Methodology

Last updated 2026-09-26

1. My Job Lifespan: what the number means

The date shown for each job is when, counting from 2026, the average employer is estimated to start visibly reducing the hiring of people for this job because of AI. It is not when half of the work changes or when the job disappears; it is when replacing people actually begins.

• The judgment is a number of years, in steps of 0.5. A job where this has already begun gets 0.5 years; a job where it is not expected within a generation gets the cap of 30 years.
• The date on screen is the judgment date plus those years, so the time left counts down every day.
• Likely range: a range from a faster case to a slower case.
• AI impact: one of Replacement (demand for people itself falls), Shrinking (teams get smaller but the job remains), AI-assisted (the tools change and demand for people holds or grows) and Shielded (protected for a long time by rules or physical work).
• Human share in 2046: the share of the core work that people are expected to still do 20 years from now.
• Replacement risk: set by the years. Under 3 years Very high, 3 to under 7 High, 7 to under 15 Medium, 15 or more Low.
• Each judgment also records a confidence level from 1 to 5.

2. How jobs are judged

For each job, an AI model scores the items below, each with a one-line reason, and then sets the years by overall judgment rather than a formula. The judgment draws on public information the model knew as of mid-2026 (the state of AI, industry trends, rules and institutions).

• AI capability now (0–10): how well the most advanced AI as of mid-2026 produces the job’s core output at expert quality without human supervision. This is judged by kinds of capability, not by model names or benchmark scores.
• Capability year: the year AI is expected to handle most of the core work well enough, given the pace of progress so far.
• Adoption year: the year industry actually starts to cut people. It weighs cost structure, adoption cycles, resistance from incumbents and unions, data and security requirements and customer acceptance, and it separates adopting tools from reducing staff.
• Barriers (0–3): licences, legal responsibility for signatures, regulation, civil-service status, collective agreements and similar protections, and how many years they actually add. If assistants and new hires are cut first even in a licensed field, the judgment says so.

Assumption: the criteria assume fast AI progress and fast adoption by industry, while taking licences, rules and physical work into account as they are. The dates can therefore be earlier than those of a more conservative outlook.

Consistency: within the same field, assistant roles are not judged later than professional roles, physical work is not judged earlier than cognitive work, and licensed jobs are not judged earlier than unlicensed ones. Reference ranges for a few jobs keep the scale aligned (for example call-center agents 0.5–2 years; plumbers, electricians and welders 20–30 years), and the results were kept from bunching at 3–5 years.

3. How each country is handled

• South Korea: 752 jobs. The list was built so that all 450 unit groups of the Korean Standard Classification of Occupations (KSCO, 7th revision) link to at least one job. The Korean judgments use “the average employer in advanced economies (the US, Korea, Japan and similar)” as the reference and also serve as the shared baseline for other countries.
• United States and Japan: jobs whose timing depends only on technology and adoption speed use the shared baseline. Jobs whose timing is changed by licensing, public-sector rules, labor law or market structure were judged again for that country. Jobs that do not exist there were removed, and jobs found only there were added. US: 790 jobs (159 judged again, 74 added). Japan: 770 jobs (164 judged again, 43 added).
• Taiwan: all 803 jobs were judged one by one for Taiwan. Where the situation was judged the same as elsewhere, the existing value was kept and that decision was recorded.
• Numbers on the Korean pages: to keep figures that were already shared unchanged, the Korean pages show the judged years multiplied by 1.02 (for example 30 years → 30.6 years). Pages for other countries use the judged years as they are.

4. Rankings

• Job rankings: jobs in a country are ranked from the fewest judged years, that is, from the job where replacement is expected to start first. #1 is the earliest. Ties go to the job with the earlier likely range.
• Major rankings: majors are ranked from the earliest expected start of the hiring decline; ties go to the major with the lower 2030 hiring-outlook score.

5. My Major Lifespan

The date shown for each major is when AI is estimated to start putting visible downward pressure on new hiring in the main career paths of that major’s typical graduates in that country. It does not mean the program will close or that all graduates will lose their jobs.

