There's a title, among our research work, that sums up a conviction: “From Knowledge Graphs to Practice Graphs.” It's not a play on words. It's a change of nature.
A knowledge graph does one remarkable thing: it maps entities and their relations. This client is linked to this project, which is linked to this decision, made by this person, on this date. It keeps the memory, the provenance, the traceability. It answers beautifully “why was this decided,” “what is connected to what,” “who said what, when.” It's a map of what you know. For traceability, reasoning over facts, searching a body of information, it's powerful.
But a knowledge graph describes. It's a representation, a map. Knowing how things are connected isn't the same as being able to act. A knowledge graph tells you the recipe exists, who wrote it, and what it relates to. It doesn't cook. It gives you the organization's memory. It doesn't give you its gesture.
The PracticePracticeA unit of know-how captured in Marylink: not a document but an executable structure (content, prompt, rules, style). Graph starts from the other end. It doesn't first keep “why you decided.” It keeps “how you work, here, now.” Each node isn't an entity on a map, it's an executable practice: a validated method, reachable by a human or an agent via MCPMCPModel Context Protocol: the open standard that connects an AI assistant (Claude, etc.) directly to your Marylink space., combinable with a style and a model, governed by roles and validation stepsValidation stepA stage in a publication's path (draft → review → approved), with rules that gate each transition..
Where the knowledge graph represents knowledge, the Practice Graph operationalizes know-how.
The distinction is exactly the one that separates knowledge from know-how. A knowledge graph is a map of what you know. A Practice Graph is a set of actions your teams and your agents execute. One is a representation. The other is a capability. One answers a question. The other does the work.
This difference becomes decisive in the age of agents. An agent plugged into a knowledge graph knows a great deal about your organization: its entities, its history, its links. But knowing isn't executing. For an agent to do things your way, it doesn't need a map of what you know. It needs access to what you know how to do, in a form it can execute. That's precisely what a typed, governed, executable practice is.
Put it simply. The knowledge graph gave you the memory. It answered the “why” and the “what.” The question that remains, the one agents make urgent, is elsewhere.
A knowledge graph tells you why you decided.
The real question, now: how do you work, here, and who, or what, knows how to execute it?