The strongest idea is that a large AI bill can raise the opposite question from the one you’d expect: Is everyone else experimenting enough? Dan’s threefold usage gap makes that idea concrete. The rest of the draft supplies the necessary discipline: Give ambitious experiments room to run, then evaluate the results and reduce waste. Higher usage isn’t itself evidence of better work.

- **Headline:** “Token Math” names the subject but hides the surprising claim. “How Many Tokens Are Too Many?” adds tension, but frames the piece around a spending ceiling when its strongest opening argument is about the possibility of spending too little. **“Your AI bill might be too low”** puts that reversal in a short, reader-facing headline. For a more explicitly attributed option: **“Every’s CEO thinks we should use more AI.”**
- **Dek:** “What’s the ROI on experimentation?” is abstract and suggests a calculation the draft doesn’t deliver. The alternative promises useful judgment, but “expensive experimentation” is heavy, and “rein your tokenmaxxing in” implies the reader is already burning tokens for sport—the behavior the next paragraph disavows. Use the dek to explain how Every balances experimentation with accountability.
- **Opening:** Lead with Dan and the usage gap. “Often,” “if,” and “might think” delay the point, while “highest spender” blurs token volume and dollar cost. The reversal is good; give it a direct first sentence. Keep the ambition question as Dan’s interpretation, rather than treating the leaderboard as proof that everyone else is underusing AI.

Suggested replacement package:

**HED: Your AI bill might be too low**

**DEK: At Every, ambitious AI experiments get room to run—and a reckoning with what the tokens bought.**

Every’s biggest OpenAI token user thinks the rest of us might be aiming too low. CEO Dan Shipper consumes more than three times as many tokens as the next-highest user on the company’s OpenAI leaderboard—and the gap has him wondering whether we’re being ambitious enough with AI.

This flows into the existing “The goal isn’t tokenmaxxing” paragraph, which explains the potential payoff. Keep its conditional “should they pan out,” and retain the later expectation that people assess results and improve efficiency. The revised opening scopes the comparison to OpenAI; the headline’s reference to a bill should not turn the threefold token gap into a claim about threefold spending. Randy’s experiment demonstrates learning and a revised setup, not a proven financial return.