Building software is getting cheaper. Judgement isn't

Building software is getting cheaper every month.

Knowing what not to build costs exactly what it always did.

I've been building websites since I was twelve, so I've seen a few "now anyone can do it" moments. Every one of them was partly true. And every one of them left the same thing behind: a lot more software, and not many more people who knew which bits of it mattered.

The cheap part was never the expensive part

If you listen to the noise around AI, the hard part of making software is writing it. Typing the code. Making the screens.

That's never been my experience.

The hard part of PagoCRM wasn't the code. It was understanding how a club actually collects fees from people who'd rather not pay on time, often run by volunteers who already have a day job. The hard part of a booking engine wasn't the calendar. It was saying no to the next client who wanted it slightly different.

None of that gets cheaper because typing gets faster.

In fact, I suspect it gets harder. When building something took a month, you thought twice before starting. When it takes an afternoon, nobody thinks at all. You just build it. And then you have to maintain it. For years.

When anyone can build anything, the scarce skill is deciding what deserves to exist.

Every feature is a promise

Here's what people forget about software. Building it is a one-off. Looking after it is forever.

We learned that the hard way with custom work. Every little tweak for one client was a line of code someone had to keep alive through every update, every PHP version, every new rule from the tax office. Eventually we stopped customising the booking engine for each client. Not because we couldn't build it. Because we couldn't keep promising it.

If AI makes features ten times cheaper to build, I don't think it makes them ten times cheaper to keep. A product is a promise you keep every month. You can generate code in seconds. You can't generate the responsibility for it.

That's why, if anything, I expect judgement to become more valuable, not less. More software in the world means more of it half-maintained, half-understood and quietly breaking at midnight. Someone will have to know which parts matter.

What small companies actually have

This is the bit that makes me optimistic.

A small company with no investors was never going to win by writing the most code. We didn't have the money for that, and we still don't. What we had was closeness. Clients who tell us their problems directly. Products born from problems we'd solved more than once, not from ideas on a napkin.

That doesn't get automated. It gets earned, one client at a time, slowly. Patient money buys you that slowness.

So when I hear that building software is about to become nearly free, I don't panic. I nod.

Good. Let the typing be cheap.

The expensive part was always somewhere else.

P.S. If you're about to build something in an afternoon, spend the morning asking who'll maintain it.