My gut tells me this pattern, which doesn’t really have a name, is going to be the pattern for working with these technologies to do complex things. The best name I can find for it is “specification repair”—see, for example, this paper. You ask a question and the LLM transforms it into a more detailed specification of a problem, then writes code to test that specification, then creates more specification, then tests that.
The first two steps of the OODA loop (Agentic loop) is to observe and orient: resulting in a specification of state and intention.