One thing this framing gets right: the constraint used to be compute. Then it became headcount (10 people to ship anything). Now it’s attention and judgment.

If AI handles the mechanical part of coding, what separates a working product from a mediocre one is taste in problem selection, ruthless scope discipline, and knowing what not to build. Those don’t scale with team size. They often get worse.

The micro teams I’ve seen succeed do one thing: they don’t try to compete on polish or features. They go narrow — solve one problem well for one audience. The opposite of the feature-accumulation treadmill.

This is wild because it inverts the startup orthodoxy of the last decade (hire fast, iterate on product-market fit with 20 people). Now you need fewer people but different people. Less execution, more judgment.