Three graphs: a memory of how your company decides
A knowledge graph captures what the company knows. A process graph captures what actually happens. A decision graph captures what was known, what AI recommended, what the human chose, and what followed.
Following a recent discussion with the team behind Frin, we exchanged views on the value of graphs and where this architecture can go with agentic AI. I believe graphs will play a major role in how we give AI real business context.
Frin already starts with a live knowledge graph. It connects the things a company needs to understand: customers, contracts, people, assets, events and their relationships, while keeping track of how that context changes over time. For AI agents, that context matters. An agent needs to know what exists, how things are connected, what is happening now and what happened before.
The graph can go further
- The Knowledge Graph captures what the company knows.
- The Process Graph captures what actually happens across people, agents and systems.
- The Decision Graph captures what was known at the moment of a decision, what AI recommended, what the human chose and what happened afterwards.
Frin makes this direction particularly interesting because AI agents can interact directly with the operational context through MCP, while working under the same permissions and access rules as other consumers.
The executive cockpit
This creates an interesting foundation for the AI Executive Cockpit I am working on. A CEO could see the company through the same connected context: which processes are running, which humans and AI agents are involved, what the agents are doing, what AI is recommending, what humans are deciding, where intervention is needed, and what results follow those decisions.
Over time, the organisation builds something extremely valuable: a memory of how it operates and how it decides.
The gap between what AI recommended and what humans chose can then become a powerful signal of how decision making actually works inside the company.
A note on truth
As knowledge graphs become a foundation for GraphRAG, agents and digital twins, structural validity alone does not guarantee that a fact is true. A triple may be perfectly valid and still be wrong. This raises a key architectural question: how do we ensure that what AI systems treat as facts is actually grounded in trusted evidence and provenance? It deserves much more attention as we build AI-native enterprises.