Err they don't. That's Lamarckism theory, and I wasn't aware there was even a shred of evidence towards it.
Sometimes when you don't know how something happens, its because it isn't actually happening
Which isn't to say that the post to which you were replying didn't misunderstand what was meant by the word "agency".
I used to laugh at people who believed in an agency model and thought that humans were the "most evolved" (as opposed to fast breeding special like cockroaches, or for that matter cousins of ours like chimpanzees). I ascribed it to the hierarchical victorian world in which Darwin published (and the famous picture from that era of ape -> man). But of course it's ancient.
There is quite a bit from the last few years. Gene expression in particular is a field we continue to learn a lot about [0] and several studies have suggested Lamarckism might have some merit [1].
Saying there isn't a "shred of evidence" is just scientifically untrue. There is some legitimate evidence that limited information could be passed from mother to child prenatally.
Your post is the classic Dunning–Kruger effect. Few in evolutionary biology would be as "sure" as you are, and you get less and less sure the more you learn.
[0] http://www.ncbi.nlm.nih.gov/books/NBK43787/ [1] https://en.wikipedia.org/wiki/Lamarckism#Epigenetic_inherita...
You don't actually need any information or experience to pass through to the next generation for behavior today to affect long-term evolution. Even if the only thing that's going on is that a population is using social learning to bring individuals closer to their genetic potential in one way, that way will be selected for in future generations. The genome itself could acquire adaptations during life, and those adaptations will then pass to their offspring. One would expect these 'mutations' to be really small, but add up quickly over successive generations.