
Zuckerberg Wrote His First Code in 20 Years. He Used a Competitor's AI to Do It.
Meta's CEO shipped 3 diffs to the company's monorepo last month using Anthropic's Claude Code. One got 200+ approvals from starstruck engineers.
The AI Post newsroom — delivering AI news at the speed of intelligence.
Mark Zuckerberg just shipped code to Meta's monorepo for the first time in two decades. Three diffs landed in March 2026. His tool of choice wasn't Meta AI. It was Claude Code CLI, the terminal-based coding assistant built by Anthropic, one of Meta's direct competitors in the AI race.
Let that sink in. The CEO of the company building Llama, the company that just froze its Mercor training pipeline over a security breach, the company spending billions on AI infrastructure, is using a rival's product to write his code. If that doesn't tell you where the AI coding wars stand, nothing will.
The Pragmatic Engineer's Gergely Orosz broke the story this week. One of Zuckerberg's diffs got more than 200 approvals, mostly from engineers who wanted bragging rights for reviewing the CEO's code. Meta's monorepo now holds close to 100 million diffs total. Zuckerberg's last meaningful contribution was in 2006, when the entire Facebook codebase had fewer than 10,000.
He's not alone. Garry Tan, CEO of Y Combinator, returned to coding after 15 years and open-sourced gstack, a Claude Code system with 23 specialist tools. Shopify CEO Tobias Lutke has been running overnight experiments with Karpathy's AutoResearch on internal data. The pattern is clear: AI coding tools are pulling founders back into the codebase.
This is a bigger deal than the anecdotes suggest. Leaked Meta documents from March show the company wants 65% of engineers writing 75% or more of their committed code using AI by mid-2026. The Scalable ML org set targets of 50-80% AI-assisted code. Zuckerberg told Dwarkesh Patel that "in the next year, maybe half the development will be done by AI."
Here's what everyone is missing: the CEO of Meta is dogfooding a competitor's product because it's better than what his own company has built. That's not a PR stunt. That's a product verdict. Claude Code is winning the developer war, and Meta's own leadership knows it.
The irony gets sharper. Anthropic just cut off Claude subscriptions for third-party tools like OpenClaw, effectively monetizing the exact demand Zuckerberg is demonstrating. The company's capacity problems are a luxury problem: too many people want to use Claude Code.
Twenty years ago, Zuckerberg force-merged a diff past a senior engineer who blocked it for formatting issues. In 2026, he's using AI to write code good enough that engineers fight to approve it. The tools changed. The ambition didn't. But the fact that Meta's founder trusts Anthropic's tool over his own company's tech? That's the story.