As far as I can tell, there isn't any special treatment of starting food resources in the AI code. At any point of the game, e.g. for Animal Husbandry (AH), the AI counts (or rather: estimates) how many Pastures it'll be able to build and by how much the Pasture yields exceed the yields of the best improvement that could currently be built on that tile. AH especially was almost always researched way too late until I tweaked the yield evaluation in version 0.95(?). I don't think AH was reearched that late in BtS and K-Mod, so I suppose I had created the problem by reducing the value that the AI assigns to health in the (very) early game.It is basic strategy to choose initial techs based on starting food resources. However, and for example, I see time after time the AI choosing something like mining when it has 2 cow resources. I understand having some flavor, but initial tech choice should almost always be based on starting food resources.
I've run another test game now, and noticed Roosevelt (starting tech Agriculture, Fishing) with a coastal start on Prince difficulty researching Hunting, Sailing, Archery and then the Wheel. He ended up working a Plains Deer Camp and Wine Farm for a while until he finally got to AH and improved his Plains Sheep. Screenshot attached, showing also Roosevelt's tech values after discovering Archery (Ctrl+Alt on his capital in Debug mode). The shown tech values unfortunately include a random portion. When viewing the values in Debug mode, that's kind of OK because the values get constantly recomputed, so one can see how randomness affects the values. For a screenshot, it's not so good. (Oh, and the right part of the screenshot is from turn 5, i.e. not contemporary with the left part.) But one can tell that AH isn't valued highly enough. Perhaps trade networks are generally valued too highly too early, which would explain Wheel > AH and early Sailing. Archery and Hunting is OK with me; hardens the AI against Warrior rushes.
I've found and fixed one more obvious issue with the yield evaluation: Values were reduced to 2/3 in order to account for owned resources that might be outside of city radii – but that never happens in the very early game. Probably still not enough. At least during the Ancient era, when performance isn't an issue, the AI should go through each city, check for tiles in the city radius where the new improvement can be built and compute the difference in yields between the new improvement and the worst currently worked tile. And then put the sum of those differences in relation to the total yield rate of all its cities.