Mukhtar Ahmad is CEO at CodeNinja, building internal intelligence engines by merging AI research and engineering to scale enterprises.getty​In March 2026, enterprise software stocks traded below the S&P 500 for the first time in two decades. The coverage that followed treated it as a market story, a correction, a repricing of growth expectations in a higher-rate environment. I think that reading misses what is actually happening, and the difference between the two interpretations has significant consequences for how enterprise leaders allocate their technology investment over the next several years.​I have spent 12 years in the operational realities of organizations in financial services, government, energy and logistics, long enough to understand how their intelligence actually works, where it lives and what happens to it when a technology engagement ends. What I am watching unfold is not a correction but the first visible consequence of a structural mistake that most enterprises have been making for a decade without realizing it.​Every time an enterprise deploys AI on infrastructure it does not own, it exposes operational data to external platforms and renews that dependency, it transfers its most irreplaceable asset, the institutional intelligence accumulated through years of operating in its specific context, into systems controlled by someone else. The market is beginning to recognize the consequence of that arrangement all at once.​The architecture of enterprise software was never designed with this problem in mind because the platforms built over the last two decades assumed a human being would sit at the center of every workflow. Systems of record existed to help humans interpret operational state, move decisions through chains of authority and understand what happened before determining what to do next. The companies that mastered this architecture built genuinely durable businesses, and for the era in which they operated, the market rewarded them appropriately.Why The SaaS Model Breaks In An Agentic AI Economy​In its report, Goldman Sachs argues that AI agents are reshaping software economics by reducing the amount of human interaction required to complete business processes and shifting value creation away from access to software toward the work software performs. As AI systems increasingly perform tasks on behalf of users, traditional per-seat licensing models face growing pressure, with vendors exploring pricing models more closely tied to usage, productivity and outcomes. The distinction is structural; traditional SaaS helps humans perform work faster, whereas agentic systems perform the work themselves and deliver the result as the product.​Consider compliance monitoring in a financial institution. Traditional software presents dashboards showing regulatory signals and routes alerts to analysts who decide what action to take. Results-as-a-Service ingests the signals, reasons over them against the organization's specific regulatory context, routes exceptions according to established escalation logic and logs decisions to governance infrastructure that the organization controls. The human moves to oversight and judgment, while deterministic execution moves to autonomous infrastructure, and what becomes critical is whether the intelligence embedded in how these decisions are made accumulates within the organization's infrastructure or within the vendor's platform. When that intelligence resides on vendor infrastructure, it resets upon closure, while the vendor retains the learning from your operations.The Emerging Battleground: Institutional Intelligence Ownership​​This matters because of a distinction that most procurement processes do not evaluate: the difference between a system of record and a system of context. A system of record stores what happened and makes it retrievable. A system of context makes what happened available for reasoning by encoding not just the event but the surrounding conditions, the sequence it belongs to, the decision logic that produced it and what it implies for what should happen next.​When an AI system asks why three procurement contracts stalled this week, a system of record returns the fact that they are pending. A system of context explains—for example—that contracts in this category, above a certain value threshold, from this specific supplier type, stall at this approval stage because the approver typically requires a specific supporting document that was not included. It then provides the resolution path. Traditional SaaS platforms, even those adding AI features, remain systems of record at their architectural core because they were not designed to expose operational context in forms that AI agents can reason over.​In my experience, enterprises can easily determine whether intelligence is accumulating within or outside their organization by examining three operational realities. When you ask your vendor whether models trained on your operational data transfer with you at the end of the contract, the answer reveals whether the intelligence you funded remains yours. When you evaluate whether your systems can expose operational context to AI agents you control, rather than only to the vendor’s platform, the architecture shows whether your institutional knowledge is accessible or effectively locked. When you assess whether you are building on infrastructure you govern or integrating with infrastructure someone else controls, the distinction makes clear whether your most valuable operational patterns live in systems you own or systems you rent.​According to IDC's enterprise spending survey, 70% of software vendors are expected to refactor their pricing strategies around consumption, outcomes or organizational capability metrics by 2028. The budget is moving decisively, and whether intelligence accumulates within the organization or on a vendor's platform is an architectural decision most enterprises make by default rather than by design.​The important thing to note is that this decision is made when the system is built, not afterward. The organizations building AI on infrastructure designed around their specific operational logic, with model weights and training datasets that remain under their governance, are the ones whose capability compounds with every cycle. According to Gartner's long-range research, "agentic AI could drive approximately 30% of enterprise application software revenue by 2035, surpassing $450 billion." The organizations that define their industries inside that market will be the ones that recognize early enough that the intelligence their operations generate is their most durable competitive asset and that the decision about where it accumulates is being made right now in every budget cycle, every vendor renewal and every AI pilot being deployed on infrastructure someone else controls.​Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?