• One value per major. It refers to bachelor’s graduates if a bachelor’s program exists; otherwise to the country’s main program (associate, master’s, professional degree and so on), which the page then states.
• It is judged as months from the base date of 24 September 2026, in steps of 6 months, looking up to 30 years (360 months) ahead, with a likely range. If the evidence is too thin the page says “Not enough evidence yet”; if the start is not expected within 30 years it says “30+ years”.
• Order of judgment: set the main career paths, find which entry-level tasks AI reduces, weigh what slows adoption or supports hiring (on-site work, responsibility, licences, demand), then set the date. It is not calculated automatically from averages of job years or a formula, and hiring declines caused by the economy or population change are not counted as AI replacement.
• By country: the 198 Korean majors were judged first and serve as baselines. For the 220 US, 239 Japanese and 249 Taiwanese majors, a baseline with the same main career path was found and compared on program structure, main career paths, licensing and hiring rules, and local demand; the value was then kept, adjusted or judged anew.
• When you pick a graduation year, the page compares that year’s graduation date (calculated as end of February in Korea, end of May in the US, end of March in Japan and end of June in Taiwan) with the estimated date to show the time left after graduation. The estimated date itself does not change with the graduation year.
• Lists of majors: Korea uses departments at four-year universities and junior colleges grouped by the Ministry of Education’s seven fields; the US uses CIP 2020 (the program classification of the National Center for Education Statistics, NCES); Japan uses faculty and department names at universities, junior colleges and vocational schools; Taiwan uses the Ministry of Education’s program list and universities’ official program pages.
• Limits: no new web research was done at the judgment stage; judgments used material already collected. There are no statistics that separate AI effects from economic, population or budget effects, so majors where those factors are large, such as education and public health, have low confidence and wide ranges.

6. My University’s Lifespan (Korea only)

Rates the next 10 years for 321 Korean universities on five levels (stable, fair, caution, warning, uncertain). It looks at six axes — recruitment, student retention, protected departments, financial strength, policy signals and location — and combines rule-based calculations for each axis with an AI model’s judgment. It uses data from the last three years and is not an official government grade or a probability of closure. Full criteria (in Korean): 대학 전망 판정 기준.

7. Where My Area Ranks (Korea only)

The Korean page “AI가 살고 싶은 지역 순위” (areas AI would want to live in) is based on an AI model researching public data for each of Korea’s 256 cities, counties and districts and scoring five items from 0 to 100 (in steps of 5) for 5, 10 and 20 years ahead.

The AI’s raw scores cluster in the middle, so the page shows a nationally comparable score produced by one formula applied to every area, period and item: item score = 60 + 1.4 × (raw score − 55), with results below 0 set to 0 and above 100 set to 100. The overall score is the weighted sum of these item scores, and grades and ranks use it. If any single item cannot be judged, the overall result for that period is withheld.

Housing prices, public safety and the natural environment are not scored yet. The scores are interpretations under shared rules, not probabilities, official ratings or measurements. Base date: 24 September 2026.

8. 3-Minute Career Quiz

Three questions about your situation and 24 questions about activities you enjoy (answered on a 5-point scale) produce scores on six interest types (Hands-on, Analytical, Creative, Helping, Persuading, Organizing), which are compared with each job’s O*NET interest profile to calculate a fit score. Candidates are only jobs judged slow to be replaced in that country (7 years or more with an AI-assisted or Shielded impact, or 12 years or more). If you choose a preparation time of 6 months or less, jobs that need a licence or exam are left out. The questions were written for this site, following the format of the O*NET Interest Profiler.

9. Survival Calculator

From your current savings, monthly spending, income and the date your paycheck stops, it calculates a simple cash flow: what you save until then and when spending afterwards uses the money up. If you pick a job, you can use that job’s estimated replacement date as the date your pay stops, and you can compare changes such as spending cuts, extra income, re-employment, investment returns and inflation. It does not predict job loss or bankruptcy, and the amounts you enter are not sent to a server or stored.

10. My Life Difficulty

Questions in 8 stages (23 to 46 depending on the person) cover 12 areas — money (income, spending, assets, debt, housing), work, public safety nets, people you can rely on, health, care, recent shocks and the future, and deprivation — and combine them around the worst conditions into a level from Lv.1 to Lv.10. The calculation runs on your device with fixed rules; answers are not sent to AI or a server (only one anonymous count, such as the level, is sent; see the Privacy Policy). The level boundaries have not yet been checked against the real distribution of users.

