Thoughts on AI engineering, distributed systems, and building at scale.
How we built a GAN-inspired harness — planner, builder, and adversarial evaluator in a structured loop — that turns Claude Code into a deterministic software factory. The methodology behind spec-driven development, the Ralph Loop, and Beads.
How we built a multi-agent system where AI investigators, a Devil's Advocate, and a Judge collaborate adversarially to verify misinformation — and what we learned about building agents that argue productively.
How spec-first development and an eval framework turned a simple Gmail parser into a self-improving expense categorizer — and why building the eval harness before the feature is the most important habit in AI engineering.
Practical insights from building AI operations agents at scale — from system design to production reliability.
How we used spec-driven AI agents to reverse-engineer 8 Java microservices into Python/FastAPI — 122 tasks completed deterministically, 578 tests passing — and what this reveals about AI's real capability frontier.