Grounding Agentic Data Platforms: How SAP's Knowledge Graph Strategy Changes the Game
At VB Transform 2026, Max McPhee didn't mince words. The senior solution advisor at SAP told VentureBeat's Rob Stretchay that most enterprise AI tools are still stuck in chatbot mode—replying politely while accomplishing nothing. The difference between those and genuinely agentic systems, he said, comes down to one thing: grounding.
"Where we're starting to see more emergent behavior of it feeling like a coworker rather than an assistant, is where we're able to provide context on the actual enterprise rather than being able to use more of the standard knowledge," McPhee said.
That's the gap keeping most enterprise AI platforms from delivering real value. And it's where knowledge graphs—paired with vector-embedded data—are becoming the backbone of any serious agentic data platform strategy.
Building Enterprise Context with Knowledge Graphs
Here's the thing about knowledge graphs that gets overlooked: they're not some abstract academic concept. They're the same kind of structure you'd use to onboard a new employee, just adapted for software that retrieves information differently than humans do.
McPhee put it plainly. "When you are onboarding a new agent, I think it's important to acknowledge how you might onboard a new employee, but tune that for an agent," he said. "The way that is really powerful is using knowledge graphs and having vector-embedded data, because that's a really easy format for an agent to be able to find and retrieve information."
IBM's definition of knowledge graphs backs this up. At their core, knowledge graphs represent a network of real-world entities—objects, events, situations, concepts—connected by labeled relationships. Nodes represent things. Edges define how those things relate. Labels give you context. Simple on paper, devastatingly complex at enterprise scale.
The practical advantage? Knowledge graphs provide the structure and grounding that agentic data platforms need to work reliably. Without them, agents stumble over internal jargon, acronyms nobody outside the company uses, and tribal knowledge that never made it into documentation. You get a chatbot that asks "What does that acronym mean?" when you're trying to process an invoice.
Wikidata, the open knowledge graph powering Wikipedia, contains over 120 million entries across 300 languages. To make that accessible to AI systems, Wikimedia Deutschland used DataStax Astra DB on IBM watsonx.data to vectorize the graph, achieving query speeds 30 times faster and cutting development time by 90% without rebuilding core infrastructure. That's the scale we're talking about when knowledge graphs stop being theoretical and start being essential.
Governance, Identity, and the Double-Permission Model
SAP's 50-year history as a process company gives it an unfair advantage when it comes to governance. "That's where SAP really has a good home, around that governance and process control," McPhee said. "We're a 50-year-old process company, modernizing that governance to be able to handle the flexibility that comes with agents running."
The twist? SAP is reviving machine learning as a guardrail. Customers run agents within defined processes, then layer in anomaly detection and ML-based validation. It's the same approach SAP long used for intelligent approval recommendations, now applied to autonomous agents that can actually execute business processes rather than just suggest them.
Then there's the identity problem, which is where most enterprise AI implementations fail. Under SAP's model, both the human user and the AI agent—SAP's Joule, embedded across cloud applications and the Business Technology Platform—must independently hold permissions to access a given system. Even if a user has S/4 access, they can't route around controls through Joule unless the assistant itself has been provisioned for that access.
This double-permission model closes a real vulnerability. Autonomous agents that bypass human authorization gates have been a security nightmare across the industry. SAP's approach at least acknowledges that agents need their own identity and their own permissions, not just a user's credentials on autopilot. For a deeper look at how the industry is evolving its trust infrastructure for agentic systems, see our coverage of agentic AI security and the evolution of trust.
The Landscape Problem: SAP Is Only 10% of Most Enterprises
Here's where things get interesting—and messy. McPhee's biggest challenge isn't building agents. It's dealing with the reality that most customers tell SAP: "You're only 10% of my landscape."
Decades of acquisitions, customizations, and non-SAP integrations mean SAP's agents can't just operate within SAP systems. They need to understand how the entire enterprise connects. That's why recent acquisitions like LeanIX (described by McPhee as "Google Maps for your architecture") and process-mining company Signavio are central to SAP's strategy. They map the non-SAP majority so agents understand enterprise-wide system interconnections.
SAP's also invested in Berlin-based automation company n8n, embedding it natively into Joule Studio—its intent-based, low-code environment for building agents. This isn't just acquisition theater. It's an acknowledgment that knowledge graphs alone don't solve the integration problem. You need orchestration tools that actually connect heterogeneous systems.
McPhee's warning about legacy infrastructure is worth taking seriously: "You're going to probably run into throughput issues, and you're kind of trying to drive a Ferrari around a dirt track. You've got to upgrade the track first if you want to drive a Ferrari."
Companies building agentic data platforms on top of aging on-premises systems are going to hit performance walls. Period. Modernizing that infrastructure isn't optional if you want autonomous agents to function at scale.
The Bottom Line
SAP's approach to enterprise AI agents isn't perfect. But it's honest about the problems: grounding, governance, and integration. Knowledge graphs provide the structure. Governance provides the guardrails. And modernizing legacy infrastructure provides the capacity.
The companies that figure this out first will have a real advantage. The ones that treat AI agents as glorified chatbots will keep paying for them without getting much back. And as the broader agentic data platform landscape continues to mature, SAP's emphasis on knowledge-graph grounding and process-level governance positions it as one of the few vendors with decades of enterprise context to draw from.
