One possibility is to combine these two options. Just mix the order.
Randomly select each user and assign him a group. Then proceed as follows:
Group 1: A B Group 2: B A
In this way you minimize the impact of the first design on the second and you see reactions of all participants to both designs.