AI made building cheap. It didn’t make demand automatic.

Anyone can build software now.

That sentence would have sounded like hype two years ago. Today it’s just true. A working application — real interface, real database, deployed on a real domain — no longer requires a team or a year. The ability to ship, once the moat around an entire profession, is now the cheap part.

So what’s scarce now?

Not code. Attention, trust, and — scarcest of all — a stranger’s money. The hard part of building something moved: it used to live between the idea and the working product, and now it lives between the working product and the first customer who pays for it.

Economics has a name for the difference. When you ask people what they want, you get stated preference — cheap to give, unreliable, polite. When someone gives up money for a thing, that’s revealed preference. It is the hardest signal in commerce to fake, because it costs the sender something. Every founder has a phone full of people who said they loved the idea. The ones who paid are a different dataset entirely.

This is why I find most AI-era portfolios unconvincing, including the ones with beautiful code. A repository proves you can produce software. It has never proven that anyone wanted the software, and now that production is cheap, it proves less than it used to. The demo is no longer the demonstration; the transaction is.

I run Pruvara with that conviction built into the foundation. Before the first client site went live, I built the measurement: a first-party event pipeline that records every visit, the source that sent it, and every call tap and form send, as it happens. Not because instrumentation is glamorous — it isn’t — but because without it, “the website is working” is a stated preference. Mine. And my opinion of my own work is the least trustworthy data I have access to.

What the measurement buys is the ability to improve from evidence instead of taste. Something changed on the site; did call taps go up? A page was rewritten; did the people it was written for start arriving? These are answerable questions, but only if you built the apparatus to answer them before you needed it. Most businesses — and most builders — bolt on analytics later, when there’s finally something they want to know, and discover the past is unrecorded.

AI compressed the build, but it did not compress the trust. Getting a real business to hand you money still takes what it took in 1995 — a credible offer, proof you’ll still exist next quarter, and someone willing to stake their name on the outcome. That work is slow, personal, and completely resistant to automation. Which means the builders who do it now have a signal that’s appreciating while every other signal deflates.

Building was the filter. Now the filter is whether anyone pays.