Does More AI Token Usage Really Mean More Productivity?

"That link is not there yet, right?" — Uber COO Andrew Macdonald on the connection between AI token usage and productivity gains.

The Token Paradox

Uber COO Andrew Macdonald stated in an interview released last week that he has not observed direct productivity improvements from increased AI token usage. While more products may be shipped, Macdonald noted it is difficult to attribute a specific productivity increase to token consumption. His comments gained over 2 million views on X.

AI tokens are the basic units processed by AI chatbots, roughly three-quarters of a word each. "Tokenmaxxing" refers to maximizing token usage to boost productivity and demonstrate capability.

The Corporate AI Rush

Macdonald's remarks come amid widespread corporate adoption of AI in the US:

• Meta has designated some employees as "AI builders" working in AI-native "pods."
• Disney and JPMorgan track employee AI use.
• Visa rewards teams for faster AI-driven development and reports monthly token spend of nearly 2 trillion.

Voices of Skepticism

Some tech professionals argue that AI adoption leads to waste:

"50% of internal token spend is completely useless." — Akshat Bubna, CTO of AI startup Modal

Engineering manager Karthik Hariharan stated that tokens were "burned for millions of dollars without any real significant ROI."

Google CEO Sundar Pichai said at Google's I/O conference that chief information officers are concerned about budget overruns on AI, adding: "I think the problem is going to get worse as we go through the year."

Investor Michael Burry called tokenmaxxing a "crazy, rushed, temporary phase" and warned that Nvidia stock could face an "aggressive" decline.

The Defense

Y Combinator CEO Garry Tan defended tokenmaxxing, stating: "We've been tokenmaxxing longer than most people."

The Numbers Tell a Story

A report from engineering intelligence company Jellyfish found a stark disconnect:

The top 10% of Claude Code users consumed about ten times the median developer's tokens but produced only about twice the output.

The report recommends tying costs to concrete metrics like pull requests rather than rewarding or penalizing raw token consumption.