Meta built an internal leaderboard ranking all 85,000+ employees by token consumption. Tiers included "Session Immortal" and "Token Legend." 60.2 trillion tokens in 30 days. An engineer suggested the real purpose was generating training data for Meta's coding models. The leaderboard was discontinued after media exposure. They were measuring volume, and the volume meant nothing.
Snapchat is putting brand AI agents directly into users' chat tabs. You can have an actual conversation with a brand, ask about a product, get a recommendation, tap to buy. Early data shows 22% more conversions than standard ads. The surface area: 10.5 billion messages a day.
Here is the problem. Meta was optimizing an old metric (token consumption) that AI made trivially easy to game. Snapchat created something so new that no metric exists for it yet. CPM, CTR, ROAS all assume a human saw something static. When the ad is an AI agent having a dialogue, what exactly are you measuring?
The companies pulling ahead are the ones who recognized that AI breaks their existing measurement frameworks and started building new ones. The companies falling behind are the ones gaming the old ones.
I wrote about this, plus the Federal Reserve getting three different answers to "how many firms use AI" and Coinbase rebuilding around one-person AI pods, in this week's Foraging. Link in comments.