KENNY CHIEN / CONSULTING / AGENTIC ARCHITECTURE & ONTOLOGY
Agentic Architecture & Ontology
The decisions that outlive any model: how your enterprise is represented, and what agents are allowed to do to it.
Models are becoming a commodity; the durable asset is the layer between them and your business. That layer is an ontology — a machine-readable model of your entities, relationships, actions, and permissions. It is what lets an agent know that "cancel the order" is a real operation with preconditions, side effects, and an owner — not a string to autocomplete.
I work with your architects to design this layer deliberately: the semantic model, the action interface agents are given, the permission system that makes autonomy safe, and the evaluation regime that keeps it honest as models and business rules change underneath it.
“Agents do not fail for lack of intelligence. They fail because nobody told them what the business is.”
What we design
Semantic core
Entities, relationships, and state machines that describe your business as it actually runs — derived from systems and operators, not org charts.
Action interface
A typed catalogue of operations agents may invoke — each with preconditions, effects, cost, and reversibility — so capability is granted, never improvised.
Permission & provenance
Who may an agent act as, on what, within which limits — and a full decision trail for when audit, legal, or an angry customer asks why.
Model-agnostic seams
The ontology binds to your systems, not to a vendor. Swap models the way you swap drivers; keep the asset.
How the work runs
Excavate
Workshops with operators and system owners. We mine the real ontology from how work actually flows — including the exceptions nobody documents.
Formalize
The semantic model and action catalogue, written down, reviewed, versioned — plus a reference implementation against one live system.
Prove
One agent, traversing the ontology in production conditions, with evals that test permission boundaries as hard as task success.
What you walk away with
An ontology document your engineers and your auditors can both read
A reference agent implementation against your live systems
A permission model that makes "give the agent access" a precise statement
An architecture review your board can act on
Frequently asked questions
What is an agentic ontology?
An agentic ontology is a machine-readable model of your enterprise — its entities, relationships, actions, and permissions — that lets agents reason about the business instead of autocompleting strings. It is the layer between commodity models and your systems, and the asset that survives model upgrades.
How is an ontology different from a database schema?
A schema describes your data; an ontology describes your business — including what operations exist, their preconditions and side effects, and who may perform them. I make the full argument in Ontology is the new schema. If you want to start mapping this yourself, the Ontology Canvas is the worksheet system I use in engagements.
Do we need to rebuild our systems before starting?
No. The ontology binds to your systems as they are, through model-agnostic seams. The work starts by excavating how work actually flows today, then formalizing it, then proving it with one agent operating in production conditions.
Architecting this now?
Share where agents sit in your current architecture diagram. I will tell you what will break first.