Writing

Practical notes on making AI agents reliable enough for work that can't be “mostly right” — LLM failure modes, orchestration, and audit-grade automation. Analogy-heavy, anti-hype.