A proprietary LLM developed to run these proprietary tools might just maybe have guards in place to prevent it from telling you any details about what it's doing, lest you learn enough to not need that LLM
This argument applies to learning to code in general. What is your estimate of the ratio between "just shipped whatever the clanker spat out" vs "interrogated the LLM about its own output†" vs "deliberately interrogates LLMs to learn"?
†Untrustworthy, because it will either double down on its own output in its context, or will flip-flop to whatever the user said .
I don't know how many users of LLMs fall into each of these buckets but I have the sense that there these distinct buckets. I don't think it's necessarily a global thing. A single user could fall into all three buckets depending on the domain they are working in. For instance for CI/CD I tend to ship clanker, for tests I interrogate LLM about output, but for code algorithms I am more likely to interrogate to learn.