How to start with AI when you are not technical

You do not need to learn AI. You need to hand over one task. That is the whole on-ramp for a non-technical person, and it takes about a week.

Most beginner advice gets this backwards. It opens with what a large language model is, moves on to prompt engineering, and by the time it reaches something you could do on a Tuesday morning you have closed the tab. Nobody learns a tool in the abstract. They learn it because something annoyed them enough.

Do I need to learn AI before I can use it?

No. You need two skills you already have: describing a task in plain words, and judging whether the result is any good. That is exactly what you do when you brief a person.

The belief that expertise has to come first is the main thing keeping smaller businesses out. In 2025, 17% of small EU enterprises used any AI technology, against 55% of large ones — and among businesses that considered AI and decided against it, the most common reason given was a lack of relevant expertise, cited by 71% (Eurostat).

That answer is fair from a company trying to build something. It is much less true for one person handing one recurring task to a chat assistant — but people read enterprise-shaped advice, conclude the door is shut, and never try the version that takes twenty minutes.

The useful question is not what can AI do. It is: what do I do every week that I resent?

Which task should I hand over first?

The most repetitive, lowest-stakes, most text-shaped thing in your week. If a candidate task fails any of those three tests, it is the wrong place to start — not because the tools cannot do it, but because you will learn nothing useful from the attempt.

That narrows the field fast. First tasks that pass all three:

• Turning messy meeting notes into the follow-up email.
• Drafting the first reply to a routine enquiry — the one you have now written 200 times.
• Cutting a long document down to the five things that actually matter.
• Turning a voice note recorded in the car into a to-do list.
• Writing the first draft of a proposal from your own bullet points.

None of that is exotic, and that is the point. The largest published study of how people actually use ChatGPT found three categories — practical guidance, seeking information and writing — account for close to 80% of all conversations (Chatterji et al., NBER, 2025). The mainstream use is mundane. Yours should be too.

How do I stop drowning in tool choices?

Choose the task first and the tool second, and start with a general assistant rather than a specialist one. Overwhelm is what happens when you go shopping without a list.

For a first task, a general chat assistant covers more ground than a purpose-built product, because you do not yet know precisely what you need. Specialist tools earn their place later, once you can tell whether the specialist is better. Free tiers are enough to run the test.

1. Name the task in one sentence, out loud.
2. Open one assistant. Do not open a second.
3. Brief it the way you would brief a capable new starter: what you want, who it is for, what good looks like, and an example of a previous one.
4. Read the output as an editor, not as a reader. Fix what is wrong.
5. Do the same task the same way next time. The saving comes from repetition, not cleverness.

Notice what is missing from that list: prompt engineering. Briefing clearly is a management skill, not a technical one, and you have been doing it for years. For the longer version of the choosing logic, what AI can actually do for a small business covers it.

What if I get it wrong?

Pick a task where being wrong is cheap, and stay in the loop as the editor. That removes almost all of the risk in week one.

The genuine hazard is that these tools are confident when they are wrong. Treat every output as a first draft from someone fast, willing and occasionally mistaken — you would not forward a new starter's draft to a client unread.

• Keep a first task away from money, tax, legal wording, medical detail, or anything a client sees unedited.
• Do not paste in client personal data, passwords or anything under an NDA before you have read the tool's data settings.
• If you could not tell whether the output was wrong, it is the wrong task to start with — you have no way to grade it.

Why does using AI feel embarrassing?

Because there is a measurable social cost, and you are not imagining it. Across four preregistered experiments with 4,439 participants, people who used AI at work both expected and received lower ratings of competence and motivation from others, and those judgements carried through into assessments of job candidates (Reif, Larrick & Soll, PNAS, 2025).

Worth saying plainly rather than pretending the worry is silly. It is not silly. It is also not a reason to stay out, because the finding is about how the method is perceived, not about whether the work is better. Nobody now asks whether you did your quote in a spreadsheet or on paper.

So judge yourself on the finished work. If the follow-up email is clearer, and it went out the same afternoon instead of four days later, your client got a better service. That is the only scoreboard that pays.

[Steve — one line here about the first time a client asked whether you had used AI, and what you told them.]

How do I know whether it actually saved me time?

Time the task. Your impression of whether it helped is not reliable evidence, and this is the most consistently uncomfortable finding in the research.

In a randomised trial, 16 experienced open-source developers worked through 246 real issues. With AI tools they took 19% longer. They had predicted they would be 24% faster, and even after finishing the work they still believed the tools had sped them up by about 20% (METR, July 2025).

That was skilled programmers on complex code they knew well, close to the hardest case, so it does not mean AI slows everyone down. It does mean it feels faster is not a measurement.

The broad numbers are equally sobering. Self-reported time savings from generative AI across the US workforce came to roughly 1.4% of total work hours, in a survey where 23% of employed people had used it for work at all in the previous week (Bick, Blandin & Deming, NBER). Averages that thin tell you nothing about your own week. Only your stopwatch does.

What does the first week actually look like?

1. Monday. Name the task. Write down how long it currently takes and how often you do it.
2. Tuesday to Thursday. Do the task with the assistant every time it comes up. Note the minutes, including the time you spend fixing the output.
3. Friday. Compare. If you saved time and the quality held, keep it — then leave it alone for a fortnight before adding anything else.
4. If it did not work, drop it without ceremony and pick a different task next week.

Two or three cycles of that and you have handed over two or three tasks, plus a real sense of what these tools are and are not good at. Better foundation than any course. If you are stuck for a first task, the answer is usually hiding in your admin — how many hours a week do you lose to admin is the place to look.