2023年春天，当乔治城大学的同学们正埋头准备期末考试时，布伦登·弗迪却在忙着实践他关于工作的新理论。
“大二那年，我早在期末考试前就下定决心要退学了，”他告诉《财富》，“所以我压根没去考场。”
那时，弗迪已经在课堂之外找到了更重要的东西。几个月前，在圣保罗的一场黑客马拉松上，他和联合创始人偶然发现了一个简单却高效的商业模式：将企业与海外技术工程师对接，处理所有中间环节，并从每笔交易中抽取少量佣金。他们的第一位客户同意以每周500美元的价格雇佣一名开发者；Mercor将其中约70%支付给工程师，剩下的留作平台服务费。
这个最初旨在连接人才的平台，很快演变成了一个更宏大的构想：建立一个人类可以帮助训练AI系统的市场——而这些AI未来或许会取代人类。如今，Mercor雇佣专业人士——包括顾问、律师、银行家和医生——来创建评估任务和评分标准，用以测试并完善AI模型的推理能力。
“大家的注意力都集中在模型能做什么上，”弗迪说，“但真正的机遇在于教会它们只有人类才懂的东西——判断力、细微差别的把握以及品味。”
在九个月内，他与联合创始人——高中同学兼辩论队队友阿达什·希瑞马斯和苏利耶·米达——将这个初步构想变成了一家年化收入达百万美元的公司。这三人的早期成功与其说是运气，不如说是一次概念验证：他们昔日辩论中磨练出的那套结构化推理，可以被编码用以教导机器如何思考。
两年后，Mercor已成长为一家估值百亿美元的公司，使得这三位创始人成为全球最年轻的白手起家亿万富翁。圣保罗那次实验的产物，已蜕变为AI时代扩张最快的初创企业之一，吸引了众多重量级投资者，他们视其为未来“人在回路”自动化领域的核心。
对弗迪而言，从大学辍学生到亿万富翁创始人的跨越，是一个理性的选择。
“上大学时，工作是件需要自律才能完成的事，”他说，“但当我创立Mercor后，它变成了让我魂牵梦绕、无法停止思考的事情。”
过去三年里，弗迪一天也未休息过。他说，即使和父母共进晚餐时，他也在思考工作，但这对他而言并不感觉像是在工作。
“当人们在不产生复利效应的事情上拼命努力时，就会精疲力竭，”他解释道，“而我每天都能看到自己时间投入的回报。”
这种心态已成为Mercor使命的哲学核心。在弗迪看来，AI并非在消灭劳动，而是在重新分配劳动。随着软件将重复性的白领工作自动化，人类将向价值链上游迁移，去教导机器如何推理、决策和创造。
“就好像我们的经济中存在一个瓶颈，即人类劳动力的总量有限，”他说，“而在未来十年，这种结构将发生根本性改变。””
Mercor如何缓解这一瓶颈？其平台允许企业发布成千上万的微任务，在真实的专业场景中——如撰写财务备忘录、起草法律简报或分析医疗图表——评估模型表现。人类评估员根据详细的评分标准对每一项输出进行打分，并将结构化的反馈提供给模型。每一次评估都在帮助AI学习人类如何决策以及如何衡量质量。
该系统的核心是APEX——即“AI生产力指数”，这是Mercor专有的基准测试，用于评估AI在执行具有经济价值的工作时的表现。APEX并非测试抽象的推理或数学难题，而是基于从投资银行家、律师、顾问和医生工作流程中提取的200项任务来评估大模型。为了构建APEX，Mercor聘请了一个重量级顾问团队，成员包括美国前财政部长拉里·萨默斯、麦肯锡前全球管理合伙人多米尼克·巴顿、法律学者卡斯·桑斯坦以及心脏病专家埃里克·托波尔。他们各自协助设计了评估标准和案例结构，以反映高风险专业工作的实际情况。
正如公司所言：“口袋里装着1万名博士固然很棒——但拥有一个能可靠帮你报税的模型则更胜一筹。”
Mercor的成功意义深远。在弗迪看来，这个新型劳动力市场可以在全球雇佣数百万人，同时加速AI的进步。
“我们或许能将三分之二的知识工作自动化，”他说，“这将是非凡的成就，因为它能让我们去攻克诸如治愈癌症、登陆火星这样的难题。”
对投资者而言，Mercor的增长故事令人难以抗拒。它正处于两大趋势的交汇点：AI的主流化以及灵活的项目制工作的兴起。每一个企业客户都会带来新的评估员，而每一位评估员都有助于优化更多模型，从而形成数据和需求齐头并进的飞轮效应。
