
Module 6 — The Safety Rules (Children's Data Above All)
**Step 1 (instruction) — The one rule that protects everything**
Public AI tools (free chatbots on the open internet) may keep what you type and learn from it. So here is the rule international guidance keeps arriving at, and the one rule from this course to remember if you remember nothing else: **never put into a public AI tool anything you wouldn't want appearing in a stranger's search results.** For a school that means: no student names, no grades attached to names, no family circumstances, no medical or behavioural details, no photographs of children, no contact details. Not even "just to reformat a list." A child's data is a trust we hold, not a convenience we spend.
**Step 2 (instruction) — Anonymise, then ask**
The work itself is almost always shareable once the child isn't. "Write feedback for this Year 6 story" with the name removed is fine; the same request with name, class, and the family situation attached is a breach. Practical habit: before pasting anything into a public tool, scan it for names, IDs, and identifying details, and strip them. And know the difference in your tools: our own platform is the opposite of a public chatbot — closed to the school community, role-scoped so each person sees only their own data, encrypted, and logged. That is why school work belongs on school systems.
**Step 3 (instruction) — In the classroom: supervised, with a purpose**
When AI enters your lessons, two school rules apply. First, the **purpose-statement rule**: every classroom AI use has a stated teaching purpose — you can say in one sentence what the AI is for in this lesson. Second, **supervision**: students use AI tools under your guidance, on age-appropriate tools, never alone with the open internet — this matches the KHDA-aligned national programme's own design, which curates tools specifically for supervised use. And model the verification habit out loud: when the AI errs in front of the class, celebrate it — "let's check that" is one of the most valuable lessons a child can watch a teacher perform.
**Step 4 (quiz)**
*Which of these is SAFE to put into a public AI chatbot?*
- A list of student names and their grades
- A child's behavioural report
- ✅ An anonymised writing sample with all names and identifying details removed
- A class photo
---
Module 7 — You and AI: The Honest Conclusion
**Step 1 (instruction) — The human-centred mindset**
UNESCO's global framework for teachers in the AI era is built on one principle before any technical skill: a **human-centred mindset** — humans set the goals, humans stay accountable, and AI serves human judgement rather than replacing it. In practice: the teacher remains responsible for every pedagogical decision, however much AI assisted along the way. "The AI suggested it" is never an explanation a professional gives. The tools change; the accountability doesn't move an inch. That is not a burden — it is precisely why teachers cannot be automated.
**Step 2 (instruction) — What actually changes about your work**
The honest forecast, repeated from research institutes to government programmes: jobs change shape rather than disappear. The hours spent on repetitive assembly — formatting, collating, first drafts, chasing — shrink; the hours that need a human — knowing a child, judging a situation, inspiring a room — become a larger share of your day. The people who thrive are not the most technical; they are the ones who learned to supervise the new intern well: ask precisely, verify honestly, and keep their own judgement sharp. You have just completed the entire toolkit for that.
**Step 3 (instruction) — Your three next steps**
First: **use ASK daily** — every question you ask it is practice in the core skill, and it can't see student data, so it's a safe training ground. Second: **apply the supervisor's rule everywhere** — AI drafts, you verify, nothing unchecked reaches a child, parent, or record. Third: **take Course 2** — "The Direction" — which shows where the UAE is taking all of this nationally, and why your school is moving with it. You started this course having maybe never knowingly used AI. You finish it knowing what it is, how it works, when it lies, how to talk to it, what an agent is, and how to keep children safe around it. That is more than most professionals in any industry can say.
**Step 4 (quiz)**
*According to the human-centred approach, who is accountable for a teaching decision made with AI assistance?*
- The AI tool
- The company that made the AI
- ✅ The teacher — accountability for pedagogical decisions always stays human
- Nobody
---