That's a very weak blanket statement, there are totally reasonable A/B tests you can run that don't deteriorate a user's experience, and the results can guide you to a better customer experience overall.
It did not mean it too seriously, of course there are also good AB tests, but there are a lot of bad ones out there. Those are what the article was about.
Obviously, it depends on what the A/B test is about. Molesting or not the customer for the sake of some shortsighted metric is a bad choice; deciding what content should go above the fold or not (e.g. Amazon places images, short description, details/specs and similar products in that order) is a good choice.
(edited for clarification)