Practical perspectives on enterprise AI — strategy, agentic & RAG platforms, governance, and getting from pilot to production.
A lesson from architecting an EAI/ETL mapping platform two decades ago: AI-generated integration logic becomes far less risky when it’s tied to metadata contracts, lineage, and validation — not introduced as free-floating code.
PLATFORMSThe workflow patterns I dragged onto a canvas 23 years ago mapped almost perfectly onto today’s AI agent orchestration. What didn’t survive were the two guarantees that made the old engine trustworthy.
STRATEGYA comment pushed back on my last article: governance isn’t a constraint on AI-native architecture, it IS the architecture. That’s right, and not the whole story — what’s the actual unit of architecture once code stops determining behavior?
STRATEGYMainframes, databases, SOA, cloud, containers, AI — every generation believed its technology was the transformation. What actually survives every migration is meaning, and that’s what architecture has always stewarded.
GOVERNANCEFive behavioral pillars only work if something enforces them. Why enterprise AI is splitting into three planes — execution, control, and evidence — and who has to own each one.
STRATEGYAI isn’t just another component in enterprise architecture — it changes what software is. Why applications, documents, and code may become implementation details.
STRATEGY“Should we train our own model?” is usually the wrong question. The value lives in retrieval, tools, and human-in-the-loop — not in custom weights.
PLATFORMSThe structural pillars govern how code connects. AI-native systems need a second, behavioral layer to stay trustworthy as they change.
PLATFORMSThe pillars still hold after two decades — but the security model, trust boundaries, and execution environment beneath them have changed completely.
GOVERNANCEThe hardest AI privacy problems show up after launch — not in the model, but in the plumbing around it.
GOVERNANCEYou rarely get burned because the model failed — but because you assumed intelligence alone was enough. Four controls that keep AI fast and the business in control.
Seven layers, structural and behavioral first principles, and six governance domains — a vendor-neutral blueprint for trustworthy, enterprise-ready AI.
CHECKLISTPressure-test whether your AI is safe to put into production across seven control areas.
Occasional, practical notes on shipping trustworthy enterprise AI. No spam, unsubscribe anytime.
I’ll never share your email.