Why Microsoft is joining industry crackdown on 'tokenmaxxing'

Microsoft has issued a warning to employees who engage in 'tokenmaxxing', as it joins the growing number of tech companies to advise against the supposedly flippant use of AI amid rapidly growing costs.

As one of the tech giants leading the AI revolution, it makes sense that Microsoft would want to do as much as it can to get its employees on board — with workers incentivized to take advantage of the tech wherever possible.

That quickly became an expensive incentive for Microsoft, however, as spending on tokens rapidly accelerated with little to actually show for the money — and a new trend known as 'tokenmaxxing' is at the heart of the issue.

Now, amid reflection of worrying costs, Microsoft is fighting back and has urged employees to 'be aware' of how they consume their tokens going forward, joining countless other companies attempting to cut out frivolous spending.

What is tokenmaxxing?

If you're unaware of what 'tokenmaxxing' is, it's a trend that has taken over the AI industry and wider tech world, with employees fighting back against competitive leaderboards in a way that bosses certainly didn't expect.

In order to encourage AI use, many companies created employee leaderboards where workers are ranked by their token usage – with tokens referred to the amount of work an AI model has to do, with one token roughly equating to around four characters, including letters and punctuation.

AI models like Microsoft's own Copilot use tokens every few characters processed, making redundant requests rather expensive (Cheng Xin/Getty Images)

It's the easiest way to measure how much someone is actually using AI, yet what it cannot quantify is the quality of the usage, leading employees to game the system and spam the models with as many meaningless commands and requests as possible to climb the rankings.

They'd incessantly ask what the weather is, enter any thought that went into their brain, or use it to automate useless tasks on repeat throughout the day — and doing so would cost the company a lot of money as a result.

How is Microsoft fighting tokenmaxxing?

It's not hard to understand then why Microsoft would want to cut this out wherever possible, and it's not just strongly-worded advice from leadership that's part of this top-to-bottom change.

"As we accelerate our use of GitHub Copilot to deliver on our goals, we all need to be aware of how we consume tokens," explained Jay Parikh, an executive vice president at Microsoft, in an email to employees obtained by 404Media.

"Tokenmaxxing is not what we are optimizing for," he continued. "I want all of us focused on maximizing outcomes that move the needle for our customers and our business. As such, we are updating our internal guidance and managing token spend with the same discipline we apply to every other critical resource."

Microsoft is imposing token limits, urging people away from tokenmaxxing, and even opting for GPT-5.6 over Copilot (Gary Hershorn/Getty Images)

Employees can now individually track how much they've spent on tokens, with the possibility for increased spending limits to be implemented at some point in the future, contrasting earlier advice that urged people to go all-in on AI.

What's more, Microsoft has made the bold move to switch out its own model, Copilot, for OpenAI's GPT-5.6 as it offers cheaper token spending and allows the company to "get greater value from our token investment."

This suggests a significant direction of power within the industry that Microsoft isn't seeing its own model as the best value, and that could then inform others who might currently be on the Copilot train.