Google had great engineering practices and infrastructure, the best of my career across several FAANGS. If only they didn't make launching products (and killing the old one) the way for advancement.
Some of Google's great build tools came about simply because they built absurdly large binaries (GWS and GFE). IE, they own-goaled themselves repeatedly and had to invent entire new technologies to get themselves out of it. THis was a consistent pattern; the much-vaunted GFS was actually a bear to run in production. Some things I really liked about working there include:
Monorepo for the entire front end and back end of Google services and products (excluding the Android OS code, and some other projects) made it really easy to get started doing interesting things without having to do a lot of work. LexicographicRangeSharder, which shards data streams using reservoir sampling, is a great example that I wouldn't have been able to build myself, but could trivially link into my code.
Blaze was an excellent build tool most of the time, and the team that build it/supported it was excellent.
The build/test platform (Forge and TAP) were both absolutely genius, and made large-scale testing easy. I was the owner of numpy and scipy and frequently had to do megabuilds to ensure that upgrades didn't break important things (IE, we forward-fixed all the dependencies that would break when the upgrade was committed).
A highly scalable version control system with serial commit numbers, making all sorts of projects involving bisection and problem hunting far easier than git.
A wide collection of very smart people who could answer nearly any technical question that came up
After returning to more conventional company, as well as a startup I really miss a lot of that. but blaze and TAP were partly necessary because of the design decisions made by early (pre-google3) engineers.
