An agent driving deterministic tooling in this way does so efficiently on whole business units. Each factory is a self-improving loop. An agent drives a recipe, and sees its effect on the collective repository set on which it is operating on.
A compilation or test failure in one repository in the set is fed back to an improvement to the recipe, which makes the provably correct change even on other repositories whose test coverage was less than ideal. This is how we scale safely.
