Anthropic says Claude writes 90% of the company's code

Anthropic has disclosed that its Claude AI writes more than 90% of the code used inside the company. CFO Krishna Rao made the statement, stressing how deeply the AI assistant is embedded in every stage of development. It is not just a number: it signals how a technology leader is restructuring its processes around artificial intelligence.

A record level of integration

The figure means Claude is no longer an auxiliary tool but the company's main code author. This is not about one-off helpers like autocomplete: the AI writes functions, tests and documentation, refactors existing code and suggests architecture improvements. Anthropic's developers work in code review and architecture mode rather than line-by-line coding.

The shift happened gradually, after years of experimentation and optimization. First, Claude helped with simple tasks; then it was taught the company's style through examples; then it was built into CI/CD so that the AI generates code automatically from tickets. Today a developer creates a ticket, outlines the requirement in two sentences, and Claude produces a first draft of the PR. A human reviews it, fixes the things only people notice (context, edge cases, performance) and merges. The process is 2-3 times faster.

Big tech has joined the race

Google, Meta* and Microsoft are also bringing AI into development, but they rarely disclose real metrics. Anthropic was the first to name a specific percentage, 90%, which automatically made it the leader in an undeclared automation contest. Other companies are either behind or afraid to reveal their numbers.

For investors, this is a signal: Anthropic is not just talking about AI in programming but has deployed it at a scale that seemed unattainable a year ago. There is an added bonus in PR and talent recruitment. Developers prefer companies where they can learn from AI rather than compete with it.

The advantages of this approach are clear:

• Iteration speed grows: less time on mechanics, more on solving hard problems
• Code review focuses on logic and architecture, not syntax
• New developers get up to speed faster, because Claude writes in the company's style
• Fewer bugs from copy-paste and inattention: the AI is consistent

What this means

AI programming has moved beyond the experimental stage and begun to reshape the industry. For developers, this means retraining: from writing syntax to becoming a systems architect, API designer and code reviewer. The developer of the future is not the one who writes more lines, but the one who can define a task, verify a solution and notice when the AI has missed something.

For Anthropic, this reliance on Claude also creates a risk, but the company considers it a price worth paying for speed and an advantage in the current moment.

*Meta is designated an extremist organization and banned in Russia.