Build something somebody actually wants, even if that somebody is one person—and even if it isn't a person at all.
Execution used to be the filter for what got built. It took months to build anything, so every desire had to be matched to a market before it justified the work. Niches had a floor, and the floor was thousands of users. But when building costs a weekend instead of a year, the floor drops to one. That is the real shift. Not faster startups: smaller ones, all the way down to a business that serves exactly one customer well.
The proof runs on my own machine. I built a research tool that reads Reddit, X, and YouTube before I write an article; it's an audience of one, by design. I never guessed at a market. I just wanted it badly enough to keep sharpening it. It now has 57,000 GitHub stars, but the stars are a side effect. It would have been worth building at zero, because its one intended user gets value from it every day. Notice what the tool actually does, though: It listens to real people saying what they actually want. My one want, it turns out, was a machine for finding everyone else's.
And the one person you're building for is not always human. I begged a founder to let me invest in their startup in this space after watching an agent sign up for its own email account, login, password, and inbox, with no human in the loop. That agent is a customer. Agents want tools they can drive, accounts they can hold, and context they can pull mid-task—and they are the fastest-growing users of what I ship.
The narrowest viable business isn't even one person. It's one person's agent.
So after automation, ignore the flood of things nobody asked for. Find one real want—yours or your agent's—and build the thing only that one needs.
Which leaves the real question. When anyone can build anything for anyone, how do you know what one person, or one agent, truly wants? Whoever answers that will not just serve markets of one. They will mint them. I have been working on my answer. To be continued...
