From Tokenmaxxing to Tokenminimizing! Meta’s Zuckerberg admits to making a mistake.

```

Meta is hitting the brakes on its AI ambitions. After months of aggressively encouraging employees to use AI tools, the social media giant is now moving to limit internal AI token consumption and address an internal crisis triggered by aggressive restructuring — with runaway costs and collapsing employee morale putting simultaneous pressure on management.

According to an internal memo reviewed by The Information, Meta disclosed to around 6,000 employees this week that the company's internal AI spending alone is expected to reach several billion dollars by 2026, with plans to introduce a token management mechanism based on budgets and quotas in 2027. Meanwhile, CEO Mark Zuckerberg admitted in another internal memo that the company had "made mistakes" while reorganizing teams around AI, and pledged to provide "meaningful roles" for impacted employees.

The successive disclosures in these memos highlight the dual pressures Meta faces in its AI transformation: on one hand, internal AI costs are rising exponentially; on the other, employee discontent from aggressive restructuring has reached a tipping point—an employee vented publicly with profanity during a livestreamed all-hands meeting attended by thousands, bringing internal conflict into the public eye.

The developments have drawn market attention. AI researcher Gary Marcus noted that "tokenmaxxing is giving way to tokenminimizing," and predicts this trend will cause Anthropic and OpenAI's Q3 revenues to fall short of Q2. For Meta, striking a balance between cost control and retaining AI talent has become the most urgent management challenge.

Runaway Internal Consumption, Record Token Use

The expansion rate of Meta's internal AI usage has far exceeded expectations. In April this year, an internal ranking dubbed "Claudeonomics" showed that Meta employees consumed 602 trillion tokens in 30 days, then rose further to 737 trillion. Named after Anthropic’s flagship product, the ranking tracked AI usage of more than 85,000 employees, listing the top 250 "super users" by consumption. The highest single user consumed as much as 28.1 billion tokens in 30 days, which, according to Anthropic’s public pricing, could cost several million dollars.

This ranking spawned the phenomenon known as "tokenmaxxing"—employees competing to rack up higher token consumption to showcase their AI usage skills. Some even instruct AI agents to run multiple tasks in parallel, artificially inflating usage. Employees could gain gamified rewards such as bronze, silver, gold, platinum, and jade medals and titles like "Session Immortal" or "Token Legend".

Meta CTO Andrew Bosworth issued a warning in April, saying "no one should use AI for the sake of using AI," emphasizing that "token usage alone is not any meaningful measure of impact." The company subsequently shut down the Claudeonomics leaderboard. Now, according to the internal memo, Meta is building a centralized dashboard called "AI Gateway" to monitor employees’ AI usage and expenses in real time, launch automated alerts for abnormal consumption, and track existing costs to forecast future spending — serving as a basis for resource planning and negotiations with vendors.

Switching to In-House Tools, Reducing Reliance on External Vendors

Another path for cost control is steering employees toward Meta’s own AI tools. The memo states Meta plans to guide employees away from third-party AI programming tools (especially Anthropic’s Claude) and toward the company’s in-house programming assistant MetaCode (formerly Devmate).

Reportedly, the newly established Applied AI Engineering (AAI) division has tasked engineers with specifically improving MetaCode’s capabilities, such as generating high-quality reinforcement learning data—by repeatedly having MetaCode solve coding challenges to train its response skills. The company noted it will still retain employee access to third-party AI models.

Meta currently faces dual financial pressures: On one hand, the company plans to spend up to $145 billion in capital expenditures this year, some of which will go toward expanding data centers, AI chips and talent reserves; on the other, investors are continually pushing Meta to get returns from its massive AI investments. Meta has launched paid subscription tiers for Facebook, Instagram and WhatsApp, and signaled it may charge businesses using its AI commercial agents. Against this backdrop, the strategic value of lowering internal operating costs becomes more prominent. It is worth noting Meta is not alone: reportedly, Uber and ServiceNow exhausted their entire 2026 annual budgets for Anthropic tools within just a few months, and several venture firms have also set employee AI usage caps, as daily token costs can reach thousands of dollars.

Mandatory Transfers Trigger Internal Crisis, Employees Protest Publicly

Beyond cost issues, a bigger internal crisis at Meta comes from AI-driven organizational restructuring. The Applied AI division, established in March 2026, now has around 6,500 engineers and product managers, with large numbers reassigned to new roles with little warning. In May, Meta cut 8,000 employees citing the AI transition, with another 7,000 reassigned to new AI projects.

Those forcibly transferred now mainly generate puzzles, write programming challenges, and conduct model test evaluations to provide data for AI model training. For engineers accustomed to developing products and launching features, this shift is widely seen as a career downgrade. One current employee described:

"Suddenly you have no direction in life, barely talk to anyone, and just mechanically repeat these tasks every week."

Another stated bluntly:

"Most people feel this work is suffocating."

Excessive flattening of the organizational structure has exacerbated the discord. In some teams in Applied AI, managers have to directly oversee about 50 employees, leading to lack of support, unclear promotion pathways, and reduced visibility to leadership. The pent-up discontent erupted in a livestreamed all-hands meeting this week: one attendee lost control, interrupted a speaker with profanity, and demanded the criticism be relayed to a certain AI executive, calling them "an asshole". Earlier, over 1,600 Meta employees had signed a petition to halt a project that collected AI training data by recording U.S. employees’ mouse clicks, keyboard strokes and screen activities. Under pressure, Meta has since slightly scaled back the project.

Zuckerberg Admits Mistakes, Leadership Responds Rapidly

Amid the intensifying internal crisis, Meta’s top management broke their silence. Instagram Chief Product Officer Chris Cox described the past months in the all-hands meeting as "difficult" and "brutal", comparing employees’ experience to "running a marathon in hail, swapping teammates halfway, with people recording the whole time". He gave an unusually sober assessment of AI:

"It’s neither a god nor a devil. It’s not as good as you imagine, nor as bad as you imagine."

Zuckerberg was more direct in his internal memo: "Given the complexity of these changes, we made mistakes." He promised to provide "as much stability as possible," announced a large-scale hackathon for July, and will adjust the management structure of the Applied AI division. According to Reuters, Zuckerberg also said there would be no company-wide layoffs this year.

Analysts note that these statements show Meta’s leadership has realized this round of restructuring is posing a serious risk to its talent pool. Engineering talent is the scarcest resource in the current AI race—if key employees consistently feel marginalized, the risk of their departure could become a major liability at a critical competitive juncture. Whether the remedial moves now underway can really stabilize morale remains to be seen.

Risk Disclosure and DisclaimerThe market is risky, invest with caution. This article does not constitute personal investment advice, nor does it take into account individual users' specific investment goals, financial situations or needs. Users should consider whether any opinions, views, or conclusions expressed herein fit their circumstances. Invest accordingly at your own risk. ```