Speed is a legitimate goal almost everywhere, and a catastrophic one in a few
specific places: the data model, the public API, the thing you promise people
will not change.
Where to spend the slowness
You get a budget of carefulness. Spending it evenly is the same as spending it
badly. The parts that are cheap to change deserve none of it; the parts that
are expensive to change deserve nearly all of it.Most teams reverse this, because the expensive parts are also the boring ones.