Ontology × Knowledge Base × Orchestration × Acceptance: A Formula for Deliverable AI Agent Applications

During a summer of teaching GIS and consulting on AI projects, the author distilled a repeatable formula for turning "cool demos" into deliverable AI applications. The formula consists of four elements: ontology, knowledge base, orchestration, and acceptance.

1. Ontology: Let AI See Your Business

The author emphasizes that ontology is not about how you view the business, but how the AI perceives it. By asking the client a probing question—"How does the AI understand your business?"—the author uncovered that AI needs a formalized, machine‑readable structure to reason about the domain. The author cites Palantir’s approach of turning expert knowledge into computable models as evidence that AI cannot generate its own business ontology.

2. Knowledge Base: Context Is the AI’s Experience

All successful AI adopters store their course materials, process documents, and business data in versioned, discoverable repositories. The author stresses that knowledge not stored in a system of record is invisible to agents. Since AI lacks memory and relies solely on context, a well‑structured knowledge base makes the agent appear smarter, analogous to how organized reference material speeds up student learning.

3. Orchestration: Write Business Logic as an AI Playbook

Static assets (ontology and knowledge base) become executable through orchestration. The author describes using prompt‑crafted commands, skill‑wrapped SOPs, tool integrations, and sub‑agents to break tasks and prevent context pollution. Although the process sounds complex, it forms a closed loop: the first configuration takes effort, but subsequent invocations are incremental.

4. Acceptance: No Dashboard, No High‑Speed Run

Without acceptance criteria and external evaluation, an AI agent is like a sports car without a dashboard—dangerous to run at speed. The author notes that complex tasks often have less than 50% reliability, so before launch one must define what counts as "complete" and attach an external benchmark that allows rollback when issues arise.

5. Summer Takeaways: A Four‑Step Checklist

Draw a business ontology diagram covering objects, actions, and rules.

Consolidate scattered documents into a versioned knowledge base.

Use prompts and skills to orchestrate a small closed‑loop workflow.

Define completion criteria, attach external evaluation, and ensure rollback capability.

The author warns against piling on tools first; clarifying the business model is the prerequisite for AI to deliver value.