Field Notes

Meta employees seeking to maxxx their productivity drove all-time AI usage highs, burning 60 trillion tokens in 30 days

Meta employees seeking to maxxx their productivity drove all-time AI usage highs, burning 60 trillion tokens in 30 days

Meta employees seeking to maxxx their productivity drove all-time AI usage highs, burning 60 trillion tokens in 30 days.

I'm really not sure what to do with that.

This is a company filled with smart people that somehow decided the best way to measure AI productivity was to count how much of it you used and give out super awesome titles like "Token Legend" and "Cache Wizard" to the big winners who burned the most AI tokens.

Not measuring what it produced, how processes changed, just... how much was used.

To no one's surprise, measuring tokens used tells you... how many tokens you used. Not whether any of them added value beyond melting ice in the fabled Northwest Passage.

Time to resolution, escalation rate, feature cycle time, deploy frequency, deal cycle length, win rate? Nope. Just token counting. Brilliant.

The other metrics can be slow, noisy, and hard to attribute to AI specifically, but that's fine. If you roll out AI across a team and don't see a meaningful shift in three months, that's a signal in itself.

Charlie Munger wins again - Show me the incentive, and I’ll show you the outcome (Even if the incentive is a bro-ified nerd title)