Scope, by inheritance.
The agent doesn’t need to be told what process it owns or which step it’s in. It’s a node attached to a process node. Move the process; the agent moves with it.
Drop an agent into the graph and it inherits its scope. Which process it owns. Which decisions it can make autonomously. Where it must escalate. Who its peers are. Below: an invoice-processing agent, in flight, on a Tuesday morning.
Fig. — A walk-through of agents traversing the org graph: scope inherited from the process node, escalation routed via the people graph, audit written back to the model.
The agent doesn’t need to be told what process it owns or which step it’s in. It’s a node attached to a process node. Move the process; the agent moves with it.
What it can decide alone, what it must escalate, who it must escalate to — read from the same policy nodes that humans read. One source of truth for both.
Every action is a write to the graph: which node fired, on what input, with what outcome. The audit trail isn’t a separate log — it’s the graph’s history.
Invoice processing is the easy one. We’ve also dropped agents into onboarding triage, policy compliance, customer-issue routing, capacity planning. The pattern is the same: connect to a process node, inherit the rest.