Register-walled.

I expect that most of the "productivity improvements" from LLMs will come from destroying the value of certain signals so that nobody bothers to generate them... with indirect new costs because now the receiver has to create new ways to try to get the data.

For example, well-written job application cover letters. When LLMs make it cheap and easy to generate ghost-written counterfeits, they will stop being valuable. Job applicants will "freed" from the time-investment of writing them.

On the other hand, employers will no longer be able to use them as an indicator for "has more than a passing interest in our field" or "can string two ideas together" or "won't embarrass us in front of clients", etc. The new mechanisms to evaluate candidates--or the cost of hiring bad candidates--will be a an covert and indirect drain on productivity caused by LLMs.