Computer science ranks highest for AI exposure in our analysis, and it still pays well. An exposure score alone cannot tell you what a degree is worth.
Majors whose work is highly exposed to AI are not necessarily losing value. Computer science has the highest AI exposure score in our seven-field analysis (65.2 of 100), yet its programs still show strong pay, with a national median of about $107,009 four years after completion, even as enrollment falls and the early-career job market gets harder. This report explains what to check instead of treating an exposure score as a forecast.
Key Takeaways
- Computer science has the highest AI exposure of the seven fields we analyzed (65.2 of 100) and still shows strong pay: about $71,458 across reporting programs one year out, about $107,009 at four years.
- High AI exposure does not automatically mean poor career prospects. Exposure measures how much AI overlaps the work, not whether the job disappears.
- CS enrollment fell 12.5% from spring 2025 to spring 2026, and early-career graduate unemployment rose to about 7.2%. Our data show those declines; they do not show that AI caused them.
- Before choosing or dropping a major, check what AI headlines skip: the skills employers value and the credentials they still require.
This report is Issue 3 in a four-part series. Issue 1 covered return on investment for individual programs. Issue 2 covered debt and completion. Issue 4 applies the same tests to trades, two-year paths, and apprenticeships. For major-level automation context, see our guide to degrees with weak ROI and high AI exposure.
The Finding: High AI Exposure Hasn't Meant Low Pay for Computer Science
Our analysis of federal education and labor data found that computer science has the highest AI-task exposure among the seven broad fields we studied: 65.2 on a 0–100 scale. That high exposure has not translated into low earnings. CS programs show an EDsmart-calculated first-year median of about $71,458 across reporting programs. Median means the middle result: half of the reporting programs sit above that number and half sit below it. The federally published national median reaches about $107,009 four years after completion.
At the same time, computer science enrollment fell about 12.5% from spring 2025 to spring 2026, and unemployment among early-career CS graduates rose to about 7.2%, up from roughly 4.3% a decade earlier. Exposure, earnings, enrollment, and hiring conditions are moving in different directions at once.
That is the point of this report. High AI exposure does not automatically make a degree worthless, and an exposure score by itself cannot tell you whether a major is worth it. AI can change a lot of the work people do without eliminating the need for people trained to do that work.
The sections below walk through the evidence: exposure scores for all seven fields, the computer science numbers in detail, and what students should check instead of treating an exposure score as a forecast.
Which College Fields Are Most Exposed to AI?
Task content varies by field. This index shows where AI overlaps the work people in each field typically do, not which majors to avoid.
We built an original EDsmart index from O*NET 28.0 Technology Skills data. We matched each of the seven college fields with two to four jobs that commonly represent that field (their O*NET-SOC codes). We then measured how much the technology used in each job overlaps with what today's AI tools can do. Each job received a score from 0 to 100 compared with other jobs in the O*NET database. We averaged those job scores to create the score for each field. The score measures overlap with AI. It does not predict how many jobs will disappear.
| Field family | Exposure index |
|---|---|
| Computer Science | 65.2 |
| Liberal Arts | 59.3 |
| Fine Arts | 47.7 |
| Engineering | 39.2 |
| Healthcare | 34.3 |
| Business | 32.9 |
| Education | 21.1 |
Computer Science ranks highest; Education ranks lowest among mapped families. That is the core paradox: high exposure does not automatically mean low value. CS exposure near 65.2 pairs with an EDsmart-calculated first-year median of about $71,458 across reporting programs and a federally published four-year national median of about $107,009. In these data, the most exposed field is also a relatively high-paying one.
Why is liberal arts second at 59.3? The occupations we map to that family are writing- and editing-heavy, such as writers, authors, and editors, and text is exactly what current AI tools generate most readily. The label is broad, though. Liberal arts graduates work across far more occupations than the codes we mapped, so treat one-decimal scores for broad families as estimates, not precise measurements. That also means the chart does not say every liberal arts major faces the same level of AI exposure.
An exposure score raises questions; it does not answer them. Don't rule out a field until you check what graduates from actual programs earn and owe.
See the interactive chart on AI exposure by major.
The Computer Science Paradox: Strong Pay, Falling Enrollment, a Harder First Job
Computer science is where the numbers pull in different directions most visibly. Typical pay looks strong. Entry-level hiring has tightened. Enrollment is falling. Each number below is real, and the story connecting them is less simple than most headlines suggest.
Earnings. For computer science programs (federal program code CIP 11.07), the federal dataset does not publish a national one-year median, so EDsmart calculated the median across reporting programs: about $71,458 one year after completion. The federally published national median four years after completion is about $107,009.
The highest single reporting program reaches about $173,344 one year out. That is the top of the range from a single school's graduating group, not a typical outcome. We cite it to show the spread across programs.
Hiring. Georgetown CEW's Major Payoff puts early-career computer science graduate unemployment at about 7.2% (2021–2023), up from about 4.3% (2013–2015). One timing detail matters: that rise began before the current enrollment decline, which complicates the tempting story that AI arrived, graduates could not find work, and students fled. Our data do not show that events happened in that order.
