The End of Software

When Inca Yupanque speckled the hills of Peru with cities of gold, it wasn’t to appease the consumer. DARPA didn’t invent the internet - the computer’s killer app - to make Americans happy. Much - perhaps most - impressive innovation and economic engineering serves gods, wars, and bureaucracies. Not the average Joe.

Why is it these things, and not man, that turn the wheels of history? Sadly, Joe is simply not very ambitious. His hedonic returns diminish and taper. Give him a great news article and a good coffee and he’s happy. Give him the best of both and he’s … maybe twice as happy? The good news is we have a hedonic treadmill, the bad news is it doesn’t move all that fast. And, there are only so many free hours in the day.

Getting into the business of selling Joe news articles is a bad idea, because even with the best news articles possible, he’d probably only read twice as many for twice the price. Not a transformational outcome. Selling Joe news articles when the barriers to entry to doing so have just cratered is an especially bad idea. Once there is a supply glut, gg no re. Unless you make more Joes, but we know how TFRs trend. There’s nowhere to grow, no cavalry to save you.

Gods on the other hand, never seem to have enough. Viracocha Contiti demanded just about as much gold and many vestal virgins buried alive as Yupanque could afford, and his greed was hardly unmatched in the Mediterranean. Armies and crowns have arms races to win and empires to rationalize. Bureaucracies have entropy to resist and companies bankruptcy to stave off.

Sell an army guns and soon it needs tanks. Sell a bureaucracy spreadsheets and next it asks for databases. These gods live in darwinian hell and need every scrap of energy to survive another day. That’s how competition works - each tech tree unlock is a temporary respite, a fleeting boon that sustains the e.g. Fortune 500 for another quarter of earnings before expectations and competition immediately adjust. Then, onto the next one. There is no satiety, no rest.

A supply glut in task logic automation - software - then, is distinctly possible for consumer verticals. Consumer software is quite possibly the next news media. But it’s impossible for B2B. If one developer could pair program an entire F500 company’s worth of software in an afternoon with GPT-5, that probably leads to a weird quarter or two and then - voila - there’s now demand for Google level infrastructure from random industrials companies in the 400-500 range and Google has moved on to measuring its monorepo in trillions of LoC.

Why? For businesses, software is a way to trade developer hours for user hours. For a short while, people pick all the low-hanging fruit but then - by logical necessity - you reach a frontier of valuable problems that need serious developer hours to crack. Ballooning complexity means this frontier has to be big - the fraction of human hours spent serving war and bureaucracy has only gone monotonically up this millenium. Today’s jobs will go away, but there will still be a ton of user hours on the table ready for SWEs to save. If AI enables an hour of developer time to save hundreds of user hours by cutting a delay, automating trivial subtasks, whatever - Google will take that trade a million times a day, even if it involves piles of LoC that would be at present unimaginable.

But yeah it’s over for to-do list apps.