Evan Armstrong: why the context layer is changing the economics of enterprise software and hiring

Tech analyst Evan Armstrong proposes viewing corporate AI not as a threat to software itself, but as a redistribution of value within the stack. According to his view, code and interface-based SaaS products are becoming cheaper, while the primary asset for companies becomes context: knowledge about processes, access permissions, and workflows.

Why SaaS Tanked

In an article titled Context is King, published February 12, 2026, Armstrong attempts to explain why the market has dramatically reassessed its attitude toward traditional cloud services. According to him, approximately $300 billion in market capitalization of software companies vanished amid fears that generative AI would simplify the creation of business applications to the level of a commodity. The investors' logic seems crude, but straightforward: if a CRM, an analytics dashboard, or an internal tool can be assembled from instructions, why pay the SaaS premium as before?

Armstrong believes the issue goes beyond the hype around code generation. He points to three deeper shifts: slowing growth rates at public SaaS companies, deteriorating gross margins in products built around AI from the start, and declining switching costs between services. Since 2023, public SaaS growth rates have, by his estimates, plummeted from 36% to 17%, while new AI applications maintain margins closer to 50–65% rather than the previous 75–85%. In such a model, code becomes cheaper, and old business valuations no longer seem sustainable.

Three Layers of the Stack

Armstrong's key idea is that corporate software is now better described not as a collection of separate applications, but as a stack of three layers. The first layer consists of accounting systems and databases, which don't disappear because agents need reliable records to operate from. The second layer comprises interface applications, where humans click buttons, view reports, and execute individual tasks. This layer, in the author's view, faces the greatest risk of becoming interchangeable.

The focus shifts to the third layer—the context layer. This is not merely agent orchestration, but accumulated company knowledge about how work actually gets done day in and day out. What matters here are not only formal regulations, but practical dependencies: who approves exceptions, where processes stall, which combinations of actions accelerate deals or help close incidents. This layer, by Armstrong's reasoning, determines whether an agent will be useful in a working environment or remain an impressive demo.

• who has access to which data
• which actions can be performed without approval
• the order in which sales, purchases, incidents, and support processes flow
• which steps more often lead to success versus failure

Such a layer transforms an agent from an executor into a system that chooses the right action at the right moment. And it accumulates: each workflow execution leaves traces that make the next run more precise. Therefore, context becomes not a checkbox document, but an asset with a compounding effect. Essentially, companies begin encoding their own way of working so that it can be used not only by people but also by AI systems.

"The context of how people work gradually becomes the context of how

AI agents work even better."

Who Gets the Layer?

The most interesting question here is not technical but strategic: who will actually own this layer. Armstrong lists several contenders. ServiceNow already has deep penetration into corporate processes, Notion controls vast arrays of internal documents and knowledge bases, and Glean is building a bridge from enterprise search to enterprise context. In parallel, OpenAI and Anthropic are moving from the agent platform side, attempting to become a semantic operating system on top of CRM, databases, and internal applications.

From this logic comes an uncomfortable conclusion for the labor market. If the context layer truly begins replacing coordination costs, money will drain not only from IT budgets but also from payroll. Pressure will fall not only on developers of simple internal tools, but also on those roles that depend on emails, approvals, task handoffs between departments, and manual process management. This is about the gradual automation of office coordination, not just another round of developer optimization.

What It Means

The main takeaway for business is straightforward: winners won't be those with simply more models or lower token prices, but those who quickly encode their processes, access rights, and internal logic into context suitable for agents. In the new corporate race, the most valuable asset becomes not the interface, but knowledge of how the company actually works.