What never changed

The timeline on this site is a list of things that changed. Ten of them, and each one felt at the time like the ground moving.

Take them in order and it is a lot of ground: a machine whose limits you could count on your fingers, then a publishing tool that turned out to be a programming tool, then a market that vanished without taking the work with it, then an entire operational discipline reduced to somebody else's console. Then three more inside fifteen years.

The honest question after a list like that is not sentimental. It is: what was any of it worth, if the expertise kept getting deleted?

The answer is that something never moved. Five habits, and I would have given the same five in 1999.

• Make systems simpler. Every year something arrives that adds a layer. The work is deciding which layers earn their keep, and removing the rest before they harden.
• Prefer open standards. Anything you cannot leave, you do not own. That was true of proprietary Unix, it was true of the platforms that turned every web property into a tenant, and it is true of whatever is being sold this year.
• Document. The version in your head is not a system. It is a single point of failure with opinions.
• Hand it on. Somebody else has to be able to run it. If it dies while you are away, it was only ever working.
• Build things that outlast the fashion. Boring that lasts beats brilliant that breaks. This is the one that gets harder with age, because after enough cycles you start being smug about it, and smugness gets the call wrong just as often as enthusiasm does.

All five are about what happens after the interesting part is over, which is where almost all of the actual time goes. The interesting part of any system is a few weeks. The rest is years of somebody keeping it alive, and that somebody is usually not the person who enjoyed the few weeks.

It is also why the list survived the shifts. Every one of them replaced a tool, a layer, or an entire job description. Not one of them changed the question of whether the thing you built can be understood, left, handed over, and kept running by a person who is not you.

The models are the current example. They are a genuine shift, the tenth on the list, and pretending otherwise would be the oldest mistake available. They also generate more code than anyone has time to read, which makes documenting, simplifying and handing over more load-bearing than they have ever been, not less. The newest thing on the list makes the oldest items on it matter more.

Thirty years is long enough to have been wrong about which things were permanent. I was wrong about several. The list is what is left after subtracting all of those.

Ask again in ten years and the timeline will be longer.
The list will be the same five.