“我们的收入增长速度在商业史上名列前茅，”弗迪平静地陈述道。
弗迪喜欢将其称为下一次工业革命。他知道人们害怕被AI取代，也经常需要回应关于训练AI取代人类工作的伦理质疑。弗迪认为，我们应该勇于面对现实。
“人们很容易陷入卢德主义的思维，将生产力提升视为坏事，因为它会导致短期失业，”弗迪说，“但每一次重大的技术革命，最终都让生活变得更好。”
弗迪指出，工业革命后，美国农业人口从占总人口的75%下降到约1%，这使人们得以解放出来从事其他各行各业的工作。
“当下的挑战在于，我们需要审慎思考未来：人类将把时间投入到哪些更高层次、更有价值的事情上，”弗迪说，“以及我们能多快地帮助实现那个未来。”（*）
译者：刘进龙
审校：汪皓
2023年春天，当乔治城大学的同学们正埋头准备期末考试时，布伦登·弗迪却在忙着实践他关于工作的新理论。
“大二那年，我早在期末考试前就下定决心要退学了，”他告诉《财富》，“所以我压根没去考场。”
那时，弗迪已经在课堂之外找到了更重要的东西。几个月前，在圣保罗的一场黑客马拉松上，他和联合创始人偶然发现了一个简单却高效的商业模式：将企业与海外技术工程师对接，处理所有中间环节，并从每笔交易中抽取少量佣金。他们的第一位客户同意以每周500美元的价格雇佣一名开发者；Mercor将其中约70%支付给工程师，剩下的留作平台服务费。
这个最初旨在连接人才的平台，很快演变成了一个更宏大的构想：建立一个人类可以帮助训练AI系统的市场——而这些AI未来或许会取代人类。如今，Mercor雇佣专业人士——包括顾问、律师、银行家和医生——来创建评估任务和评分标准，用以测试并完善AI模型的推理能力。
“大家的注意力都集中在模型能做什么上，”弗迪说，“但真正的机遇在于教会它们只有人类才懂的东西——判断力、细微差别的把握以及品味。”
在九个月内，他与联合创始人——高中同学兼辩论队队友阿达什·希瑞马斯和苏利耶·米达——将这个初步构想变成了一家年化收入达百万美元的公司。这三人的早期成功与其说是运气，不如说是一次概念验证：他们昔日辩论中磨练出的那套结构化推理，可以被编码用以教导机器如何思考。
两年后，Mercor已成长为一家估值百亿美元的公司，使得这三位创始人成为全球最年轻的白手起家亿万富翁。圣保罗那次实验的产物，已蜕变为AI时代扩张最快的初创企业之一，吸引了众多重量级投资者，他们视其为未来“人在回路”自动化领域的核心。
对弗迪而言，从大学辍学生到亿万富翁创始人的跨越，是一个理性的选择。
“上大学时，工作是件需要自律才能完成的事，”他说，“但当我创立Mercor后，它变成了让我魂牵梦绕、无法停止思考的事情。”
过去三年里，弗迪一天也未休息过。他说，即使和父母共进晚餐时，他也在思考工作，但这对他而言并不感觉像是在工作。
“当人们在不产生复利效应的事情上拼命努力时，就会精疲力竭，”他解释道，“而我每天都能看到自己时间投入的回报。”
这种心态已成为Mercor使命的哲学核心。在弗迪看来，AI并非在消灭劳动，而是在重新分配劳动。随着软件将重复性的白领工作自动化，人类将向价值链上游迁移，去教导机器如何推理、决策和创造。
“就好像我们的经济中存在一个瓶颈，即人类劳动力的总量有限，”他说，“而在未来十年，这种结构将发生根本性改变。””
Mercor如何缓解这一瓶颈？其平台允许企业发布成千上万的微任务，在真实的专业场景中——如撰写财务备忘录、起草法律简报或分析医疗图表——评估模型表现。人类评估员根据详细的评分标准对每一项输出进行打分，并将结构化的反馈提供给模型。每一次评估都在帮助AI学习人类如何决策以及如何衡量质量。
该系统的核心是APEX——即“AI生产力指数”，这是Mercor专有的基准测试，用于评估AI在执行具有经济价值的工作时的表现。APEX并非测试抽象的推理或数学难题，而是基于从投资银行家、律师、顾问和医生工作流程中提取的200项任务来评估大模型。为了构建APEX，Mercor聘请了一个重量级顾问团队，成员包括美国前财政部长拉里·萨默斯、麦肯锡前全球管理合伙人多米尼克·巴顿、法律学者卡斯·桑斯坦以及心脏病专家埃里克·托波尔。他们各自协助设计了评估标准和案例结构，以反映高风险专业工作的实际情况。
