OpenAI's assisted coding tool reached 5 million weekly users, disclosed alongside the acquisition that gives it persistent cloud agents.
Weekly use, rather than total accounts created, is the measure that counts. It indicates incorporation into the working routine rather than curiosity. At that scale, assistant-generated code stopped being an exception in repositories and became a meaningful share of what reaches production.
The consequence for code review is concrete. Review was designed assuming someone wrote each line and can explain it. When a substantial part of a change comes from an accepted suggestion, the reviewer's question shifts: not only whether the code is correct, but whether whoever submitted it understands what it does when the unusual case arrives.
Add the background work the acquisition enables and the problem sharpens: code produced while nobody is watching needs audit logging to be reviewable at all.
None of this is an argument against the tool. It's the observation that review practice has to catch up with who is actually writing.