Blog / Build with AI /
Teaching non-developers to build with AI
Handing someone an AI coding tool and saying “go build something” is like handing someone a chainsaw and saying “go make furniture”. They’ll produce something. It may not be what they intended, and it may not be safe.
My argument: the best way to teach non-developers to build with AI is to teach judgement before technique. Prompts and tools change every few months. What lasts is knowing what to build, how to check it and when to stop.
Training non-technical people is something I’ve done in a related field: writing documentation and teaching colleagues and clients to run content systems. At Royal Arts, clear CMS training and documentation improved our publishing efficiency by 30%. The principle that made it work is the same one I’d use here: people learn by doing a real job, with a safety net.
Why teaching this is different
AI hides a lot of complexity, which is both the appeal and the risk. A beginner can get a working page in ten minutes. What they don’t get automatically is the understanding that tells them when it’s wrong. So the teaching has to rebuild that understanding on purpose.
If you’re teaching a colleague, a client, a student or your team, here’s what I’d put in the course.
1. Start with a real, small problem
Not “learn to code”. Something the person actually wants:
- A calculator for their own business.
- A tracker for something they currently manage in a messy sheet.
- A simple page that collects form answers.
A real problem keeps motivation high and makes the lesson stick. Small examples worth borrowing: a shipping cost calculator, a loan and savings calculator or a bookmarks manager.
2. Teach them to describe before they ask
The first skill isn’t prompting. It’s describing: who uses this, what goes in, what comes out, what must never happen. Have learners write that on paper before opening any tool. Then show how much better the AI’s first attempt gets. How to describe a tool to AI so it builds the right thing is the longer version.
3. Teach thin slices
Show them the difference between asking for a whole app and asking for one small step. Let them experience both. The big request usually breaks and they can’t tell why. The small one works, and they know exactly where they are.
4. Teach checking as a habit
This is the lesson I’d spend the most time on. Give learners a deliberately flawed output, such as a calculator with a wrong formula, and ask them to find the error. Then give them these questions to carry forward:
- Does it do what I asked on a normal example I can check by hand?
- What happens with an empty field, a zero or a very large number?
- Did anything change that I didn’t ask for?
A learner who checks is safer than one who prompts brilliantly.
5. Teach what not to paste
Spell it out: no passwords, no API keys, no customer lists, no client documents without permission. Many beginners don’t realise a prompt leaves their computer. A few minutes on this prevents real mistakes. Security basics for AI-built web apps covers more of the ground.
6. Teach them to read the output a little
You don’t need fluency. You do need enough to see the shape: here’s the HTML, here’s the part that does the calculation, here’s where the data is stored. Ask the AI to explain its own code line by line, then have the learner explain it back in their own words. If they can’t, the code is too complicated for them to own.
7. Teach them when to ask for help
A responsible builder knows where their ability ends: payments, user accounts, anything with sensitive data, anything on a live business site. At that point you bring in a developer. Asking a developer to review AI-written code shows how to do it.
What to leave out at first
- Frameworks and tooling. A single HTML file teaches more than a build system.
- Model comparisons. Pick one tool and learn it.
- Prompt “tricks”. Clear writing beats tricks.
- Anything that needs a server, until they’re ready.
A four-week outline
| Week | Focus | Outcome |
|---|---|---|
| 1 | Describe a problem; build a one-page tool with fake data | A working page they can explain |
| 2 | Real data; checking and edge cases | A tool they trust |
| 3 | Fixing and changing one thing at a time; keeping versions | Confidence to edit |
| 4 | Privacy, sharing and handing over | A tool they could show someone |
The counter-argument
“Why teach non-developers to build at all? Hire a developer.” Fair for anything serious. But for small, personal or internal tools, the cost of a developer outweighs the benefit, and people with a problem often understand it better than any outsider. How much coding do you need to know to build with AI looks at where the line sits.
There’s also a benefit that goes beyond the tool. People who build a little understand what developers deal with, which makes them better clients and colleagues. Building with AI made me a better problem solver touches on that.
Key takeaways
- Teach judgement first: describing, checking and knowing limits.
- Use real, small problems.
- Keep steps small, and keep versions.
- Make “what not to paste” explicit.
- Know when to call a developer.
Related reading
- Git basics for non-developers before you build with AI
- Testing AI-written code when you are not a developer
- GitHub Copilot explained for non-developers
If you were going to teach one person this month, who would it be, and what would their first small tool be?