Forget the SaaSpocalypse. Here's How Vertical AI Actually Wins. Part II

Part II: Building the Moat

By Mike Becker

In Part I, we covered why the SaaSpocalypse selloff, while directionally right about workflow wrappers, badly misprices what’s happening for Vertical AI companies, and what it demands operationally: getting data infrastructure right, aligning your organization to AI, and evolving your business model. Here we get to the harder question: what does durable competitive advantage actually look like in this environment, and how do you build it?

Context will be king in the agentic era. The intelligence layer is likely to commoditize. The execution layer is the prize. The major tech companies and AI labs will fight to control the overall orchestration of this enterprise layer, but vertical players can build durable businesses owning a ‘node’ within it. Here’s what it takes:

Vertical Focus

A foundation model knows language. A Vertical AI company knows what things mean inside a specific domain and that distinction is what drives superior results. This ‘semantic understanding’ is the moat, and it shows up several ways:

• Taxonomy and ontology design: how you structure entities, relationships, and decision trees in a domain. That’s years of iteration with real users and real edge cases, not something a general-purpose platform can shortcut.
• Output calibration: knowing what confidence threshold justifies a recommendation versus a flag for human review. A general model doesn’t know where the liability line is.
• Error mode awareness: understanding which mistakes are catastrophic versus tolerable. In radiology, a false negative is existential. In marketing copy, it’s a minor annoyance. Vertical AI companies build their entire product logic around those asymmetries.

This knowledge is tacit. It lives in customer conversations, support tickets, and the judgment calls domain experts made. A new entrant with a better base model still has to earn all of it the hard way, one customer deployment at a time.

Cornered Data

Not just data you have access to, it’s data you are uniquely positioned to continuously capture. The ‘what’ and ‘why’ of how businesses actually operate inside your vertical. An AI lab can build a demand forecasting tool in weeks. What it cannot build is a tool that knows this company’s forecast requires regional VP sign-off, applies a non-standard seasonal adjustment for the Asian market, and has historically over-weighted promotional lifts by 12% because the marketing team systematically over-promises.

The counterargument, that AI labs accumulate cross-enterprise data at scale through their own deployments, misses the point:

• Breadth of cross-company patterns does not equal depth of single-company operational memory.
• One tells you what most enterprises do. The other tells you what this enterprise does, why, and what the exceptions are.
• Observing API inputs and outputs is not the same as understanding the organizational context that produces them, including who established a policy, what operational logic underlies an exception, and why a pattern exists.
• Vertical AI players need to capture some level of cross-company pattern and then go deep on company data to build a moat. Their service level in this later effort can be a moat in and of itself.

Vertical AI companies embedded in the workflow capture the context of decisions, not just the decisions themselves. That contextual depth is what makes accumulated data actionable rather than merely descriptive and what a general-purpose platform, however well-resourced, is unlikely to replicate from the outside.

Embedded Best Practices

Standardizing and encoding the right way to do a job inside your vertical is what enables the shift from selling tools to selling outcomes, from “we help you do X faster” to “we are the platform that executes X correctly.” That’s a fundamentally different value proposition, and a fundamentally different price and retention conversation.

There’s a second-order implication that doesn’t get enough attention. As the cost of software development approaches zero via AI-assisted engineering, the old constraint of limited resources forcing you to build only the highest-ROI features is disappearing:

• Vertical AI companies can increasingly say yes to nearly every customer customization request (as long as they don’t run counter to best practice).
• The product roadmap starts to look less like a resource allocation problem and more like a customer service function.
• Each customer’s specific workflows, exceptions, and preferences can be encoded at a depth that was previously uneconomical.

As Sequoia’s “Services: The New Software” frames it: sell the work, not the tool. The moat implication runs deeper. When you can afford to say yes to everything, your embedded knowledge of how each customer actually operates becomes genuinely unique.

System of Action

This is the destination: the transition from recording the work to performing the work. Every Vertical AI company should be asking honestly — are we on a credible path to owning a node within the execution layer in our domain, or are we building a better record-keeping system?

Two structural advantages make this position durable:

• Dual-deployment flexibility: agents can run on top of existing legacy systems or natively on the Vertical AI’s own platform. Customers with tens of millions locked into legacy infrastructure don’t have to rip it out, they layer new agentic workflow on top. Once orchestration intelligence accumulates, the legacy system underneath becomes increasingly interchangeable.
• Orchestration logic ownership: every AI interface (chatbot, voice assistant, UI agent) needs to know not just how to act, but what to do in the specific context of this company’s operations. Which approval path applies. Which exceptions are routine. Which actions carry real liability. That logic accumulates with every customer deployment.

Move Fast — The Flywheel Logic Is Real

The early mover advantage here isn’t just about market share, it’s structural. Context accumulates on a curve measured in years, not months. Companies that are ahead in capturing the why behind their customers’ workflows are building flywheels that are genuinely hard to catch up to. The market window for establishing this position is measured in quarters, not years. That’s not hyperbole; it’s the conclusion we’ve reached watching how fast the competitive environment has shifted in the last twelve months.

A Final Thought

What we see when we cut through the noise is not an extinction event for software, but a control transition, reshuffling which companies own the execution layer in the industries they serve. Vertical AI companies with deep workflow knowledge, cornered data, and a credible path to system-of-action status are better positioned than the market currently appreciates. But positioning alone doesn’t win. Execution against these priorities is what separates the companies that will define their industries from the ones that will be footnotes in someone else’s case study.

The intelligence layer is commoditizing. The execution layer is the prize. Go build it.

TL;DR

• The $285B “SaaSpocalypse” selloff is right about one thing: workflow wrappers and thin AI apps are genuinely vulnerable. If your product is a model plus a wrapper plus a workflow, the moat was temporal, not structural.
• But the selloff misjudges purpose-built Vertical AI companies, conflating them with workflow wrappers.
• The intelligence layer is commoditizing. The $2.9 trillion prize is the execution layer: who owns the orchestration logic, the domain context, and the system-of-action position in their industry.
• Winning requires getting five things right: clean data infrastructure, an AI-ready organization, a business model that reflects how AI delivers value, a four-part moat, and speed.
• The four-part moat: vertical focus and last-mile product depth; cornered data (the accumulated context of how and why your customers operate); embedded best practices; and system-of-action positioning.
• Vertical AI companies that move fast on these priorities are better positioned than the market currently appreciates. The ones that don’t will be footnotes in someone else’s case study.
• The window to establish this position is measured in quarters, not years.

For further reading, D’Ornano’s three-part series: “The $285 Billion ‘SaaSpocalypse’ Is the Wrong Panic,” “Decoding Anthropic’s $380 Billion Valuation,” and “Figma’s Orchestration Play” is worth your time, as is Sequoia’s “Services: The New Software.” D’Ornano’s framing on the system-of-record to system-of-action transition and Sequoia’s on the shift from selling tools to selling outcomes are two of the sharper pieces of analysis we’ve seen on this topic.