Claude Code now accounts for 4% of all public GitHub commits, according to SemiAnalysis analyst Dylan Patel, who projects that figure will exceed 20% by the end of 2026. The statistic appeared in blogger TheZvi's fifth "Claude Code, Claude Cowork and Codex" roundup, published March 9, 2026. New York Times tech columnist Kevin Roose compared it to early pandemic case counts — a small number on a steep curve.
On revenue, SemiAnalysis analyst Doug O'Laughlin reported that Anthropic's quarterly ARR additions have now overtaken OpenAI's. The constraint on Anthropic's growth, per O'Laughlin, is compute availability, not demand. TheZvi argues Claude Code represents something more than an incremental LLM improvement: in his framing, it operates at a distinct "agentic layer" where model outputs are organized into coherent software development workflows rather than delivered as one-off responses.
Anthropic has shipped a dense set of new features. Recent Claude Code and Claude Cowork updates include Agent Teams for multi-agent coordination, Fast Mode for the Opus 4.6 model, scheduled task support, <a href="/news/2026-03-14-direnv-git-worktrees-parallel-agents">git worktree integration</a>, and new slash commands — /simplify and /batch among them. Windows support has arrived for Claude Cowork.
The hackathon results TheZvi highlights make a specific point about who benefits. Winners included a personal injury attorney, a cardiologist, an electronic musician, and an infrastructure worker. One software engineer placed among them. <a href="/news/2026-03-15-developer-builds-cutlet-language-with-claude-code-without-reading-code">The tools are letting people outside traditional engineering roles ship working software</a> — and, at hackathon speed, beat engineers doing it.
The piece is not uniformly bullish. TheZvi flags concrete risks: malware injection into agent-generated code, accidental infrastructure wipeouts via tools like Terraform, and API credential misuse. These are operational hazards that come with delegating more execution authority to coding agents, and TheZvi treats them as current problems rather than hypotheticals.
A separate section covers TheZvi's dispute with what he calls the "Department of War" — his rhetorical label for the U.S. Department of Defense — over Anthropic's designation as a supply chain risk. He shelved the dispute after what he described as a productive exchange with Undersecretary Emil Michael.
Anthropic and OpenAI are running hard at the same target. The commit share data, the ARR trajectory, and the product velocity are all pointing the same direction: the company that owns agentic coding owns the next layer of software development.