Have you noticed how the same degree that used to feel like a golden ticket now comes with a longer job hunt and a thinner first paycheck? I keep hearing that question from parents, recent grads, and even hiring managers who cannot quite explain the shift. A new look at early-career records after late 2022 points to a blunt pattern: the college majors most exposed to generative tools saw weaker immediate employment and lower starting pay than fields sitting farther from the technology. That is not a small wrinkle. Researchers compared it, in size, to the earnings hit that comes with graduating into a large recession.
What The New Graduate Data Actually Shows
The core finding is simple enough to say out loud and harder to sit with if you just finished a technical program. Since the public arrival of widely used generative systems in late 2022, graduates in the most exposed tenth of majors were about five percentage points less likely to land work soon after commencement. Their first-quarter earnings also dropped by roughly thirteen percent relative to classmates in less exposed fields. In my experience, families hear “tech degree” and still picture a bidding war. The records tell a colder story for the first job, not the tenth.
The analysis drew on administrative files covering a large slice of bachelor’s degrees awarded from 2016 through 2024. That window matters. It includes several pre-tool cohorts and several post-tool cohorts, which lets the comparison isolate a break rather than a slow drift. The sharpest drops showed up in the three majors most tied to the technology: computer science, computer and information systems, and computer engineering. Fewer of those graduates found work right away. Those who did often started at lower pay.
This earnings decline is comparable in magnitude to the earnings losses associated with graduating into a large recession.
I do not treat that line as theater. Recessions usually hit almost everyone. Here the bruise is concentrated. That concentration is why the result feels like a shock even if the broader youth labor market was already cooling for other reasons.
Why Starting Pay Fell So Fast
Two channels explain most of the pay drop. First, industries that used to bid aggressively for these graduates started paying less at the entry gate. Second, more new graduates drifted into lower-paying sectors. About half of the earnings decline came from that industry mix shift. Retail trade and accommodation and food services show up in the story not because those fields suddenly became dream destinations, but because they absorbed people who could not clear the usual high-pay door.
That mix effect is easy to miss if you only watch headline salaries at a handful of famous firms. Averages move when the destination map changes. If a larger share of a cohort takes a first role in a thinner-margin industry, the whole group’s first-year numbers sag even if a lucky minority still lands well. Perhaps the most interesting aspect is how quickly that mix changed after 2022. Markets do not usually reprice a whole major in a couple of cycles unless hiring managers rewrite what “entry level” means.
Over time the gap narrowed. Two years after graduation, the earnings difference between the most and least exposed majors shrank to about five percent. Job switching also rose, which hints that some people climbed after a messy start. There is not yet enough distance in the data to say they fully catch the cohorts who entered before generative tools became everyday workplace software. That unfinished catch-up is the part I would watch if I had a child in a computer program right now.
The Majors In The Blast Radius
It is tempting to lump every campus program into one panic headline. The records do not support that. Exposure is uneven. Computing fields sit at the top of the list because a large share of early tasks in those jobs can be drafted, checked, or sped up by current tools. That does not mean the work vanished. It means the first rung got narrower. Employers can ask a smaller junior team to cover more of the grunt layer, then hire fewer people at the old starting rate.
I have found that students hear “exposed” and think “obsolete.” Those are not the same word. A major can be highly exposed and still valuable if the graduate can do the parts machines handle poorly: messy requirements, security judgment, systems that fail in the real world, and communication with non-technical clients. The market is not deleting computer science. It is getting pickier about who gets paid like a computer scientist on day one.
| Group | Early job signal after 2022 | Pay pattern |
| Most AI-exposed majors | Lower chance of work right after graduation | About 13% lower initial quarterly earnings versus less exposed fields |
| Computing trio | Largest drop in placement and pay | More movement into lower-paying industries |
| Less exposed majors | Comparatively steadier first foothold | Smaller relative hit at the starting gate |
| Same cohorts two years later | More job switching | Gap narrows to roughly 5% |
Read that table as a snapshot, not a life sentence. Two-year catch-up is real. Full catch-up is unproven. If you are advising a sophomore, that distinction should change the conversation from “switch majors tonight” to “build proof you can work above the tool.”
