Or are we quickly approaching an apex where the people running these companies realize AI cannot completely replace human developers?

The boosters are loud and the failures are made silent, until the failures become impossible to ignore. Tens to hundreds of millions of dollars spent on AI in enterprise software companies need to show RoI at some point, and the promises of executives leaning on AI's productivity improvements to make their accelerated release schedules need to actually make the releases. Its currently a very dangerous career move to publically apply critical thinking to LLM capabilities. I think the math sort of does itself.