A company announced dedicated monthly credits for agent builders, effective from mid-June, creating a usage limit separate from the regular subscription plan.
The separation acknowledges a concrete difference in pattern. Conversation with an assistant is irregular use, with pauses between messages; an agent in operation fires many calls in sequence, without pause, to complete a task.
Charging both from the same limit penalised precisely those building something more advanced, hitting the plan ceiling while doing the work the tool was sold for.
There's a predictability effect that matters to anyone putting agents into production: dedicated credits allow planning how much automated work fits in a month without compromising the team's interactive use.
For small operations, the design is copyable even without such a plan. Separating interactive spend from automated work spend inside your own system is what reveals where the money is actually going.