He provides a defense of his youtube video here. These arguments are very similar to the ones I made on Martin Luther King day about racial disparities - a literature I'm more familiar with than the gender disparities literature.
The point is, late stage discrimination is the tip of the iceberg when it comes to disparities, although it's often the most trumpeted. Steve and I do make slightly different points. Mine emphasizes the fact that when we define the problem as "the portion of the disparity that is not accounted for by legitimate observable variables, like education, intelligence, etc.", we miss the fact that inequality in those "legitimate observable variables" is often not acceptable either - often because of earlier-stage discrimination or inequality that we would not approve of. So only focusing on the "unexplained" variation at the later stages downplays the problem of racial disparities.
Steve's video's point is more highlighting the fact that the problematic disparities we do see are - for lack of a better term - "compounded" inequalities that occur well before the employer, rather than "discrimination" by the employer. Steve's post's point is that in saying what he did in the video, he did not mean to imply there aren't problems further upstream (the problems I spent time discussing in my post the other day).
An excellent source on this kind of compounding of inequality is Charles Tilly's book Durable Inequality, as is Oliver and Shapiro's book, Black Wealth/White Wealth. I read both in an economic sociology course I took with Deirdre Royster at William and Mary, and to this day are very important for how I think about inequality and racial disparities.