Editorial illustration for Anthropic says Claude writes >90% of its code; AI pause button urged

Anthropic says Claude writes >90% of its code; AI pause...

Anthropic says Claude writes >90% of its code; AI pause button urged

By AI Daily Post Edited by Brian Petersen, Editor-in-Chief

June 5, 2026 • 2 min read

Anthropic says its flagship model, Claude, now writes more than 90 percent of the company’s production code—a stark shift from the way the startup has traditionally built software. While the headline number grabs attention, the firm also notes that Claude is handling a growing share of research tasks and is “closing in on human‑level judgment” in certain domains. Engineers in the second quarter of 2026 are reportedly shipping an average of eight times as much code per day as they did in 2024, a productivity jump that the company attributes largely to the model’s output.

Yet the surge raises questions. Anthropic’s internal report, released through its Anthropic Institute, warns that fully autonomous AI self‑improvement—recursive design of successor systems—has not yet materialised, but “could come sooner than most institutions are prepared for.” In response, the company is lobbying for a verifiable, global development pause, arguing that a unilateral halt by any single lab would be insufficient. The tension between rapid capability gains and calls for a coordinated slowdown defines the current debate.

Anthropic says Claude now writes over 90% of its code and wants the world to have an AI pause button Key Points - Anthropic has shared internal data showing that Claude now writes more than 80 percent of the company's production code, a major shift in how Anthropic builds its own software.

— Anthropic, Anthropic says Claude now writes over 90% of its code and wants the world to have an AI pause button - THE DECODER

Why this matters

We see Claude now generating the bulk of Anthropic’s production code—over 80 percent according to internal data—signaling a tangible shift in how the company builds its software. Yet the claimed eight‑fold productivity jump is likely exaggerated; a March 2026 internal survey of 130 staff members put the median boost at four times, and Anthropic itself admits the true figure may sit lower. Claude’s growing competence in research tasks and its approach to human‑level judgment further blur the line between tool and collaborator, prompting us to question how much of the engineering pipeline can safely be delegated to an LLM.

At the same time, Anthropic’s call for an AI “pause button” underscores lingering concerns about speed outpacing oversight. For developers, this raises practical questions about code quality, maintainability, and reliance on AI‑generated artifacts. Founders must weigh short‑term efficiency gains against long‑term governance challenges.

Researchers are left with an open question: does Claude’s progress represent a sustainable advancement, or a temporary spike that will plateau once the novelty fades? The answers remain unclear.