// thoughts
Where I write about AI, engineering, and building in public.
Sep 21, 2026 — 12 min read
I raced TypeSafe’s jev against four generative models on support ticket triage under load, scored on cost per correct on-time decision. The answer was not the one I set out to find.
Coming Soon
Everyone's racing to build bigger models. I'm more interested in what happens when the model is smart enough but doesn't know what you need.
Coming Soon
Sometimes a regex, a lookup table, or a well-designed rule engine will outperform any LLM. Here's how I decide what tool fits.
Coming Soon
51 files. 14,750 lines of knowledge infrastructure. Here's the framework I built to prevent agents from writing plausible but wrong code.