The recurrent exposure to herbicides in agricultural landscapes forces weeds to adapt in a race against extinction. What role newly arising mutations and pre-existing variation play in this evolution of herbicide resistance is critical for developing management strategies. Here, we present a model of rapid adaptation in response to strong selection, capturing complex life cycles of sexual and asexual reproduction and dormancy in a perennial weed. Using a multitype Galton-Watson process, we derive the probability of herbicide resistance evolution and the waiting time distribution until resistant plants appear in the field. We analyse the effect of seed bank dynamics and details of the reproductive system in defining the probability and timing of resistance adaptation in Sorghum halepense. Further, we investigate key factors determining the primary source of adaptive variation. We find that even small fitness costs associated with resistance reduce adaptation from standing genetic variation. For herbicide resistance inherited in a (incompletely) dominant fashion, self-pollination also diminishes standing variation for herbicide resistance by increasing the homozygosity. Our study highlights the importance of seed banks for weeds' adaptive potential, preserving genetic information of forgone selection and prolonging the period in which the population can adapt.