From System of Record to System of Action: The Shared AI Strategy of the Platform Giants

When six competing corporations independently reach for the same strategy in the same year, that is not a coincidence — it is a signal about where the market is heading. That is exactly what I observed in the 2026 enterprise keynote season: SAP, Microsoft, Salesforce, Oracle, ServiceNow and Workday each made the same strategic pivot, in their own language.

I have spent much of my career at the intersection of business strategy and enterprise architecture. From that vantage point I read the season not as a product show but as a strategy paper — and it distils into five patterns every decision-maker should know. For the full overview of the season see this article; here I go into the strategic depth.

The paradigm shift in one sentence

The platforms stop being mere systems of record and become systems that execute. SAP calls it the “Autonomous Suite”, Oracle the move from “system of record” to “system of innovation”, ServiceNow speaks of the “system of action”, Salesforce of the “Agentic Enterprise”. The common denominator: software that doesn’t just know, but acts. What this means commercially — a shift of the unit of value from access to outcome — I described in “From SaaS to GaaS”. These five patterns are the technical-strategic scaffolding beneath it.

Pattern 1: Data and context are the moat

The most-repeated sentence of the season was, in spirit: “The language model doesn’t know your business data.” Oracle put it most directly — “It all starts with data. Your data.” SAP built a dedicated context layer with a knowledge graph and “Company Memory”, Salesforce grounds its agents in Data 360, Microsoft in the IQ layer spanning web, business and work data.

The strategic punchline: the generic language model becomes a commodity, while your proprietary context becomes the actual competitive advantage. A company that hasn’t got its data in order can buy the most powerful model available and still get mediocre results. Why SAP is pouring more than a billion euros into its data strategy I unpacked in this analysis.

Pattern 2: The semantic layer becomes mandatory

Data alone is not enough — the AI has to understand what it means. That is why a term that was long a niche topic surfaced in almost every keynote: the semantic layer. Oracle presented “Semantic Models & Ontologies” that explain how to reason over the data. Salesforce showed a semantic layer as a shared data vocabulary across agents and teams. SAP relies on domain models, Microsoft on the ontology layer of Fabric.

Strategically this means: authority over definitions and metrics — what is an “active customer”, how does a “closed order” count — becomes the foundation of trustworthy AI. Why context is the real API of the AI era, rather than the model, I laid out in “Context Is the New API”.

Pattern 3: Governance is architecture, not a feature

ServiceNow coined the most memorable formula: governance is “not a feature, it’s the whole ball game”. And indeed every vendor built a dedicated control plane — ServiceNow’s AI Control Tower, SAP’s AI Agent Hub, Microsoft’s Agent 365, Workday’s Agent Passport, Oracle’s shift of security down to the data layer.

That is an architectural statement: if you let loose agents that act on their own, you have to treat them like employees — with identity, permissions, audit and a way to switch them off. Governance here is not a downstream compliance checkbox but a precondition for speed. The ROI and risk side of this bet I cover separately in “Agentic AI in the Enterprise”.

Pattern 4: Open protocols tie it all together

Three acronyms ran through all six keynotes: MCP (Model Context Protocol), A2A (agent-to-agent) and, in part, A2UI. They are the traffic rules by which agents call tools and talk to one another — across vendor lines. Workday agents appear in Microsoft Copilot and Google Gemini, Salesforce becomes usable “headless” outside its own interface, ServiceNow’s Action Fabric opens the platform to any AI.

A second pattern within the pattern is worth noting: Anthropic’s Claude is a partner at five of the six vendors. The market is converging not only on protocols but on models. Why context — not the model — becomes the real API of the AI era I explored in “MCP Explained”.

Pattern 5: Building agents goes low-code

Every vendor rolled out a workbench for building agents — SAP Joule Studio 2.0, Microsoft Copilot Studio, Salesforce Agentforce, Oracle’s Private Agent Factory, Workday’s Developer Agent. The promise everywhere: natural language, low-code to pro-code. Strategically this democratises development — and shifts the contest from “who can code” to “who understands their processes well enough to hand them to an agent”.

What the strategy means for your platform choice

Three sober conclusions follow for decision-makers. First: invest in your data foundation and semantic layer before anything else — they are the precondition on which every agent initiative otherwise founders. Second: assess platforms not only by feature scope but by their governance layer and their openness to MCP/A2A — because pure lock-in plays run against the market trend. Third: your ERP, CRM or HCM remains the brain — the agent layer is only as good as the processes beneath it.

These trade-offs are rarely technical and almost always strategic. If you want to translate the 2026 platform strategy into your specific architecture and maturity level, get in touch — translating vendor announcements into sound decisions is exactly what I do.