11. My Job-Market Rank

36 questions (split into five paths by the first answer: student, graduating soon, looking for a first job, changing jobs while employed, returning to work) produce scores on three axes — school (a tier, not the school’s name), experience and skills — which are compared with a synthetic group modeled on all job seekers in that country (students, new graduates and experienced workers) to assign Tier 1 to 10. The weight of the three axes differs by country (school, experience, skills: Korea 25/40/35%, US 15/45/40%, Japan 25/30/45%, Taiwan 20/40/40%). Goal conditions, field conditions and age are not part of this tier; they feed a separate “how hard is my goal” result. It is not a probability, and reference values such as the mix of job seekers contain many estimates. Answers are calculated only in your browser and are not sent to a server.

12. Data sources

• Korean Standard Classification of Occupations (KSCO), 7th revision (Statistics Korea): used to define the scope of the Korean job list.
• US Standard Occupational Classification (SOC 2018): used to link jobs to US statistics and O*NET data.
• O*NET 29.1 (US Department of Labor, Employment and Training Administration): task lists and interest profiles by occupation. The job interest scores in the career quiz come from here.
• US Bureau of Labor Statistics (BLS) Employment Projections 2024–2034: the “Growth 24–34” figure on job pages. It is a US value linked through SOC codes and is shown for reference on the Korean, US and Japanese pages. Taiwan pages leave it blank when there is no Taiwanese figure.
• Microsoft Research “Working with AI” data (arXiv 2507.07935, CC BY 4.0): AI applicability scores by occupation. When the baseline (Korean) jobs were judged, the judging AI was also shown a reference value built from this score (a percentile adjusted by the O*NET share of physical tasks) together with the BLS growth rate, with the instruction not to follow it and to judge independently. The score is not part of any formula that computes the years.
• Taiwan: facts about licences, rules and programs were checked against public sources such as the Laws & Regulations Database (全國法規資料庫), the Directorate-General of Budget, Accounting and Statistics (行政院主計總處), the Workforce Development Agency (勞動部勞動力發展署), the Ministry of Examination (考選部), the Ministry of Health and Welfare (衛生福利部) and the Ministry of Education (教育部).
• Universities: Korea’s university information disclosure service (대학알리미), education statistics, and official announcements from the Ministry of Education and institutions. Sources for each school are listed on its result page.
• Regions: resident registration population statistics (Ministry of the Interior and Safety), Statistics Korea’s KOSIS, the school information service (학교알리미), the Health Insurance Review and Assessment Service, local government and education office materials, local finance disclosures, and news reports. Sources for each area are kept in the judgment records.

This page includes information from the O*NET 29.1 Database by the U.S. Department of Labor, Employment and Training Administration (USDOL/ETA). Used under the CC BY 4.0 license. O*NET® is a trademark of USDOL/ETA. YOU’RE NEXT (Opening Step) has modified all or some of this information. USDOL/ETA has not approved, endorsed, or tested these modifications.

13. Who made the judgments

Data preparation and judgments were done by AI models. Afterwards, separate AI review steps and automated checks (format, value ranges, ordering within a field, links to evidence) found problems that were then corrected. No item has yet been reviewed by a person, such as a subject expert or someone who lives in that country. The explanations on job and major pages are also written by AI using only the judgment data, and they are checked automatically so that no number outside that data is added.

14. Limits and reporting errors

• Every outlook is an estimate and is uncertain. It can change as technology, the economy and rules change, and new judgments can change the numbers.
• AI judgments can be wrong. The model may misread material, miss recent changes or rely on something that is not true.
• The numbers describe the average course for a job, major or area. Even within one job, results can differ a lot by company, location and personal experience.
• If you find an error, email 2024opennext@gmail.com with the page address, what is wrong and, if possible, a source. For university data, use “정정 요청” (request a correction) on each university’s result page. We review reports and fix what needs fixing. Details: Contact.

About · Methodology · Terms of Use · Contact · Privacy Policy
YOU’RE NEXT · Operated by Opening Step · 2024opennext@gmail.com