In the age of AI, anyone can build an app. Here’s what actually makes software work.

Something real has changed. With AI, a working demo of an app can be generated in an afternoon — an interface, a database, a few features, all standing up faster than anyone thought possible a year ago. The barrier to producing software has genuinely fallen through the floor.

It’s tempting to conclude that building software is now easy. But there’s a gap that AI hasn’t closed at all: the distance between a demo that looks impressive and a system that quietly runs your business, day after day, without breaking. That gap is where the value always lived — and it just became the whole game.

Code got cheap. Judgment didn’t.

What AI made abundant is production — generating screens, wiring up basic features, imitating existing products. What stays scarce is everything around it: knowing which problem is worth solving, understanding how a real business actually works, deciding what not to build, and getting a team to adopt the result. AI can generate a thousand apps. It can’t tell you which one your business needs.

A demo is not a system

A demo only has to work once, on the happy path, in front of an audience. A system has to work on a bad day — when the data is incomplete, the customer is upset, two staff do the same step at once, and the internet drops mid-transaction. Most AI-built demos look great and quietly fall apart under real use, because handling the messy reality is the hard, unglamorous part that no prompt writes for you.

What actually makes software work

Strip away the tools, and the software that succeeds tends to share five things:

• It solves the right problem — chosen deliberately, not the first idea someone had.
• It fits the workflow — it follows how the business already works, so people don’t fight it.
• It’s reliable under real use — errors, exceptions, and edge cases are handled, not ignored.
• People actually adopt it — it makes their day easier, or it becomes shelfware.
• Someone is accountable after launch — a system nobody owns slowly rots.

What to ask whoever builds your software

If “I can build it with AI” is the whole pitch, that’s a signal to dig deeper. The questions that separate a builder from a demo-generator are about judgment, not tools:

• What business decision or workflow does this actually improve?
• What happens when the software is wrong, or the process hits an exception?
• How will my team be brought along so they actually use it?
• Who owns the system, the data, and the fixes after it launches?

The question is no longer who can build it. It’s who can decide what to build — and make it work in a real business.

Why this matters more, not less, as AI improves

It’s natural to assume that as AI gets better, the human judgment matters less. It’s the opposite. The more code becomes a commodity, the more the value concentrates in the parts AI can’t do: diagnosing the real problem, designing the workflow, deciding what should and shouldn’t be built, and taking responsibility for whether it actually improves the business. Cheap production makes good judgment more valuable, not less.

That’s the standard we hold ourselves to. We use AI heavily — it makes us faster. But it doesn’t define what we do. We start with the business, use technology as the lever, and measure the outcome. Anyone can generate an app now. Making one that a business can actually run on is still the hard part, and still the point.