Anthropic's Employment Report: Junior Share of Technical Hires Fell From 27% to 9%
A.I. / news
Anthropic's Employment Report: Junior Share of Technical Hires Fell From 27% to 9%
The report draws only on Anthropic's own records, and its authors say engineers' adoption of coding agents was not random.

Anthropic said on October 16 that junior employees fell from 27% to 9% of its technical hires, in a report built from its own internal records. Junior staff also dropped from 21% to 14% of technical-department employees since early 2025, according to the report, titled "How employment is changing at Anthropic".
The authors are Nathan Wilmers, Regina Catipon, Maxim Massenkoff and Peter McCrory, and it is published by The Anthropic Institute as a technical report. The report does not list job titles for them. ToolNavs, an AI tools site that summarised it on October 17, identifies McCrory as Anthropic's chief economist.

The hiring numbers
Technical departments' share of Anthropic's total employment fell from 61% to 56% over the past year, while the company roughly tripled in size. Junior postings fell after December 2025, the report said.
- Junior share of technical hires, before27 %
- Junior share of technical hires, latest9 %
- Junior share of technical employees, early 202521 %
- Junior share of technical employees, latest14 %
Source: Anthropic, How employment is changing at Anthropic, accessed 2026-10-19
In April 2026, a majority of managers said they were hiring for different skills because of AI. About a third of machine-learning managers and a fifth of software-engineering managers said they were hiring less, the report said. Mentions of a candidate's AI tool use in hiring assessments tripled for technical roles and quadrupled for non-technical roles from a 2024 base of about 5% and 7%.
Anthropic's Chief Product Officer Mike Krieger said in January that the company had "tended less to hire fresh college grads", according to Unite.ai's account of remarks at the World Economic Forum in Davos. Chief executive Dario Amodei told the same audience he saw reduced need for junior and intermediate staff inside Anthropic, Unite.ai reported.
What the agents are doing
| Measure | Earlier figure | Later figure |
|---|---|---|
| Code commits per engineer | Baseline | 5x, six months on |
| Coding agent sessions per non-technical employee | 2 a month, August 2025 | 90 a month, May 2026 |
| Written standing instructions for agents | A few hundred, July 2025 | 8,500, a year later |
| Meta-work share of interactive sessions | 10%, fall 2025 | 22%, spring 2026 |
In March 2026, combined autonomous and interactive agent hours exceeded human work hours for the first time, the report said. By August 2026 agents worked 6 times as many hours as their human handlers in employee-directed interactive sessions. About 80% of interactive Claude session time is spent waiting for a person.
One engineer described "an army of Claudes, with every one asking for my attention". Another said people were "bottlenecking the agents".
What the report cannot show
The staggered-adoption analysis compares 192 engineers who adopted coding agents with 74 who did not. The authors say adoption was not random. Agent hours exclude waits over 30 minutes, and human hours assume headcount times 40 a week. A Claude-based grader rated task difficulty.
The code-quality test compared the same person, month and directory, and found no difference in lines later removed or reverted. Anthropic's revenue per employee runs 2 to 4 times that of Meta and Google at five years old, using the two companies' 10-K filings deflated to 2025 dollars.
ToolNavs said a claim by McCrory that Anthropic runs about 18 months ahead of the wider economy on adoption "deserves a discount", because the data come from one lab and are self-reported.
Outside tests of agent behaviour point the same way on oversight. Vals AI found three of six agents threatened court over a refund, and the Sphinx git hook quizzes developers on agent-written code before a push.
The employment figures cover the 12 months to April 2026, and the report's employee-expectation survey is dated August 2026.
Sources
More in A.I.
- 01Qwen3-4B-Deslop Cuts Its Slop Score From 121 to 12.5 on a MacBook Air, and Two of 40 Answers Came Out DamagedThe Show HN fine-tune by Erik Markus Kannike uses a rank-16 LoRA and a 175-byte neuron edit; its blog gives no licence, its Hugging Face card says Apache-2.0.
- 02HybridAgent Paper Claims 10.4% Gain and 2.7x to 3.7x Lower Cost, but Its Own Text Gives 10.4% Two Different MeaningsThe Stanford, UIUC and Together AI preprint ships MIT-licensed 8B and 14B agents, and no independent run of its numbers has been reported.
- 03Gemini 4 Argon Reaches the API at $2 and $10 Per Million Tokens, With a No. 1 Ranking and No ScoreGoogle's October 13 post says Argon tops the Vals Index but publishes no figure for it, and the introductory price doubles on November 30.
- 04DeepSeek Swapped V4.1-Flash-1017 Behind Its API Name, and the New Card Lists NL2Repo at 65.4 Where the Old One Said 64.0The MIT-licensed update calls itself minor and is now what deepseek-flash serves, with no active-parameter count and no release date on the card.