The hardest engineering skill I've developed is knowing when to stop optimizing. There's always another microsecond to shave, another edge case to handle, another abstraction to introduce.

The best codebases I've worked with share a common trait: they have clear boundaries about what they don't do. The worst ones try to be everything and end up being nothing well.

This applies doubly to developer tools. The ones that survive decades (Make, grep, curl) do one thing and compose well. The ones that try to be platforms tend to collapse under their own weight.