Tokenmaxxing is the name used in 2026 in LLMs (so far) for a trend that began in February 2026: large employers pushing for maximal use of AI by their staff. Simon Willison illustrates it with headlines about Meta making AI adoption a formal part of performance reviews, Microsoft wanting almost every employee to use AI, and Uber boasting that 90% of its engineers were using AI in their workflows. A few months later the same companies were reported cracking down on employee token use, telling engineers that tokenmaxxing was “not what we are optimizing for”, and capping AI spending.
Usage
Willison’s explanation for the reversal is cost: tokenmaxxing went straight up and straight back down because agents turned out to be expensive. Where a year earlier it was hard to spend more than 1,000 in a day doing real work. He presents the episode as part of the evidence that AI hit product-market fit in 2026, primarily through coding agents.
The term describes an organizational pattern rather than an individual one; Willison’s name for the individual counterpart — the compulsion to keep one’s own agents busy — is AI Mania.