The strongest idea is the reversal: Every’s biggest token user wonders whether everyone else should use more. The rest of the draft gives that provocation substance: ambitious experiments can be worth the cost, including failures that yield useful lessons, but spending still needs scrutiny. Lead with that tension.

**Headline:** “Token Math” names the subject without giving readers a reason to open. “How Many Tokens Are Too Many?” is clearer, but frames the story around a spending ceiling. The more surprising question is whether people are using too few. Make that the hook, while preserving Dan’s uncertainty about what the gap means.

**Dek:** “What’s the ROI on experimentation?” is generic and suggests a calculation the draft doesn’t deliver. The alternate gets closer to the actual tension, but “rein your tokenmaxxing in” is cumbersome and puts unexplained jargon in the inbox. Use the dek to explain how Every balances experimentation with accountability.

**Opening:** “Often if … you might think” delays the news with a tentative generalization. Start with the three-times gap. Also, “highest spender” implies a dollar ranking; the evidence here is a token leaderboard. “Thinks everyone else should” is firmer than the paragraph’s concluding question. Keep the distinction between a striking usage gap and Dan’s interpretation of it.

Suggested replacement:

**HED:** You Might Be Using Too Few AI Tokens

**DEK:** At Every, ambitious AI experiments get room to run—then scrutiny over what they cost, produce, and teach.

Every CEO Dan Shipper uses more than three times as many OpenAI tokens as the company’s next-highest user. That gap has him wondering whether the rest of us are being ambitious enough with AI.

This opening moves naturally into the existing paragraph distinguishing experimentation from tokenmaxxing. It preserves the measured comparison and presents underuse as a possibility, without treating token consumption as proof of productivity.

For a less reader-directed headline: **The Case for Using More AI Tokens**.
