There's a corresponding barrier in cultural evolution. Cakes contribute to recipes, but mainly through selection. Anywhere where reverse engineering is complex and difficult, something like the cultural equivalent of the Weismann barrier comes into play. It is hard to dispute the idea that there is less of a barrier in the cultural domain, though. The difference between the domains is partly because there is more reverse engineering in cultural evolution. Genetic engineers can reverse engineer designs originally coded in DNA as well, but for various reasons, they rarely bother. There's plenty of secret computer software in the world, but only a small number of secret genomes. So, people are forced to reverse engineer software from web sites, while few are forced to reverse engineer gene sequences from organic phenotypes.
It's important to recognize that Weismann's barrier is also pretty seriously compromised even if you ignore cultural transmission. As I explained in my article on Lamarckian inheritance in cultural evolution:
Some acquired characteristics are inherited while others are not. If Weismann had chosen a different trait - for example, stress - he might well have drawn the opposite conclusion.
Another problem for Weismann's barrier is genetic engineering. These days a scientist can transfer whatever information they like into the genome. Acquired characteristics, information from somatic cells, detailed sequence information, genes from other species, prime numbers - whatever they like, really. The permeability of Weismann's barrier is increasing as time passes and biotechnology improves.