The Morphing of the Knowledge Economy

In the original Only the Paranoid Survive, Andy Grove described one of the most consequential structural transformations in the history of business: the computer industry’s shift from vertical to horizontal.

For decades, the computer industry was organized vertically. If you were IBM, you made everything — the chips, the hardware, the operating system, the applications, the sales channel. You controlled the stack from silicon to service contract. Your customer bought a system, and you owned every layer of it. DEC worked this way. Wang worked this way. Every major computer company worked this way. It was the natural order of things, and it seemed permanent.

Then the PC revolution came, and the industry morphed. It went horizontal. Intel made the chips. Microsoft made the operating system. Dell assembled the hardware. Lotus and then a thousand others made the applications. No single company controlled the full stack anymore. Instead, each layer became its own competitive battlefield, with its own winners and losers.

The companies that understood this structural change — Intel, Microsoft — became the most valuable companies on earth. The companies that clung to the old vertical model — DEC, Wang, most of the minicomputer industry — were destroyed. Not because their products were bad. Not because their engineers were incompetent. But because they were organized for a structure that no longer existed.

I read Grove’s account of this transformation for the first time as a graduate student at Stanford in 1998. I read it again in my home office in 2024. And I felt a chill, because I recognized the same pattern unfolding — not in the computer industry this time, but in the entire knowledge economy.

For sixty years, the knowledge economy operated on a structural assumption so fundamental that nobody bothered to state it: humans think, machines execute.

The division of labor was clean. Humans decided what to build, what to write, what to recommend, what to analyze, what to approve. Machines stored the data, ran the calculations, rendered the interface, transmitted the output. You hired humans for judgment. You bought machines for throughput. The entire apparatus of modern business — the org charts, the job descriptions, the compensation structures, the training programs, the management hierarchies — was built on this division.

Software-as-a-Service, the model on which I had built my career and my company, was the purest expression of this structure. A SaaS product was a tool that made human cognitive labor more efficient. It organized information so humans could find it faster. It automated routine steps so humans could focus on complex ones. It enforced rules so humans did not have to remember them. But at the center of every workflow, there was a human, clicking, reading, deciding, approving.

And then, in the space of about two years, that structural assumption broke.

The new structure is something like this: AI drafts, humans curate. AI generates, humans evaluate. AI proposes, humans decide. The human is still in the loop — for now, and probably for a long time to come — but the human’s role has moved from production to supervision. From writer to editor. From analyst to reviewer. From coder to architect. From procurement specialist to procurement supervisor who oversees an AI agent that does the procurement.

This is not a semantic distinction. It changes the economics of every organization. And it changes the structure of every industry built on selling tools to knowledge workers.

Let me make this concrete by describing what happened to the software industry between 2023 and 2026, because it is the structural transformation I lived through, and because it is the closest analogue to Grove’s vertical-to-horizontal shift that I have ever seen.

The Old Structure: Vertical SaaS

In the old world — my world, the world I built Arcline in — a SaaS company owned the full experience. We built the data model. We built the workflow logic. We built the user interface. We built the integrations. We sold by the seat. A customer bought Arcline the way an enterprise in 1985 bought a minicomputer from DEC: as a complete system, vertically integrated, controlled by a single vendor.

This model had enormous advantages. It was predictable. It was sticky. Once a customer was implemented, switching costs kept them locked in for years. Revenue was recurring. Margins were high. Wall Street loved it. Every enterprise software company in the world — the CRMs, the ERPs, the IT service platforms, Arcline, and a thousand others — was organized around this structure.

The New Structure: Horizontal AI

The new world is horizontal. It has distinct layers, each with its own competitive dynamics, its own economics, its own winners. And like the PC industry’s horizontal layers, these new layers do not respect the boundaries of the old vertical stacks.

At the bottom sits the raw intelligence — the foundation models built by Anthropic, OpenAI, Google, Meta. This is the Intel of the new era: enormous fixed costs, massive scale advantages, and a small number of players who will dominate for a generation. I was never going to build a foundation model. Nobody in their right mind outside a handful of extraordinarily well-capitalized research labs was going to build one. This layer was decided.

