The decision to create separate monthly credits for agent builders is, above all, a market data point: consumption from that kind of use grew enough to justify its own billing line.
Companies don't separate billing for elegance. They separate when one group's consumption pattern distorts the general plan's arithmetic, hurting both sides.
In this case the distortion is clear. A subscriber chatting with an assistant consumes irregularly and predictably; someone running agents fires calls in sequence and exhausts the limit quickly doing the work they bought the tool for.
Separation benefits both: those chatting don't pay for the infrastructure of those automating, and those automating gain predictability to plan how much work fits in a month.
For small operations, the pattern works as a diagnosis. If AI spending grew in a way nobody can explain, there's probably automated work running with nobody tracking how much it consumes.