Ask an AI tool for a booking form that emails customers a confirmation and you will get one, often within the hour. You may not get the rule that stops two people booking the same slot, because you didn’t ask for it. You may get a login page that checks someone is signed in but never checks whose bookings they are looking at. And you will almost certainly get code that works on the handful of examples you tried, because those are the examples you tried.
None of that shows on the screen. The form looks finished and does what you described. The problems sit in what you didn’t describe.
What AI tools are good at
We build AI into the systems we make for clients, and we use it in our own work, so none of this is a case against it. The tools are good at what they do. An owner with no coding background can describe a booking tool on a Monday and have something running by Wednesday. A member of staff can wire up an automation that saves them an hour a day. Five years ago both of those would have meant hiring someone.
Close enough is often all a job needs. A first draft of an email, a summary of a long document, a rough internal tool used by two people who know its quirks: if the result is a little off, someone notices and corrects it, and the cost is a few minutes.
Where it goes wrong
The trouble starts when output that is only close goes in front of people who don’t know it. A tool can only build what it was told, and nobody describes the cases they haven’t thought of.
What makes this hard is that the gap is invisible. An exposed key in the browser, or a page that leaks one customer’s details to another, looks exactly like a working app until somebody finds it, and the somebody is rarely the owner.
Who answers for it
An AI tool has no memory of why it wrote the code the way it did, and it takes no responsibility for the result. When a developer wrote the software, the person responsible was the developer, or the agency behind them. When it came out of a chat window, the responsibility sits with the business, which is now running code nobody there has read. We go into that properly in who is responsible for software built with AI.
We spent years building our own processes around knowing who owns each piece of work, well before AI coding tools arrived, which is a story told in why we built our own client dashboard. The tools have made the question more urgent, because they have made it so much cheaper to produce software that nobody has checked.
The person is in charge of the AI
Alexander’s view is that the person is in charge of the AI, not the other way round. In practice, you decide what the tool is for and write that down before you start, and you read or test what comes back instead of taking the tool’s word that it works. What goes live is your decision as well, since the AI has no idea what would embarrass your business.
The second half of his view is to automate what you already understand. If you can explain a process clearly to a new member of staff, you can explain it to an AI tool and check what it gives you. If you can’t, the tool will fill the gaps with guesses that look reasonable, and you will have no way of spotting which parts are guessed. That is the same pattern behind most “we need AI” requests turning out to be process problems.
Getting a developer to read it
This is why we started offering AI Sign-Off, a developer review for software built with AI. An owner keeps everything good about building with AI, and before customers rely on the result, a developer reads the code, tests it against a written description of what it should do, fixes what fails and signs off the version they reviewed. If the owner wants us to stay accountable as it changes, that continues on a support retainer.
Some owners will do a first pass themselves, and that is a sensible place to start. Our guide on getting a vibe-coded app ready for real users is written for exactly that.
If you have something built with AI that is about to matter more than it did when you made it, tell us what it does and who will be using it.
Written by
Alex
CEO
I’m a software developer and CEO of Digital Royalty, helping growing teams scale their SaaS platforms without losing quality, visibility, or control. I focus on building structured, maintainable systems with clear processes, reporting, and accountability. With over a decade of experience across agency and in-house roles, I specialise in delivering long-term, scalable solutions that support complex, evolving products.