Apple puts its staff on an AI allowance. Budgets need a unit of value.
Apple has cut some engineers’ daily AI spending limits from $300 to $200 and may cut further, Bloomberg reports. Daily caps control spending, but they say nothing about value. The firms that get AI economics right will measure cost per completed task and fund the work that pays back.
What happened
Apple cut the daily AI spending limit for some engineers from $300 to $200 in October and may cut some to $100, Bloomberg’s Mark Gurman reported on 4 November. The cuts come, Bloomberg notes, “even as the company seeks to put AI at the center of its operations.” Many non-engineering groups already have a $100 daily limit, and some teams could be capped at $25. Apple’s staff use Claude, ChatGPT, Gemini and other models, and some teams are being asked to switch to cheaper models or on-device AI. Engineers have the largest budgets because AI coding consumes many tokens. Employees who need more can ask executives for an increase. Some employees said few people use their full $300 limit, so the savings may be small. The report links the cuts to wider cost reviews under chief executive John Ternus, who took over on 1 September.
My view
A daily allowance is a reasonable first move and a poor long-term answer.
Bloomberg’s framing captures the tension. A company can’t put AI at the center of its operations and ration it by the day for long. One of those has to give, and what decides which is whether the firm can show what the spending buys.
It’s reasonable because AI spending is new, variable and hard to see. Coding agents in particular can run long sessions that read and reread large codebases, and a few heavy users can account for most of the bill. A cap stops the surprises. It also tells staff that AI isn’t free, which changes behavior on its own.
It’s a poor long-term answer because a cap measures the wrong thing. It counts dollars per person per day. It says nothing about what those dollars produced. An engineer who spends $200 to close a week of backlog and an engineer who spends $200 rerunning the same failing prompt look identical on the budget report. Cut both to $100 and the firm saves money on the second and loses value on the first.
Measure the task, not the person
The better unit is cost per completed task: per resolved ticket, per merged change, per reviewed document, per closed claim. That requires two things most firms don’t yet have. First, a definition of the task and of “done,” agreed by the people who own the work. Second, a way to attribute model spending to the task rather than to a seat or an account. With both, the conversation changes from “who is spending too much” to “which work pays back, and which should stop.”
Model choice is an economic decision
The report that some teams are being moved to cheaper models or on-device AI is the more interesting part. Routing work to the least expensive model that does the job well is how AI costs fall without cutting value. It only works if someone has tested which tasks each model handles well enough, and if the routing is a policy rather than each employee’s guess.
Caps can become a quiet pullback
There is a risk in rationing by job title. If the people with the most valuable uses hit the cap first, adoption stalls in exactly the places it was working, and nobody decided that it should. A cap with a fast, visible exception path is fine. A cap that becomes the default answer is a pullback by another name.
Why it matters
For the industry. Apple won’t be the last. As AI moves from pilots into daily work, finance teams will ask for the same control they have over cloud spending: budgets, attribution and accountability. Firms that only have per-person caps will end up rationing. Firms that can show cost and value per task can decide where to spend more.
How it connects to our work. AI Token Economics measures cost per completed task and separates committed capacity from cash, so a budget can follow the work that pays back instead of a flat allowance. Adoption by Design covers the other side: redesigned work and honest measures, so the value a budget is buying is visible. Three steps before tightening caps:
- Define the task and “done.” Agree the unit of work with its owner before measuring its cost.
- Attribute spending to work. Tag model calls to the task, team and agent, not just the account.
- Route by policy. Test which models handle which tasks well enough, and send each task to the cheapest one that does.
Sources
- Apple cuts internal AI usage budgets in move to reduce spending, Bloomberg, 4 November 2026 (subscription)
- Apple is putting employees on an AI budget, MacRumors, 4 November 2026, reporting Bloomberg’s story
Linked sources are the record. Check them for the full reporting.