What drives (drove) most of it is that as a product matures and the user base grows, polishing smaller features and chasing 0.1% wins can meaningfully drive metrics to a point that having a team focus just on a single user flow is a net positive. Enabling this often means a more modular codebase and system architecture so people aren't stepping on each other's toes. There are also exploratory features that are unlikely to pan out, but at the scale of 500M users, if they do, it's huge. A lot of this work might be wrong, but some of it isn't, and you won't know what's what until you try.
At 500M users, you also need more spam detection, account protection, content moderation, etc.
You can run status quo Twitter with fewer engineers, but don't expect incremental growth, and expect more issues like the recent SEC mishap.