# Editorial feedback

## Strongest idea

The most interesting claim is not “token math” in the abstract. It is the paradox that Every’s heaviest OpenAI user—at more than three times the usage of the runner-up—thinks the company may be underusing AI. The draft earns that provocation with a more nuanced argument: high usage is not inherently good, but an expensive or even failed experiment can be a sound investment if it yields useful knowledge, a benchmark, or a more efficient way of working.

That distinction should drive the opening: judge experiments by what they produce and teach, not by token volume alone.

## Headline and dek

“Token Math” is short, but generic and bloodless. It suggests the story will offer a calculation or ROI formula, which the draft does not. “What’s the ROI on experimentation?” repeats that promise while withholding the article’s surprising answer. The alternate headline is clearer but still broad, and a question is less forceful than the claim the reporting supports. The alternate dek has energy, but “tokenmaxxing” makes the package feel more insiderish and jokey than the underlying management question warrants.

### Recommended package

**HED: The Case for Spending More Tokens**

**DEK: Every’s top OpenAI user consumes more than three times as many tokens as the runner-up—and thinks the rest of us may not be experimenting ambitiously enough.**

The headline states the counterintuitive claim cleanly; the dek supplies the evidence and preserves “may,” which matters because Dan is raising a question rather than proving that everyone should increase usage.

A slightly more analytical alternative:

**HED: When More Tokens Are Worth It**

**DEK: How Every distinguishes useful AI experimentation from expensive waste.**

This is accurate, but less distinctive. If the article were reframed around its most vivid detail, “What Every Learned From Burning 4.5 Billion Tokens” would be the strongest headline in the draft’s material. That choice would require bringing Randy’s experiment forward, however, and is therefore more than an opening polish.

## Opening paragraph

The current paragraph takes too long to reach its hook. “Often if…” is throat-clearing, the conditional construction is awkward, and “highest spender at your company” is broader than the known fact: the evidence is an OpenAI token leaderboard, not necessarily total AI or dollar spending. “Consumes” is clinical, while sentences two and four make nearly the same point. The paragraph also stops at Dan’s question instead of setting up the draft’s real standard for evaluating an experiment.

### Suggested replacement

> Dan Shipper uses more than three times as many OpenAI tokens as anyone else at Every—and he thinks the rest of us may need to use more. Not for tokenmaxxing’s sake: He wants people to test ambitious new ways of working that, if they pan out, could more than justify their cost. The token math comes afterward: What did the experiment cost, what did it produce, and what did we learn?

This starts with the contradiction, keeps the comparison tied specifically to OpenAI tokens, and carries the reader naturally from the headline’s provocation to the article’s qualified thesis. “May need to” preserves the exploratory nature of Dan’s view; “if they pan out” preserves the draft’s qualification that expensive experimentation is not automatically worthwhile.

One small consistency note for the later copy: “caped the orchestrator” should be “capped the orchestrator.”