Mark Zuckerberg has started writing code again. According to sources cited by The Pragmatic Engineer's Gergely Orosz, the Meta CEO has submitted three diffs to the company's monorepo - his first code contributions in roughly 20 years - and is described as a "heavy user" of Anthropic's Claude Code CLI.

The revelation lands at the same time Meta is aggressively pushing AI-assisted coding across the entire company. Internal documents show the Creation org, responsible for Messenger, WhatsApp, and Facebook, has set a first-half 2026 target requiring 65% of engineers to produce more than 75% of their committed code using AI tools. The Scalable Machine Learning team has an even steeper goal of 50-80%.

Zuckerberg is not the only tech leader picking up the keyboard again. Y Combinator CEO Garry Tan claims to be shipping 10,000-20,000 lines per day part-time using his open-source Claude Code skill suite "gstack," while simultaneously running YC full-time. A Fast Company analysis of Tan's output found significant code bloat, raising questions about whether raw line count is a meaningful metric in the age of AI-assisted development.

Meta's Reality Labs division has undergone the most aggressive AI-driven restructuring, abolishing traditional job titles in favor of small "AI-native pods" and rebranding employees as "AI builders." While Meta says headcount won't be affected, employees are openly anxious that the same tools making them more productive could eventually replace them.

The broader signal is clear: when the people running the biggest tech companies are personally using AI coding tools daily, the shift from optional experiment to core workflow is complete.