Open research on AI, work and education

How AI is changing work, and how people and colleges can keep up.

Workforce 2100 studies how AI is changing work, from today's jobs to the long horizon of this century. We forecast which occupations, academic programs and places change most as AI spreads. We measure what AI is really worth once human checking and errors are counted. And we build frameworks and instruments to teach and assess AI fluency. Papers, models and data, all in the open.

830occupations modeled to 2035
1,432academic program families mapped
51states and DC compared
10papers and studies, shared openly

What we study

Four connected areas of work. A framework explains how AI enters work. We use it to measure the effect on the economy, then on jobs and the academic programs that prepare people for them. The last area turns what we learn into how we teach.

All research →
  1. Framework
  2. Economy
  3. Jobs & programs
  4. Education
01 · FRAMEWORK

How AI enters work

How can we understand, and predict, where AI takes on work?

A framework for the impact of AI as it moves from generative tools to agents that act on their own and, next, to robots. AI takes on a task when it is cheaper than a person after paying for review and for errors. How much people trust it depends on what an error costs and how complex the work is. The human side of that equation is AI fluency: judging what to delegate and checking the result.

02 · ECONOMY

What AI is worth, and where

How much value does AI really create, and who feels it first?

We apply the framework across the U.S. economy. Only about a fifth of AI's headline value is realizable once checking and errors are counted. Automation could remove about 13.6% of labor demand by 2035 before growth, and the wages at stake are highest in high-wage states. We also show how more fluent reviewers raise the value AI delivers.

03 · JOBS & PROGRAMS

Where work is moving

Which jobs grow, shrink or change, and what does that mean for the programs that train for them?

We trace the impact to 830 occupations, showing how much of each job AI takes on and who should stay in charge of the decisions. Then we map it to 1,432 academic program families at every award level, sorting them into Invest, Reinvent and Review tiers so colleges can plan their portfolio.

04 · EDUCATION

Teaching for AI fluency

How should we teach, assess and design programs for work shared with AI?

An academic framework that matches AI use to the stakes of a task and to what the learner still needs to learn, with an instrument to measure that judgment. We test it with leaders and practitioners, track how colleges commit to it, and turn it into new curriculum, starting with an Applied AI associate degree.

How we work

Labor-market modeling

Task-level models built on federal employment and wage data, O*NET task data and observed AI use, run across many scenarios with ranges, not single numbers.

Document analysis

Coding college mission statements, strategic plans and core competencies across U.S. community colleges to track institutional commitments.

Surveys & instruments

A diagnostic for academic leaders and an expert review of our AI fluency assessment. Take part in a survey →

Open science

Preprints, pre-registered studies and public data sets with DOIs, so others can check, reuse and extend the work.

Work to 2035

Search 830 occupations and 1,432 academic programs.

See how much of each job's work AI is likely to take on by 2035, who should stay in charge of the decisions, which programs to invest in or reinvent, and which states feel the change first.

Our framework

AI fluency is a judgment, not a course.

Most AI programs stop at the second rung: knowing what the tools are and how to use them. Work now asks for the third: deciding when to delegate to AI, checking the result and answering for it. That is what we teach toward and what we measure.

RUNG 3
AI Fluency

Deciding what to delegate to AI, checking the result, and answering for it.

RUNG 2
AI Literacy

Knowing what AI tools are and how to use them.

RUNG 1
Information Literacy

Finding, evaluating and using information.

Featured papers

All research →
Workforce & programsPreprint · SSRN

Where Work Is Moving: Occupations, Academic Programs and Places in the 2035 AI Workforce Scenario

Thalassinidis · 2026

A roadmap for higher education: which occupations shrink or grow under the 2035 AI scenario, what that means for academic programs, and which places feel it first.

Framework + instrumentPreprint · EdArXiv

AI Fluency Under Delegation: A Consequence-Based Framework and Assessment Instrument for AI Fluency in Agentic Work

Thalassinidis & Nathanail · 2026

A framework for AI fluency when AI agents do part of the work: deciding what to hand over, checking what comes back, and owning the outcome. It comes with an assessment instrument to measure that judgment.

Questions

About the research

Who is behind this site?

workforce2100.org is the research site of Dr. Angelo E. Thalassinidis, with Dr. Lance Bowen, his fellow cofounder of Workforce 2100. Papers written with co-authors credit all their authors. See the About page.

Is the methodology public? Can I reuse it?

Yes. The methods, our own data and the code are published under a Creative Commons Attribution license (CC BY 4.0). Anyone can read, test and build on them, as long as they credit the source. Third-party data keep their own terms; see License and data credits.

Why give it away?

Because trust matters more than secrecy, and because the pace of change ahead is likely to outrun any one organization. Locked-up methods will not help colleges, states and workers adapt in time. Open methods let everyone check the work, learn faster and move forward together.

Are the papers peer reviewed?

They are posted as preprints, with data and code on OSF where available. A preprint has not been through journal review unless the paper says so.

How certain are the numbers?

They are estimates from our model and today's data, with ranges, not predictions or advice. Every figure rests on stated assumptions, and the Work to 2035 page shows the ranges and the slow-recovery case.

What is not included yet?

Robots, humanoids and other physical automation; agentic AI, which carries out multi-step work on its own; wages responding to AI; and local plans such as a new plant or a base closing. Occupations and programs tied to physical work may look less exposed than they will.

How often is it updated?

Quarterly, as new federal data are released, such as county jobs by industry and how fast employers are adopting AI. Each update is dated, and the full model is re-run once a year when the BLS employment projections, OEWS wages and IPEDS completions arrive.

Using the data

Why is my labor market a commuting zone, not my county?

Many people live in one county and work in another. A commuting zone groups the counties that share a labor market, so it is the closest match to the market a college serves.

What do Invest, Reinvent and Review mean?

They are signals from the model, not recommendations. Invest: our model expects demand for graduates to grow 10% or more by 2035. Reinvent: between −10% and +10%, where the work itself is changing. Review: expected to fall more than 10%.

What is the AI line?

According to our model, the point where AI can do half or more of graduates' task time. Programs above it likely need new curriculum whatever their group, including growing ones. Our research suggests the work there shifts from doing tasks to directing AI, checking its work and owning the result; see AI Fluency Under Delegation and The Price of Judgment.

My program looks different from what we see locally. Why?

The AI share uses national task data for each program family, graduates do not all work where they studied, and federal data lag by a year or two. Local knowledge matters. If you see something different, please write; it helps improve the next edition.

Can I use a profile in a board presentation or grant?

Yes. Please credit "Angelo E. Thalassinidis, Workforce 2100 (workforce2100.org)" and the edition date.

I found an error. What should I do?

Email athalass@gmail.com with the page and the number. Corrections are listed in the change log.

Going further

Can you look at my institution, region or state in more depth?

The profiles use public data and the same layout for everyone. If you need more, such as your own institutional data, a region you define, or a specific question your board is asking, email athalass@gmail.com and we can look at whether and how it can be done.

Can I take part in the surveys?

Yes. The current studies are listed on Surveys, with how participation works on Research participation.

Can I collaborate on the research?

Yes. Researchers, colleges and agencies interested in working together can email athalass@gmail.com.

Contact

Questions, collaborations or media requests

We welcome questions about the research, requests to use our data or instruments, and invitations to collaborate or speak.