Small-scale systems are easy to build and maintain. So long as the details can all fit in the heads of a small number of programmers, it’s relatively easy to shuffle things around to meet requirements and verify that all the pieces interact properly.

Large-scale systems are a different story. Many cooks in many interdependent kitchens necessitate strong, assertable rules that allow programmers to reason about the unknowable. These rules provide a baseline level of order, but to be truly useful, they need to be predictable: different programmers should be able to invent similar or identical rules by deriving them from a small set of core principles, such that everyone can make reasonable predictions about the high-level behavior of code they haven’t read.

Software engineering at this level is an art, whose core mission is to find the right abstraction - one that naturally offers guidance and solutions for the problems that need to be solved (especially the ones that don’t exist yet). The wrong abstraction is painful and error-prone. The right one is a never ending stream of goodness from which all answers flow.