正如公司所言：“口袋里装着1万名博士固然很棒——但拥有一个能可靠帮你报税的模型则更胜一筹。”
Mercor的成功意义深远。在弗迪看来，这个新型劳动力市场可以在全球雇佣数百万人，同时加速AI的进步。
“我们或许能将三分之二的知识工作自动化，”他说，“这将是非凡的成就，因为它能让我们去攻克诸如治愈癌症、登陆火星这样的难题。”
对投资者而言，Mercor的增长故事令人难以抗拒。它正处于两大趋势的交汇点：AI的主流化以及灵活的项目制工作的兴起。每一个企业客户都会带来新的评估员，而每一位评估员都有助于优化更多模型，从而形成数据和需求齐头并进的飞轮效应。
“我们的收入增长速度在商业史上名列前茅，”弗迪平静地陈述道。
弗迪喜欢将其称为下一次工业革命。他知道人们害怕被AI取代，也经常需要回应关于训练AI取代人类工作的伦理质疑。弗迪认为，我们应该勇于面对现实。
“人们很容易陷入卢德主义的思维，将生产力提升视为坏事，因为它会导致短期失业，”弗迪说，“但每一次重大的技术革命，最终都让生活变得更好。”
弗迪指出，工业革命后，美国农业人口从占总人口的75%下降到约1%，这使人们得以解放出来从事其他各行各业的工作。
“当下的挑战在于，我们需要审慎思考未来：人类将把时间投入到哪些更高层次、更有价值的事情上，”弗迪说，“以及我们能多快地帮助实现那个未来。”（*）
译者：刘进龙
审校：汪皓
In the spring of 2023, while his classmates at Georgetown were cramming for finals, Brendan Foody was busy testing out his new theory of work.
“I knew I wanted to drop out before finals my sophomore year,” he told Fortune. “I just didn't go to finals.”
By then, Foody had already found something he couldn't learn in a lecture hall. A few months earlier, at a hackathon in São Paulo, he and his co-founders had stumbled onto a simple but powerful model: match companies with skilled engineers abroad, handle the logistics, and take a small cut of each deal. Their first client agreed to pay $500 a week for a developer; Mercor paid the engineer roughly 70% and kept the rest as a service fee.
What began as a way to connect talent soon evolved into something more ambitious: a marketplace where humans could help train the AI systems that might one day replace them. Mercor now hires professionals---consultants, lawyers, bankers, and doctors---to create “evals” and rubrics that test and refine models' reasoning.
