[BoorishBears]

Why would you add a second function for the answer object when you can return an answer object in the same response as the chain of thought?

Overall your second approach makes for really terrible UX and dramatically weakens the performance at the task unless you go and repeat every single definition along the way: ensuring you now have X copies of the prompt spread across the code base and have blown up your token count.

Once you get to that level of granularity between calls, you've pretty much fall back into doing a slower more expensive version of NLP pre-ChatGPT.