On 4 November, Bloomberg reported that Apple had lowered the daily AI spending cap for some of its engineers from $300 to $200, and that more cuts are planned. Apple declined to comment. The story got attention because Apple is also pushing its staff to use AI more. But the more useful detail for most teams is simpler: a large engineering organization now manages AI tool use with a per-person daily budget, and it is adjusting that budget as it learns.
If your team has given everyone a coding agent and a company card, this is a good moment to ask what your own numbers look like.
What was reported
According to Bloomberg's sources, as summarized by 9to5Mac and The Next Web:
- some engineers' daily cap went from $300 to $200 in October;
- Apple plans further cuts for some engineering teams, possibly to $100 a day;
- many teams outside engineering already had a $100 daily cap, and Apple has considered lowering it to $25 for some of them;
- employees who need more can ask senior leaders for an exemption;
- some teams are being encouraged to use cheaper models, or models that run on the device.
The budgets cover tokens across the tools staff can use, which reportedly include Claude, ChatGPT and Gemini. It is not clear which tools the limits apply to. One detail in the reporting is telling: few workers were using their full $300 allowance. The $100 and $25 figures are plans and options, not policies that have taken effect.
Apple is not alone. The Next Web lists other reported limits: Microsoft set division-level AI budgets in July, Tesla caps staff at $200 a week, Uber limits each AI coding tool to $1,500 a month, and Accenture caps Claude at $29 a month.
Why this matters for a smaller team
A cap of $200 a day is roughly $4,000 a month for an engineer who uses all of it on every working day. Most won't, and that is the point: the spread between people can be very wide. A developer who uses an agent for quick questions spends little. One who runs several long agent tasks in parallel, each rereading a large codebase, can spend many times more.
That spread is what makes AI spend hard to plan. A flat subscription per seat used to make developer tools predictable. Usage-based agents are closer to cloud compute: cheap for most people, expensive in a long tail, and easy to waste without anyone noticing. Apple's move looks less like a cut and more like an organization finding out what normal use is, then setting limits around it.
How to budget agent spend without slowing people down
- Measure before you cap. Pull a month of usage per person and per tool from each vendor's admin console. Look at the median and the top 10%. Apple's first cap was apparently above what most people used, which is a sensible place to start.
- Set a default with an easy exception. A soft cap that warns, plus a quick way to ask for more, catches runaway loops without punishing the engineer who is legitimately running a big migration. Apple's exemption path is the right idea, though senior leadership approval is heavier than most teams need.
- Match the model to the task. Not every task needs the largest model. Use a smaller, cheaper model for routine edits, searches and summaries, and keep the top model for planning and hard debugging. Several teams at Apple are reportedly being nudged the same way.
- Cut wasted context. A lot of an agent's cost goes into reading context. Clear project instructions, a tidy repository and short, focused tasks mean the agent spends fewer tokens finding its way around and more on the actual change.
- Put limits on unattended runs. Agents in CI or background sandboxes should have a maximum number of turns or a time limit. A loop that retries a failing test all night is an easy way to get a surprising bill.
- Judge spend by outcome. Tokens per engineer is an input. Merged pull requests, review time and incidents are what you actually care about. A higher bill that clearly shortens delivery is a good trade. A high bill with nothing to show for it is the one to look at.
The takeaway
AI coding spend is turning into a normal budget line, managed like cloud costs: measured, given sensible defaults, and reviewed as usage changes. Apple's numbers are not a benchmark for your team, since its scale and its needs are very different. The habit is worth copying, though. Know what your team spends, know who is at the edges and why, and set limits that stop waste without getting in the way of real work.
Sources: Bloomberg, “Apple cuts internal AI usage budgets in move to reduce spending” (4 November 2026), 9to5Mac (4 November 2026), The Next Web (4 November 2026).