Consideration on Intent-Driven Development

In my in-depth study of hashtag#intent and hashtag#spec-driven development, I'm coming up with important insights regarding interactions between humans and AI. Thinking about intent, it's vague by nature, lacking technical details but ready to capture the needs of the customer (generally the person expressing the need).
The intent then evolves as knowledge increases, and the AI ​​supports this process, reasoning, providing different perspectives, and stimulating conversation. The intent grows and improves.
In this sequence of phases, as well as during subsequent phases where code is discussed, there's a concrete risk to consider: hashtag#cognitiveoffloading, meaning the human losing in-depth knowledge of the system because someone else is making decisions.
A framework can help mitigate this problem by generating shared consensus on the issue and keeping all (human) stakeholders on the same page (and perhaps doing other things in parallel).
Then, whether the problem actually exists depends on what we're building: an MVP? A PoC? A complex application?
If we rely on the intent to create an application, we assume the risk that the #AI hashtag will make decisions we're not aware of. On a PoC, I can afford that, but elsewhere, not so much. I believe this is an example of why the importance of humans will be in the ability to govern the system.