Meta doubles down on AI: Zuckerberg uses Claude to write code, while 85k employees burn through tokens like crazy to chase performance targets

Author: HIBIKI, Crypto City

Meta doubles down on AI: Zuckerberg is using Claude Code to write code Meta, the tech giant behind social platforms like Facebook, Instagram, and Threads, has recently been channeling the company’s resources fully into the generative AI space. Meta CEO Mark Zuckerberg has started using an AI coding tool, Claude Code, to write code in person, breaking a yearslong record of not directly developing himself. Meanwhile, foreign media also reports that Meta’s internal teams have recently sparked a token-spending competition, with many engineers burning through large amounts of tokens to boost their personal performance metrics (KPIs).

AI coding is all the rage; founders return to the development front lines In March 2026, Zuckerberg submitted three code diffs to Meta’s single repository, his first substantive code contribution in 20 years. Zuckerberg is using the Claude Code CLI, an end-terminal programming assistant developed by Anthropic, and one of his submissions received approvals from more than 200 engineers. His actions reflect how AI coding tools are drawing corporate founders back into systems development. Y Combinator CEO Garry Tan also returned to writing code after 15 years and open-sourced a system that combines Claude Code. According to internal documents leaked by Meta in March 2026, the company set ambitious targets, planning to have 65% of its engineers use AI to write more than 75% of their code by mid-2026.

Image source: flickr, Photo by Niall Kennedy ｜ Meta founder Mark Zuckerberg’s speech at the Facebook F8 developer conference in September 2011

A Token-burning contest inside Meta turns KPIs into a show To drive adoption of generative AI applications, a phenomenon has emerged inside Meta that ties token usage to productivity. Tokens are the smallest unit processed by large language models to handle text; in Chinese, they are often referred to as “symbol units” or “tokens.” The Information reports that inside Meta there appeared a leaderboard called Claudeonomics, tracking AI token consumption by more than 85k employees. The data shows that employees consumed as many as 60 trillion tokens in just 30 days, with the top-ranked user averaging 281 billion tokens consumed. The leaderboard assigns titles such as Token Legend to encourage employees to integrate AI tools into their daily work. Forbes reports that Meta CTO Andrew Bosworth has mentioned that a top engineer consumed a token amount equivalent to their annual salary; Nvidia CEO Jensen Huang has also said that he would worry if an engineer earning $500k failed to consume tokens worth $250k. However, this KPI system that drives internal competition for token consumption also brings downsides. To boost performance numbers, some Meta employees leave AI agent programs running for hours idly, wasting computing resources. In addition, treating employees’ token consumption directly as a productivity metric makes the consumption behavior turn into performance for show, posing a challenge for performance evaluations that lack support from tangible business outcomes.

Learning from metaverse failures, Meta’s next challenge in the AI race Before making a major push into AI, Meta’s metaverse bet ended in failure. The company once invested about $8 billion to build the virtual world Horizon Worlds and VR/MR equipment, even changing the company name to “Meta,” yet still failed to reach the user scale expected by the market. In a comment section on a social media platform, Lily Liu, president of the Solana Foundation, also took a pessimistic view of the virtual economy models behind blockchain games and the metaverse, which lacked substantive content support in the past.

Image source: Meta ｜ In the initial version of Meta’s metaverse platform Horizon Worlds, the virtual depiction of Zuckerberg shown

Now that Meta is shifting its focus to AI, it is actively laying out its market strategy. In addition to launching its own large language model, LLaMA, it is also gradually advancing an AI model project called “Avocado.” A recent Axios report also revealed that Meta acquired Moltbook, a so-called “AI version of Reddit” proxy community, and that Moltbook founders Matt Schlicht and Ben Parr will join Meta’s team.

• **Related report:**The lobster community was bought! Rumor says Meta acquired Moltbook, and the founder successfully transitioned from media to AI

Outside observers are also watching whether Meta can avoid repeating the metaverse’s pattern of overinvesting without meaningful applications, and whether it can turn the current internal token-consumption frenzy and acquisition deals of startups like Moltbook into real products with commercial value—so that it can gain a foothold in the fiercely competitive generative AI market.