Your calculation is correct if the error bars are defined to be something like "maximal deviation from the center value" without specifying how the error-distribution looks like. In particular the result value $AB+ab$ is just the center of all possible values. it is NOT "the most likely value" nor is it "the mean value" (nor "median value" either). Its not very meaningful in a statistical sense.
In order to get any of these properties, you have to specify how the error-bars on $A$ and $B$ should be interpreted. Common choices include "normal distrubtion with mean $a$ and standard deviation $a$" or maybe "uniform distribution over the range $[A-a,A+a]$". Depending on this choice, the result will change.