Creates a implementation plan using a graph of components instead of markdown docs and then implements each component in a loop before connecting them. My target was to figure out how to make LLM code more maintainable and a computation graph seemed to be the way. What do you think, any merit to the idea and has anyone tried a similar path?