What we learned shipping.
Written by the engineers who did the work, about the decisions that were genuinely hard. No thought leadership, no SEO filler — if we didn't learn it in production, it isn't here.
The most valuable thing an AI agent does is refuse
Every team benchmarks what their agent can do. Almost nobody benchmarks what it does when it shouldn't — and that's the behaviour that decides whether it survives contact with production.
Priya Raman
Principal AI Engineer
Sixteen milliseconds: how we cut a frame budget in half
Ashfall shipped at a locked 60fps across three platforms. Getting there meant treating the frame budget as a contract rather than a target.
Dana Okonjo
May 30, 2026We migrated a 22-year-old mainframe with the lights on
Two vendors said it couldn't be done without downtime. Here is the architecture that made fourteen months of live migration boring.
Marcus Feld
May 12, 2026Offline is not an edge case
Most mobile apps treat connectivity loss as an error state. For six thousand field clinicians, it's the normal state — and designing for that changed the entire architecture.
Elena Cortez
April 22, 2026We stopped interviewing for algorithms and started interviewing for taste
Two years ago we replaced our technical interview with a single question: show us something you made, and tell us what's wrong with it.
Ilya Novak
March 28, 2026In XR, a dropped frame isn't a bug report — it's motion sickness
Comfort is the hard requirement in spatial computing, and it constrains your architecture far earlier than most teams expect.
Sofia Marchetti
March 4, 2026Get the next one by email.
One a month. The engineering decisions that worked, the ones that didn't, and what we'd do differently.