# Your team's AI now comes with a daily allowance.

Agencies and Apple are rationing AI by the day. The interesting question is not the size of the cap but who decides when someone has earned a bigger one.

Published by Fiction Tribe, 2026-11-16. Canonical: https://velocity.fictiontribe.com/field-notes/your-ai-budget-is-a-daily-allowance

Key figure: $50 daily AI spend limit per person at media agency PMG.

## The short version

- Bloomberg reported that Apple caps engineers' AI token use at $200 a day, with non-engineers possibly cut to $25.
- PMG limits staff to $50 a day. Dept caps by team. Roast has capped tokens per person since March.
- Brainlabs skips the hard stop. Hit your tier and you show your work to your department head.
- Four agencies are not a market. But compute is becoming a budget line, and somebody has to own it.

AI was sold to teams as unlimited. It turns out it has a meter, and the meter now has a manager.

Digiday's Sam Bradley set out on 10 November how agencies are rationing what their staff spend on AI. The numbers are small and specific, which is what makes them useful. They describe a new line in the org chart that nobody drew on purpose: how much machine each person is allowed to use.

## Who is capping AI use, and by how much?

A handful of agencies and, reportedly, Apple. Citing a 4 November Bloomberg report, Digiday says Apple has put a hard cap of $200 a day on its engineers' token use and may reduce non-engineers' allowance to $25. US media agency PMG sets a $50 daily limit through its internal Alli4U portal. Roast, a performance agency, has capped tokens per employee since March, with a set allotment inside its Claude subscription and higher caps in busy periods.

> **$50** daily AI spend limit per person at PMG, set through its internal portal. Digiday, 2026

Dept caps at team level rather than per person. Its global EVP Jonathan Whiteside put the case for a constraint plainly: "If you say you have a limit of $30, $25, $10 a day, you'll find people become very creative in how they get this stuff done based on that constraint."

## Why would anyone ration a productivity tool?

For control, and sometimes for creativity. PMG's Dillon Larberg said the limit "provides us a human in the loop moment to meet with teams or meet with individuals." Staff who hit the cap more than once can be given a bigger budget or advice on using different models.

That is the part worth noticing. The cap is not really about money. It is a prompt for a conversation about whether the spend is producing anything.

> A token cap measures how much someone used. A checkpoint measures what they made with it.

## Which model should a team copy?

The one that ties budget to reviewed work. Brainlabs CEO Daniel Gilbert described checkpoints instead of hard caps: "When someone maxes out their spend for that tier, they show their work and their output to their department head." The constraint stays. The judgement moves to someone who can tell good output from busy output.

A flat daily number does the opposite. It treats the person writing a strategy and the person batch-producing variants as the same cost. Roast's John Campbell said it is "up to the individual to use those tokens responsibly", which is honest, and also leaves the hard question with the individual.

## How far should you trust this as a trend?

Carefully. This is four agencies' practices and one secondhand report about Apple. Dept's figures were offered as examples, not as its policy. None of it tells you whether capped teams produce better or worse work. Read it as an early signal of where governance is heading, not as a benchmark to set your own number against.

## What should a marketing or ops leader do about it?

Three things, before finance does them for you.

**Make compute a visible line.** Put AI spend on each project or team with a named reviewer, the same way you would budget freelance hours. If nobody owns the number, a blunt cap will arrive from above.

**Tie the allowance to output, not seniority.** Copy the checkpoint idea. When someone hits a tier, look at what they made. Raise the limit for the work that earns it.

**Ask your agencies the same question.** What share of their fee is machine time, and who reviews it? Expect procurement to ask it if you do not ask it first.

The meter is already running. The only choice left is whether it is read by finance or by someone who knows what good work looks like.

## Questions people ask

### Are companies capping employees' AI token use?

Some are. Digiday reported on 10 November 2026 that PMG limits staff to $50 a day, Dept caps usage by team and Roast has capped tokens per employee since March. Bloomberg reported that Apple set a $200 daily cap for engineers, with non-engineers possibly dropping to $25.

### What is the alternative to a hard AI token cap?

A review checkpoint. At Brainlabs, staff who max out their spend tier show their work and output to their department head instead of hitting a hard stop. It ties the budget to the quality of the work rather than to a fixed number.

### Is AI token capping a market-wide trend?

Not yet on this evidence. The reporting covers four agencies and one secondhand report on Apple, and Dept's $30, $25 and $10 figures were given as examples rather than policy. It is an early signal worth planning for, not a standard.

## Sources

1. [The case for and against AI token capping](https://digiday.com/marketing/case-for-against-ai-token-capping/), Digiday, 2026-11-10. Sam Bradley. Four agencies' practices plus Apple's cap as reported by Bloomberg on 4 November. Dept's figures are examples, not policy.
