Workers in six occupations do 79 tasks that O*NET, the US Department of Labor’s database of the tasks each occupation involves, does not list, Mercor said on 27 October 2026. Mercor puts that work at about $140 billion a year in wages. The finding comes from TaskNet, an open-source task catalog that Mercor built with researchers from Stanford, the O*NET Resource Center and the National Bureau of Economic Research.
Economists use O*NET to estimate which jobs AI will change. “If the lists are incomplete, so are the estimates,” Mercor wrote on X.
Short explainer · Mercor TaskNet
A vertical news Short narrated from the published text of this piece and related FourWeekMBA pieces.
Business Pill · Data labeling explained
An evergreen Business Engineer explainer on data labeling, the human work behind an AI model. It explains the general idea and does not discuss TaskNet or O*NET.
The key insight: As we read it, TaskNet’s point is that AI exposure scores depend on which tasks a job list includes, and Mercor finds 79 tasks across six occupations that O*NET does not list.
TaskNet covers six occupations so far. Mercor’s chart counts the tasks in each TaskNet list with no O*NET counterpart. They are 23 of 44 for software developers (52%), 18 of 36 for personal care aides (50%), 17 of 38 for ER physicians (45%), 10 of 32 for lawyers (31%), 9 of 39 for elementary teachers (23%) and 2 of 13 for dishwashers (15%).
Many of the missing tasks are work that is done but not often counted, the blog post says. Its examples include elementary school teachers giving basic first aid and ER physicians calling patients with test results. The post cites the behavioural economist Rory Sutherland, who calls this the “doorman fallacy”: assuming a job is made up only of its most visible tasks.
The $140 billion is Mercor’s estimate. Its chart note says it multiplies the estimated share of working time on the new tasks by BLS May 2025 employment and mean wage for the six occupations.
Tasks per occupation in TaskNet: those with an O*NET counterpart (TaskNet total minus new tasks, our arithmetic) and the full TaskNet list, from the Mercor and Stanford Digital Economy Lab chart in Mercor’s TaskNet blog post, 27 Oct 2026.TaskNet against O*NET for six occupations: 79 tasks with no O*NET counterpart, 5 of them newly created or reshaped by AI adoption, and about $140B a year in time-adjusted wages by Mercor’s estimate. Source: Mercor blog, 27 Oct 2026.
What Was Posted on X
Workers in six occupations do 79 tasks the government’s official occupational taxonomy has yet to add. That work is worth about $140B a year in wages.
That taxonomy is called O*NET, the U.S. Department of Labor’s database of the tasks each occupation involves. Workers use it to help with job transitions, and economists use it to estimate which jobs AI will change. If the lists are incomplete, so are the estimates.
Today we’re releasing TaskNet, an open-source task catalog built on O*NET and updated for how AI is changing work.
Built with researchers from Stanford, the O*NET Resource Center, and NBER. Thank you to @robinsonaisaac @erikbryn @diyi_yang @jennjwang @andreas_h0wpt @Digeconlab @stanfordnlp @stanfordhai @stanfordAIlab for contributing to this work.
Mercor posted the release on 27 October. The post says workers in six occupations do 79 tasks that O*NET, the US Department of Labor’s task database, has yet to add, worth about $140B a year in wages, and that TaskNet was built with researchers from Stanford, the O*NET Resource Center and NBER. This piece uses the post for the headline figures and the quote on incomplete lists.
How the Interviews Worked
TaskNet uses a language model as the interviewer. It asks workers about their tasks, follows up on vague descriptions and confirms the task list with them. Mercor says it has completed over 3 million AI interviews since 2023.
O*NET analysts reviewed all 229 candidate statements and found that they all fit the O*NET center’s standards for a valid task, the post says, though 21% were rewritten. On a holdout set of 1,184 task reports from 57 workers, TaskNet’s lists covered 89.3% of reported tasks on average, against 78.6% for O*NET.
Five Tasks Created or Reshaped by AI
Tasks in TaskNet with no O*NET counterpart, by occupation (each icon = 1 task), from the Mercor and Stanford Digital Economy Lab chart; ~$140B a year in wages is Mercor’s estimate.
TaskNet found five tasks that were newly created or reshaped as a direct result of AI adoption. Most come from software engineering, the post says, such as designing prompts and context pipelines or writing evals for AI systems.
The post also cites law, a non-technical field. Lawyers reported that they now spend time reviewing and revising AI-generated legal content for accuracy, compliance and supporting authority before filing or circulation.
What It Does to AI Exposure Scores
The fuller lists change the AI exposure scores. Mercor’s chart shows average RL feasibility, how feasible it is to train AI on a task with reinforcement learning, using the method of Tomei and Klein Teeselink. It rose by 9.6 points for lawyers, 9.1 for software developers, 8.1 for elementary teachers and 2.0 for ER physicians. It fell by 6.5 for personal care aides and stayed at 0.0 for dishwashers.
For lawyers, software developers, teachers and physicians, the added tasks were mainly clerical work around documenting, communicating and keeping records, the post says. For personal care aides, the new tasks were more physical or relationship based, such as helping clients dress or caring for a client’s pet.
TaskNet also ranked tasks by how much better the best person is than the average person. Among the software tasks easiest for AI to learn, only 34% of developers say top performers clearly outperform the average developer. For the more difficult to learn tasks, 62% say so.
What’s new: a task list built from worker interviews finds work that the government’s list leaves out, and the gaps move AI exposure scores up for lawyers, developers and teachers and down for care aides.
The Structural Read
As we read it, the missing work is mostly the unglamorous part of a job: the post says the added tasks for lawyers, developers, teachers and physicians were mainly clerical work around documenting, communicating and keeping records.
As we read it, that clerical work is where the exposure scores rose, by 9.6 points for lawyers and 9.1 for software developers on Mercor’s chart.
As we read it, care work moved the other way: the added tasks for personal care aides were physical or relationship based, and their score fell by 6.5 points.
Mercor, on X, 27 October 2026
“If the lists are incomplete, so are the estimates.”
Three Implications
FOR ECONOMISTS AND POLICYMAKERS Mercor notes that economists use O*NET to estimate which jobs AI will change; TaskNet’s lists covered 89.3% of reported tasks on its holdout test, against 78.6% for O*NET.
FOR EMPLOYERS The post says the most AI-exposed software tasks were the ones with the lowest returns to additional expertise.
FOR WORKERS Mercor says workers use O*NET to help with job transitions, and TaskNet adds tasks such as reviewing AI-generated legal content.
The framework opens: “In any product cycle, there are two curves to follow.” One is what the technology can do; the other is what users can actually do with it.
As we read it, TaskNet adds a third measure to that picture: what workers actually do all day, task by task, and Mercor’s point is that scores built on an incomplete list are themselves incomplete. The framework was written before TaskNet and does not discuss Mercor or O*NET.
What Is Not Established
TaskNet covers six occupations, and the $140 billion is Mercor’s own estimate built on its estimated time shares. We read Mercor’s blog post and its charts; we could not open the working paper on SSRN, so the method details here are as the blog describes them. Mercor does not say how the exposure changes would look across the wider economy.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
Mercor and academic researchers say O*NET misses 79 tasks across six occupations, worth about $140 billion a year in wages by Mercor’s estimate. Adding them raised AI exposure scores for lawyers, software developers and teachers, and lowered them for personal care aides.
A note on sourcing. Figures are from Mercor’s blog post “Introducing TaskNet: A New Taxonomy of Work” and the three charts in it, and from Mercor’s post on X, both 27 October 2026. The $140 billion is Mercor’s estimate. We did not read the SSRN working paper. This is not investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
