Workers who can drive the tools are being repriced faster than schools can mint them. The agents that would replace them are still, mostly, a physics problem.
AiSignal staff · 2 September 2026 · 8 min read
PwC read nearly a billion job ads and found a number that should have stopped every HR all-hands: the wage premium for AI-skilled workers went from 25% in 2024 to 56% in 2025. That is not a gentle upskilling curve. That is a market screaming that it cannot find the people who know how to sit the new machine.
McKinsey says 88% of enterprises now use AI in at least one function. Deloitte says only about 20% of them can point to revenue growth from it. Both can be true. Adoption is a checkbox. Advantage is a practice. The premium is being paid to the people who can turn the first into the second — and to the people who can keep the lights on while everyone else prompts.
The agent that is not here
It is fashionable to skip from Copilot to mass displacement without visiting the power bill. A token-to-salary conversion published this spring put the current global factory at roughly 1,300 fully autonomous agent-equivalents. The median path to 1.2 million is 2028. Even the loud scenarios do not put a dent in global knowledge work until the end of the decade, and they assume the chips arrive, the models generalize, and the lawyers nap.
The bottleneck is not imagination. It is watts per worker-equivalent.
None of this is comfort. 1,300 is a 2026 number. The compute curve is not linear. The correct stance is not panic and it is not smugness. It is a census: how many people, at what premium, driving how many machines, against a factory that doubles on a timescale shorter than a labor contract.