True expertise on a large, complex code base does not just take a few months; competence is best modeled as growing over long timespans, not as something that quickly reaches an "expert"-level ceiling. This is all the more true when-- as is often the case-- the codebase is a complicated legacy thing, full of accumulated "technical debt" and unneeded complexity of all sorts. This is not to say that 100% pair programming is always the answer, but doing it to some extent is likely warranted.