Anthropic, the developer of Claude chatbot, said that the personnel structure and working style at this startup have changed significantly after applying AI.

In a report released over the weekend, Anthropic said that trends at the company could signal "how AI will change the way work is organized, recruitment and labor needs across the economy", especially when the company completely shapes its work process around AI.

Specifically, Anthropic has reduced the number of low-level, junior employees. From the beginning of 2025, the proportion of low-level employees in the technical department will decrease from 21% to 14%. The recruitment rate of this group even dropped sharply from 27% to 9%. Technical staff is also currently reduced from 61% to 56%.

Anthropic attributes the trend to the natural expansion of business functions as companies grow in size, and programming tools help non-technical employees create their own software.

In addition, the report shows that the average meeting time per employee increased by 20% compared to last year, and the group of technical experts increased by 16%. The company explains this is due to the importance of "trust and responsibility between people", and employees also repeatedly emphasize that "face-to-face coordination" is indispensable.

According to the report, AI applications help Anthropic operate more efficiently, as shown by "the increase in codes per person, number of orders per salesperson, and number of contracts processed". While admitting more code doesn't necessarily mean better code or greater business impact, the company hasn't seen any signs of a decline in code quality.

The report also quoted Anthropic employees as saying there was constant pressure to "work as fast as possible and use Claude for everything". In the April survey, 27% of employees said they became more productive thanks to Claude, and 20% said they "moved from actually working to managing Claude models."

"When you babysit, each child brings you problems with equal demands for attention, and Claudes are the same," compares an engineer. However, human attention and judgment are becoming a bottleneck to productivity, as Claude generates huge amounts of text that needs to be sorted.

In investor Patrick O'Shaughnessy's Invest Like the Best podcast that aired in May, Anthropic Chief Financial Officer Krishna Rao said Claude changed the way office workers work every day. Within companies, AI is increasingly performing intellectual work, from software engineering to financial reporting. Meanwhile, humans move into the role of monitoring, evaluating and strategizing.

According to Rao, more than 90% of Anthropic's code is written by Claude Code, the AI ​​programming tool developed by the company. The finance department is undergoing a similar transformation using Claude for reporting. The company's monthly financial review process is even 90-95% complete before humans evaluate and interpret the results. Internal reports that used to take hours to complete now only take 30 minutes.

However, actively applying AI also brings some problems. "What's noticeable in the Claude Code team is that after a while, the work starts to feel lonely because everyone interacts so much with agents," Fiona Fung, who runs the AI ​​tool development team Claude Code and Claude Cowork, said on Lenny's Podcast at the end of June.

As a solution, Fung's team tries to restore some of what is lost due to programming habits with AI agents. They organize programming lunches, hackathons and "shared creations" for engineers to sit close together and learn work processes, increasing communication within the team.

Thu Thao summary