That is a new piece in the May AER by Paul Beaudry, David A. Green, and Benjamin M. Sand, here is the clincher scary paragraph:
The data reveal a clear break in 2000. Between 1992 and 2000, each successive entry cohort has a higher share in cognitive occupations at the outset of their working lives, with the proportion increasing by 0.1 between the 1994 and 1998 cohorts. After 2000, with the exception of the difference between the 2004 and 2006 entry cohorts, each successive cohort has a lower share in these occupations, with the share at entry for the 2010 cohort being approximately the same as for the 1990 cohort. Given all the attention that has been paid to growing demand for cognitive skills, this complete reversal is striking.
Do you know of an ungated version? Here is a related Brookings piece of research (pdf). Here is a related piece by Acemoglu, Auor, Dorn, Hanson, and Price.