June 21, 2026, (Inside AI) — Tech giants are slamming the brakes on employee AI use after a brief, costly binge. The era of ‘tokenmaxxing’—maximizing AI token consumption—has given way to ‘tokenminning,’ a push to minimize usage and control spiraling costs.
The Rise and Fall of Tokenmaxxing
Earlier this year, companies like Meta and Amazon encouraged workers to use AI tools as much as possible. They even set up leaderboards tracking token use, a unit roughly equal to a word fragment. Employees embraced the trend, dubbing it tokenmaxxing.
But then the bills arrived. Providers like Anthropic and OpenAI charge not just monthly subscriptions but per-token fees. With tens of thousands of workers, costs exploded. Meta cited an ‘exponential increase’ in costs when announcing limits last week.
Costs Spiral as AI Habits Shift
Simple tasks use a few hundred tokens. Complex coding requests can consume tens of thousands. The rise of AI agents—autonomous systems that work for hours—has pushed monthly per-engineer costs into tens of thousands of dollars.
Read: Run a Local LLM on Mac Mini: Slash OpenClaw API Bills to Zero
Anthropic’s newest model, Fable, costs twice as much as its predecessor. Many employees default to the most powerful models for all tasks, even when cheaper ones suffice. Rob May, CEO of Neurometric, said CEOs measured AI savviness by token volume, promoting waste over efficiency.
Companies Seek ROI and Set Limits
Uber blew through its annual AI budget in four months and imposed monthly caps. Walmart set tool-specific limits. Meta and Amazon dismantled their leaderboards. The pivot underscores a hunt for measurable returns.
Andrew Macdonald, Uber’s COO, said in a podcast:
“If you’re not actually able to draw a direct line to how many useful features and functionality you’re shipping, that trade becomes harder to justify. That link is not there yet.”
Marc Benioff, CEO of Salesforce, now tracks ‘agentic work units’ to measure output, not just consumption. Meta aims to slash spending while maintaining results, even pushing its internal MetaCode tool over third-party options.
The Path to Tokenminning
Experts advocate using cutting-edge AI only for complex tasks. Andy Markus, AT&T’s chief AI officer, said his teams save up to 90% by using less advanced models for most actions. May’s ‘Tokenminning Manifesto’ calls for strategic model selection.
Read: MIT and Microsoft’s Murakkab Slashes AI Agent Energy Use by 73%
The shift may dent revenues at AI providers, which hit records during the tokenmaxxing peak. But for now, companies are prioritizing fiscal discipline over unchecked experimentation.
Shamil has a low tolerance for AI that only works in controlled demos. He covers agentic systems and generative AI with the skepticism of someone who has seen too many “revolutionary” tools quietly disappear six months after launch.
