At a glance

Yes, usually, for a genuine MVP. The whole point of a minimum viable product is to find out whether anyone wants the thing before you spend real money, and AI has made that dramatically cheaper.

The problems start when the MVP succeeds and quietly stops being an MVP without anyone redesignating it.

Yes, when

  • You are testing whether anyone wants it. The purpose is learning, not longevity. If the answer is no, the code was never going to matter.
  • The users are you and a few friendly people. Small numbers, forgiving audience, direct line to you when something breaks.
  • It is an internal tool. Staff only, no customer data, and a manual workaround exists if it falls over on a Wednesday.
  • You would genuinely throw it away. Not "in principle" — actually. If the honest answer is you would rebuild from scratch once it worked, build it however you like.
  • Nothing sensitive is stored. No payment details, no health information, no personal data beyond an email address.

In those situations, using AI is not a compromise. It is the correct call, and taking three months and a five-figure budget to test an unproven idea would be the mistake.

Be careful, when

  • It holds other people's personal data. The moment you store information about identifiable people, a mistake stops being your problem and becomes theirs, with obligations attached.
  • Money moves through it. Payments, invoicing, balances. Anything where being wrong has a number attached.
  • Real customers depend on it. Not friendly early adopters, but people who will be genuinely stuck if it is down.
  • It needs to work in two years. Someone will have to change it, and they will need to understand it.
  • You could not explain how it works. If nobody on your side can, you are dependent on a tool rather than an owner of an asset.

Careful does not mean do not. It means know that the invisible engineering has probably not been done, and plan for it rather than discovering it.

The rule of thumb

Two questions. Answer them honestly and you have your answer.

If this software is completely wrong, who pays for it? If the answer is only you, build it with AI and enjoy the speed. If the answer includes customers, staff or anyone whose data you hold, get someone to look at it before it goes live.

Would I be relaxed if a stranger read the code? Not because they will, but because it is a good proxy for whether anyone actually understands what you have.

That is it. Not the technology, not the tool, not how the code was produced. Just: who carries the cost of it being wrong.

The trap nobody sees coming

Almost nobody gets into trouble by making the wrong decision here. They get into trouble because the right decision expired.

You built an MVP. Correctly, sensibly, with AI, for very little money. It worked. People liked it. You added a payment page, because customers asked. Someone stored an address. A year later there are two hundred accounts, real revenue, and software still carrying every assumption of a fortnight's experiment.

There was no moment where anyone decided to run a business on a prototype. It simply happened, one reasonable step at a time, which is exactly why it is so common.

What to actually do

Build the MVP with AI. Genuinely. Move fast, spend little, find out if you have something.

Write down what would change the answer. Fifty paying customers. Storing card details. First member of staff who is not you. Pick the trigger now, while it is theoretical and you are calm about it.

When you hit the trigger, get it reviewed. Not rebuilt. Reviewed. It is usually days rather than months, and it tells you which of the invisible things are actually missing.

Then fix what matters and carry on. Most of these applications are perfectly good foundations once the unglamorous parts are added.

Our longer take on the decision, with the trade-offs laid out, is here: should you build your software with AI?

The underlying principle has not changed since the first post in this series: AI is a brilliant typist and a terrible architect. For an MVP, typing is most of the job. That stops being true precisely when it starts working.

Think you might have hit your trigger? Our AI Code Audit is a fixed-price £495 review that tells you plainly what you are dealing with, or get in touch for a no-pressure conversation.