“Everyone's been focused on what models can do,” Foody said. “But the real opportunity is teaching them what only humans know---judgment, nuance, and taste.”
Within nine months, he and his co-founders---high school friends and debate teammates Adarsh Hiremath and Surya Midha---had turned that fledgling idea into a company with a $1 million revenue run rate. The trio's early success was less a fluke than a proof of concept: that the same structured reasoning they once practiced on the debate circuit could be codified to teach machines how to think.
Two years later, Mercor has become a $10 billion company, turning the trio into the world's youngest self-made billionaires. The product of that São Paulo experiment had transformed into one of the fastest-scaling startups of the AI era, attracting major investors who view it as a linchpin in the future of human-in-the-loop automation.
To Foody, the leap from college dropout to billionaire founder was rational.
“When I was in college, work was something I had to be disciplined to do,” he said. “When I started Mercor, it became something I couldn't stop thinking about.”
Foody still hasn't taken a day off in three years. He says even when he's at the dinner table with his parents, he thinks about work, which, to him, doesn't feel like work.
“People burn out when they work hard on things that don't feel compounding,” he explained. “I see the ROI of my time every day.”
That mindset has become the philosophical core of Mercor's mission. In Foody's view, AI isn't eliminating labor: it's reallocating it. As software automates repetitive white-collar tasks, humans will move up the value chain, teaching machines how to reason, decide, and create.
“It's like we have this bottleneck of only so much human labor in the economy,” he said. “That shape is going to change radically over the next decade.”
How is Mercor alleviating the bottleneck? Its platform allows enterprises to commission thousands of micro-tasks that measure model performance in real professional contexts: writing a financial memo, drafting a legal brief, or analyzing a medical chart. Human evaluators grade each output against detailed rubrics, feeding structured feedback back into the model. Every evaluation helps AI learn how people make decisions, and how they measure quality.
At the center of that system is APEX---the AI Productivity Index, Mercor's proprietary benchmark for assessing how well AI performs economically valuable work. Rather than test abstract reasoning or mathematical puzzles, APEX evaluates large models on 200 tasks drawn from the workflows of investment bankers, lawyers, consultants, and physicians. To build it, Mercor enlisted a heavyweight advisory group that includes former Treasury Secretary Larry Summers, ex-McKinsey managing partner Dominic Barton, legal scholar Cass Sunstein, and cardiologist Eric Topol. Each helped design the evaluation rubrics and case structures to mirror the realities of high-stakes professional labor.
As the company puts it: “It's great to have 10,000 PhDs in your pocket---it's even better to have a model that can reliably do your taxes.”
The implications of Mercor's success are sweeping. In Foody's eyes, this new labor market could employ millions of people globally while accelerating AI progress.
“We'll automate maybe two-thirds of knowledge work,” he said. “And that'll be incredible, because it lets us do things like cure cancer and go to Mars.”
For investors, Mercor's growth story is irresistible. It sits at the intersection of two seismic shifts: the mainstreaming of AI and the rise of flexible, project-based work. Each corporate client adds new evaluators, and each evaluator helps refine more models, creating a flywheel of both data and demand.
“We have one of the fastest revenue ramps of any company in history,” Foody said matter-of-factly.
Foody likes to describe it as the next industrial revolution. He knows people are afraid of being replaced by AI, and constantly fields questions on the ethics of training AI to replace jobs. Foody argues we ought to just bite the bullet.
“It's easy to fall into a Luddite mindset and see productivity gains as bad because they cause short-term job losses,” Foody said. “But every major technical revolution has ultimately made life better.”
After the industrial revolution, the economy went from 75% of Americans working as farmers to about 1%, and that freed people to do everything else, Foody said.
“The challenge now is to be thoughtful about what comes next: the higher, better things humans will spend time on,” Foody said, “and how quickly we can help make that future real.”