The creator of the Claude assistant, Anthropic, has drastically reduced the proportion of junior profiles among its new technical hires. An internal study published by the company itself confirms that this figure fell from 27% to 9% in just over a year, a fact that reopens the debate on the future of entry-level jobs in the technology sector.
The reason behind this transformation has a name: the company's artificial intelligence agents are already taking on a workload that is six times that of human employees in some areas. The news is relevant because Anthropic is one of the most influential companies in the development of generative AI, and its internal practices often anticipate trends that are later discussed across the rest of the tech industry.
From 27% to 9%: the drop in junior hiring
According to the Anthropic report, which has monitored the presence of junior profiles in its technical teams since the beginning of 2025, the overall proportion of these workers within the staff dropped from 21% to 14% during that period.
The most severe impact is observed in new hires. While in 2025, 27% of technical hires were junior profiles, that figure plummeted to 9%. Vacancies specifically opened for this type of candidate came to represent only 6% of the total.
The company notes that this decline intensified especially starting in December 2025. In contrast, Anthropic's non-technical departments maintained a stable situation, confirming that the adjustment is concentrated in areas where Claude plays a more direct role.
Agents working six times more hours
The study attributes this change to the growing operational capacity of the company's artificial intelligence agents. Since March 2026, these systems have accumulated more effective working hours than Anthropic's own human employees.
In August 2026, the proportion reached a 6-to-1 ratio across the company. In research departments, where automation is more intense, that figure soared to 19-to-1.
Tasks traditionally associated with entry-level profiles, such as writing code or locating bugs, are increasingly being delegated to Claude. The report itself identifies a clear pattern: the reduction in junior hiring was most pronounced precisely in positions where AI agents resolve the most complex work.
An uneven productivity leap by area
The Anthropic study documents notable productivity increases in different departments. Software engineers sent 5.1 times more code commits than before. Data scientists executed 3.1 times more queries, and the legal team processed 3.1 times more contracts.
However, the increase is much more modest outside of strictly technical areas. The sales department closed 1.7 times more deals, recruiting hired 1.4 times more people, and accounting managed 1.4 times more invoices.
This contrast illustrates the underlying mechanism: tasks that previously served as training for a newly hired engineer, such as debugging an error or writing a delimited function, are now executed directly by an agent. The profile that traditionally learned by performing that work now finds fewer entry opportunities.
The caveats included in the study itself
The Anthropic report explicitly acknowledges that the company is a unique case. It is a software company with extremely intensive use of artificial intelligence, whose results are not directly extrapolatable to the economy as a whole.
Furthermore, the rapid growth of the business and the voluntary adoption of these tools by employees could be distorting the figures. For the moment, the economy in general does not reflect measurable productivity improvements that can be clearly attributed to artificial intelligence.
The warning about the “tragedy of the cognitive commons”
Researcher Nolan Lovett, affiliated with the NATO Special Operations University, has described in an academic paper what he calls a "tragedy of the cognitive commons." According to his approach, when a company replaces entry-level positions with AI systems, it keeps all the immediate efficiency gains, while the future cost is shared among those who will need experienced talent in the future.
Lovett argues that professionals only acquire real experience after years of practice in those same junior positions that are now beginning to disappear. For this reason, he places the full impact of this phenomenon between 2030 and 2045, as the experts currently available were trained between 5 and 20 years ago.
In his analysis, the fields most exposed to this risk are software engineering, financial analysis, and legal research. His forecast, however, depends on people continuing to perform that type of work in the near future. Anthropic has previously suggested that programmers and call center workers could be forced to migrate to trades such as electrician or nurse.
What this means for those starting their career
The data published by Anthropic offers some useful guidelines, although they should not be interpreted as a guarantee applicable to the entire labor market.
Those in technical training should compare job offers based on the type of tasks they are actually assigned. Positions where the most complex work already falls to an AI agent are precisely those that have cut the most junior spots.
For those looking for work in the sector, it is advisable not to interpret Anthropic's situation as an exact reflection of the entire tech industry, as the study itself warns that its results are not extrapolatable without nuance.
Those who hire or manage teams should consider who will occupy senior positions in five to twenty years, given that Lovett directly links the training of future experts to the continued existence of entry-level positions in organizations.
A first job with fewer classic learning tasks
For those starting their career in programming, financial analysis, or legal research, the first job may include fewer traditional learning tasks, as an AI agent may be executing them. It is advisable to identify which part of the daily work involves supervision and independent judgment, versus that which consists of repeatable and automatable execution.
For those hiring, the 9% figure marks how far a particularly AI-intensive company has gone, but it does not necessarily imply that the rest of the market will follow the same path or at the same pace.
Frequently Asked Questions
How much has junior hiring fallen at Anthropic?
According to the company's internal study, the proportion of junior profiles among new technical hires went from 27% to 9%, with specific vacancies for this profile falling to 6%.
How many hours do Anthropic’s AI agents work compared to humans?
Since March 2026, AI agents have accumulated more working hours than human employees. In August 2026, the ratio reached 6-to-1 company-wide, and up to 19-to-1 in research departments.
Is Anthropic’s situation extrapolatable to the rest of the labor market?
Not necessarily. The company's own study warns that it is a special case, as it is a software company with extremely intensive use of artificial intelligence, so its results do not automatically reflect what is happening in the economy as a whole.
What is the “tragedy of the cognitive commons” that Nolan Lovett mentions?
It is an academic concept that describes how companies that replace entry-level positions with AI benefit from immediate efficiency gains, while the cost of the lack of experienced talent is shifted to the future, affecting those who will need trained senior professionals several years from now.
Which professions are most exposed according to this analysis?
According to researcher Nolan Lovett, software engineering, financial analysis, and legal research are among the fields most exposed to this phenomenon of entry-level position reduction.
When is the full impact of this trend expected to be felt?
Lovett places the full impact between 2030 and 2045, as current experts were trained 5 to 20 years ago in entry-level positions that are now beginning to disappear in AI-intensive companies.
