New frontiers in O*NET analysis
O*NET is the U.S. government's official catalog of occupations. Workers in more than 900 formal U.S. jobs — radiologists, sailors and marine oilers, education administrators, switchboard operators — answer detailed surveys about what they do all day. Their answers become job profiles, organized into a few standard buckets: skills (transferable capacities like Active Listening), knowledge (bodies of subject matter), abilities (cognitive and physical traits), work activities (the general things people do on the job), and thousands of occupation-specific task statements (the concrete day-to-day stuff). Analysts add ratings on importance and frequency. Roughly 200 occupations get re-surveyed each year. We use version 30.2, released this February.
It is a very serious federal machine for turning work into rows in a spreadsheet.
O*NET replaced the Dictionary of Occupational Titles in the late 1990s. It is now the default substrate for saying what a job is. The task-based model of technology and work — Autor, Levy, and Murnane in 2003, Frey and Osborne in 2013, Eloundou et al. in 2023, the follow-ups since — treats a job as a bundle of tasks and asks which tasks machines can do. The pattern is always the same: translate work into tasks, score the tasks, infer the future. If you want to make a serious quantitative claim about work, you eventually find yourself using O*NET primitives.
We live in an era of AI automation. So we built a pipeline that takes anything — a transcript, a YouTube clip — and gives back its O*NET vector. Pass in a scene and get back which skills and abilities and knowledge domains it depicts and at what frequency.
The first thing we pointed it at was — obviously — Seinfeld.
Why Seinfeld?
You've heard the show described as "a show about nothing." That's not quite right. It is a show where nobody has material problems.
Jerry is implausibly wealthy from standup. George drifts in and out of employment without visible financial panic. Elaine has middle-class publishing jobs, but money is rarely the binding constraint. Kramer doesn't work and somehow thrives, mostly through schemes and Jerry's refrigerator. They eat out constantly, take cabs, and do not check bank balances.
What occupies them is each other. Their problems are interpersonal. Their stakes are social. Their work, when they have any, is being around other people.
And yet they are constantly annoyed.
This is roughly the best-case AGI transition. Material basics become automated and abundant. Humans spend their time on what automation misses: identity proofs, status games, service failures, one more trip to the notary. The policy question is not only what gets automated. It is what people do after.
We model Seinfeld, then, as a 174-episode simulation of a post-scarcity quartet. It is — in other words — an important window into the future of work.
Analysis
O*NET insists every human activity can be nested inside an occupation, then a task, then a work activity, then a skill. Concretely: an occupation is a job title like "Waiters and Waitresses." A task is a concrete thing that occupation does ("Take orders from patrons for food or beverages"). A Detailed Work Activity, or DWA, is O*NET's medium-grain bucket — clusters of similar tasks that recur across many occupations ("Take customer orders"). A skill is the underlying capacity those activities exercise (Active Listening, Speaking). We apply that framework here.
The pipeline has two stages. First, for each episode, Claude Sonnet 4.6 sees the O*NET-SOC occupation catalog — the ~900 federal codes O*NET uses to classify jobs — and asks: which occupations are clearly portrayed, and by whom? Second, for each detected occupation, the model sees that occupation's O*NET profile — tasks, detailed work activities, skills — and asks which profile items appear scene by scene.
One output row is not a worker. It is not an hour of labor. It is a narrower claim:
In this episode and scene, this character presents as this O*NET occupation and appears to perform or discuss these specific O*NET tasks, detailed work activities, or skills, supported by this evidence quote.
Across all 174 episodes, that produced:
| Output | Count |
|---|---|
| Occupation mentions | 837 |
| Unique O*NET-SOC occupations | 189 |
| Evidence rows | 11,712 |
| Task rows | 5,024 |
| Detailed Work Activity rows | 2,392 |
| Skill rows | 4,296 |
Call it an occupational census of the work the show makes visible.
Results: the economy of nothing
So what are the jobs that populate the post-AGI Seinfeld economy? The top occupational groups:
| SOC major group | Evidence rows | Share |
|---|---|---|
| Arts, Design, Entertainment, Sports, and Media | 3,531 | 30.1% |
| Management | 1,630 | 13.9% |
| Sales and Related | 1,244 | 10.6% |
| Food Preparation and Serving Related | 753 | 6.4% |
| Office and Administrative Support | 730 | 6.2% |
| Protective Service | 665 | 5.7% |
| Personal Care and Service | 625 | 5.3% |
| Healthcare Practitioners and Technical | 520 | 4.4% |
| Transportation and Material Moving | 468 | 4.0% |
| Business and Financial Operations | 354 | 3.0% |
Against Bureau of Labor Statistics (BLS) May 2025 employment shares — i.e. how the actual U.S. workforce is distributed across these same SOC groups — Seinfeld bends sharply toward work done in public.
