Meta has executed a sharp strategic pivot from aggressively promoting artificial intelligence adoption to strictly regulating internal consumption. Following months of encouraging staff to utilize AI tools, the social media giant is now imposing limits on internal token usage to address uncontrolled expenditures and severe morale degradation caused by rapid restructuring. An internal memo reviewed this week informed approximately 6,000 employees that projected spending on internal AI use alone for 2026 is expected to reach billions of dollars. Consequently, the company plans to implement a comprehensive token management system based on rigid budgets and quotas starting in 2027.

Concurrently, CEO Mark Zuckerberg admitted in a separate communication that the organization 'made mistakes' in leveraging AI to drive team reorganizations, promising to secure 'meaningful positions' for affected personnel. These disclosures highlight the dual pressures facing Meta: exponentially rising internal AI costs and critical employee dissatisfaction stemming from aggressive structural changes.

The scale of internal AI consumption had previously far exceeded management expectations, fueled by a gamified ranking system known as Claudeonomics. Data compiled by Woofun AI shows that in April, Meta employees consumed a total of 60.2 trillion tokens within 30 days, a figure that subsequently surged to 73.7 trillion tokens. This system tracked the usage of more than 85,000 employees, identifying the top 250 'superusers' based on consumption volume. The highest individual user consumed 281 billion tokens in a single 30-day period, a volume that would cost millions of dollars at Anthropic's public pricing. This environment fostered a phenomenon termed 'tokenmaxxing,' where employees competed to inflate their usage to demonstrate AI proficiency, with some instructing agents to run multiple tasks simultaneously. Incentives included gamified medals ranging from Bronze to Jade and titles such as 'Session Immortal' and 'Token Legend.'

Recognizing the inefficiency of this approach, CTO Andrew Bosworth warned in April that 'no one should use AI just for the sake of using it,' asserting that token consumption is not a valid measure of influence. The company subsequently dismantled the Claudeonomics ranking system. Internal memos now reveal the construction of a central dashboard called AI Gateway to monitor employee AI usage and expenses in real time. This system will introduce automatic warnings for abnormal consumption patterns and track current costs to forecast future expenditures, providing a data-driven basis for resource planning and supplier negotiations.

Furthermore, Meta is actively steering employees away from third-party tools, particularly Anthropic's Claude, toward its internally developed programming assistant, MetaCode, formerly known as Devmate.

To enhance MetaCode's capabilities, the newly established Applied AI Engineering department has assigned engineers to generate high-quality reinforcement learning data by having the tool repeatedly solve programming challenges. Despite this push for internal tools, the company stated it will continue to allow access to third-party AI models. This cost-control strategy is critical given Meta's dual financial pressures: a planned capital expenditure of up to $145 billion this year to expand data centers, AI chips, and talent reserves, alongside investor demands for returns on these massive investments. Meta has already introduced paid subscription tiers on Facebook, Instagram, and WhatsApp and signaled intentions to charge companies for its AI commercial services. Woofun AI notes that the strategic value of reducing internal operating costs has become paramount in this context.

Meta is not isolated in this trend of curbing AI spending. Reports indicate that Uber and ServiceNow exhausted their annual Anthropic tool budgets within the first few months of 2026, while several venture capital firms have imposed strict limits on employee AI usage due to daily token costs reaching thousands of dollars. Beyond financial constraints, Meta faces internal crises driven by organizational restructuring tied to its AI transformation. The Applied AI department, established in March 2026, currently employs approximately 6,500 engineers and product managers, many of whom were forcibly reassigned without prior notice. In May, Meta laid off approximately 8,000 employees under the pretext of advancing its AI transformation, while another 7,000 were transferred to new AI-related projects.

Engineers previously accustomed to product development and feature deployment are now primarily tasked with generating puzzles, writing programming challenges, and conducting model testing to provide training data.

This shift is widely perceived as a demotion, exacerbated by excessive flattening of the organizational structure. In some teams within the Applied AI department, each manager oversees approximately 50 employees, leading to a lack of support, unclear promotion paths, and difficulty in gaining management attention. The accumulated dissatisfaction erupted during a live meeting this week when a participant interrupted the speaker with vulgar language, demanding that attendees convey criticism to an AI executive whom he called a 'jackass.' Previously, more than 1,600 employees had signed a petition demanding the termination of an internal project that tracked mouse clicks, keyboard inputs, and screen operations for AI training, forcing Meta to scale back the initiative.

In response to escalating internal crises, senior management has issued a series of public statements. Instagram CTO Chris Cox described the recent period as 'difficult' and 'harsh,' comparing the employee experience to 'running a marathon in hail, being forced to change teammates halfway, while someone is constantly filming you.' Zuckerberg's internal memo was more direct, acknowledging that '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 initiated adjustments to the management structure of the Applied AI department. Reuters reported that Zuckerberg also stated he did not expect any further company-wide layoffs this year. Woofun AI analysis suggests these statements indicate that senior management recognizes the restructuring poses a serious threat to its talent pool. With engineering talent being the most scarce resource in the AI competition, the risk of core employees departing could have significant consequences at critical competitive moments, leaving the efficacy of these remedial measures to be tested in the coming quarters.