What Students Say They Would Do Again
A separate, much smaller survey of graduates adds a human layer the administrative files cannot. None of the computer science or information technology respondents said the technology made them wish they had picked a different major. Contrast that with communications, journalism, and media graduates, where forty-three percent said they would choose differently after weighing the same shift. Biological sciences followed at twenty-eight percent. Psychology sat at twenty-two percent.
Among the full sample, seventy-seven percent still called college the right choice. Twenty-two percent said they would have picked another major with what they know now. Those numbers come with a warning label I will not bury: group sizes were tiny, often one or two dozen people per major cluster. I would not bet a household budget on a poll that small. I would treat it as mood, not measurement.
Still, the mood is useful. Technical graduates can feel the hiring freeze and still defend the major. Media graduates can feel the same tools eating drafts and wish they had a different base. That split matches what I hear in conversations. People closest to the code often believe the tool is a lever. People closest to the blank page often believe the tool is a rival. Both can be true in different offices.
AI Is Not The Only Headwind For Young Workers
It would be sloppy to pin every weak campus-to-job handoff on one technology wave. An earlier look at workers aged eighteen to twenty-four found that falling job openings did more damage than tool-related displacement between 2023 and 2025. Growth in roles tied to the technology appeared to crowd out some new entrants, especially recent graduates, but that channel was smaller than the broad drop in labor demand.
Recent graduates still posted a national employment-to-population ratio around seventy-nine percent in early 2026. That is not a collapse. It is a market that stopped rolling out the welcome mat. Economists have described a low-hire, low-fire climate. Firms keep the people they have. They post fewer seats. They feel less pressure to take a chance on someone with a thin resume. Newcomers pay the price because newcomers are the people who need the posted seat.
The technology is not wiping out work across the whole economy. It is raising the bar for young workers who need a first foothold, in a market where employers advertise less, offer less, and hire less.
That framing feels right to me. If openings shrink and screening software gets stricter in the same season, a graduate can do everything “correctly” and still wait. The wait then pushes some people into lower-paying industries, which is exactly the mix shift sitting inside the earnings drop.
How A First Job Market Tightens Without A Classic Recession Label
Classic downturns are loud. Headlines shout layoffs. This episode has been quieter. Many firms did not empty the building. They just stopped adding the extra junior analyst, the extra support engineer, the extra content assistant. If you already had a desk, you might have felt busier. If you needed a desk, you felt invisible.
Generative tools turbocharge that quiet freeze in specific majors. A manager who once needed three juniors to clean data, draft tickets, or write first-pass code can now ask two people plus a model. The third seat never opens. No dramatic firing email. Just a requisition that never gets approved. Multiply that choice across hundreds of teams and you get a five-point drop in early placement without a cinematic collapse.
Pay follows the same logic. When the supply of newly minted specialists meets a thinner stack of junior seats, starting offers soften. Some employers still pay well for rare proof of skill. The average does not. Averages include the people who took retail or hospitality work to keep cash flowing while they hunted. Those choices are rational. They also pull the cohort mean down.
What “Exposure” Means In Plain Language
Exposure is not a moral grade. It is a measure of how much of a field’s early tasks overlap with what current systems can already draft, summarize, classify, or generate. High overlap does not equal zero human value. It equals a different bargaining position for the person with only classroom proof.
- High exposure often means junior tasks are easier to automate or accelerate.
- Medium exposure often means the tool helps, but clients still need a person in the loop.
- Lower exposure often means physical presence, regulated judgment, or messy human contexts still dominate.
Campus catalogs rarely teach bargaining position. They teach methods. Methods still matter. Bargaining position decides whether those methods buy a strong first contract. That is why two students with the same transcript can have wildly different first-year outcomes. One shows a portfolio that survives a live problem. The other shows homework that a chatbot can mimic in ten minutes.
The Industry Mix Shift Nobody Wanted To Discuss
Let’s talk about the half of the pay drop that comes from sector switching. It is the least glamorous part of the paper and, to my eye, the most honest. When the preferred industries pause, people still need rent. They take what is open. Open work is not evenly paid.
