Vincent Gusdorf warns that the corporate trend of 'tokenmaxxing' is hitting its limits as AI costs rise without a proportional increase in productivity.
Vincent Gusdorf, head of AI analytics at Moody’s Ratings, reports that the corporate trend of 'tokenmaxxing'—maximizing the usage of tokens to generate AI work—is reaching a point of diminishing returns. While tech leaders previously celebrated high token consumption as a sign of high performance, many businesses are now experiencing sticker shock as bills pile up without a corresponding spike in productivity. Palantir CEO Alex Karp expressed frustration with the trend, noting that many American businesses are becoming 'livid' about paying for tokens that create little value. Experts suggest that companies are moving toward a 'model routing' approach, where simpler tasks are assigned to cheaper models while complex tasks use more powerful ones. This shift reflects a broader realization that high token usage is not always the best metric for success, much like how lines of code once served as a productivity metric in software development. Companies are now being urged to use AI more disciplinedly, choosing the right model for the specific task at hand.