Meta’s AI re-stakes: Zuckerberg uses Claude to write code, and employees wage a token consumption battle to hit their KPIs

Meta is doubling down on AI, with Zuckerberg personally using Claude Code to write code. To hit KPIs, an internal token-consumption frenzy has reportedly erupted, leading to chaotic wastage of resources. Drawing lessons from the metaverse’s $80 billion loss, Meta is actively acquiring startups, hoping to turn the technology into tangible value.

Meta is doubling down on AI: Zuckerberg is using Claude Code to write programs

Meta, the tech giant behind social platforms such as Facebook, Instagram, and Threads, has recently been putting all of its company resources into the generative AI field.

Meta CEO Mark Zuckerberg has begun personally using an AI code-writing tool to write code with Claude Code, breaking his record of not participating in direct development for many years.

According to external media reports, a token-consumption contest has also recently broken out inside Meta. Many engineers boost their personal performance metrics (KPIs) by consuming large amounts of tokens.

AI code-writing is all the rage, founders return to the development frontline

In March 2026, Zuckerberg submitted three code diffs to Meta’s single repository. This is his first substantive code contribution in 20 years.

The terminal-based coding assistant he uses is Claude Code CLI, developed by Anthropic. In one submission, it received approvals from more than 200 engineers.

His actions reflect how AI code-writing tools are attracting corporate founders back to system development. Y Combinator CEO Garry Tan has also returned to coding after 15 years, and open-sourced a system that combines Claude Code.

According to internal documents leaked by Meta in March 2026, the company has set aggressive targets, planning that by mid-2026, 65% of engineers will use AI to write 75% or more of their code.

Image source: flickr, photographed by Niall KennedyMeta founder Zuckerberg speaking at the Facebook F8 developer conference in September 2011

Meta holds a token-consumption contest internally, KPIs become mere theater

To promote generative AI applications, Meta has developed an internal phenomenon linking token usage to productivity. A token is the smallest unit that large language models process text; in Chinese, it is often referred to as “token” or “word token.”

《The Information》 reports that an internal leaderboard called Claudeonomics has appeared inside Meta, tracking AI token consumption for more than 85,000 employees. The data shows that employees consumed as many as 60 trillion tokens in just 30 days, and the user with the highest ranking averaged 281 billion tokens consumed.

The leaderboard also introduced titles such as Token Legend to encourage employees to integrate AI tools into their day-to-day work.

《Forbes》 reports that Meta’s Chief Technology Officer Andrew Bosworth previously mentioned that a top engineer consumed a number of tokens equivalent to their annual salary. NVIDIA CEO Jensen Huang has also said that if an engineer earning $500,000 fails to consume tokens worth $250,000, he would be concerned.

However, this KPI system built around “internal competition” for token consumption also has drawbacks. To drive up their performance numbers, some Meta employees leave AI agent programs running idle for hours, wasting computing resources.

Moreover, directly treating employees’ token consumption as a productivity metric turns consumption behavior into theater, making performance evaluation face challenges due to the lack of support from tangible business outcomes.

Learning from the metaverse’s failure experience, Meta’s next challenge in the AI race

Before making a major investment in AI, Meta’s bet on the metaverse ended in failure. The company previously spent about $80 billion to build the virtual world Horizon Worlds and VR/MR devices, and even changed the company name to “Meta,” but ultimately still failed to reach the user scale the market expected.

Lily Liu, President of the Solana Foundation, also shared a pessimistic view in the comment section of a social media platform when discussing the development of blockchain games and the metaverse.

Image source: Meta Meta’s metaverse platform Horizon Worlds—in the initial version, Zuckerberg’s virtual likeness shown on display

Now that Meta is shifting its focus to AI, it is actively expanding its market presence. In addition to launching its own large language model LLaMA, it is also gradually advancing a named AI model initiative called “Avocado.”

A recent report from 《Axios》 also revealed that Meta has acquired Moltbook, an agent community hailed as an “AI version of Reddit.” Moltbook founders Matt Schlicht and Ben Parr will join Meta’s team.

• **Related report:**The lobster community was bought! It’s reported that Meta acquired Moltbook, and the founder successfully transitioned from media to AI

The outside world is still watching to see whether Meta can avoid repeating the metaverse’s mistakes—overinvesting without real applications—and whether it can convert its current internal token-consumption frenzy and acquisition plans for startups such as Moltbook into actual products with commercial value, thereby establishing a foothold in the highly competitive generative AI market.