Retail and food service are not insults. They are functioning labor markets with hours, training, and a paycheck this Friday. They are also not the implicit promise printed on a lot of computing brochures. If a larger share of an exposed cohort starts there, first-year earnings fall even if those workers later climb. The climb is visible in higher switching rates. The delay still costs compounding raises, retirement matches, and the simple confidence that comes from a clean start.
I keep coming back to compounding because early-career pay is not only about this quarter. It sets the base for the next offer. A thirteen percent hole at the start can fade to five percent after two years and still leave a scar in lifetime earnings if promotions stack on a lower platform. We do not have the long panel yet. We should stop pretending the long panel will automatically be kind.
Why Computing Felt Safe And Then Did Not
For a decade, families treated computing as weatherproof. Enrollments swelled. Bootcamps multiplied. High school counselors pointed anxious kids toward code the way an earlier generation pointed them toward accounting. The logic was tidy: software eats the world, so study software.
Then the tool that writes software became cheap. Not perfect. Cheap. Cheap changes staffing math faster than perfect ever could. A messy assistant that drafts seventy percent of a task can still erase a junior seat if a mid-level person will clean the last thirty. That is an ugly sentence. It is also how a lot of teams now talk in private.
Does that make the major a trap? I do not think so. It makes the generic version of the major a trap. Generic means no internship trail, no shipped project, no domain, no proof you can sit with an ambiguous brief. Specific still travels. Security, embedded systems, infrastructure, research-adjacent work, and roles that mix code with regulated industries still look different from “I completed the standard course list.”
What Employers Quietly Changed After 2022
Hiring conversations shifted in three practical ways. First, postings asked for evidence of tool fluency and then discounted candidates whose only samples looked machine-made. That double bind is real. You must use the systems. You must also show judgment the systems lack.
Second, probation got less patient. A junior who needs six months of hand-holding is a harder sell when a model can cover the first draft on day two. Teams still train people. They train fewer people. They want the trainee to arrive half-formed.
Third, compensation bands for raw newcomers softened while pay for scarce specialists stayed firm. That split produces the average decline without contradicting the friend who got a strong offer. Both stories can sit in the same graduating class. Averages hide the fork in the road.
A Practical Playbook For Current Students
If you are mid-degree, panic is a poor advisor. A plan is better. I would start with proof, not slogans.
- Ship work a stranger can inspect. Class projects that never leave the learning portal do not count.
- Pair the major with a domain: health operations, logistics, energy, public records, finance operations, or another messy field.
- Collect references who saw you solve a live constraint, not just a clean homework set.
- Practice explaining tradeoffs out loud. Tools draft. People decide.
- Treat the first role as a bridge if needed, then switch with evidence instead of shame.
None of that requires abandoning a computing path. It requires refusing the fantasy that the diploma alone still opens the old door at the old price. The door exists. The doorman got pickier. That sentence will annoy some readers. It will also save a few people a year of confusion.
Advice For Parents Who Already Paid The Tuition
Parents often want a binary answer: stay or switch. The better question is whether the student can add a second signal before graduation. A minor, a clinic, a research assistant slot, a campus job that touches real users, a summer that produces an artifact. Switching late can work. Switching late without a new signal just resets the clock.
I would also lower the temperature at the dinner table. A delayed first role is not a character flaw. In a low-hire climate, delay is common. The expensive mistake is drifting for eighteen months with no portfolio and no network while waiting for the old market to return on its own. Markets return. They rarely return as the same market.
What This Means For Career Switchers And Mid-Skill Workers
Not every reader is twenty-two. Plenty of adults are staring at the same tools from a mid-career desk. The graduate findings still matter to them because they show how firms now value junior-shaped work. If your role looks like a stack of tasks a model can first-draft, you are in a softer bargaining spot even if your title sounds senior.
The healthier response is not to mock the tools or worship them. It is to move toward work that includes accountability. Someone has to own the error. Someone has to face the client when the draft is wrong. Someone has to know when the output is confidently false. Those someones still get paid. The people who only produce volume feel the squeeze first.
