A Knowledge Graph Can’t Capture Every Phone Call.
“The carrier just called. Tomorrow, Site B can arrange extra transport for only 150 units.”
Your AI’s recommendation assumes 300.
How does that new information become a revised plan?
It’s been a while since my last post in this Agentic AI series.
Since founding Nex & Bridge, I’ve been busy with proposals, speaking engagements, and getting projects underway. I’m grateful to have secured projects with two major Korean enterprises—in financial services and manufacturing—to help them put Agentic AI into practice. These opportunities came sooner than I expected, and I’m excited to begin.
Back to that phone call.
In the illustrative demo below, two sites need to ship 1,200 units. Site A can ship 500. The initial recommendation assigns 700 to Site B, including 300 units of additional transport capacity.
That plan fulfills all demand at a dispatch-day transport cost of $1,780.
But the planner has just learned that only 150 units of additional capacity will be available tomorrow.
The planner shares this update through the cockpit’s chat interface. The assistant identifies the proposed condition:
Site B · Additional transport ≤ 150 units · Tomorrow only.
The ontology-based knowledge graph supplies the business context: the sites, orders, transport services, and constraints involved. The conversational layer lets the planner add a temporary condition to this scenario while keeping the base KG unchanged.
After the planner confirms the interpretation, the planning calculation produces a revised recommendation:
• Ship 500 units from Site A and 550 from Site B. • Fulfill 87.5% of demand. • Spend $1,420 on dispatch-day transport. • Flag the remaining 150 units for replanning.
The planner reviews the cost and shortfall, then approves the revised dispatch instructions.
Building this kind of system has reinforced something for me: much of the work lies in making the business explicit.
How do processes actually run? What does the data mean? Which objectives take priority? Which constraints can be relaxed, under what conditions, and by whom?
AI helps us with the engineering. In my experience, clarifying and structuring these business questions often takes more effort than building the application itself.
That is becoming a core capability in Agentic AI transformation: translating business reality into structured knowledge and decision logic that AI can work with—and people can inspect, challenge, and improve.
How much of your organization’s decision logic is explicit—and how much still lives in people’s heads?