The weight of a good default

When you ship software with a default, you are making a claim about what normal looks like. The default font size says: this is how large text should be. The default sort order says: this is what is most important. The default privacy setting says: this is how much of yourself you should share.

Most teams treat defaults as implementation details — the thing you ship before you have enough data to know what users prefer. I think this is backwards. The default is the product. Everything else is an override.

Defaults are a moral position masquerading as a technical decision.

This has become harder to ignore as software has gotten more complex. A default setting in a machine learning system is not just a convenience — it is a claim about what good output looks like, baked into every interaction until someone changes it. Most people do not change defaults. This is the entire point.

I am not arguing for paralysis — defaults have to exist. But I am arguing that the question “what should the default be?” deserves at least as much design attention as the question “what features should we add?”

The feature is optional. The default is mandatory.