The usual reason AI adoption fails: people expect it to work on its own.
Software labelled 'AI' is not an employee, and AI is not only chat. It doesn't decide what matters, judge what's good, or run itself. Left undirected, it just adds noise:
- too much manual effort still hanging around underneath the AI layer
- too much back-and-forth with a tool nobody's actually directing
- too much work living in one person's head, AI or no AI
- too much inconsistency between whoever happens to be using it
- too much experimenting without a clear win to show for it