Anthropic disclosed that Claude authored more than 80% of code merged into its production systems as of May 2026, up from low single digits before its in-house coding agent launched in February 2025 — an early sign of recursive self-improvement in AI development.
"Claude-written code was somewhat worse than human-written code at Anthropic in late 2025, is roughly at parity today, and we expect it to be strictly better within the year," according to an excerpt from the report published by the Anthropic Institute.
The productivity gain is steep. Anthropic's typical engineer merged eight times as much code per day in the second quarter of 2026 as in 2024, with the human now directing and reviewing while Claude does the writing. On the company's internal benchmark — asking AI to optimize code that trains a small model — Claude Opus 4 averaged a 3x speedup in May 2025. By April 2026, its Mythos Preview model reached 52x. A skilled human needs four to eight hours to achieve a 4x improvement.
The findings carry implications beyond Anthropic. If the trend continues, AI systems could design and build their own successors, compressing development timelines in ways the industry has not yet priced in. Anthropic, which recently filed a confidential IPO registration with a $965 billion private valuation and a $47 billion revenue run rate, has built its brand around safety — and faster development raises the stakes for both commercial opportunity and risk management.
The company runs the same test on every model. Shown a session before a researcher took a wrong turn, Mythos Preview picked a better next step 64% of the time, up from 51% for Opus 4.5 in November 2025. Research judgment — the ability to choose which problems matter most — remains harder to automate, and Anthropic cautioned that Claude has not yet shown that taste.
Claude Code Drives Enterprise Revenue
The coding capability has become Anthropic's most visible commercial product. Claude Code, its software development platform, generated $2.5 billion in annualized revenue, and business subscriptions have quadrupled since the start of the year. Enterprise users account for more than half of that revenue, CEO Dario Amodei told CNBC. Overall, about 80% of Anthropic's business comes from enterprise customers.
The company's revenue run rate reached $47 billion in early May, up from $30 billion in April and $10 billion a year earlier. Anthropic recently raised $65 billion at a $965 billion valuation, surpassing OpenAI's reported $852 billion valuation from March.
Security Breakthroughs Add Another Revenue Line
Anthropic's Mythos Preview frontier model identified thousands of zero-day vulnerabilities, including some in every major operating system and web browser. That prompted Project Glasswing, a partnership with Nvidia, Amazon Web Services, Apple, Google, Broadcom, Microsoft, Cisco, CrowdStrike and Palo Alto Networks to plug the security holes.
The cybersecurity application opens a new commercial channel for Anthropic beyond coding and general-purpose chatbot usage. It also positions the company as a critical infrastructure provider rather than just a consumer AI brand — a distinction that could matter when investors evaluate its IPO.
What It Means for Investors
Anthropic's disclosure that Claude is accelerating AI development challenges the assumption that human researchers will remain the bottleneck in frontier model progress. If the curve continues bending, the cost and time required to build each successive generation of AI could shrink dramatically.
For investors, the key question is whether this trend benefits Anthropic's private valuation or creates risk for incumbent cloud providers that rely on GPU sales. Nvidia, whose H100 and B200 GPUs power much of the current AI training infrastructure, could face demand shifts if models become efficient enough to require less compute per capability gain. Anthropic's IPO, expected as soon as next month, will give public market investors their first chance to price these dynamics directly.
This article is for informational purposes only and does not constitute investment advice.