Widespread use alone isn't conclusive proof of productivity, granted. But your alternative seems to require assuming thousands of companies are all independently making the same stupid decision, spending real money on tools that provide no value, while you've correctly identified what they've all somehow missed.

Corporate adoption can be wasteful, and some uses absolutely are poor. But that doesn't establish that LLMs don't improve productivity. At most, it shows the gains are task-dependent and easy to misuse.