KENNY CHIEN / IDEAS
Ideas worth arguing with
Working theses from real enterprise deployments — argued in public, revised when they lose.
What are these ideas? They are the four working theses behind my consulting and coaching practice — on agentic ontology, on orchestration as the new unit of engineering, on forward deployed engineering (FDE), and on quality in the age of generation. Each one was formed inside real enterprise deployments, not composed for publication.
I hold them firmly and loosely at once. Firmly, because I bet engagements on them. Loosely, because a thesis that cannot survive a strong counterargument is not worth betting on. Every essay ends with the best objection I know and my answer to it. If you have a better objection, send it.
A thesis you cannot defend in public is not a thesis. It is a hope.
The essays
Ontology is the new schema
Databases described your data. Ontologies describe your business. Agents need the latter.
02Prompting is not programming. Orchestration is.
The unit of engineering is shifting from the function to the agent loop.
03The FDE mindset
Why the best consultants now ship code from inside the building.
04Slop is a choice
AI does not produce mediocrity. Unexamined taste does.
Questions about these essays
What does “ontology is the new schema” mean?
It means the durable asset in enterprise AI is a machine-readable model of your business — entities, relationships, actions, permissions — rather than the database schema. A schema describes how data is stored; an ontology describes what it means and what an agent may do with it. Agents need the latter to act safely. Read the full argument.
Is prompting a real engineering skill?
Prompting is a useful input skill, like writing a clear ticket, but it is not where the engineering lives. The engineering lives in orchestration: the loop of planning, tool calls, checks, recovery, and evals around the model. That is the argument of the essay on orchestration over prompting.
What is forward deployed engineering (FDE)?
Forward deployed engineering is the practice of consulting by shipping software from inside the client's team, against real data and real users. Palantir built its delivery model on it. The deliverable is a production system and a team that can run it, not a deck. The FDE mindset makes the case; the FDE engagement page shows how I run it.
Does using AI mean accepting lower-quality work?
No. The same model produces slop on one team and production-grade work on another; the difference is taste, review culture, and evals. Mediocrity is a process failure, not a model property — which is why slop is a choice.
How often are these essays revised?
They are working theses, so they change when a deployment teaches me something or a reader lands a better counterargument. All four were last revised in July 2026. Each essay ends with the strongest objection I know and my answer to it.
Want to argue with one of these?
Send me the strongest counterargument you have. The essays improve when they lose, and I credit good objections.