SAP on Business AI: Knowledge Graphs, Agents and the People Who Supervise Them
Opening ALIGN Chicago 2026, SAP's Doug Freud explained how SAP uses knowledge graphs to ground agents in its ERP software, why it is becoming a business AI company, and why employees must learn to supervise agents.
By The AI Enterprise Magazine Team3 min read
Key takeaways
- Protocols made agents practical. MCP and A2A gave agents standard ways to reach data and each other.
- People become supervisors. Freud said employees will shift from doing work to supervising agents.
- Knowledge graphs ground agents. SAP uses them to prevent hallucinations and to plan and execute actions.
- Context lives in the business process. SAP is building fine-tuned models for domains like finance and supply chain.
- Governance stalls pilots. Training people and answering their concerns matters as much as the technology.
The speed of change in enterprise AI over the past two years has been hard to describe, according to Doug Freud of SAP. For a company whose software runs core business processes, he said, the question of what agents are good at is "existential." Freud opened the ALIGN AI Executive Summit Chicago 2026 on July 15 with a keynote sponsored by SAP titled AI's Endgame: When Your Data Knows, Your Apps Act, and Your Agents Decide.
What changed to make agents practical?
Freud walked through a series of milestones. Anthropic's release of the Model Context Protocol (MCP) in November 2024 gave agents a standard way to reach data in files and applications without an army of developers maintaining API connections. Lower-cost open models showed that powerful models no longer required a frontier lab. Google Cloud introduced the agent-to-agent (A2A) protocol, which he called important infrastructure for agents working together.
The shift that matters most for business leaders, he said, is what it means for people. "Humans are going to have to learn to supervise agents that are doing the work," Freud said. He recommended a simple model for explaining agents to business audiences: SPAR, for sense, plan, act and reflect. Few people he speaks with have a clear picture of what agents do, he said, and the "act" step raises the hardest governance questions.
Why does SAP rely on knowledge graphs?
Turning a natural language request into the right query or action inside enterprise software is hard, Freud said. Knowledge graphs, a database structure that maps how things relate, power everyday experiences such as social media feeds and the company summaries that appear in search results. They are also central to how SAP grounds AI in its applications.
SAP's main ERP application, S/4HANA, sits on a database with 125,000 tables, Freud said. Without a knowledge graph, getting agents to work reliably inside it would be extremely difficult. SAP first used knowledge graphs to keep large language models from hallucinating, then learned that agents need them to plan and execute as well. Freud said building the knowledge graph for S/4HANA took 800 people eight months.
What is SAP changing as a company?
Freud said SAP is moving from an application company that builds systems of record to a business AI company. That shift includes what SAP calls autonomous domains, fine-tuned models for areas such as finance and spend, supply chain management, human capital management and customer experience, as well as industry-specific versions.
The goal is to understand business processes at the persona level. When a user asks a specific question, the system must translate it into the right API call, retrieve the right data, reason about it and choose the right tools to act.
Why do pilots stall?
Freud said many proofs of concept never leave the pilot stage, and technical problems are only part of the reason. Governance is the other part. Some governance questions are technical, such as knowing which agents are running. The more important ones are about the business and its people.
Organizations must train employees to move from doing a job to supervising the agent that does it, and people vary widely in how readily they make that shift. Employees whose work is being automated are also asking, spoken or not, what their role becomes. Whatever a board may hope about cutting jobs, Freud said, companies will always need people, and assuming otherwise "is a fool's errand."
Where to hear more in 2027
Enterprise leaders will keep comparing notes on agents and governance at the 2027 ALIGN AI Executive Summits, invitation-only gatherings in eight U.S. cities: Atlanta, New York City, Dallas, Chicago, San Francisco, Boston, DC and Los Angeles. Registration is coming soon. See the ALIGN AI Executive Summits page.
Frequently asked questions
Why does SAP use knowledge graphs for AI?
Freud said they ground language models in SAP's complex data, which spans 125,000 tables in S/4HANA, and help agents plan and execute.
What is SPAR?
A model for explaining agents: sense, plan, act and reflect.
Why do AI pilots fail to scale?
Freud cited governance and people issues, including training employees to supervise agents, alongside technical problems.
This article is based on remarks by Doug Freud, SAP at ALIGN AI Executive Summit Chicago 2026 on July 15, 2026. More coverage: ALIGN AI Executive Summits.
Filed under: Agentic AI, Context, Semantics & RAG, People, Culture & Leadership
