Writing
Practitioner notes on enterprise AI, cloud-native systems, and engineering leadership.
Series
Recent posts
What AI-Native Actually Looks Like — A Working Definition
What AI-native architecture actually looks like: five operational characteristics that define a binary boundary — you've crossed it or you haven't.
When You Have 10 Agents in Production: Spec-Driven Development at Scale
Scaling AI agents in production with spec-driven development: stop asking 'does each agent work?' and start asking 'does the system still do what I designed?'
Prompt Injection, MCP, and the Trust Boundary Problem
MCP standardizes tool connections — and attack surfaces. Enterprise prompt injection via MCP starts with retrieved data. Trust boundaries as code, not policy.
Cloud, On-Prem, or Edge: The Real Decision Framework for AI Workloads
Cost is the wrong frame for AI workload deployment. Three variables actually decide: data gravity, inference latency, and compliance surface area.
LLMOps in Practice: Observability That Doesn't Lie to You
LLMOps observability patterns for production AI: four instrumentation approaches that tell you if the system is right — not just whether it's running.
The Case Against the Monolithic Agent
Single agent vs multi-agent architecture in enterprise systems: the monolith fails the same way it always did — just faster, because agent errors compound.