Writing
Essays on AI-first development, metadata-driven architecture, healthcare technology, and what I'm building.
-
The mind that writes the model
If the durable thing the AI should author is the typed model — then who authors it, and what happens when the author can remember, learn, and reason about what it built? This is what I'm exploring: a cognitive system that authors and evolves a governable metadata model. Two projects, one bet.
-
The LLM should write the model, not the code
Everyone is racing to make AI write applications. One camp has the LLM emit code — maximally expressive, totally ungovernable. Another has it emit flows in a fixed vocabulary — safe but boxed in. The durable answer sits in the seam between them: have the AI author the typed model the application is generated from. The model is the asset; the code is regenerable.
-
The prompt is code — and yours is drifting too
Most teams ship their LLM prompts as strings — assembled imperatively, scattered across services, untested, bloating token cost no one can see. MetaObjects makes the prompt a declared, deterministic, testable artifact: snapshot-testable, cache-stable, and drift-checked at build time. The prompt is code, so treat it like code.
-
The 2025 AI stack and the architecture it's missing
Context engineering, schema-driven agent design, knowledge-graph grounding, the Model Context Protocol — the AI industry is building useful pieces. None of them is the architectural layer the problem actually needs.
-
I thought AI killed MetaObjects. Then AI drift made it essential.
I built MetaObjects in 2001 to generate an entire application stack from one metadata model. I set it aside in 2022 — partly because I'd moved into bigger leadership roles, partly because low-code platforms like Boomi and OutSystems were doing similar work. Then I watched AI drift wreck consistency across my own side projects and at CareMetx, and realized metadata is the spine AI was always missing.