Just because AI can build it doesn’t mean it works

Anyone can build software now. Describe what you want, and an AI coding tool will write it. In an afternoon you can have a working app, a dashboard or a booking tool that looks like it took a team months.

That’s genuinely remarkable, and we use it every day. But there’s a gap that the demos don’t show. A working demo and a product people can rely on are two very different things. The space between them is where most of the real work lives, and AI hasn’t made that work disappear.

What vibe coding is good at

“Vibe coding” is the name for building software by describing it to an AI and accepting what comes back, often without reading the code. For some jobs it’s perfect.

We use it constantly to prototype. A rough version of a tool, a clickable mock-up of a screen, a quick script to test an idea. Something that used to take a designer and a developer weeks now takes an afternoon. We build it, put it in front of real people, learn what works, and throw most of it away.

That’s the right use. A prototype’s job isn’t to last. Its job is to teach you what to build.

Where it falls apart

The trouble starts when the prototype becomes the product.

We learned this ourselves. A smart, motivated team member with no coding background set out to build a software project with AI. The AI wrote line after line of code. Nobody involved could tell whether it was any good, spot what was broken, or ask the right questions to fix it. Two days and $2,000 in usage costs later, there was nothing usable.

The AI did exactly what it was asked. The problem was that nobody could judge the output.

The public examples are worse. In 2025, one founder launched a paid app built with “zero hand-written code”. Within days, people were getting past his paywall, maxing out his API keys and filling his database with junk. He couldn’t fix it quickly because he didn’t understand what had been built. It isn’t a one-off: Veracode’s 2025 study found AI-generated code introduced security flaws in 45% of its tests.

The 10% and the 90%

Writing the code was never the hard part of building software. It’s maybe 10% of the job. The other 90% is everything that makes a product dependable:

• Security. Who can see what, and what happens when someone tries to break it.
• Edge cases. The customer who does the thing nobody expected, at 2am, on an old phone.
• Integration. Making it work with the systems the business already runs on.
• Maintenance. Someone who understands it when it breaks in six months.
• Trust. Real users relying on it every day, and it not letting them down.

AI is brilliant at the first 10%. It has made the other 90% more important, not less, because now far more things get built that were never designed to survive.

What it takes to move things forward

The software that genuinely changes how a business runs isn’t built in an afternoon. At R6 Automate we’ve helped self storage facilities move from fully staffed to staffless operation. AI and automation made that possible. But the code was a fraction of the effort. The real work was designing for every situation a customer or operator might hit, testing it until it held up, training people and earning their trust.

That’s the pattern we see everywhere. AI makes good developers dramatically faster. It doesn’t turn non-developers into developers, any more than a nail gun turns someone into a builder.

How to use AI to build, safely

1. Prototype freely. Use AI to test ideas fast and cheaply. Call it a prototype out loud so nobody gets attached.
2. Decide what’s disposable before you build it. The first version is for learning. Plan to throw it away.
3. Put an expert between the prototype and production. Anything customers, money or private data will touch needs someone who can read the code and judge it.
4. Know which door you’re at. Reversible experiments can move fast. Anything irreversible, regulated or customer-facing deserves the slow, careful treatment.

AI has made building cheap. It hasn’t made judgment cheap. If you’re weighing up whether to build something with AI, or you’ve built something that’s started to creak, talk to us. We’ll tell you honestly whether it’s ready.