Post by Tayyab Tariq

The AI-era equivalent of measuring engineers by lines of code may turn out to be measuring them by token burn. You optimize for what you measure. If token usage becomes the signal, teams will find ways to use more tokens. That does not necessarily mean they are building better systems. Jensen Huang recently argued that a $500,000 engineer should be using roughly $250,000 in AI tokens, which is exactly the kind of framing that can turn tool consumption into a proxy for performance (Huang, of course has an incentive to push in this direction:) ). The best engineers often create value by narrowing the problem, not expanding it. They remove unnecessary work, reduce complexity, and avoid creating more code and decisions than the team can actually absorb. AI usage should go up when it improves outcomes. But treating higher token consumption as inherently better engineering is how you reward motion and mistake it for progress.