The key measure of leverage for AI-powered developers is no longer GPU utilization (FLOPs) but the volume of tokens processed by agents. Karpathy feels nervous when his token subscriptions are underutilized, indicating he's the bottleneck, not the system.

The proliferation of AI leaderboards incentivizes companies to optimize models for specific benchmarks. This creates a risk of "acing the SATs" where models excel on tests but don't necessarily make progress on solving real-world problems. This focus on gaming metrics could diverge from creating genuine user value.

Public leaderboards like LM Arena are becoming unreliable proxies for model performance. Teams implicitly or explicitly "benchmark" by optimizing for specific test sets. The superior strategy is to focus on internal, proprietary evaluation metrics and use public benchmarks only as a final, confirmatory check, not as a primary development target.

A trend called "tokenmaxxing" is emerging in Silicon Valley, where companies like Meta use leaderboards to track employee AI token usage. This reflects a corporate bet that higher token consumption correlates with increased productivity, turning AI usage into a new, albeit gameable, performance metric for engineers.

Jensen Huang reframes AI compute as a productivity investment, not a cost. He would be "deeply alarmed" if a $500,000 engineer used less than $250,000 in tokens, comparing it to a chip designer refusing to use CAD tools. This sets a radical new benchmark for leveraging AI in high-skilled roles.

Heavy use of AI agents and API calls is generating significant costs, with some agents costing $100,000 annually. This creates a new financial reality where companies must budget for 'tokens' per employee, potentially making the AI's cost more than the human's salary.

Jensen Huang argues that elite AI engineers should not be constrained by compute costs. He proposes a heuristic: if a $500k engineer isn't consuming at least $250k in tokens annually, their talent isn't being leveraged effectively. This reframes compute from a cost center to a critical force multiplier.

Labs are incentivized to climb leaderboards like LM Arena, which reward flashy, engaging, but often inaccurate responses. This focus on "dopamine instead of truth" creates models optimized for tabloids, not for advancing humanity by solving hard problems.

AI tools can generate vast amounts of verbose code on command, making metrics like 'lines of code' easily gameable and meaningless for measuring true engineering productivity. This practice introduces complexity and technical debt rather than indicating progress.