Anthropic's IPO filing says its Claude AI wrote over 90% of its new code in September. Here is what that means for businesses, and what it does not prove.
Anthropic, the company behind the AI assistant Claude, says Claude wrote more than 90% of the new code added to its software in September. The number comes from the company's IPO filing, the paperwork a company submits when it plans to offer shares to the public.
Anthropic staff used Claude more than 250 times as much in September as they did in January. In April, the value of that internal use passed the company's monthly pay bill, not counting stock-based awards. In short, the AI is now woven into daily work at the company that makes it.
The comparison with payroll can mislead. Anthropic valued its internal use at the prices it charges outside customers. That is not what it actually costs the company to run the AI for its own staff. So the filing does not show that AI is a bigger cash expense than salaries.
The filing suggests a company can get a lot more done without hiring at the same pace. The article notes that Anthropic's revenue has grown much faster than its headcount.
It also points to a new way to think about pay. AI use is measured in tokens, which are small units of text that an AI model processes. Tracking tokens shows how much employees rely on AI. Some companies may start treating access to AI tools as part of a job's value, alongside salary, bonuses and stock.
A high share of AI-written code does not mean engineers are no longer needed. Software still needs design decisions, security reviews, debugging, testing and someone who is accountable. AI-written code can contain errors and security flaws, so its value depends partly on how well people check it.
Costs also shift rather than vanish. More AI use calls for more computing power, specialized chips, electricity and data centers. A company may spend less on some kinds of human work and more on the systems that run the AI.
Anthropic sells the same technology it is using, so its own heavy use doubles as a showcase. The filing shows that AI is changing how a leading AI developer organizes its workforce and builds software. It does not show that machines have replaced human labor.
AI can now do a large share of routine work, but human review for errors and security flaws still matters.
World models, AI that learns physics from video, are being built into home robots and driverless cars, and your own videos may help train them.
Read more: AI that learns how the physical world works is coming to robots and carsA new Alibaba-led test of real coding requests found the best AI agent fully finished just 26.4 percent of tasks, so human review still matters.
Read more: AI coding helpers fully finish only about 1 in 4 real tasksA University of Chicago law professor asked Claude to fix page numbers. It deleted his syllabus and a year of notes instead. Dropbox saved him, not the AI.
Read more: An AI assistant deleted a law professor's year of lecture notes
