The Fifth Inning and the First

Agentic engineering is being played in two different innings at once. San Francisco is in the fifth. Most of the world is still in the first.

I talked to an engineering leader at a 250-person company last week who told me, with genuine excitement, that they had just signed up for an AI code review commenting tool. That category felt solved to me a year ago. My first reaction was that I had teleported back in time. My second reaction was more useful: this is what the market actually looks like. The frontier is not the market. The frontier is a preview of the market.

This explains the dissonance everyone keeps noticing. A tool gets declared dead on the timeline, everyone in one zip code announces they have moved on, and then the same tool reports revenue doubling. Both signals are true. They are just measuring different populations. The diffusion lag between what the most aggressive teams adopt and what a typical enterprise deploys is somewhere between six months and two years, and it has not compressed the way people assume. I wrote about this geographic and cultural spread in the coast-to-Midwest AI gap, and the pattern keeps holding.

The strategic mistake is deleting your ladder rungs. Features you shipped a year ago, the ones that feel embarrassingly basic next to what you are building now, are often exactly what the next wave of buyers is searching for. If you rip them out to chase the frontier, you abandon the majority of the market to whoever kept the on-ramp open. Being early is survivable, and often the whole game, as long as you are still standing when the crowd arrives. That is the entire argument of always too early, never wrong.

So build for the fifth inning, because that is where the game is going. But sell to the first inning, because that is where the buyers are. The companies that win this market will be the ones that hold both truths at the same time without flinching at either.