Software engineering at Google & CS education with Titus Winters
Source
newsletter.pragmaticengineer.com
Author
Gergely Orosz
Date
Why it matters
A veteran who helped write the modern software-engineering curriculum argues much of it is outdated, with concrete critique of how CS education maps (or fails to map) to real engineering practice.
Key takeaways · AI-distilled
Google's 2010 cleanup of a roughly 250M-line C++ codebase predated LLMs: a MapReduce parsed every build target into an AST, matched exact call patterns, generated edits, ran old and new APIs side by side, sharded changes by directory, blocked the old pattern, then deleted it.
Per Winters, Google made a test mandatory with every change around 2007, after Search, which almost every change touched, slipped from monthly to quarterly releases. Search later returned to its faster cadence.
Every Google code change needs two sign-offs: a code owner with domain knowledge and a reviewer certified in that language's readability. Winters says more than 10,000 engineers were enrolled in the readability program at any given time when he left.
Winters says engineers at other companies describe the same pattern since AI adoption: code output up hugely versus a year ago, outages up about 30%, and the number of features shipped flat. He still expects net gains from projects too ambitious to attempt before.
Winters says the CS curriculum has barely changed since about 2000 and omits testing and verification. Professors tell him students below a B grade now tend not to understand code at all, and he would grade students' prompts and tests rather than their code.