Above that sits the infrastructure — the cloud providers who run the models at scale, who provide the compute and the serving layer and the APIs that make the intelligence accessible. AWS, Azure, Google Cloud. This is the PC hardware layer: essential, competitive on price and performance, and ultimately a scale game. I was never going to compete here either.

Then comes the layer that I believe will determine the shape of the industry for the next decade — and the one that barely existed two years ago. I think of it as the orchestration layer: the platforms that provide the governance, security, and trust infrastructure allowing AI agents to act on behalf of enterprises. They manage the handoffs between AI and humans. They enforce policies. They maintain audit trails. They handle the messy, unglamorous reality of deploying autonomous systems inside organizations that have compliance requirements and legal obligations and customers who need to trust that the system is doing the right thing. This layer is the Microsoft of the new era — not because of any particular company, but because of its structural position. Whoever wins here will sit between the intelligence below and the applications above, controlling the most strategically valuable chokepoint in the stack.

And at the top sit the applications — the products that solve specific problems for specific users in specific domains. Procurement. Customer support. Legal review. Financial analysis. These applications look nothing like the SaaS products they are replacing. They do not sell by the seat, because there is no seat — the AI does the work. They sell by the outcome. Per contract reviewed. Per anomaly detected. Per ticket resolved. Per procurement cycle completed.

This last shift is the most radical in the entire stack, and the one that keeps me awake at night: the move from seat-based to outcome-based pricing. When you sell by the seat, your revenue scales with your customer’s headcount. When you sell by the outcome, your revenue scales with your customer’s work volume. These are fundamentally different businesses, with fundamentally different economics, and the transition from one to the other is as wrenching as the transition from perpetual licenses to subscriptions was in the 2010s — maybe more so.

I looked at this new horizontal structure and I saw my company clearly for the first time. Arcline was a vertical SaaS company in a world that was going horizontal. We owned every layer: the data model, the workflow logic, the UI, the integrations. That had been our strength. It was now our vulnerability.

Because in the new structure, each layer is contested by specialists. The foundation model layer is contested by Anthropic and OpenAI and Google, and I was never going to compete with them. The infrastructure layer is contested by AWS and Azure, and I was never going to compete with them either. The agent platform layer is being fought over by a new generation of companies building orchestration and governance tools, and Arcline had no position there at all.

Which meant that our only viable future was at the top of the new stack — AI-native applications for procurement. But to compete there, we had to abandon the vertical integration that had defined us for thirteen years. We had to stop building our own workflow engine and start building on top of someone else’s orchestration platform. We had to stop selling seats and start selling outcomes. We had to stop thinking of ourselves as a platform and start thinking of ourselves as a solution — one that happened to be powered by layers we did not control.

Grove described the exact same realization at Intel. The vertical model was dying. Intel’s future was in one layer — microprocessors — and they had to let go of everything else to win in that layer. The emotional difficulty was not in seeing the new structure. It was in accepting what it meant for the old one.

I looked at my org chart. I had teams building workflow engines that would be obsolete within two years. I had integration engineers maintaining connections that foundation models could generate on the fly. I had UI designers crafting screens that AI agents would bypass entirely. I had a pricing model — per seat, per year — that assumed a human on the other end of every interaction.

Seventy percent of my engineering organization was building for a structure that was disappearing.

The numbers confirmed what I already felt. Enterprise spending on generative AI had gone from essentially nothing in 2022 to almost twelve billion dollars in 2024 to thirty-seven billion in 2025. That was not incremental growth. That was a landslide. And the composition of that spending told the real story: the fastest-growing category was not model APIs or infrastructure. It was applications — the AI-native products that were replacing the SaaS tools that companies like mine had been selling for a decade.

Everywhere I looked, the structural shift was accelerating. CEOs of the largest enterprise software companies in the world were publicly declaring that they no longer operated in a “SaaS neighborhood” but in an “enterprise AI neighborhood.” CTOs were announcing that “the platform becomes the product” — acknowledging, in public, that the thing they had been selling for fifteen years was being subsumed by a new architecture. Every major enterprise software company was scrambling to reposition itself for the horizontal structure — because the ones that did not reposition would go the way of DEC and Wang.

The morphing of the knowledge economy was not a prediction. It was not a trend to monitor. It was happening, in real time, in the financial statements of every software company in the world.

And Arcline was still organized for 1985.