Best Way to Handle Categorical Data With Random Forests

I was surprised when, during video 1, the bulldozers dataset was not transformed using dummy variables, but instead was used instead.

Is this always going to be the case, or was it due to the differences in compute times between one vs. the other?