A token-usage KPI, and your manager is right
A manager set a KPI on tokens consumed. Absurd as a target, telling as a signal: AI fluency is now tested in interviews, even where the product has nothing to do with AI.
A manager in South Korea set his team a KPI on tokens consumed.
A middleware engineer who barely uses AI wrote to Gergely Orosz to complain about a manager "blinded by hype." Orosz's answer: your manager is trying to help you.
I laughed, then I looked at the numbers next to it.
At Higgsfield, which went from $1M to $1B in annualized revenue in 18 months, each person spends on average more than $10,000 a month on models. Nearly 400 people, more than $4M a month. One employee burned more than $30,000 in a week to fix an asset-organization problem on his own. The founder calls that a good week, and expects $50,000 to $100,000 a month for his best engineers within a year.
A token KPI is absurd as a target. But as a signal, it says something: how far a person has built agents into the way they work.
That's why it spills over into hiring. Indranil Ghoshal, nine years in product, notes that "AI fluency" is now tested in interviews, even at companies whose product has nothing to do with AI. Rarely announced, it slips into the normal questions.
His advice to candidates seems right to me, on both sides of the table: prepare an honest AI story. Where it saved you time, and where you pushed back on it.
