Loading page content
Loading page content
Insights
Writing from the engineering and design teams — the reasoning behind our architectural choices, and the things we got wrong before we got them right.
Most business websites lose people for four reasons, and all four are measurable in about twenty minutes with tools that cost nothing.
2 min readThe demo worked. Everyone was impressed. Eighteen months later it is still a demo. Here is the pattern behind almost every stalled AI project we are asked to rescue, and the four things that separate the ones that ship.
3 min readA grounded assistant that occasionally invents an answer is worse than no assistant at all, because people stop being able to tell the difference. Refusal has to be a first-class outcome, not an edge case.
3 min readSoftware that assumes connectivity fails precisely where the data is created — underground, in the pit, on the route. Retrofitting offline support does not work; it has to be the first architectural decision.
2 min readCamera-based monitoring projects usually fail for consent reasons, not technical ones. Designing the privacy architecture first is what makes them acceptable — and demonstrable beats promised.
2 min readEvery team agrees the site should be fast. Almost none of them have a mechanism that stops it getting slower. The mechanism is the entire point.
2 min readMost content platforms are judged on what they let editors do. The more useful question is what they prevent — publishing a page with no meta description, a missing alt attribute, or a link that 404s.
3 min readSubscribe
No newsletter cadence theatre. We send something when we have something worth reading, and you can leave in one click.