Tech workers made the most of AI. Now he tries to minimize it.
Earlier this year, the message from tech companies to employees was clear: Use as much artificial intelligence as possible in your work.
The staff called it “tokenmaxxing”, with a token referring to a unit of AI use that was roughly equal to a word fragment. Meta and Amazon employees even competed on leaderboards that tracked token usage.
Then came bills from companies like Anthropic and OpenAI, which provide AI tools—and they weren’t cheap. Now the era of tokenmaxxing seems to be over.
Meta told staff last week that it would soon scale back its use of AI after seeing an “exponential increase” in costs. In May, Uber said it had surpassed its projected annual AI spending in just four months, and had imposed certain monthly limits on AI coding tools. Walmart has also set limits on various AI tools. And Amazon and Meta took down the tokenmaxxing rankings.
In other words, “tokenminning”, short for “token minimization”, is now in.
The turnaround, which has occurred in just a few months, underscores that the use of artificial intelligence remains in flux as people try to figure out how to best use the tools.
“The biggest problem is that it’s all changing so fast, people and companies don’t know what to do,” said Rob May, chief executive of Neurometric, a start-up that helps companies make better use of artificial intelligence, and author of the book “Tokenmining Manifesto.”
“Executives who didn’t know how to measure their employees’ AI smarts thought, ‘Who’s using the most tokens?'” he said, adding that the philosophy ended up favoring volume over efficiency.
OpenAI and Anthropic offer subscriptions that cost $10 to $200 per month to use their AI models; when subscribers reach their usage limit, they will be cut off. However, most of the revenue comes from offering tools to companies like Meta, Shopify, and Amazon, which pay not only subscription fees but also tokens used by their tens of thousands of workers. So the more tokens used, the more money the AI costs.
A simple task like asking an AI to summarize the transcript of a company meeting can require several hundred tokens. More complex requests, such as writing code to create a new product or feature, can use tens of thousands.
The cost of using AI models has increased as they become more powerful and consume more tokens. Anthropic’s latest AI model, Fable, is twice as expensive as its predecessor, Opus. While there are cheaper models, many employees have become accustomed to using the most powerful models for everything, May said.
The ways in which people use AI have also changed. Instead of simply conversing with AI chatbots, engineers are deploying AI “agents” that can work on complex tasks for hours. As a result, engineers can use tens of thousands of dollars worth of tokens each month.
Many companies said they are trying to be more strategic about AI spending after not seeing a clear return on their investment.
“If you can’t really draw a straight line to the amount of useful features and functionality you’re delivering, it’s going to be harder to justify the business,” Andrew Macdonald, Uber’s chief operating officer, said in a recent statement. podcast interview. “The link isn’t there yet.
That doesn’t mean companies won’t continue to spend big on AI. Meta told employees it was on track to spend billions this year using AI, but it wanted to “find places where we can spend less while achieving similar or better business results.” Marc Benioff, CEO of Salesforce, the enterprise software company, said his company plans to spend hundreds of millions on AI this year, but is now eyeing “agentic work units” instead of tokens. The new metric is meant to measure output, not just usage.
Meta and Walmart’s limits on employee AI use were previously announced Information and Bloomberg.
It’s unclear how tokenmining might affect Anthropic and OpenAI’s bottom lines. At the peak of the tokenmaxxing era this year, AI companies reported record revenues driven by the use of coding tools. Last week, Meta told its engineers to use its internal coding assistant MetaCode instead of third-party tools whenever possible.
Meta declined to comment, Anthropic did not provide comment and OpenAI did not respond to a request for comment. (The New York Times has sued OpenAI and Microsoft, alleging copyright infringement of news content related to artificial intelligence systems. The lawsuits have denied those claims.)
The clear way forward for companies, Mr May said, is to use high-end AI only for complex tasks that require it, and to replace cheaper models with other examples.
Companies can save up to 90 percent by opting for less advanced AI models, said Andy Markus, AT&T’s head of AI. He said his engineers use the most powerful AI models for some tasks and the less powerful ones for most other actions.
“There’s an ebb and flow,” he said. “We’ve found that for most use cases, the latest, greatest frontier model isn’t needed.”
Kalley Huang contributed reporting.