Anthropic: Claude Wrote 80% of May Code

Anthropic revealed on June 4, 2026, that Claude authored more than 80% of the code merged into Anthropic’s own codebase in May 2026, marking a dramatic acceleration in AI-assisted development that has raised concerns about the path toward recursive self-improvement. In a report titled “When AI Builds Itself,” Anthropic argued the world needs a verifiable, multi-country option to slow frontier AI before recursive self-improvement stops being theoretical. The disclosure comes as the company prepares for a potential public listing and as its revenue trajectory reaches unprecedented levels.

AGI announcement presentation with speaker at podium, displaying artificial intelligence concept with neural network visua...

The Speed of AI-Driven Code Production

Before Claude Code launched in research preview in February 2025, the share was in the “low single digits”, demonstrating how rapidly the technology has evolved. The average engineer in the second quarter of 2026 was merging 8 times as much code per day as in 2024, though Anthropic acknowledges this metric has limitations.

The report flatly states that lines of code are an imperfect measure, counting quantity rather than quality, so the 8× figure “is almost certainly an overstatement of the true productivity gain”. When surveyed internally, the median estimate from 130 employees in March 2026 pegged the output boost from Mythos Preview at 4x.

The human element has shifted dramatically. One employee stated, “it’s now been ~5 months since I last wrote any code myself”, illustrating how roles have transformed from writing code to directing AI systems. Leadership estimates the total share, including scripts and experimental code, at more than 90 percent.

Technical Capabilities and Complex Problem-Solving

Claude’s performance on challenging engineering tasks has improved substantially. On highly complex, open-ended engineering problems where clear specifications are initially absent, Claude’s success rate climbed to 76% in May 2026, a 50-point increase in a six-month window.

In practical applications, the results have been significant. Claude delivered more than 800 fixes in April 2026 that cut a class of API errors by a factor of 1,000, work that would have needed four years for a human according to the supervising engineer. Anthropic’s internal Mythos Preview model achieved a 52x speedup on AI model training code optimization, while a skilled human developer typically requires four to eight hours of manual refactoring to achieve a mere 4x speedup on the exact same codebase.

Where Humans Still Hold the Edge

Despite these advances, Anthropic emphasizes that human judgment remains critical. The company stated, “The comparative advantage of humans as of right now is still in seeing the bigger picture and thinking beyond the confines of the immediate task”.

Claude Mythos Preview suggested the better next step in 64 percent of cases where human developers took a suboptimal detour, up from 51 percent for Claude Opus 4.5 six months earlier. However, the critical bottleneck is what Anthropic calls “research taste,” the ability to pick the right problems and spot dead ends early.

The Call for Coordinated Action

Researchers Marina Favaro and Jack Clark wrote that a worldwide frontier slowdown “would likely be a good thing,” but only if US and Chinese labs (and others near the frontier) stop together under rules outsiders can verify. This represents a departure from previous unilateral pause proposals.

Favaro and Clark wrote that “a meaningful slowdown or pause would require multiple well-resourced labs at or near the frontier, in multiple countries, agreeing to stop under the same conditions”. A unilateral pause by a single company would be easier to implement but would have limited impact, primarily shifting leadership rather than fostering broader global deliberation.

The company said, “We are not there yet, and recursive self-improvement is not inevitable,” but warned that this shift could come faster than many institutions are ready for.

Business Context and Industry Reaction

The announcement arrives during a period of rapid growth for Anthropic. Anthropic’s run rate is on track to reach $50 billion in annualized revenue by the end of this month, up from $9 billion at the end of 2025. The governance pitch was timed one week after Anthropic confidentially filed for IPO.

The proposal has drawn skepticism from some quarters. David Sacks, a venture capital investor and informal adviser to President Trump, has accused Anthropic’s leaders of running a “regulatory capture agenda”, suggesting concerns about potential impacts on open-source AI models. Others have suggested that Anthropic’s warnings about the dangerous potential of its own tools could also be a marketing move.

Broader Industry Implications

The company describes the current phase as “Autonomous Agents” where agents execute code independently, debug live environments, and delegate multi-hour work streams to specialized sub-agents. This represents a shift from earlier stages of manual writing and chatbot assistance.

Enterprise leaders aiming to match Anthropic’s technical velocity cannot afford to ignore psychological dynamics, as achieving an 80 percent automated codebase requires more than purchasing API tokens, it demands a total cultural overhaul and the implementation of rigorous, automated verification guardrails.

Key Facts

Claude authored more than 80% of code merged into Anthropic’s codebase in May 2026
Average engineers in Q2 2026 merge 8 times as much code per day as in 2024
Claude’s success rate on open-ended engineering problems reached 76% in May 2026, a 50-point jump in six months
Before February 2025, Claude Code contributions were in the “low single digits”
Anthropic’s revenue run rate is on track to reach $50 billion by end of June 2026, up from $9 billion at end of 2025
Report published June 4, 2026 by Marina Favaro and Jack Clark