Enrollment. National Student Clearinghouse data show computer science (CIP 11.07) enrollment fell about -12.5% from spring 2025 to spring 2026 (217,482 to 190,285 students). The broader computer and information sciences family (CIP 11) fell about -8.4% (628,770 to 575,747).
The enrollment data show a decline. They do not establish that AI caused it. Application cycles, hiring headlines, tuition costs, and how many students programs can accept all move enrollment too.
“AI has automated tasks that I used to see junior Computer Science hires doing its first year on the job. The candidates getting placed are now those who can review AI-generated code with an eye for catching errors. That’s the kind of judgment your data shows – the premium on top of just doing work - is the ability to apply good judgment.”
— Devin Hornick, Co-Founder & Partner, KORE1; President, Orange County Technology Professionals
High typical pay and a harder first job market can both be true. Licenses, accreditation, clinical training, and finishing the degree still matter even when AI reshapes day-to-day tasks.
“AI exposure, earnings, enrollment, and hiring do not move together in a simple way.”
Computer science makes that visible.
An exposure score is not a verdict, and strong typical pay is not a guarantee. Finishing the degree and earning the required license may still be necessary before students can enter many fields.
What Should Students Check Instead? Skills and Credentials
If an exposure score cannot tell you whether a major is worth it, what should? Two questions most AI headlines skip: Do employers still pay for the skills this field builds? And do employers still require a credential (a degree, license, or certification) to enter the field? In this series we call those the skill premium and the credential premium. They are categories we use to organize evidence. This analysis does not calculate either premium as a single number.
On the pay side, job-level data show how wide the gaps can be. In BLS May 2024 OEWS data, the median wage across STEM jobs runs near $103,580 vs. $48,000 for non-STEM jobs. That is a pay difference between groups of jobs. It is useful context, but it is not a measured skill premium, not a major-by-major map, and not proof that a degree alone unlocks pay.
College Board Education Pays 2026 (Figure 2.9) reports worker medians by field of work, not program outcomes after graduation. Performing arts workers earn near $44,000; computer science workers near $87,000.
The Lumina–Gallup 2026 State of Higher Education AI report finds about 57% of students using AI daily or weekly for coursework. Tools spread fast. Credentials and licensed pathways still gate many jobs.
Credentials employers require
Many fields still control entry through official program approval (accreditation), supervised clinical hours, required professional licenses, and hiring paths that ask for a degree, even when day-to-day tasks change. Education is the clearest example in our data: it ranks lowest on exposure (21.1) and still requires a license to teach.
One honest caveat: this analysis does not measure a credential premium directly, for example by comparing similar workers with and without a degree. Treat it as a question to ask about your target field, not a number we calculated.
“It’s less about what colleges should be teaching students, and more about how. Each quarter, I change one or more lessons because students' use of LLMs is evolving so quickly in pursuit of "the right answer." The colleges that best prepare students will require leveraging AI in project-based coursework, supported by increasingly powerful models, and developing the durable problem-solving skills employers will increasingly demand. This must be the new signal of a higher ed credential.”
— Jeff Eyet, Faculty Lead for AI Curriculum, Campus; Lecturer, UC Berkeley Haas School of Business
Skills That Remain Common Across Occupations
When AI handles more routine analysis, which human skills still show up in job requirements? We counted how often each skill appears across occupation profiles. The counts below measure how common a skill is. They do not show whether AI has changed the pay attached to it.
O*NET rates how important each skill is to each occupation. We counted how often a skill appears among the five most important required skills per profile, across 1,016 occupation profiles. We capped the count at five to focus on the skills O*NET rates most central to each job. One occupation lists several skills, so counts overlap. This is a baseline for talking about what schools teach, not a live feed of employer demand.
| Skill | Occupations listing in top 5 |
|---|---|
| Active Listening | 738 |
| Speaking | 619 |
| Critical Thinking | 575 |
| Reading Comprehension | 567 |
| Monitoring | 277 |
| Writing | 227 |
| Operations Monitoring | 186 |
| Social Perceptiveness | 157 |
| Operation and Control | 135 |
| Service Orientation | 105 |
These results do not show that AI has increased the pay attached to these skills. They show that active listening, speaking, critical thinking, and reading comprehension remain widely required across occupations even as AI capability expands. Their broad use across occupations is why they are worth watching, and why they sit at the center of the skills question above.
“As AI becomes more capable, professionalism is becoming increasingly valuable—and noticeable. That means more than workplace etiquette; employers are looking for graduates who communicate clearly, think critically about AI-generated information, write persuasively, and navigate cultural differences with awareness and judgment. AI can produce an answer or recommendation, but we still need to determine whether that work is credible, appropriate, and compelling for a particular audience. Those distinctly human judgments will increasingly separate capable professionals from people who simply use the latest tools.”
— Casey Evans, Interim Dean, Kogod School of Business, American University
“Discernment is holding a standard and applying it to work you did not produce. AI output looks finished, which is why judging it is now the scarce skill, and why critical thinking sits in 575 occupations in EDsmart's data. Test yours: ask your question again with the opposite premise. If the answer flips, you learned about the tool, not the topic. Writer Julia Galef calls that instinct the “scout mindset,” wanting the accurate answer over the agreeable one.”
