Anthropic, the company behind the AI assistant Claude, says that in September Claude generated more than 90% of the code contributions inside the company. The figures come from Anthropic's own internal data, as reported by Brief IA, and have not been checked by an outside party.
The numbers
- Claude wrote over 90% of code contributions in September.
- Daily output per engineer in Q3 was over 45 times the 2024 daily average.
- Staff grew from 940 to 5,300 in under two years, more than fivefold.
- Regrettable turnover was about 1% in the 12 months ending September 30.
Regrettable turnover means people leaving whom the company would have preferred to keep. Anthropic says it uses Claude to grow without hiring in step with its workload. It also says the workforce could have grown even more without Claude.
How the work is changing
Most Anthropic employees work in research, engineering or product development. In software engineering, they increasingly oversee and check what Claude produces instead of typing code line by line. The company says the tool lets it finish tasks that once would have needed much larger teams.
The risk Anthropic names
The company also admits it leans heavily on senior leaders and key technical staff. It warns that if they left, operations could be disrupted and new work could slow down. Retention could get harder if their beliefs change or another organization seems a better fit.
One example from the wider industry: researcher Jacob Coxon left Anthropic last month. He believed Anthropic and OpenAI were moving too fast toward more powerful AI without a solid plan to control it.
Culture under growth
Anthropic describes a low-ego culture, where the company's interest comes before status, credit or seniority. It says this shortens the path from a research idea to a product. Organizational psychologist Jamie Shapiro finds the approach sound, since good ideas can come from anyone. She adds that hiring, recognition and accountability get harder to manage as a company grows.
Why it matters
For business owners, the lesson is not about writing code. It is about how much routine work a small team might hand to AI while people review the results. A fair first question is which tasks in your own business could be handled that way.
The takeawayAnthropic reports that AI now writes most of its code, so it is worth asking which of your own tasks an AI tool could draft while your people check the work.