Writing
Build logs, not hot takes.
Engineering notes from building production LLM systems — agent architectures, guardrails, evals, and the cost discipline that keeps them running.

25 Sept 2026 · 20 min read
The 2 a.m. question: what Jev gets right, where it breaks, and how I would use it in an enterprise
A principal engineer's field notes on TypeSafe's Jev, the first System One model. How it works, what the official documentation says it cannot do, what happened when I put it behind a real incident-triage workflow, and where decision models belong in an enterprise architecture.
21 Sept 2026 · 9 min read
The boring work is the real work: what OpenAI's Codex story taught me about automation
OpenAI quietly published how their own engineers use Codex to automate the repetitive operational grind. As a principal AI/ML engineer, here is what actually matters in it, why the design is smarter than it looks, and how I would push it further.

17 Jul 2026 · 4 min read
How I let an LLM write SQL against my database — safely
The four-layer guardrail that lets a Gemini-powered manufacturing copilot write its own SQL without ever being able to hurt the data — and why the human stays in the loop.