Anthropic’s engineers are not yet being replaced by AI that can code.
Anthropic CEO Dario Amodei told Salesforce CEO Marc Benioff on Wednesday at the annual Dreamforce conference that, while Claude AI is now creating 90% of the code for most teams at the firm, humans are still needed.
According to Amodei, he predicted that 90% of code will be created by AI models within six months. Some people believe the forecast is incorrect, yet it is now totally accurate inside Anthropic and a number of other organizations with whom they operate.
Benioff then inquired about whether that proportion would climb and whether Anthropic would now require fewer engineers.
Amodei stated that people shouldn’t “misinterpret” Claude’s capacity to build features and assist with solving of long-running bugs.
If Claude is writing 90% of the code, you will often want just as many software engineers. You could need more, because they can provide additional leverage, according to Amodei.
They can concentrate on the 10% that is revising the code, writing the 10% that is the most difficult, or directing a team of AI models. So what happens is, you know, you wind up being ten times more productive, Amodei explained.
Amodei said that it was about “rebalancing” rather than replacing.
Anthropic teams aren’t the only ones coding with AI. In March, Garry Tan, the president and CEO of the startup accelerator Y Combinator, stated in an X post that almost a quarter of the entrepreneurs in the company’s 2025 winter batch are using AI to generate up to 95% of their code.
According to a new Stanford research, entry-level software engineering positions are already being impacted by the development of AI coding tools, which may discourage young people from pursuing careers in the industry and cause a talent pipeline breakdown.
According to the Stanford researchers, employment for developers aged 22 to 25 had decreased by over 20% as of July 2025 compared to its peak in late 2022, when ChatGPT was introduced in November 2022.
However, more seasoned workers are far less vulnerable to the effects of AI coding tools.
