Welcome
This is for anyone who uses a computer for work but has never really used AI — and feels a bit wary of it. There's no coding here, and no jargon. By the end you'll be able to use AI for everyday tasks safely: getting help without handing over things you shouldn't, and checking answers instead of trusting them blindly.
Two habits run through everything: check the answer, and protect your data. Learn those two and you're ahead of most people already using AI.
Each module takes 5–10 minutes and ends with a small "now you try." Do them in order, at your own pace. It's fine to do a module sitting next to a colleague or family member.
Module 1 — What AI is, in plain terms (and where it gets things wrong)
The big idea. An AI assistant is a very good guesser of words. You type something, and it predicts the most likely helpful reply, one word at a time. That makes it brilliant at writing, explaining, and rephrasing — and surprisingly bad at facts and arithmetic, because it isn't looking anything up or doing real maths. It's predicting what an answer probably looks like.
Why this matters. The tricky part is that a wrong answer looks exactly as confident as a right one. There's no wobble in its voice when it's guessing. So the skill isn't trusting it or distrusting it — it's knowing when to check.
The one habit to start with: let AI help with words, and check anything that's a number or a fact you'll rely on.
A quick example. Ask an assistant to "rewrite this email to sound friendlier" and it'll do a lovely job — words are its home turf. Ask it "what's 17% of $4,310 and what was our total last March?" and you should pause: the percentage it might fumble, and "last March" it simply cannot know. Same tool, two very different levels of trust.
Open any AI assistant. Ask it one wording question ("give me three friendlier ways to say 'please reply by Friday'") and one number question ("what is 23% of 880?"). Notice which answer you'd happily use as-is, and which one you'd want to double-check with a calculator.
AI predicts words; it doesn't know facts or do reliable maths. Wrong looks the same as right. Words: go ahead. Numbers and facts: verify.
Module 2 — How to ask so you get something useful
The big idea. Good results come from clear asking, not secret tricks. If you'd struggle to hand the task to a new colleague with the instructions you gave, the AI will struggle too.
Four things a good request carries:
- What you want ("write a short reply declining a meeting").
- A bit of background ("it's from a supplier, and I want to stay polite").
- The format ("three sentences, friendly tone").
- An example, if you have one ("similar to how I'd say…").
It's a conversation. You rarely get the perfect answer first try, and that's normal. Read what comes back, then say "shorter," "warmer," "now make it sound like me." Each nudge gets you closer.
A quick example. "Write something about our delivery delay" gives you something generic. "Write a 3-sentence apology to a customer for a 2-day delivery delay, warm but professional, offering a discount on their next order" gives you something you could almost send.
Take a vague request you might normally type, like "help with my email." Rewrite it with the four parts above. Run both versions and compare — the difference is the whole skill.
Task, background, format, example. Then refine. Clear asking beats clever asking.
Module 3 — Keeping your information safe
> Please note: this module is general education, not legal or privacy advice. The rules differ by workplace and by country — when in doubt, check with yours.
The big idea. Some things are fine to share with an AI tool; some really aren't. The good news is there's a simple rule for telling them apart.
Why it matters. When you type into most public AI tools, your words travel to a company's computers elsewhere. That's fine for "rewrite this paragraph." It's not fine for a customer list, someone's personal details, passwords, or anything confidential — because you've now sent it somewhere you don't control.
A rule of thumb you'll remember: would you email this to a stranger? If not, don't paste it into a public AI tool.
A safer option, in one sentence. If you want AI's help with real, sensitive data, you can make "practice data" — made-up information that has the same shape as the real thing but none of the real details — and work with that instead. (You'll do this hands-on in Tier 2 with our Sandbox tool; for now, just know the option exists.)
Here's a short list. Sort each into "fine to share" or "keep private": (a) a paragraph from a public web page you want summarised; (b) a spreadsheet of customer names and phone numbers; (c) a draft blog post; (d) a colleague's salary. (Answers: a and c are fine; b and d stay private.)
Public AI tools send your words elsewhere. Would-you-email-a-stranger is your test. Sensitive data has a safer path: practice data.
Module 4 — Letting AI speed up your spreadsheet work
The big idea. AI can make the work you already do faster — as long as you check it. This is where Module 1's "check the numbers" habit earns its keep.
Three things it's great at:
- Explaining a formula you didn't write ("what does this
=VLOOKUP(...)do?"). - Drafting a formula you need ("give me a formula to add up column C only where column B says 'paid'").
- Cleaning messy text and dates ("turn these into a consistent date format").
The check, every time. When it gives you a formula or a cleaned column, spot-check one or two rows by hand. If the AI's total and your manual total of three rows disagree, trust your three rows.
A quick example. You paste a column of dates written ten different ways. The assistant offers a tidy version. Before you use it, glance at a few — did "3/4" become April 3rd or March 4th? That five-second check saves an hour later.
Take a small, non-sensitive spreadsheet (or make three rows up). Ask AI to explain or write one formula, use it, then verify the result against a manual count of a couple of rows.
AI speeds up explaining, drafting, and cleaning. Always spot-check. Your manual check beats its confident answer.
Module 5 — Your first safe "vibe code"
The big idea. You can describe a small improvement to your own work in plain English and have AI build it for you — that's all "vibe coding" is. And you can do it in a way your IT team can say yes to.
Without the scary word: you tell the AI what you want ("make me a little tool that renames these files by date"), it produces something, and you read back what it made and try it. You don't need to understand every line — but you do test it.
Doing it safely. Practise on a piece of your own process that isn't confidential, or on practice data — never paste private documents into a public tool. When you've built something useful and checked that it works, you have a small, data-safe win you can show your manager or IT — which is how you earn permission to do more, rather than working around anyone.
Why you still test it. Same as the number rule, in miniature: the AI wrote it, so you confirm it does what you wanted on a small example before trusting it on the real thing.
A quick example. "Write me a step-by-step I can follow to merge these three spreadsheets into one." You read it, try it on copies, and if it works, you've just saved yourself a recurring chore — safely.
Think of one small, repetitive part of your work. Describe it to an AI in plain English and ask for a simple way to do it faster. Read what it suggests. (You don't have to run anything yet — the goal is to feel how approachable this is.)
Vibe coding is just describing what you want in plain English. Do it on non-confidential work or practice data. Test before you trust. A safe, verified win earns you a yes.
Where to next
You've now got the two habits that make AI safe to use at work — check the answer and protect your data — and a first taste of building. Tier 2 takes this further: you'll make your own practice-data sandbox and build a real, working automation you can use.
This material reduces risk; it doesn't remove it. Edge, not guarantee.