The software factory

Software development is turning into a factory. Agents pick up a signal, a bug, a request, a metric that moved, and carry it through building, testing, and shipping, then watch what happens and start again. People don’t leave the work; they go from building the software to building the factory that builds it. The cost of turning an idea into working software is falling fast, and it won’t climb back.

The job I do took its shape in the opposite world. Product management never had a clean definition, but under every version of it sits one fact: building was slow and expensive, so someone had to choose what got built. Roadmaps, prioritization, the quarterly plan, the art of saying no, most of it is machinery for rationing engineering time. We were good at it because the build was the bottleneck. Take the bottleneck away and a lot of the job goes with it.

For anyone who likes to build, this reads as pure liberation, and a lot of it is. But cost was also what forced us to think. When building is free, you can build before you understand, and the factory will happily answer a question nobody asked. Making something work used to be the hard part; now it’s the easy one. What’s scarce is knowing what’s worth making at all. People call that taste, but taste only picks between things already in front of you. The rarer thing is vision: knowing which things belong there in the first place. With it, speed is a gift. Without it, you just fail faster.

I’m glad about that, because it hands the work back to building. I came up believing the way to know if an idea is any good is to build a rough version and look at it, and for years that instinct had to be rationed like everything else. Now, if you have some vision and some taste and you like making things, the era is on your side. If you enjoyed building before, you’ll enjoy it more now, with most of the friction gone. What changes is how you build, not whether. The catch is the factory itself, still early and uneven: some companies will be good at it, some won’t and will fall behind, and doing it yourself takes real investment before it pays. But it’s a bet I like. It puts the part I care about back at the center, and asks me to learn a new way of doing what I’d have done anyway.

Monetization, where I’ve worked for a while, gets a head start. It always ran on bets: form a thesis, ship the smallest version, read the numbers, and expect most experiments to lose. That’s about to become how everything works. And when a swing costs little, it invites bigger ones. A bet that was once too expensive to justify is now a sprint you can afford to lose. The instinct I learned there, find the smallest change that moves the most, is finally free to go looking for the largest.

None of this makes product less necessary. It strips it back to what it always was.