I definitely agree with the core concept here - I think the the future isn't "make it so" prompts but crafting architecture and only sprinkling in nondeterminism where necessary. If a process can benefit from a DAG, it has no reason to give control to an LLM, so finding where to make the handoff is key.

I will say this has the stink of going too deep into the AI brain. It supports everything and can do everything and has a million options, but it lacks specificity and purpose. Humans need a simple "how does this help me immediately" to get excited about trying something out, not a laundry list of features when the use case isn't obvious.