Blog / Build with AI /
The skills that matter most when building with AI
Every few months a new list appears: “10 prompts that will change your life”. Most are out of date by the time you’ve saved them.
My argument: the skills that matter most when building with AI aren’t about the AI. They’re about thinking clearly, working in small steps and checking what you’ve got. Tools and models will keep changing. These skills carry across every one of them.
Here are the six I’d practise first, and why.
1. Describing the problem clearly
Everything starts here. The AI can only build what you describe, and vague descriptions give vague results.
Describing well means knowing:
- Who uses it and what they’re trying to do.
- What goes in and what comes out.
- What must never happen.
- What “done” looks like.
It’s a writing skill as much as a technical one. People with a background in content, support or training often do this well because they’re used to explaining things to a reader who doesn’t know the context. How to describe a tool to AI so it builds the right thing goes through it.
Practise: before each build, write the brief in five sentences. Read it back as if you were the AI. What would you have to guess?
2. Slicing work into small steps
Big requests produce big, tangled results. Small ones produce something you can follow.
Slicing means asking: what’s the smallest piece that I can see working? Build that, check it and move on. It’s the main defence against the loop where every fix breaks something else.
Practise: take your next idea and write the first three slices before you start. Slice one should take under half an hour.
3. Checking results
AI output is confident whether it’s right or wrong. Checking is the skill that turns “it looks fine” into “it works”.
Good checking is concrete:
- Test with an example you can calculate by hand.
- Try the empty case, the huge case and the odd case.
- Compare with the source data.
- Test on a phone as well as a desktop.
- Look at what changed, not only what you asked for.
Prompting for tests along with the code helps, but your own eyes still matter most.
Practise: every time you get output, find one thing that’s wrong or missing. There’s nearly always something.
4. Reading what the AI gives you
You don’t have to become a developer. You do have to be able to read the shape of the code or the structure of the output well enough to ask good questions.
That means being comfortable asking, “What does this part do?” and “Why did you choose this approach?” and noticing when an explanation doesn’t hold together. How to ask AI to explain code line by line is a good habit builder, and reading code you did not write helps with the rest.
Practise: ask the AI to explain one function from your project each day, then explain it back in your own words.
5. Judging what’s worth building
AI makes it easy to build things nobody needs. The scarce skill is deciding what’s worth the effort and the upkeep.
Questions that help:
- Does this solve a problem I (or someone I know) actually has?
- Is there already a good tool that does it?
- What will it cost me to maintain?
- What’s the smallest version that would prove it’s useful?
AI makes producing more easy, but producing more isn’t automatically progress. When AI is the wrong tool for the job and building one feature well vs ten features badly are about this skill.
Practise: before building, write one sentence: “This is worth building because…”. If you can’t, wait.
6. Finishing and maintaining
A tool that works once on your laptop isn’t a product. Finishing means error messages, empty states, instructions and a way to fix things when they break. Maintaining means keeping it working as the world changes.
This is the least glamorous skill and the one that separates hobby projects from useful ones. Time-boxing builds so they actually finish is a practical approach.
Practise: finish one small thing completely, including a short note on how to use it, rather than starting three new ones.
What about prompting?
It matters, but less than people think, and it’s mostly the first skill in disguise. A clear description, examples of what you want and a statement of what to avoid beat any magic phrase. Giving AI examples: the fastest way to better output is the one prompting habit I’d put at the top.
A skill I’d add if you work on websites
Safe working. Staging copies, backups, version history and the habit of never testing on the live site. A web admin already has this, and it transfers directly. If you’re new to it, auditing your WordPress plugins before they slow you down shows the care that site work demands, and the same care applies to AI-built code.
The counter-argument
Someone will say: “Soon the models will handle all of this, so these skills won’t matter.” Models will get better at slicing and checking their own work, and that’s welcome. But a better model still needs a clear goal, and someone has to decide whether the result is good enough. Those are the first and fifth skills on the list, and they stay human.
A self-check
Rate yourself from one to five on each:
| Skill | 1–5 |
|---|---|
| Describing the problem | |
| Slicing the work | |
| Checking results | |
| Reading the output | |
| Judging what’s worth building | |
| Finishing and maintaining |
Work on your lowest score first. It’s usually checking or finishing.
Key takeaways
- Tools change; thinking skills don’t.
- Describe clearly, work in small steps and check everything.
- Learn to read what the AI gives you.
- Decide what’s worth building before you build.
- Finish things, and plan for maintenance.
Related reading
Which of the six would you rate lowest today?