Skip to content
Practical technology, explained simply

Artificial intelligence · making things

Building products with AI without programming

AI makes trying out an idea cheap without coding, but it does not decide for you. How to go from idea to a prototype you can check, in seven steps.

Ibrahin · Published on 2 September 2026 · 3 min read

Text drafted with artificial intelligence. The topic, the angle and the final review are Ibrahin's. How AI is used here

A laptop showing a cassette-shaped music player, surrounded by pencil sketches of that same screen, a cassette tape and a pair of headphones.
Illustration, not a real screenshot.

What has happened

Today you can put together a simple app, a form that answers by itself or a customer assistant just by describing what you want, without writing code. That has made trying out an idea far cheaper. What hasn't changed is the hard part: knowing whether anyone needs it and whether the result is reliable.

And since August 2024 the EU Artificial Intelligence Act has been in force; it applies in phases and sets obligations according to the risk of what you do with AI.

Why it matters to you

Because it's very easy to build something nice that nobody uses, or that works in the demo and fails with real data. A convincing AI answer can be wrong, and if your product hides that, your users pay for it. And if it needs personal data to work, data protection law applies from day one, prototype included.

What you can do

1. Describe the problem without mentioning technology

Complete this sentence: “Someone needs ___ and today they solve it by ___”. If the answer is “I want an AI app”, the problem is still missing. Talk to people who would use it and look for examples of a need that already costs them time, money or patience, not for them to say the idea is interesting.

2. Reduce the prototype to a single flow

One clear input, one process, one output you can check and a way to correct or repeat. For example, taking some notes and returning a task list. You don't need users, payments, dashboards or ten integrations to find out whether the output helps.

3. Decide what the AI can and can’t do

Write correct examples and cases it must reject. If it summarises, how it handles incomplete data. If it classifies, what it does when it isn't sure. If it drafts, a person reviews before sending. The product should show doubt or ask for review, not hide it.

4. Test with made-up data

Use fictional or public examples and check where the input is stored, who sees it, which service processes it and how it's deleted. If you'll eventually need personal data, decide beforehand which data is essential, who can access it and how long it's kept.

5. Measure before adding features

Run ten or twenty tests and note how many answers work without correction, which errors repeat, how long it takes, what each use costs and when someone decides not to use it. A new feature won't fix an accuracy problem you don't understand yet.

6. Decide where a person has to be involved

If the output can affect money, health, employment, privacy or reputation, human review isn't optional. Nor is it if whoever receives it can't easily check it. And explain where AI is involved and what responsibility stays with the person.

7. Choose the tool last

Compare price, limits, export, privacy, integrations and maintenance, not the flashiest demo. Ask what happens if the provider changes the plan or disappears, and keep content and decisions outside the tool.

The recommendation

Choose a single flow you can explain in a minute, test it with five people and write down the failures without dressing them up. Only then decide whether it's worth building more. And always keep test accounts, data and permissions separate from real ones.

Sources

Review

Responsible for publication: Ibrahin. Date: 2026-09-02. Last checked: 2026-09-02.