Writing

Notes from building things. Agents, evals, systems, and the product details nobody asks about.

11 Oct 2026 · 3 min
Building Otto: my agents live on a cloud box now
Why I stopped running coding agents on my laptop, and the boring decisions that made a phone-first agent app work.
11 Oct 2026 · 3 min
Stop grading your agents yes/no
Why hybrid RAG still breaks on compliance documents, and how I evaluate agents with scores, reasons and boring deterministic checks.
11 Oct 2026 · 3 min
Halving LLM costs without making anything dumber
Batching, caching, and why your agent's tool design is secretly a billing decision.
11 Oct 2026 · 3 min
30 iterations to draw like a human
How I rebuilt a hand-drawn video engine from scratch for SketchPen, and what 'looks hand-drawn' actually means.