A big problem with all AI at work punditry right now is that it all rests on data from the pre-agentic era (which is basically just now ending) and we have very little information about what has been happening since the Claude Code moment. So everything now requires some caveat.