Infrastructure · 28 Aug 2026

What is a token ceiling per employee worth?

A response to a token-demand model replaced a simple per-seat cap with workers, agents, active hours, intensity and spawned work.

A fixed annual AI budget per person works if the product being purchased is a seat. It is less helpful when one worker can set several agents running, each for a different number of hours and each able to delegate work of its own.

The proposed accounting starts with the number of workers, agents per worker, active hours, tokens per hour and additional work spawned. Apply budget constraints to what people can pay for and power constraints to what can be served. Those are scenario inputs. They do not establish a forecast simply because they can be multiplied.

Publication and sources

Original on LinkedIn ↗
← Back to the archive