A practical way to learn AI at work is to use it on your own tasks every day and practice a few specific decisions: which jobs to hand to it, how to ask for what you want, what to show it, and how to check what comes back. Courses and videos teach you the words. The skill comes from making those calls yourself and seeing what happens.
Start with the tool you are allowed to use
Before you paste anything into an AI assistant, find out what your company allows. Many companies have an AI policy that names the approved tools and says what may and may not go into them. If yours has one, read it. If it doesn’t, ask your manager or IT team before you put work documents into a chat.
You don’t need to pay to practice. Claude and ChatGPT both have free plans. Pick the one your workplace uses, so what you learn carries straight over to your job.
Learn the few words that change what you do
You don’t need a technical background, but a few terms explain most of the surprises people hit:
- Prompt: what you type to the AI. Better called a brief, because it works best when it reads like instructions to a new colleague.
- Hallucination: when AI states something false as if it were true. Anthropic’s own guide says even the most advanced models can sometimes generate text that is factually incorrect. It sounds just as confident when it is wrong.
- Context window: all the text the AI can look at while it answers, including your messages, any files you add and its own reply. Anthropic describes it as the model’s “working memory”, and notes that more isn’t automatically better: very long chats can make answers less accurate.
Our guide to AI terms explained covers more of them in plain language.
Practice five everyday decisions
Almost everything you do with AI at work comes down to five decisions. Get good at these and the rest follows.
1. Is this a job for AI?
AI is good at first drafts, summaries, rewording, pulling points out of a long document, and turning rough notes into something tidy. It is weaker at exact arithmetic, current facts it can’t look up, and anything where one wrong detail does real harm. Before you start, ask: if the answer is slightly wrong, will I notice, and does it matter? If you wouldn’t notice and it would matter, check it closely or use a different tool. A spreadsheet is still the right place for the sum.
2. Say what you want back
The most common mistake is a one-line request followed by three rounds of “no, shorter” and “no, more formal.” Say up front what you want back (an email, a table, five bullet points), who it is for, any limits (length, tone, what it must not say) and what counts as done. Our guide to writing a good prompt goes further.
3. Give it what it needs to see, and nothing it shouldn’t
For work tied to your facts, AI needs the source in front of it. If you ask it to summarize the meeting but don’t paste the notes, it will fill the gap with something plausible. Give it the document, the figures, the earlier email. Leave out what it doesn’t need, especially passwords, personal details and anything your policy keeps out. And remember that a new chat may not remember the last one: what it keeps depends on your assistant and its settings. See how to give AI the right context.
4. Check it before it goes anywhere
Read the answer against what you know. Check every name, number, date and quote. Look for anything it added that you didn’t give it. If something matters and you can’t confirm it, don’t send it. Our guide on when to trust what AI tells you lists the mistakes to look for first.
5. Save what you repeat
When you find yourself writing the same instructions for the same job every week, write them down once. Some assistants let you save instructions as a reusable “skill”: a short file that tells the assistant how to do one job your way. Our guide to what a Claude skill is shows how.
A simple way to start this week
Pick one small, low-risk task each day: a reply you would normally write, a document you need to skim, notes you need to tidy. Do it with AI. Before you start, write down what a good result looks like. Afterward, compare. Did you have to rewrite it? What did it get wrong? What would you tell it differently next time?
Keep a short note of what worked. After a week or two you will have a handful of briefs you reuse and a clear sense of which jobs are worth handing over.
How to tell you are getting better
Hours spent and courses finished don’t tell you much. Look at what you do instead:
- You rewrite AI’s answers less, because your first request was clearer.
- You catch its mistakes before anyone else sees them.
- You know which jobs not to hand over, and you say why.
- You start a new chat on purpose, with the background it needs.
- You reuse instructions instead of typing them again.
For more on measuring this, see what AI fluency is, and how to measure it.
Mistakes to avoid
- Trusting a confident answer. Confidence tells you nothing about whether it is right.
- Only watching. Videos and articles help, but you learn the judgment by making calls on real work.
- Starting with something high-stakes. Practice on work where a mistake is cheap, then move up.
- Pasting everything in. More text isn’t better. Give it what the job needs.