| SOC major group | Seinfeld | U.S. | Ratio |
|---|---|---|---|
| Arts, Design, Entertainment, Sports, and Media | 30.1% | 1.3% | 23.2x |
| Legal | 2.5% | 0.8% | 3.1x |
| Personal Care and Service | 5.3% | 2.1% | 2.5x |
| Protective Service | 5.7% | 2.4% | 2.4x |
| Management | 13.9% | 7.2% | 1.9x |
It underrepresents the parts of the economy that are large but narratively dull: production, education, office administration, transportation, healthcare support, computer and mathematical work. Office and administrative support is 11.4% of U.S. employment; in Seinfeld it is 6.2% of evidence rows. Production is 5.5% of employment; in Seinfeld it is 0.2%. Computer and mathematical work is 3.4% of employment; in Seinfeld it is near zero.
Job zones analysis
Job Zones are an important part of any O*NET analysis. They cut the data differently. O*NET uses them as a five-point preparation scale that combines education, experience, and on-the-job training: Zone 1 occupations need little or no preparation (dishwashers, ushers), Zone 2 some (waiters, security guards, retail clerks), Zone 3 medium (electricians, paralegals), Zone 4 considerable (managers, accountants), Zone 5 extensive (lawyers, doctors, researchers). If Seinfeld were a fantasy of elite professional life, it would be overloaded on Zones 4 and 5. It is not.
The show is overweight in Job Zone 2: occupations that need some preparation, but not a four-year professional credential. Zone 2 is 52.0% of Seinfeld evidence against 44.6% in the baseline. Zone 3 is underweight. Zone 4 is only modestly overweight. Zone 5 matches the baseline.
So the post-AGI Seinfeld future is not a world where everyone becomes a lawyer, doctor, software engineer, or researcher. The visible work is moderately trained, public-facing, socially exposed, and institutionally specific. The guard. The waiter. The clerk. The retail salesperson. The building superintendent. The person with just enough authority to make the interaction difficult.
The largest bucket, Arts/Entertainment, is both right and misleading. Jerry is a comedian. The show is full of auditions, standup, movie sets, musicians, writers, performers, and people doing bits. Actors alone account for 1,985 evidence rows across 46 episodes.
If the question is "what work does Seinfeld depict?", entertainment belongs. If the question is "what in-universe labor market surrounds the four mains?", set it aside and the economy becomes concrete: managers, salespeople, waiters, clerks, police officers, security guards, doctors, lawyers, cab drivers, receptionists, cashiers.
Not a knowledge economy. A transactional Manhattan service economy. People sell things, serve things, enforce rules, answer phones, move bodies, and say no.
The top occupations by episode count:
| Occupation | Episodes |
|---|---|
| Actors | 46 |
| Entertainers and Performers, All Other | 41 |
| General and Operations Managers | 32 |
| Waiters and Waitresses | 28 |
| Chief Executives | 27 |
| Retail Salespersons | 27 |
| Musicians and Singers | 25 |
| Postal Service Mail Carriers | 23 |
| Writers and Authors | 22 |
| Police and Sheriff's Patrol Officers | 17 |
| Lawyers | 15 |
| Security Guards | 13 |
Newman is a mail carrier. Jackie Chiles is a lawyer. Elaine is in publishing. The coffee shop has waitresses. The labels are not surprising; the distribution is. Retail, food service, managers, owners, clipboards, petty authority.
Protective Service is the tell. Police officers, security guards, detectives, and quasi-detectives recur because Seinfeld plots are norm violations: parking rules, library rules, store rules, building rules, theater rules, restaurant rules. A tiny law of social life gets broken. An institutional worker appears.
Task analysis
The second stage attaches occupations to O*NET task statements. For non-actor occupations, the common tasks are a compressed description of the show:
| Occupation | O*NET task | Rows |
|---|---|---|
| Retail Salespersons | Recommend, select, and help locate or obtain merchandise | 56 |
| Writers and Authors | Write fiction or nonfiction prose | 54 |
| Retail Salespersons | Answer questions regarding the store and its merchandise | 53 |
| Postal Service Mail Carriers | Deliver mail to residences and business establishments | 49 |
| Retail Salespersons | Greet customers and ascertain what each customer wants or needs | 38 |
| General and Operations Managers | Perform personnel functions, such as selection, training, or evaluation | 37 |
| Waiters and Waitresses | Take orders from patrons for food or beverages | 36 |
| Waiters and Waitresses | Serve food or beverages to patrons | 32 |
| Lawyers | Represent clients in court or before government agencies | 18 |
Here perhaps we see O*NET lacking some richness. It does not say "George is humiliated in a store." It says retail work is recommending merchandise, answering questions, greeting customers, explaining products, processing payment. It does not say "the restaurant has become a theater of resentment." It identifies order-taking, serving, complaint handling.
