Build what you would use for a decade.

Most software is built to be shipped, not to be lived with. The first question I ask before starting anything is whether I would still want to use it in ten years — not whether it would impress someone in ten days.

A decade is a useful filter because it strips away the fashionable. The tools and frameworks will change; the temperament of the work should not. What you build to last is built differently from what you build to launch.

The strange thing is that the things I built to last have lasted — and the things I built to ship have mostly disappeared, sometimes even from my own memory.