In February the market wiped out $300 billion of software value in a single session, and someone coined the word SaaS-pocalypse. I follow Steve Eisman — the Big Short investor — and his version of the thesis is the cleanest: AI agents replace seat-based software, so seat-based revenue is in trouble; and the AI vendors themselves have no moats, so a price war is coming for them too.

I made my version of the first half in The Future of AI — part 2: the build-vs-buy line is moving back toward build, because AI collapsed the cost of building. What I want to add here is about the second half — the moats. Because the loudest voices in this debate contradict each other, and I think they are all right.

Two moats, not one

SAP’s CEO Christian Klein says the applications will keep mattering because they “encode decades of process logic that no foundation model can learn from public data alone.” Vista Equity Partners makes the analyst version of the same point: incumbents own decades of proprietary structured data that defines what counts as a valid transaction, a compliant outcome, an exception.

Bessemer says the opposite: the two deepest moats of the systems-of-record vendors — high implementation cost and custody of the customer’s data — are exactly what AI erodes, and the legacy vendors are “vulnerable for the first time in decades.”

They are not arguing about the same moat.

There are two, and the debate conflates them. The knowledge moat is the accumulated process logic behind the walled garden — thirty years of implementation guides, configuration rules, edge cases documented one painful project at a time. Only SAP’s AI has full access to that material. No competitor’s model can train on it, and no amount of vibe coding reconstructs it from outside. AI makes this moat more valuable, not less: an agent without that context is a very confident intern.

The friction moat is the cost of implementation — historically two to five times the license fee. This one AI is demolishing, and Bessemer is right about it. But look at who was collecting the toll: not SAP. The system integrators were. Implementation fees flow to Accenture, Deloitte, and the offshore pyramid, not to the software vendor.

Enter Jevons

The Jevons paradox: when the cost of using a resource falls, total consumption of it rises. Cheaper coal did not shrink the coal market; it grew it.

Apply it here. If implementation cost collapses, SAP’s effective total cost of ownership collapses with it — and implementation cost was the single biggest reason mid-size companies never bought S/4. Cheaper implementation does not shrink SAP’s market. It expands it, onto an intact knowledge moat.

The compression lands somewhere else: on the SI headcount pyramid. And that industry already knows it — Cognizant reports 47% of sales on outcome-based contracts, the whole sector is repricing away from billing by the hour. When the people who sell effort stop pricing by effort, believe them.

One honest caveat. Friction was also a switching cost, and it protected the incumbents. If implementing gets ten times cheaper, so does migrating off SAP, and so does implementing a challenger. Jevons expands the market while thinning the walls around everyone in it. The knowledge moat has to carry more of the defensive load than it ever has.

What this means if you invest

Do not short “SaaS” as a category. Sort by moat type. Vendors sitting on real proprietary knowledge — SAP, Veeva, the systems where decades of domain logic live behind the wall — can come out of this stronger. Per-seat products whose value is workflow convenience, with no knowledge a competitor’s AI cannot replicate in a weekend, are dead men walking. The integrators get repriced, not removed.

Eisman is right about the category average. But the average is what you short. The moats are what you own.