Very much agreed, and keep in mind how much we might Jevons Paradox ourselves if we just try to count heads. I use LLMs a lot for basic background research at work, and the way I describe it to coworkers is "Imagine you have an unpaid intern, at 9am on their first day, who has no experience in your field, but is very widely read, and can give you very fast top-of-mind thoughts." This kind of research probably, on its own, improves my productivity 10-20% by freeing up time that I can spend on other things. I'm in a field where human time is expensive and reducing human time probably enables more clients to afford buying our product by more than enough to offset labor savings from rising productivity.