Before this election AB was still wrong, but very systematically so. Seemed that you could make a fairly accurate prediction of the actual result by correcting for the systematic error in consistent ways, eg in 2015 and 2010 I think AB got the Tory vote wrong by the same amount, Lib dems about right both times, and Labour wrong by the same amount both times.
Similarly "wrong", but predictably so, about the EU referendum. So yeah it's always been wrong but wrong in a consistent way. Until last night, when AB's Tory voters decided to stick with May even despite all her blunders (this last not a supposition, but straight from the horse's mouths, as it were), unlike what was seen in the rest of the country.
So yes. AB's poll was wrong. Predicting that the Tories would come in front of a Labour party that, until two weeks ago, was in an absolute shambles, takes no skill at all. I could have polled myself and told you that. What matters isn't the lead the Tories had, but the scale of it. A prediction isn't "Tory lead" but "Tory lead by so many seats". And AB was out by easily a hundred, probably more.