The Catch-Up Question That Still Has No Clean Answer
Two years after graduation, the pay gap shrinks. Switching rises. That is hope with a footnote. Hope, because people are not frozen in the first bad match. A footnote, because we cannot yet see whether the delayed group fully closes on the pre-2022 cohorts across five or ten years.
Why might they not close? Lost months of relevant practice. Weaker first networks. Raises calculated from a lower base. Hiring managers who quietly treat a messy first year as a signal. Why might they close anyway? Tools raise output for people who learn to direct them. Tight markets eventually reopen seats. Ambition does not vanish because the first offer was ugly.
I lean slightly cautious. Not gloomy. Cautious. Early scars in earnings data have a habit of lingering even when people later look “fine” on paper. If later studies show full convergence, I will be glad to say so. We are not there yet.
How To Read The Headlines Without Getting Played
Some coverage will claim the technology killed the college premium. The files do not say that. Most surveyed graduates still called college the right call. Some coverage will claim computing is finished. The files do not say that either. They say the most exposed majors took a concentrated hit at the starting line after 2022, especially in computing, while the broader youth market was already dealing with fewer openings.
Hold both facts. A concentrated hit can coexist with a still-useful degree. A still-useful degree can coexist with a worse first contract. Adults can handle that nuance. Marketing departments often cannot. Your job, if you are choosing a path, is to ignore the marketing and look at first-job mechanics.
A simple way to score a major in this market: 1. How many junior tasks overlap with current tools? 2. How easy is it to show proof beyond a transcript? 3. How many industries will buy that proof? 4. How painful is a delayed first role in that field? 5. What second skill turns exposure into leverage?
If you cannot answer those five items, you are choosing on vibe. Vibe was enough in a hot hiring year. It is sloppy now.
A Note On Fairness And Who Absorbs The Shock
Shocks like this do not land evenly. Students who can afford an unpaid summer, a sixth course, or a family-funded job search will paper over a weak campus-to-work handoff. Students who need wages immediately will take the open job in a lower-paying sector. That is how an economy-wide tool wave becomes a household story about who gets time to wait.
I do not have a tidy policy speech to park here, and I will not pretend otherwise. I will say this: if universities sell computing as a sure escalator, they should also publish first-job dashboards that include time-to-offer, industry mix, and two-year earnings, not just senior-year internship logos. Sunshine would not fix the labor market. It would stop some expensive surprises.
What I Would Tell A High School Senior This Week
Pick the work you can stand to get excellent at, then inspect the first-job market with adult eyes. If that work sits in a high-exposure major, go in with a builder’s mindset rather than a passenger’s. If that work sits farther from the tools, do not assume safety. Lower exposure is not a hammock. Every field has its own squeeze.
College can still be a good bet. The survey majority said it was. The better bet is college plus proof plus a plan for a slow first year. That package is less romantic than the old brochure. It matches the records we have.
And if you already graduated into this mess? You are not late to a party that ended. You are early to a market that changed the cover charge. Gather evidence. Switch if the first seat is a dead end. Do not wait for a headline to declare the old path reopened. Headlines are slow. Rent is not.
The Quiet Conclusion Hidden In The Numbers
The shock is not that machines exist. The shock is that the majors sold as machine-proof took the first punch. Employment right after graduation slipped. Starting pay slipped more. Half of that pay slip came from people landing in thinner industries. Two years later the gap shrank but did not vanish from view. Young workers also faced a broader drought of openings that had little to do with chat windows.
Put those pieces together and you get a labor market that is less dramatic than a science-fiction purge and more punishing than a normal campus season. The bar moved. The welcome mat shrank. The degree still has value if the graduate can work above the draft. That last sentence is the whole article, stretched across a few thousand words because families deserve the texture, not just the scare.
I will keep watching the two-year and five-year marks. If the catch-up becomes complete, the story softens. If the scar remains, the story hardens into a warning about first contracts in exposed fields. Either way, the useful move is the same today. Treat exposure as information. Build proof. Assume the first foothold may be steeper than the catalog promised. Then climb anyway.
