At Anthropic, AI agents slow down junior hiring

An internal report from Anthropic describes a clear shift: AI agents now work more than humans and the share of junior positions is declining. The company is also questioning negotiation rules and how to distribute gains.

Anthropic sees its AI agents take a central role in daily work, to the point of exceeding human hours internally, while junior hiring contracts sharply in technical teams. At the same time, the company consults law firms to define “fair” negotiation standards before agents start bargaining prices, handling claims, or improving contracts. This dual evolution matters because it shows both how AI is settling into organizations and who risks paying the price long term. A research report published on October 16 and 18, 2026 by Anthropic and relayed by the Financial Times, then CNBC, provides a very concrete insight.

Les agents font plus d’heures que les humains, et ça change la hiérarchie du travail

According to Anthropic’s internal report, agent working hours surpassed those of employees in March 2026, then reached a ratio of 6 to 1 in August 2026 across the company. In research teams, this ratio even rises to 19 to 1. In other words, the machine no longer just “supports” human work: it partly engulfs it, forcing employees to spend time monitoring, managing, and correcting these agents.

The phenomenon is not just theoretical. Engineers produced 5.1 times more code commits, data scientists 3.1 times more queries, and the legal team processed 3.1 times more contracts. On the commercial side, sales increased by 1.7 times. The report specifies, however, that Anthropic is a special case, a company highly exposed to AI, and these results do not mechanically translate to the rest of the economy.

Pourquoi les postes juniors reculent-ils autant ?

The answer partly lies in work reorganization. In technical departments, the share of junior positions dropped from 21% to 14% since early 2025, while the share of new junior hires fell from 27% to 9%. Junior job offers declined mainly from December 2025. Anthropic explains this decline by the increased capacity of its agents, able to handle more code, bug fixes, and maintenance tasks.

The report also highlights a shift in sought skills. Hiring managers emphasize autonomy and judgment more, less on narrow technical expertise. Among technical candidates, mention of creative use of AI tools tripled in evaluations between 2024 and 2026. For non-technical profiles, it quadrupled. The message is clear: knowing how to use the tool now counts almost as much as knowing how to do the work manually.

Quand l’IA monte en puissance, la négociation aussi

Anthropic does not stop at internal tasks. According to the Financial Times, the company consults law firms to establish fair negotiation standards before AI agents start negotiating on users’ behalf. These systems could bargain prices, handle complaints, or improve contractual offers. The issue is significant: the report recalls that in experiments, more powerful models achieve better deals, which could widen gaps between savvy users and others.

The company is already recruiting Conceptual Reasoning Fellows, including lawyers, for four months of testing on law, philosophy, cognitive science, economics, and political science questions. The announced salary ranges from $100,000 to $167,000. This research relates to a very concrete problem: in fields where the right answer is not always clear, like law, evaluation cases and quality criteria must be created. Justin Curl, a member of Anthropic’s technical team, sums up the puzzle in a simple phrase: “notoriously hard to define quality.”

La productivité grimpe, mais le goulot d’étranglement devient humain

The internal report describes a less glamorous side effect than productivity curves: the human becomes the bottleneck. Agents produce so much content that, in some non-technical departments, reading their outputs would take more than half the workweek. As a result, employees spend more time doing “meta-work”: writing instructions, supervising pipelines, reviewing results, and building tools to manage other tools. It’s a bit like management, but with colleagues who never sleep and demand validation every three minutes.

Anthropic also says it has implemented an early warning system to identify jobs threatened by AI and promote internal redeployment. The report supports the idea that a company can share productivity gains, provided redistribution and mobility mechanisms are planned. CNBC reports that Anthropic wants to reserve 5% of its shares for a new Public Benefit Fund, subject to shareholder approval, with a possible 2-point increase for every 50% rise in company value above an undisclosed baseline.

By the numbers

  • 6 to 1 — agent hours / human hours ratio at Anthropic, August 2026.
  • 21% to 14% — share of technical juniors since early 2025.
  • 27% to 9% — share of new junior technical hires since early 2025.
  • 5.1 times — increase in engineers’ code commits, Anthropic 2026 report.
  • $100,000 to $167,000 — announced salary for Conceptual Reasoning Fellows.

To read