Building with LLMs

I am writhing this as Claude Opus 4 is writing a detailed implementation spec for an application I am developing. One of the interesting side-effects of working with LLMs to create applications is that you need to find ways to pass the time as the LLM is busy working away at a detailed spec. It is very tempting to start multiple threads of inquiry. To have different LLMs test out different approaches. You are an orchestrator and you find yourself with empty time as the worker LLMs are figuring out the details.