IT之家 10 月 17 日消息,人工智能公司 Anthropic 昨日(10 月 16 日)发布报告称,在公司成立约 5 年后,其人均雇员创收已达 Meta Platforms 和谷歌在各自高速增长期的 2 至 4 倍。
这一数据基于 Anthropic 截至 2026 年 4 月的内部财务与员工规模记录,并与 Meta、谷歌成立五年时的公开财报数据开展通胀调整后比较。
报告表明,Anthropic 工艺部门初级员工占比已从 2025 年初的 21% 降至 14%,初级工艺岗位招聘占比更从 27% 骤降至 9%。约三分之一的机器学习经理和五分之一的软件工程经理表示,因 AI 浮现而减少了招聘。与此同时,公司总体仍在扩张,销售、财务和招聘等岗位增长更快。
AI 带来的不只是岗位收缩,也涵盖效率跃升。使用编码智能体后,工程师人均代码提交量达到 6 个月前的 5 倍;律师处理更多合同,财务处理更多发票,销售人员成交更多订单。报告追踪 192 名使用编码智能体的工程师后发现,其代码产出在使用后显著上升,且后续被删除或回滚的代码比例未见恶化。
报告称,到 2026 年 8 月,员工部署的交互式 AI 智能体工作时长已达人类员工的 6 倍;早在 3 月,全公司智能体总工时就首次超过人工工时。IT之家附上相关图片如下:
但“人机协作”也带来新负担。报告称,即使非工艺员工,阅读其 Claude 智能体全部输出也可能占去半个工作周;员工正从“执行者”转向“智能体管理者”,负责写指令、搭建监控面板、审查 AI 产出。2026 年春季,这类“元工作”已占员工交互式 Claude 会话的 22%,高于前一年的 10%。
招聘标准随之改变:管理者更看重候选人的自主性、判断力及创造性使用 AI 的能力,而非狭窄的领域专长和具体执行能力。Anthropic 同时建立了岗位风险预警机制,并优先在内部重新调配受影响员工。
不过,人均创收并非衡量企业健康度的唯一指标。外部研究机构 Epoch AI 此前估算,Anthropic 在 2026 年 4 月的人均年化营收约为 900 万美元,显著高于 Meta 的约 255 万美元和谷歌的约 211 万美元;但该机构也提醒,其员工数量与营收数据并非同期披露,估算存在不确定性。
本文来源:https://www.ithome.com/1/014/195.htm
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
IT House, Oct. 17 (Xinhua) -- Anthropic, an artificial intelligence company, released a report yesterday (Oct. 16) saying that its per capita employee revenue has reached two to four times that of Meta Platforms and Google in their respective high-growth periods after the company was founded about five years ago.

This data is based on Anthropic's internal financial and employee size records as of April 2026, and is compared with the inflation-adjusted public financial data at the time of Meta and Google's establishment five years ago.
The report shows that the proportion of junior employees in the Anthropic process sector has fallen from 21% in early 2025 to 14%, and the proportion of junior process job openings has plummeted from 27% to 9%. About one-third of machine learning managers and one-fifth of software engineering managers say hiring has been reduced by the emergence of AI. At the same time, the company is still expanding overall, with faster growth in roles such as sales, finance and hiring.

AI brings not only job shrinkage, but also efficiency leaps. With coded agents, engineers submit five times as many codes per capita as they did six months ago; lawyers handle more contracts, finances handle more invoices, and salespeople close more orders. After tracking 192 engineers using coding agents, the report found that their code output increased significantly after use, and the proportion of code that was subsequently deleted or rolled back did not deteriorate.

According to the report, by August 2026, employees had deployed interactive AI agents to work six times as long as human employees; as early as March, the total number of agent hours across the company exceeded man-hours for the first time. The relevant pictures are attached to the IT House as follows:
But “human-machine collaboration” also brings new burdens. Reading the full output of its Claude agent can take up half a workweek, even for non-technical workers; employees are shifting from “executors” to “agent managers” responsible for writing instructions, building monitoring panels, and reviewing AI output, the report said. In the spring of 2026, such “meta-work” already accounted for 22% of employee-interactive Claude sessions, up from 10% the previous year.

Recruitment criteria change: Managers value candidates' autonomy, judgment, and ability to use AI creatively, rather than narrow areas of expertise and specific execution. Anthropic has also established an on-the-job risk early warning mechanism and prioritizes internal redeployment of affected employees.

However, per capita income generation is not the only measure of a company's health. Epoch AI, an external research firm, previously estimated that Anthropic's annualized revenue per capita in April 2026 was about $9 million, significantly higher than Meta's annualized revenue of about $2.55 million and Google's annualized revenue of about $2.11 million, but it also cautioned that its number of employees and revenue data were not disclosed at the same time, and there was uncertainty in the estimation.
This article is from: https://www.ithome.com/1/014/195.htm For copyright issues, please write to: 13951431913#139.com or submit a review on the right.