The AI budget has the wrong owner
As agents begin to represent projects rather than individual employees, AI budgets may need to follow the work.
Apple is reportedly cutting daily AI allowances for some employees, even as it pushes them to use AI more. Engineers who had limits of $300 per day have already been moved to $200, with further reductions planned.
I'd like to say that this makes sense...in some world, somewhere. e.g. Give each person an allowance, measure what they consume, and cut the allowance when the bill gets too large. But my goodness I can't make this make sense in my head.
Peter Steinberger's question of if the individual is still the right unit is great:
"As capabilities increase you want to work with and support an agent that represents your project, that runs loops and evolves your codebase. The humans merely curate the automations."
Apple feels like they're on a pre-agentic cost model.
If an agent (or swarm of agents) represents a project, its work may cross people, systems, and days. The person is no longer simply consuming AI through a chat window. They are deciding what work should continue, reviewing what comes back, and correcting the automation when it drifts.
A daily allowance attached to that person tells you how much they spent. It does not tell you whether the project received anything useful. And I don't think it even tells you if that individual person was successful utilizing their allowance.
The budget should follow the work instead. Give the project an AI budget, an accountable owner, a clear outcome, and conditions that stop the work when it is no longer producing something worth reviewing.
Individual limits may still be useful as a backstop. They seem like a poor way to govern agents that increasingly belong to a project rather than a person.
Cost controls still matter. But if agents change the unit of work, we probably need to change the unit of budgeting with them. Apple's move here is bizarre or out of touch, to me.