AI infrastructure
Vector stores, embedding pipelines, model serving, and the platform boundaries that have to exist before enterprise AI does anything useful.
About
I work on the engineering underneath enterprise AI.
I work on the engineering layer behind enterprise AI: the identity, governance, platform, and runtime patterns that determine whether systems hold up in production.
Most enterprise AI conversations stop at the model layer. The harder problems sit below it: identity, trust boundaries, governance enforcement, data platform readiness. That's the infrastructure that decides whether agents can actually work inside a real enterprise, and that's what I cover here.
I write for people who need both views at once: what the system actually looks like inside, and what breaks when it's wrong. AI engineering, AI security, and the operating architecture between them.
Coverage
Six threads run through most of what I write.
They show up across essays, diagrams, and briefings. Different angles on the same engineering territory.
Data engineering
Medallion layers, data contracts, lakehouse architecture, and the pipeline work that turns raw enterprise data into something AI can actually trust.
Agent runtimes
Durable execution, orchestration patterns, state management, failure recovery, and the runtime scaffolding that keeps agents auditable in production.
AI security
Agent identity, scoped tool access, prompt injection, trust boundaries, and the governance layer that keeps AI systems usable inside enterprise constraints.
Decision automation
Rules engines, ML scoring, LLM reasoning, and routing logic. I write about why these need to stay separate and what breaks when they get collapsed into one opaque layer.
Agent commerce
Autonomous procurement, agent-to-agent negotiation, machine-to-machine payments, and what enterprise commerce looks like when AI handles the buy side.