Perplexity：Teammate 每周 agent 工时一周增长至 11 倍
Perplexity 自己晒了内部数据：Teammate 的 agent 工时一周涨到 11 倍，1/5 员工每周多出 40 小时 agent 工时。8000 万到 1.8 亿美元的年化劳动价值对比 1000 万美元 token 成本，这组 ROI 数字对企业买家和做 agent 产品的开发者都很有参考价值。
Perplexity CEO Aravind Srinivas 透露，内部产品 Teammate 产生的 agent 工时达到一周前的 11 倍，周环比增速超过 50%（此前一周为 7 倍）。使用 Teammate 获得额外 8 小时以上工作日的员工数量自 9 月以来翻了三倍以上，1/5 的员工每周已获得 40 小时以上的 agent 工时，相当于多出整整一个工作周。Srinivas 估算，这些 agent 工时对应的年化劳动力价值约为 8000 万至 1.8 亿美元，而消耗的 token 成本仅约 1000 万美元，投入产出比在 9 到 19 倍之间，具体倍数取决于岗位类型。
Teammate 的机制是为每位员工配置可自主执行任务的 agent，承担编写代码、调研、处理文档等日常工作，从而直接以工时计量 agent 对组织的贡献。Perplexity 用"agent 工时/员工工时"的比率作为内部指标，且这一数字每周持续上升。
这类第一方数据是企业级 agent 产品少见的量化披露。相比 OpenAI、Anthropic 等厂商多以基准测试展示 agent 能力，Perplexity 直接给出劳动力价值与 token 成本的换算，为其他企业评估 agent 部署 ROI 提供了参照。开发者与企业管理者在评估 agent 工具时，可关注这种以实际工时而非演示效果为口径的衡量方式。
A week later, it’s 11x more agent hours per week from Teammate. Growing more than 50% week over week (was 7x last week). The number of employees getting an extra workday (8+ hours) and extra work week (40+ hours) from Teammate has more than tripled in a week since September. That’s roughly $80M to $180M a year in labor value against about $10M in tokens, so 9x to 19x depending on the job. Aravind Srinivas @AravSrinivas The median Perplexity employee now gets 7x more agent-hours per week from teammates than in September. And this has been consistently up every single week. 1 in 5 are already at 40+ agent-hours a week, which is basically a full extra workweek. 🔗 View Quoted Tweet📊 22 ⚡
This report is categorized under AITOP's "Products" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include Perplexity, Teammate, agents, enterprise AI.
Source attribution: curated from Aravind Srinivas (@AravSrinivas). Cross-source corroboration: 11 independent sources have covered this development (including TestingCatalog, Decoder, OpenAI Blog), confirming industry-wide signal strength.
