Try the following: set up on a tripod and shoot something static - not something that can move like the 'eyes' examples you posted previously. Use live view to autofocus with the aperture set wide open and compare the result to the same shot taken using normal (phase detect) autofocus, again with the aperture set wide open.
Is the point of focus for the two images clearly different? Are the results reproducible? If the answer is yes then I propose that your technique has not suddenly disappeared overnight and that the source of the issue definitely lies with regular (phase detect) autofocus.