People who are serious about email don't have this problem. The feature isn't useful much as medication for metabolic syndrome isn't useful to athletes. More generally, statistical filters I can't train are useless -- like gmail's spam filter. That's one reason I don't use gmail (I keep an account for testing). I host my own mail and my own statistical filter, which takes care of three nines of anything that isn't real correspondence with no false positives. This was a solved problem in 2003. But rather than provide features that help users learn to train their filters, google followed the other webmail providers with a shared filter. That leads to a high false positive rate (despite rigorously training it since 2004, my gmail spam box has about one false positive a month).
Then there's automatic signature hiding, hiding of addressees, lack of support for constant-width typography, integration of bullshit from Plus (like insisting on autocompleting names from Plus rather than what's been previously sent and received), many facets of the new "compose experience", etc.
Google have made it clear they intend to be the McDonalds of software. Too bad... I used to be such a fan. But since they make ad-supported software, the McDonalds equilibrium was bound to obtain eventually.