AI as the horse of the carriage

I like this kind of analogies to highlight certain aspects of human ai collaboration. This one is relating to my own experience to optimize earlier models by providing better context and optimizing my prompts, gradually increasing the scope of the tasks to keep them within the models capabilities. At some point the models became soo good that the scope of the tasks grew over my own capabilities. I simply was not able to fully understand the output and I lost confidence in my own decision making. Therefore, I am simply going backwards up the hill I initially climbed upwards already by firing incremental prompts and assessing AI output in chunks. I am the human in the loop slowing down the ai generation in order to keep navigating towards my actual goal with full awareness on my orientation. Practically, I utilize git as much as possible and try to adapt some sort of version control for all sorts of tasks. Prompting in pieces, keep staging and unstaging files around and eventually submitting a commit. Trying to be a good carriage driver, I pull the reins to limit my powerful horses to a speed that I can manage cognitively.