The Market Just Told You Something Important
When SaaS valuations shed trillions in a week, the instinct is to focus on what’s being disrupted. Which vendors are vulnerable? Which stocks to sell? Which products are about to be replaced by AI agents?
But the more important question is: what survives? What thrives?
Brad Gerstner shared the data on this week’s All-In Podcast: software is trading at 3.9x forward revenue — all-time lows. Investor models that used to project cash flows thirty years out now only “count on” seven to fifteen. The market is pricing in deep uncertainty about which software businesses have durable value.
David Sacks offered the counterpoint: enterprise systems like Salesforce are battle-tested across thousands of large customers and millions of edge cases. You don’t replace that overnight with AI-generated code. The moat isn’t the CRUD interface — it’s the institutional trust built over decades.
Both are right. And the synthesis reveals the real opportunity: the winners in the next era of enterprise software will be platforms that combine transformative AI capability with enterprise-grade trust. Intelligence alone isn’t enough. Governance is the differentiator.
The Moat Problem
Here’s what the market correction revealed about software moats:
Thin application layers are vulnerable. If your product is primarily a user interface on top of a database, AI can replicate it. Code generation tools can build CRUD applications rapidly. The interface layer has been commoditized.
Data depth is durable — but only if it’s organized. Having years of enterprise data is an advantage, but only if that data is semantically organized in ways that AI can reason over. Petabytes of unstructured documents in SharePoint aren’t a moat. An ontology-grounded knowledge graph with provenance, temporal context, and semantic relationships — that’s a moat.
Governance is the ultimate differentiator. Any platform can generate impressive demo outputs with a language model. Shipping that to production in a regulated enterprise? That requires a fundamentally different architecture.
As Sacks noted, the greatest threat isn’t that AI replaces enterprise systems wholesale — it’s that value capture moves upward to the AI orchestration layer. The vendors that retain value will be those with depth that AI agents actually need: domain-specific semantics, governed execution frameworks, and verifiable trust infrastructure.
Why “Move Fast and Break Things” Fails Here
Silicon Valley’s default operating model — iterate quickly, ship probabilistically, fix in production — has never worked in regulated industries. AI amplifies this incompatibility.
When a language model hallucinates a customer service response, it’s embarrassing. When it hallucinates a clinical recommendation, it’s dangerous. When it hallucinates a compliance determination, it’s a regulatory violation. When it hallucinates a claims adjudication, it’s potential fraud.
Regulated enterprises don’t operate in a world where “mostly accurate” is acceptable. They need:
- Provenance for every answer. Not “the model thinks,” but “based on these specific sources, with this evidence chain, under these policy constraints.”
- Audit trails for every action. Not just what the AI did, but why it did it, what alternatives it considered, what policies it applied, and what evidence it relied on.
- Policy enforcement that’s deterministic. Not hope that the model follows the rules, but architectural constraints that make rule violations impossible.
- Rollback capability. When an automated action turns out to be wrong, the ability to cleanly reverse it — not apologize for it.
These aren’t nice-to-have features bolted on after the AI works. They’re architectural requirements that must be built from the foundation up.
Three Moats That Matter
For enterprise AI platforms serving regulated industries, three moats determine who wins:
The Semantic Moat
Most AI applications rely on RAG — retrieval-augmented generation. Chunk your documents, embed them in a vector database, retrieve relevant fragments, feed them to a language model. It works for simple question-answering. It fails for enterprise reasoning.
RAG retrieves fragments. It doesn’t understand relationships. It doesn’t track provenance. It doesn’t model time (which version of the policy was in effect when this decision was made?). It doesn’t enforce constraints (which data sources are authoritative for this type of question?).
A semantic fusion fabric operates at a fundamentally different level. It organizes enterprise knowledge into ontology-grounded structures where every fact has provenance, every relationship has context, and every assertion can be traced to its source. This isn’t just retrieval — it’s comprehension. And it compounds over time: the more knowledge the fabric ingests, the richer the semantic connections become, and the more reliable the reasoning.
This moat deepens with use. A competitor can copy your interface in a week. They can’t replicate years of semantic knowledge organization overnight.
The Execution Moat
Reasoning about data is one thing. Acting on it is another.
Most AI systems are read-only. They analyze, summarize, and recommend. But enterprise value comes from execution: processing the claim, updating the record, triggering the workflow, coordinating across systems.
Safe execution in enterprise environments requires infrastructure that most AI platforms simply don’t have:
- Typed action registries — every action the system can take is explicitly defined, versioned, and constrained
- Policy gates — pre-execution validation against business rules, compliance requirements, and risk thresholds
- Idempotent operations — actions that can be safely retried without creating duplicates or inconsistencies
- Verification steps — post-execution confirmation that the action achieved its intended result
- Rollback patterns — the ability to cleanly reverse actions when needed
This execution infrastructure is what separates a demo from a production system. Building it takes time, expertise, and deep understanding of enterprise operational requirements. It’s a moat because it’s hard to build and impossible to fake.
The Trust Moat
Trust isn’t a feature — it’s an operational track record.
For regulated enterprises, trust means:
- Deployment flexibility. Can the platform run in your VPC? On-premises? In air-gapped environments? Or does it require sending your data to someone else’s cloud?
- Compliance alignment. Does the platform understand FDA, HIPAA, CMS, NCQA, GxP? Not just as checkboxes, but as architectural constraints that inform how the system operates?
- Evaluation infrastructure. How do you know the system is performing correctly? Continuous evaluation, regression testing, drift detection, and performance monitoring — built in, not bolted on.
- Operational maturity. Upgrade paths, rollback procedures, incident response, access controls, encryption at rest and in transit — the unglamorous but essential infrastructure that regulated enterprises require.
This moat is built through years of deployment in regulated environments. It’s earned through audits, reviews, and proven operational reliability. No amount of funding or engineering speed can shortcut it.
The Pricing Revolution
Friedberg raised another point on All-In that connects directly to governance: the shift from per-seat pricing to value-based pricing.
When AI moves from assisting workers to completing work to doing impossible work, the pricing model must evolve. You don’t charge per-seat for an agent that processes ten thousand claims overnight. You charge for the outcome: claims processed accurately, within policy, with full audit trails.
But here’s the critical link: value-based pricing only works when the outcomes are verifiable and trustworthy. An enterprise will pay premium prices for a claim processed correctly, a safety signal detected accurately, or a field service issue resolved on the first visit — but only if they can verify the outcome, trace the reasoning, and audit the decision.
Governed AI enables value-based pricing. Ungoverned AI doesn’t, because the enterprise can’t trust the outcomes enough to pay for them.
This creates a powerful flywheel: governed platforms deliver verifiable outcomes, enabling value-based pricing, generating higher revenue per deployment, funding deeper governance infrastructure, which enables more autonomous operation, which delivers more value. Ungoverned platforms are stuck selling per-seat access to chatbots.
What Comes Next
The SaaS disruption of February 2026 will be remembered as the moment the market recognized a structural shift in enterprise software. Not the death of SaaS — but the birth of a new value layer where intelligence, governance, and execution converge.
The enterprises that thrive will be those that choose AI platforms built for the world they actually operate in: regulated, complex, multi-system environments where every decision matters and every action has consequences.
The technology exists today to build systems that reason across enterprise knowledge, execute governed workflows autonomously, and maintain complete auditability. The question isn’t whether this future arrives — it’s whether your enterprise is positioned to capture it.
Intelligence is table stakes. Trust is the differentiator.