Anthropic released research findings showing that the rapid adoption of internal AI agents significantly boosted productivity and altered hiring structures (reducing entry-level technical roles). Employees, through firsthand experience, are more optimistic than the general public about the pace of AI capability advancements, but their views on the economic impact mechanisms are similar to those of the public.Author and source: Peter McCrory
Today, we are releasing two new research findings from within Anthropic:
First, explore how the work methods and hiring practices of a company at the forefront of AI applications are evolving.
Second, how Anthropic employees expect AI to transform the overall economy.
Our research found that both the general public and Anthropic employees expect AI capabilities to rapidly improve in increasingly complex knowledge-based tasks over the coming years.
However, Anthropic employees have higher expectations for the speed and scope of progress.
Expectations regarding progress and awareness of broader economic impacts are influenced by real-world experience.
At Anthropic, AI agents have rapidly gained adoption within the company this year: by August, the total usage time of interactive AI agents had reached six times the amount of employee work hours.
The widespread adoption of AI agents has coincided with an overall increase in company productivity: more code written, higher sales per sales representative, and increased volume of legal contracts processed.
Meanwhile, Anthropic has reduced the proportion of entry-level technical roles in its hiring and is increasingly emphasizing autonomy and judgment in decision-making, while de-emphasizing requirements for technical expertise and specific implementation skills.
We also found evidence that Anthropic employees hold views similar to the general public on how capabilities shape broader economic outcomes: those who expect AI to take over more of their own work also have similar estimates of its impact on unemployment.
Overall, there is significant disagreement about the future progress of AI and its economic impact. This observation has prompted us to develop a structured framework for comparing different future scenarios:
We are releasing this research because Anthropic is approximately one and a half years ahead of the broader economy in AI applications, making internal developments at the company a potential leading indicator for the wider world.
We also hope this research will provide policymakers, researchers, and workers with the time to prepare for future economic impacts, take proactive measures, and ensure that everyone benefits from AI.