No Moat

A founder somewhere this morning looked at a competitor's product, opened his agent, and typed three words: clone this site. By lunch he had it — the layout, the flows, the pricing page, the onboarding, a working backend. Not a mockup. A product. The thing the other company spent three years and a Series A assembling, reproduced over a sandwich. That is the death of SaaS in a single sentence, and everything below it is footnotes.

"Moat" is Buffett's word — the structural thing that keeps competitors out after you've won: switching costs, network effects, deep integrations, a brand, a hoard of proprietary data, a distribution channel nobody else can rent. Software-as-a-service was sold for two decades as a business of moats. But look hard at what most of those moats actually were, and for the median company it was this: a competent interface wrapped around a database and some business logic, defended mostly by the fact that rebuilding it was expensive and slow. Delete "expensive and slow" and there was never a moat. There was a lead time.

The cost of producing software just fell through the floor and kept falling. When the marginal cost of a thing approaches zero, so, eventually, does its price — this isn't a hot take, it's the oldest curve in economics, the same one that came for recorded music, stock photography, and long-distance calls. Now it has come for applications themselves: billions of them, stood up in an afternoon, cloned by the next person that same afternoon. Scarcity was the thing you were charging for. Scarcity is the one feature no one can ship anymore.

And here the sword shows its second edge — the one that cuts the hand holding it. Everyone is prompting the same handful of frontier models, and those models, asked to build a scheduler or a CRM or a landing page, regress to the mean, because regressing to the mean is what they are: engines for the statistical median. So a million founders receive, in parallel, the same architecture, the same feature set, the same pricing psychology, the same tasteful gray UI. The tool that lets everyone build guarantees everyone builds roughly the same thing. Differentiation was already hard; now it is being actively averaged out of the population, one identical suggestion at a time.

Meanwhile the customer is working out that he doesn't need to buy any of it. Why rent the SaaS when you can ask your own agent to spin up the exact narrow tool you need — tuned to you, owned by you, deleted when you're done? The entire SaaS proposition was "rent our software because building your own is impractical," and the whole thing rested on that one word, impractical. Disposable software, personal software, software as a sentence you say once, quietly dissolves the reason the subscription ever existed. You don't rent what you can conjure.

Distribution, the last refuge, is draining on its own timeline. Ask yourself honestly when you last googled something and clicked a blue link, instead of just reading the answer the model handed you. Search traffic is the water supply that fed a generation of SaaS growth — the content marketing, the SEO funnels, the whole machinery of ranking for a keyword and converting the click. The model answers in place now; it does not pass the click along. The mouth of the funnel is being welded shut, and the companies built to stand under it are noticing the drought about a quarter too late.

But the deepest cut is the one almost nobody is pricing in, because it kills a religion, not just a business. The entire discipline of product — the UI, the UX, the funnels, the onboarding, the delight, the pixel-craft — assumes a human looking at a screen and deciding. That assumption has an expiry date. When transactions move agent-to-agent — my agent calling your agent's endpoint, negotiating, settling, with no human watching a single frame of it — the screen has no audience. Machines don't need buttons, or onboarding, or a tasteful empty state. They need a protocol. The interface layer that SaaS spent twenty years perfecting becomes a thing rendered for precisely no one.

So the people still thinking in products — still perfecting the settings page, still A/B testing the shade of the button — are polishing brass on a ship whose passengers have already stepped across to a different vessel. In an agent-mediated economy the only human-facing interface left is your own agent; everything behind it is machines talking to machines in a language with no visual design at all. The winning surface stops being a screen and becomes an endpoint. Beauty stops being a moat. It stops being visible.

What survives, then? Only the moats that were never made of code: a genuinely proprietary pile of data no one can regenerate, a regulated license, physical infrastructure, real-world trust and liability, a network of actual humans who actually show up. Everything else — the clever wrapper, the nice UI, the clonable feature set, the SEO funnel — was a sandcastle at low tide, and the tide is a technology that makes building free, ideas identical, customers self-sufficient, distribution silent, and the screen itself obsolete. They called it a moat. It was always just the time it took someone else to type "clone this site." And now that takes an afternoon.