If AI Can Write Code in Seconds, Why Does a Custom Build Still Take Weeks?

A client showed me a demo the other week. He'd typed a description of his booking system into an AI tool and, within a minute, had something that looked like a working app. Forms, a calendar, a bit of styling. He asked, reasonably, why my quote for the real thing was six weeks and not six minutes.

It's a fair question, and worth answering properly, because the confusion is understandable. If a machine can write a screen's worth of code instantly, why does commissioning actual software still take so long?

Writing code was never the slow part

Typing out the code for a form, a database table or a page of styling has never been where the real time goes on a project. Even before AI tools existed, that part was fast for anyone experienced. What takes time is everything around it: working out what the software actually needs to do, for this specific business, with its specific quirks.

Take a booking system, since that's what the demo was. The happy path, someone picks a time, fills in a form, gets a confirmation, is genuinely simple. An AI tool can knock that out convincingly. But a real booking system for a real business has to answer questions like:

• What happens if two people try to book the same slot at the same second?
• What if the customer cancels twelve hours before, versus twelve minutes before? Does the deposit get refunded, partly refunded, or kept?
• What happens when a recurring weekly job falls on a bank holiday?
• How does a card payment failure get handled without losing the booking or double-charging someone?
• What does the tradesperson see on their phone if they've got no signal on site?

None of that shows up in a quick demo. All of it shows up the first month a real business relies on the system. I built exactly this kind of thing for a trades booking and payments app, and the deposit logic, cancellation rules and reminder timing took far longer to get right than the booking form itself ever did.

AI is genuinely useful for the boring bits

I'm not against the tools. I use them myself for boilerplate, for generating a first pass at a migration, for suggesting test cases I might not have thought of. They save real hours on the repetitive parts of the job. What they don't do is understand your business, your customers, or the specific way things go wrong in your industry. That understanding has to come from somewhere, and right now it comes from a person sitting down with you, asking questions, and then checking the output carefully. That checking step matters more, not less, when AI is involved. Code that looks right and code that is right are different things, and the gap between them is exactly where an experienced developer earns their fee. I've written before about who actually checks AI-generated code before it goes anywhere near a live system, and the answer has to be someone who understands both the code and the business it's serving.

Where the six weeks actually goes

On a typical project, the time breaks down roughly like this:

• Scoping and questions. Understanding the actual process, not the idealised version of it. This is usually where I find the edge cases that would have caused problems later.
• Data modelling. Getting the structure of the information right so it doesn't need rebuilding in six months when the business grows or changes.
• The unglamorous middle. Payments, permissions, validation, error handling. None of it is visible in a demo, but it's most of what makes software trustworthy.
• Testing with real, messy data. Not the tidy example data a demo uses. Real customer names with apostrophes in them, real bookings that overlap, real files that don't upload cleanly the first time.
• Deployment and handover. Making sure it actually runs reliably somewhere, and that you understand how to use it.

A demo skips almost all of this. It's not dishonest, it's just showing you the 10% that was always quick, not the 90% that was always the actual work.

What you're really paying for

When a business commissions bespoke software from me, they're not paying for someone to type code faster than a machine can. They're paying for someone who will sit with the awkward questions until the answers are right, who will still be reachable when something breaks in production, and who takes ownership of the whole thing rather than handing over a pile of generated files and moving on. That's the part AI can't do. It can't be accountable for a payment that went missing, or explain to your accountant why the numbers don't reconcile, or pick up the phone when a customer's booking has gone wrong on a Saturday morning. I can, because it's my name on the work and nobody else's.

If you're weighing up whether a project genuinely needs custom software or could get by with something off the shelf, it's worth working through the trade-offs properly rather than guessing. My build vs buy calculator is a quick way to see where you actually land before committing to either route.