Anthropic's internal development teams now use Claude AI to write nearly all code for the company's products, with company officials describing it as effectively 100 percent of new code. Mike Krieger, Anthropic's Labs chief, revealed the shift during the Cisco AI Summit earlier this month, stating:
"Right now for most products at Anthropic it's effectively 100 percent just Claude writing."
He added that the company has created "all the right scaffolds around it to let us trust it."
The development follows CEO Dario Amodei's prediction last March that AI would write 90 percent of code within three to six months. Boris Cherny, head of Anthropic's Claude Code division, hasn't written any code manually in more than two months. In a January social media post, Cherny stated that all his code now comes from Claude Code and Opus 4.5 models, writing: "I shipped 22 pull requests yesterday and 27 the day before, each one 100 percent written by Claude."
An internal Anthropic survey from August 2025 shows engineers and researchers use Claude in 60 percent of their work, achieving a 50 percent productivity boost. More than half of employees use the AI for debugging daily, while 27 percent of Claude-assisted work involves tasks that wouldn't have been done otherwise.
The AI now serves as what Krieger calls a "super tough grader" for developers.
"I'll put up pull requests and Claude will come back and say 'here are all the security vulnerabilities' and 'here's also how it could be refactored,'" he explained.
The company's latest Claude Opus 4.6 model demonstrates advanced security capabilities, finding more than 500 previously unknown high-severity security flaws in open-source libraries with minimal prompting. Each vulnerability was validated by either Anthropic team members or outside security researchers. This follows demonstrations of Claude's ability to autonomously build complex software projects like a complete C compiler.
Anthropic's internal transformation coincides with broader enterprise adoption. Goldman Sachs has worked with embedded Anthropic engineers for six months to develop autonomous agents for accounting and client vetting. The U.S. General Services Administration struck a OneGov deal offering Claude access to all three government branches for $1 per agency.
Syracuse University became one of the first U.S. institutions to provide campus-wide Claude access in September 2025. Cognizant plans to deploy the AI to up to 350,000 associates across corporate functions and engineering teams.
The company's Claude Cowork agent, launched in January, triggered significant market reactions. Legal software stocks including Thomson Reuters and Wolters Kluwer dropped sharply after Anthropic unveiled legal plugins. The tool can consolidate lawsuit documents into chronological exhibits and automate tasks across legal, sales, and marketing functions.
Anthropic recently introduced a fast mode for Claude Code that processes requests 2.5 times faster than standard versions. Powered by Claude Opus 4.6, the feature targets high-intensity development tasks like debugging and codebase reviews under tight deadlines.
Despite the rapid internal adoption, external software companies report lower AI coding percentages. Microsoft CEO Satya Nadella said in April 2025 that AI generated about 30 percent of code at the software giant. Industry counterparts like Google have increased AI-written code volumes but remain below Anthropic's near-total adoption rate.
The shift represents what Krieger describes as "Claude writing Claude," with the company's AI systems now developing their own successors.
"We're moving extremely quickly," he told Cisco's Jeetu Patel during their February summit discussion.
