> assuming they do the research and know what they are doing.

This is the assumption that has almost always failed and thus has lead to the banning of AI code altogether in a lot of projects.