We can go further. O*NET links every task to one or more Detailed Work Activities, and links those DWAs to skills. That gives a built-in ladder from the very specific ("Serve food or beverages to patrons") to the more abstract ("Perform for or work directly with the public") to the underlying capacity ("Service Orientation"). Roll the Seinfeld evidence through those links and compare it to a current-economy O*NET baseline — i.e. what mix of skills and activities you'd get if you weighted every U.S. occupation by employment:
| Skill implied by represented tasks/DWAs | Seinfeld | U.S. | Ratio |
|---|---|---|---|
| Active Listening | 14.0% | 4.1% | 3.4x |
| Speaking | 10.6% | 4.0% | 2.6x |
| Writing | 9.4% | 3.5% | 2.7x |
| Social Perceptiveness | 8.9% | 3.6% | 2.5x |
| Service Orientation | 8.2% | 3.5% | 2.4x |
| Critical Thinking | 8.2% | 3.9% | 2.1x |
The ratio column is a Revealed Comparative Advantage (RCA): the Seinfeld share divided by the baseline share. A value above 1.0 means the show over-indexes on that skill relative to the broader economy; below 1.0 means it under-indexes.
At the work-activity level, the headline is almost too neat: Performing for or Working Directly with the Public is 19.6% of the Seinfeld task/DWA signal against 2.4% in the baseline. "Communicating with People Outside the Organization," "Assisting and Caring for Others," and "Selling or Influencing Others" are overweight. "Working with Computers," "Processing Information," "Updating and Using Relevant Knowledge," and "Organizing, Planning, and Prioritizing Work" are underweight.
The SOC table says which jobs appear. The task/DWA comparison says what those jobs are doing. Not STEM. Not generic "high-skill cognitive work." Not even management in the LinkedIn sense. Direct public contact. Listening. Speaking. Reading the room. Performing institutional scripts. Moving conflict from one conversational state to another.
The mundane task is stable. The social situation explodes. Someone returns a jacket. Someone wants a table. Someone wants a marble rye. Someone is unhappy with a rental car reservation. The task is ordinary. The interaction is pathological.
Discussion: the aristocracy of nothing
The four main characters account for 5,748 evidence rows, or 49.1% of the corpus after name normalization. We expected more. Half the visible work belongs to the supporting cast.
The mains are occupational tourists. Jerry's rows are mostly standup and performance: Actors, Musicians and Singers, Writers and Authors. Elaine picks up writing and editing. George is management and executive fantasy. Kramer is the purest case. He does not have a job, but he has 337 unique occupation-task pairs: founder, sales rep, actor, recreation attendant, mail-adjacent operator, athlete, tour guide, business agent. His unemployment is not inactivity. It is occupational volatility.
The supporting cast is different. They enter with an institutional function. Puddy fixes cars. Jackie Chiles knows the law. Newman delivers mail. Bookman investigates overdue books. Waitresses take orders. Salespeople sell. Security guards enforce boundaries.
Two populations share one show. The four mains live in the post-scarcity quadrant, where every interaction is a status game or relational puzzle. The supporting cast still lives in the wage economy, where a specific competency meets a specific transaction.
That is a society in transition. Some people have escaped the need to work. Most people have not. They meet in coffee shops, dry cleaners, and parole hearings. They speak different languages. The mains talk about the world. The supporting cast does things in it.
If AGI moves more people into the main-character condition — materially unconstrained, socially preoccupied, no longer disciplined by ordinary employment — the remaining visible jobs look less like back-office information processing and more like public-facing institutional friction. High-touch service. Specialists. Authority figures. Physical-world operators. People who know the rule, hold the key, fix the thing, move the object, represent the institution.
That is the larger-economy claim. A post-AGI Seinfeld future is not the current labor market with fewer spreadsheets. Many occupations that dominate employment become narratively invisible: production, routine administration, transportation, computer work, healthcare support, education. What remains visible happens at the boundary between people and institutions.
Less factory floor or office park. More restaurant host stand, returns counter, clinic intake desk, courthouse hallway, security checkpoint, theater lobby, superintendent's office, customer service line that never quite resolves the problem.
Further research
There are caveats with this work, as there always are. Evidence rows are not workers or hours. Scene boundaries are imperfect. Occupation detection is stronger than task matching: median confidence is 0.90 for occupations, 0.65 for task/DWA/skill rows. This is not the labor market of Seinfeld measured perfectly. It is a structured, auditable map of the work the show depicts.
And what it depicts is not nothing.
It depicts a dense service economy wrapped around four people free from labor-market discipline. The mains experience work as customers, performers, schemers, social strategists. The supporting cast experiences work as tasks: take the order, sell the coat, deliver the mail, enforce the rule, file the complaint, represent the client.
That is why the show feels like a post-scarcity fantasy and a workplace comedy at the same time. The mains live after work. Everyone else is still on the clock.