In data-driven product development, they use A/B experiments to determine if a new feature will improve some key metric in the long run and also will not impair other key metrics. Several teams may run tens of experiments at the same time. In mature products, 80%-90% of these experiments' outcomes are zero or negative and their code is wiped out to keep the codebase unclogged.

When I was working in such a company, I warned people in the interview that most likely 80% of the code they write will be thrown away. It is surprisingly hard to cope with.