AI automates the tasks employees 'hate,' not the ones that 'make money.'
A few days ago, GeekPark reported that Microsoft, which has placed heavy bets on AI, quietly discontinued Claude Code licenses for most of its employees internally.
This is quite bizarre because one of the biggest selling points of this wave of AI adoption for enterprise users is 'increased efficiency.' If it can increase efficiency, why would Microsoft stop its employees from using Claude Code?
Microsoft is not the only one doing this. 'Tightening token usage' and no longer encouraging employees to go all out on Vibe Coding has become the new trend among Silicon Valley giants.
Uber spent its entire year's AI token budget in four months. Salesforce writes a check to Anthropic for about $300 million annually. An AI consultant revealed that one of his clients had a monthly AI spend as high as $500 million. Meta even quietly took down its internal 'tokenmaxxing leaderboard'—a board originally designed to encourage employees to use AI more.
Now, companies are doing something unthinkable a few years ago:
Limiting, and monitoring, employee use of AI.
Why are major companies shifting their stance?
"Tokenmaxxing," A Reflection of the Times
To understand today's cost crisis, we must first understand what 'tokenmaxxing' is.
This term started gaining popularity around 2025, literally meaning 'maximizing token usage.' Behind it lies a management logic—since the company spent big money on AI tools, employees should use them frantically. The more you use, the more 'digitally transformed' you prove to be. The less you use, you're wasting resources. As a result, many companies set usage quotas, leaderboards, and even performance reviews, pushing employees to use AI.
And the result?
Employees started using the company's enterprise-grade AI models to check the weather, write birthday greetings, and ask what to eat today.
A study of 2,444 companies found that for every dollar a company spends on AI tokens, $0.44 is used to fix bugs generated by AI, $0.27 to rewrite AI-produced code, and $0.11 is consumed in review and merge delays.
In other words, behind every dollar of AI procurement cost lies nearly 80% in hidden losses.
Investor Shruti Gandhi used an apt analogy: "A tokenmaxxing enterprise is like a company measuring productivity by keeping all the lights on—spending more money doesn't equal producing more."
More ironically, most of these companies have no idea what their employees are using AI for, let alone whether the completion of those tasks brought about any change because of AI.
This 'money-burning race' burned from 2024 into 2025, finally igniting this year. JPMorgan issued a sternly worded report with a title uncomfortably blunt—'AI Token Costs Are Eating Up Internet Profits'.
Shopify, Spotify, ServiceNow, and Roku all mentioned in their earnings calls that AI has become a major pressure point on operating expenses. The overall industry sentiment is starting to shift from 'how cool it is to use AI' to 'is this money well spent?'
When CEOs Start Questioning ROI
Only 14% of CFOs say they can see a clear, measurable return on AI investment.
Uber's Chief Operating Officer, Andrew Macdonald, said something very candid in a podcast—they find it difficult to link the productivity gains of individual employees to the overall business impact on the company. "If you can't see how AI helps you push more valuable features to users, token costs are even harder to justify."
This statement highlights the core of the enterprise AI dilemma: Improving individual efficiency does not equal increasing company profits.
An employee writes weekly reports three times faster with AI, but company revenue remains unchanged. An engineer generates code twice as fast with AI, but the code 'churn rate'—the proportion of code abandoned or rewritten—increases by 800%.
Microsoft's former Chief AI Officer, Sophia Velastegui, said something that makes many managers uncomfortable: "Most people default to automating the tasks they dislike, not the tasks most valuable to the company."
Put simply, companies are automating the tasks employees 'hate,' not the ones that 'make money.'
This is not a technical problem; it's a problem of priorities. It's also why about 30% of generative AI projects get abandoned at the proof-of-concept stage—costs are unclear, value is unclear, so the boss naturally stops paying.
Salesforce CEO Marc Benioff's approach is quite representative. Faced with an annual Anthropic bill of $300 million, his expectation is an 'intelligent router': something that can judge which queries are worth using a top-tier model for and which can use a cheaper, smaller model.
This idea itself isn't novel—as early as the cloud computing era, 'pay-as-you-go' and 'resource optimization' were standard practices. But this wave of AI came too fast; everyone bought first and thought later, only now starting to catch up.
Rational Return, or Prelude to Winter?
Microsoft recently canceled most enterprise licenses for Claude Code, with the official reason pointing to cost factors. This has sparked considerable discussion within the industry—after all, Microsoft itself is the largest investor in OpenAI, while simultaneously cutting subscriptions to a competitor. How much of this is cost consideration and how much is strategic planning is hard to say.
But regardless, it represents a signal: enterprises are starting to vote with their feet.
Harness and CloudZero both released AI cost management tools almost on the same day—May 28th. One focuses on real-time monitoring of AI spending and ROI, while the other launched an 'AI Financial Control Plane' to help companies link every dollar of AI spending to specific business outcomes.
The emergence of these two products itself illustrates the problem: there is market demand, and it's urgent.
Starting in April this year, HubSpot adjusted the pricing model for its AI agents, no longer charging by token, but instead charging by 'conversations resolved' or 'leads generated'—a directional shift aligning the seller's interests with the buyer's actual output. ServiceNow is making similar adjustments. AI vendors are realizing that if they continue to sell 'usage' instead of 'results,' enterprise clients will eventually push back collectively.
Is this adjustment a necessary growing pain for AI industrialization, or the prelude to a larger crisis?
I tend to think it's the former. But one detail is somewhat concerning: Global AI software spending is projected to reach $2.59 trillion in 2026, a 47% year-on-year increase. Yet, at the same time, 94% of engineering leaders say key ROI metrics are still missing. More money is being spent, but no one knows where it's burning or if it's worth it—if this contradiction isn't resolved, the next 'tokenmaxxing moment' is only a matter of time.
A Fortune magazine analysis put it bluntly: "Tokenmaxxing is easy; redesigning workflows is hard." What most companies are doing now is optimizing existing processes, not reinventing business models. This is where the real value of AI lies, and it's also a place most enterprises haven't reached yet.
A rational return is a good thing. But after this rational return, companies still need to answer a more difficult question: Should AI be a hammer for our business, or a new framework for thinking?
If you only use AI to do old jobs faster, the bill will eventually force you back to face this question.
This article is from WeChat public account "GeekPark" (ID: geekpark), author: Huilin Dance King, editor: Jingyu
