Field notes from nine years of shipping software for other people. Essays on what works, short-form scribbles on what doesn't, a reading list of the books we keep giving away. No newsletters. No growth hacks. Just the record.
Figma is a great sketch pad. It is not a specification. We stopped treating design reviews as "approve-to-build" gates about four years ago. What we do instead: if a decision matters — if it affects how the system feels to use, or how it performs, or how it scales — we prototype it in real code against real data, and we make the decision against the prototype. Lorem never leaves our machines.
This is slower than a Figma iteration. It is also the reason we have never delivered a feature that designed beautifully and shipped poorly. Legibility in production is worth ten reviews in Figma.
Request the full essay →On the difference between a Figma screen and a thing users can click.
What we learned running a partner-owned engineering studio.
The operating dashboards we publish to every client — and to ourselves.
Running RAG in production — chunking, evals, latency, cost.
Design and engineering as one practice, not two.
Why documentation is the shape of the software.
A data layer opinion, finally committed to paper.
A template we actually use, not one that lives in a wiki.
What we actually cache, what we refuse to cache, and why.
Apr 2026A budget only matters if a missed one costs someone something.
Apr 2026A long RFC is usually two short RFCs.
Mar 2026We ask the same four things every time.
Mar 2026