Long form thinking on architecture, debugging, AI in production, and the messy intersection of code and business. The trade-offs, failures, and decisions that do not fit into tutorials.
One email when something new goes up. No spam, unsubscribe anytime.

A client's support agent worked perfectly in the demo, then refunded three customers twice in its first week. Here is how I turned that flaky prototype into production-ready AI agents using idempotency, validation, guardrails, and full observability.

Many AI products fail not because of poor models, but because of poor architecture decisions. This guide explains the real difference between AI agents vs AI workflows, and how to design scalable AI systems that work reliably in production.