The value of software is racing to zero

For most of my career, writing software was the expensive part. You needed a team, months of work, and a lot of money before anyone could use what you built.

That is no longer true. I can describe a tool in a few sentences and have a working version before my coffee gets cold.

And increasingly, nobody describes the tool at all. You ask for a job to be done — reconcile the accounts, plan the trip, find the cheapest supplier — and if the job needs a piece of software, the machine writes it along the way. The software is a by-product. Nobody decided to make it, and nobody needed to know how.

Software always had a marginal cost of zero. Copying it was free. What’s new is that the cost of making it is heading the same way.

When something is free to copy and almost free to make, its price drifts towards the cost of making it. That is where software is going.

To be clear, software isn’t becoming less useful. I use more of it than ever. It’s the part you can charge for that’s shrinking.

So where does the value go? To whatever the code can’t replace. I see three things so far.

The first is compute. Every piece of generated software runs on chips someone has to manufacture, install and cool. You can’t prompt a GPU into existence.

The second is energy. Under the compute sits power, and data centres already queue for years to get a grid connection.

The third is regulated work. A model can write a loan agreement or suggest a diagnosis. It can’t hold a banking licence or answer for the diagnosis when it’s wrong. Where the law demands that someone is accountable, the value stays with whoever holds the licence.

I suspect the list is longer. I’ve written before that even bad ideas work when you get the distribution right. When everyone can build the same product in an afternoon, distribution is what separates them. Trust, proprietary data, and anything that touches the physical world are other candidates.

What they have in common is that they’re scarce, and you can’t generate them.