The one question that decides every SaaS company's AI fate: does AI thicken your moat, or thin it?

Here begins the heart of the book. Everything so far — revenue quality, cash flow, moats — was built in a world before AI. Now that world is being rewritten, and the temptation is to answer with a mood (“AI changes everything!” / “it’s overhyped”). The useful frame is older and sharper: Clayton Christensen’s distinction between sustaining innovation, which makes existing leaders stronger, and disruptive innovation, which topples them from below. Which one is AI, for any given software company?

That’s not a question you answer for “AI” in general — you answer it company by company, with one test: does AI thicken this company’s moat, or thin it? And the answer splits into scenarios. In the optimistic one, AI is sustaining for quality incumbents: the stronger AI gets, the more valuable the two things it runs on — proprietary data and distribution — become, and those are precisely what strong incumbents already own. Foundation models are commoditizing fast on many tasks (everyone can call a powerful model), so wherever the capability a task needs is widely available and substitutable, the model alone is unlikely to be a durable moat — which pushes the advantage toward whoever holds the data and the customer relationship. On this reading, AI makes the strong stronger.

But that’s a scenario, not a law, and the chapter is careful not to let it harden into one. Frontier capability, cost, latency, and reliability can still differentiate at the model level, so “models are fully commoditized” is not a settled fact. And there’s a darker scenario, where AI is genuinely disruptive — where AI-native newcomers rebuild a category from scratch, or customers use AI to do in-house what they used to buy. The honest position is probabilistic: value the moat (in most cases it still holds), while reserving real wariness for “this time is different.”

A single answer is impossible because AI’s effect is local, not global. On one workflow it can commoditize a feature that used to be a differentiator; on the next, it can make a company’s proprietary data suddenly far more valuable. So “is AI good for software?” is the wrong altitude — the useful analysis happens one company, one workflow, one moat at a time, and the same technology can be a tailwind for one incumbent and an extinction event for its neighbor.

The trap Christensen warned about is that incumbents are structurally blind to disruption — their customers, profits, and org all point them at sustaining the present. So the discipline for the rest of Part Four is this: for every SaaS company, don’t ask “is AI good or bad?” Ask the sharper thing — thicken or thin? — and remember that a large impact does not automatically equal classic disruption. That one question organizes everything that follows: margins, pricing, and the battle for the workflow entry point.

This is Chapter 15 of “Compounding in the Cloud.”