You are not going to write the code. But you might be paying for someone to build you a booking tool, a quoting calculator or a little app, and increasingly they will be using AI to do it. That changes what you need to know. You do not need to learn programming. You do need to understand enough to ask the right questions and tell solid work from a demo that looks great and falls over the first week.

We build software this way ourselves. Tools like QS Quoter, our roof-quoting tool and TradeDay were built with AI doing a large share of the work. So this guide is written from the inside, on which model matters, what good looks like, and how to judge what you are paying for.

The model matters more here than anywhere

For most AI jobs you can get away with a mid-tier model. Building software is the exception. Writing reliable code that handles real-world mess needs careful, sustained reasoning, so it wants the most capable model available. In the Claude family that is the top tier, Opus 4.8 or Fable 5 (dated July 2026), which are particularly strong at coding. OpenAI’s strongest GPT option is the other name you will hear for this work.

Why does this matter to you, the person paying? Because the cheaper, faster models will produce code that looks fine and quietly breaks on the awkward cases. If whoever is building for you is using a bargain-tier model to keep their costs down, that is a fair question to raise. The cost difference on the model is small next to the cost of a tool that fails in front of your customers.

The 80/20 trap you are paying for

Here is the pattern to understand, because it is where money gets wasted. AI can produce something that works, a demo, remarkably fast. Getting from a working demo to a reliable tool is the hard part, and AI does not do that part on its own.

The first 80 per cent is the exciting bit: the screens appear, the button does something, it looks like software. The last 20 per cent is the unglamorous work that decides whether it survives contact with real users: testing, handling the odd inputs people actually type, dealing with errors gracefully, keeping data safe. A demo skips all of that. A finished tool does not.

So when you see an impressive first version quickly, be pleased but not fooled. The right question is not “does it work in the demo” but “does it still work when a real customer does something unexpected”. That gap is where the real effort, and the real value, sits.

How to commission well without coding

You can judge the work through questions, not code. Ask:

  • Has it been tested, and how? “It worked when I tried it” is not testing. You want to hear about the odd cases, the wrong inputs, what happens when something fails. A serious maker has checked these.
  • Who owns the code, and can it be handed over? You should own what you paid for and be able to give it to someone else to maintain later. Make sure you are not locked in to one person’s head.
  • What happens when it breaks, and who fixes it? Everything breaks eventually. Know who you call, what it costs and how fast.
  • Where does the data live, and is it safe? If it holds customer details, this is your responsibility as much as theirs. Ask plainly.
  • Can I see it working on a real task, not just a polished demo? Ask to watch it handle a genuine, slightly awkward example from your business.

Good answers to these tell you more than reading a line of code ever would.

The honest bit

AI has made building small business tools far cheaper and faster than it was, which is genuinely good news if you have an idea for one. It has not removed the need for a careful human who tests properly, thinks about what goes wrong and takes responsibility for the result. The model does the typing. The judgement is still human, and that is what you are really paying for.

Last reviewed: July 2026. The specific model names above are examples of each tier as it stood in July 2026. Coding models improve quickly, so re-check the current best options twice a year rather than assuming last year’s pick still leads.

Your next step

Trying to work out which model your project actually needs? Our Which AI model should I use? picker walks you through it in a few questions and gives you a starting point with a caveat. To compare the families side by side by what each is best for, speed and cost band, see the model comparison table.

The free resources shelf also has a one-page plain-English glossary to help you follow the conversation with any maker. And if you would rather talk through your idea with people who build these tools for a living, come to our AI Automation Masterclass in Manchester. Laptops open, plain English, no fluff. Tickets are normally £20. This one’s free, a limited-time offer to launch the series.