— Michelle Marlowe, Director of Education, Pocket Prep
A Six-Question Program Checklist
Before you treat an AI exposure score as a verdict, run every program on your shortlist through six questions.
- Exposure: Where does this field rank on our exposure index, and which tasks does AI actually overlap?
- Skills: Do employers near you still pay well for the skills this field teaches, not only in national headlines?
- Required degree or license: Do you need to finish the program to enter the field?
- Earnings and debt: What do graduates from this program typically earn, and how much debt do they leave with? U.S. Department of Education program data show both.
- Completion: What share of students finish, and what happens if they stop partway through?
- Pay over time: Does pay improve years after graduation, the way CS pay does in our data (about $71,458 in year one; about $107,009 nationally at year four)?
“Colleges should teach students to question answers, not just produce them. AI’s fluency breeds false confidence: a polished response feels finished even when it’s wrong. … Students who ask ‘how do we know that’s true?’ of AI, of textbooks, even of their professors, will graduate with the judgment employers actually pay for.”
— Jason Robinovitz, CEO/COO & General Counsel, JRA Educational Consulting & Score Academy
High exposure with strong pay and a required credential is a different situation than high exposure with weak pay and no license requirement.
What to do next
College still matters, but an AI exposure score alone should not decide which major you choose. Check the skills employers value and the credentials they require, then check what individual programs pay, how many students finish, and what graduates owe from Issue 1 and Issue 2. Issue 4 extends the same tests to non-four-year pathways.
- Run the checklist on every program on your shortlist before you treat exposure as a verdict.
- Pair exposure with pay and debt from U.S. Department of Education program data, not AI headlines alone.
- Ask how many students finish and get hired in fields where a degree or license still controls entry.
- Build the judgment skills that complement AI: checking AI's work, communicating clearly, and knowing your field deeply, not tool use alone.
Limits
- O*NET exposure scores describe task overlap with current AI tools, not forecasts of job loss or hiring risk. Seven field families map to selected O*NET job codes from Technology Skills data.
- Each field family maps to a small set of representative occupations (two to four O*NET codes). One-decimal scores for broad families like liberal arts or business are estimates, not precise measurements.
- Enrollment and unemployment figures describe declines; they do not establish that AI caused them. The rise in CS graduate unemployment predates the current enrollment decline.
- “Skill premium” and “credential premium” are organizing categories, not quantities this analysis measured.
- Federal 1-year program figures use national medians when published. Computer science has no published national 1-year median, so we report the EDsmart-calculated median across reporting programs, about $71,458. The top single reporting program reaches about $173,344.
- O*NET skill counts use the first five required skills per occupation profile (1,016 occupations). Counts overlap; one job can list several skills.
- College Board Education Pays medians measure full-time workers by field of work. They are not program-level earnings after graduation.
- Clearinghouse CIP 11.07 (computer science) and CIP 11 (computer and information sciences) measure different enrollment scopes.
- BLS STEM vs. non-STEM medians are occupation-wide (May 2024 OEWS), not major-specific.
- Georgetown CEW CS unemployment (7.2% in 2021–2023 vs. 4.3% in 2013–2015) uses early-career graduate surveys, not program medians from U.S. Department of Education data.
- We do not yet publish a national underemployment rate by major.
- O*NET skills describe work requirements, not real-time job-posting demand from employers.
Sources
We merged an original O*NET exposure index with U.S. Department of Education program outcomes, BLS occupation medians, Clearinghouse enrollment data, Education Pays worker benchmarks, and Georgetown CEW unemployment series for computer science graduates. The exposure index converts each occupation's overlap with current AI tools into a percentile rank (0–100) and averages the mapped occupations in each field family. EDsmartData pulls for O*NET, U.S. Department of Education field-of-study data, BLS OEWS (May 2024), and Clearinghouse Spring 2026 appendices were frozen in June 2026; prose and expert quotes were updated in August 2026.
- EDsmartData.com — O*NET exposure index and U.S. Department of Education program outcomes.
- O*NET 28.0 Technology Skills and occupation skill profiles.
- Bureau of Labor Statistics Occupational Employment and Wage Statistics (May 2024).
- College Board Education Pays 2026; Georgetown CEW Major Payoff on computer science graduate unemployment.
- Lumina Foundation–Gallup State of Higher Education AI Report (2026).
- National Student Clearinghouse enrollment appendices (Spring 2026 CIP trends).
- Expert commentary: Quotes from higher-education and workforce experts who responded to an EDsmart inquiry, reproduced from their written submissions and trimmed for length; omissions are marked with ellipses and clarifying insertions with brackets. Respondents: Jeff Eyet (Campus; UC Berkeley Haas), Devin Hornick (KORE1), Casey Evans (American University Kogod School of Business), Michelle Marlowe (Pocket Prep), and Jason Robinovitz (JRA Educational Consulting & Score Academy).
Expert quotes are reproduced from written submissions, trimmed for length; omissions are marked with ellipses and insertions with brackets.