AI Is Making Implementation Cheap. Taste Isn't.

For most of software's history, the bottleneck was building. An idea had to be worth weeks of someone's time before it was worth trying, and that cost acted as a filter. Plenty of bad ideas died because nobody could afford them.

That filter is going away. A working version of almost anything is now an afternoon. Which sounds like pure good news until you notice what the filter was doing.

Cheap to make is not the same as worth making

When building is expensive, the scarce skill is building. When building is cheap, the scarce skill is knowing which of the ten things you could ship by Friday deserves to exist.

That is taste, and I mean something specific by it. Taste is the ability to look at a working thing and tell whether it is right. It is noticing that the flow has one step too many, that the empty state is apologising when it should be helping, that the feature solves a problem nobody has twice.

A model can produce the tenth variation instantly. It cannot tell you that the third one was the answer.

What this changes in practice

I spend less time producing and more time choosing. A typical afternoon now includes three versions of an interaction where there used to be one, and the work is deciding between them with reasons I can say out loud.

It also raises the floor. If everyone can ship something that functions, then functioning stops being a differentiator. The products that stand out will be the ones where somebody cared about what happens in the first five seconds, and in the error case, and on a phone held in one hand.

The uncomfortable part

Taste is slower to build than a codebase. It comes from looking at a lot of work, shipping some, and being wrong in public often enough to recalibrate. There is no shortcut I have found, and I am suspicious of anyone selling one.

So that is where I am putting my time. The implementation will keep getting cheaper. The judgment about what to implement is the part I expect to be paid for.