Nobody Owns the Gap

This is a compilation rather than a study. We did not run a survey and we did not collect anything ourselves. Everything below is published, everything is cited, the whole table is downloadable at the top of this page, and if you think we have read it wrong you have what you need to say so.

We compiled it because we kept hearing two claims that cannot both be true, from people who were both obviously being honest.

1. The two claims

The first comes from graduates. There are no entry level jobs. You apply to a hundred and hear back from four.

The second comes from the people who hire. We cannot find anyone. Every candidate needs six months of supervision before they are worth the salary.

Read together, they sound like one side must be exaggerating. The data says neither is.

112,720

computing bachelor's degrees awarded in the US in 2022-23, up from 51,696 in 2013-14

7.0%

unemployment among recent computer science graduates

7.8%

unemployment among recent computer engineering graduates

38%

share of employers who say it is easy to find candidates with the right skills

2. Supply more than doubled

The National Student Clearinghouse counts computing bachelor's degrees at 51,696 in the 2013-14 academic year and 112,720 in 2022-23. That is a 118 percent increase in nine years, in one field, in one country.

Everything worked exactly as the advice said it would. A generation was told to learn to code, and it did, at scale, at its own expense.

3. Outcomes went the other way

The Federal Reserve Bank of New York publishes unemployment and underemployment for recent graduates broken down by major, drawn from census data. In the most recent release, recent graduates as a whole sat at 5.6 percent unemployment and 42 percent underemployment, the latter being the highest since 2020.

Computer engineering came in at 7.8 percent. Computer science at 7.0 percent. Both sit inside the worst ten majors in the country.

Look at the other end of that chart before drawing a conclusion, because that is where the actual finding is. Special education, 0.7 percent. Elementary education, 1.2 percent. Nursing, 2.1 percent. Engineering technologies, 1.7 percent.

What those fields have in common is not that they are easier or better paid. It is that in every one of them, the qualification is the licence. A nursing degree ends with a person who can be put on a ward. A teaching degree ends with a person who can be put in front of a class. Somebody, at some point, decided that graduating and being employable had to be the same event, and then built the supervised hours that make it so.

Computing has no such convention. A computing degree ends with a person who understands computing, which is a different thing from a person who can be handed a production system on Monday. The industry has always known this and has always quietly assumed somebody else would fix it.

4. And employers are not lying either

The 2025 Cengage employability survey found that 76 percent of employers were hiring the same number of entry level workers or fewer than the year before, and that only 38 percent said it was easy to find candidates with the right skills. On the other side of the same survey, 48 percent of graduates said they did not feel prepared to apply for entry level roles in their own field, and of those, 56 percent named job specific skills as the gap.

So the shortage is real and the surplus is real, because they are shortages and surpluses of different things. There is a surplus of people who have studied software. There is a shortage of people who can be trusted with software. The distance between those two is roughly a year and a half of supervised, real, consequential work.

That year and a half is the gap. It has no owner.

Universities cannot own it, because the work has to be real and commercially consequential, and a university does not have any. Employers will not own it, because the person is least productive precisely when they are most expensive to supervise, and because whoever pays for that year risks watching the trained engineer leave for someone who did not. Bootcamps compress it rather than closing it, which is why their placement numbers hold up in a hot market and fall apart in a cold one.

Nobody owns the eighteen months between a graduate and an engineer, so nobody is accountable when it does not happen.

5. The compiled table

Unemployment among recent graduates by major, from the New York Fed series. Best and worst ten, plus the all-major figure for reference.

Recent graduates are defined as ages 22 to 27 holding a bachelor's degree or higher. Underemployment for the same cohort, all majors, is 42 percent. Source: Federal Reserve Bank of New York, The Labor Market for Recent College Graduates, drawn from 2024 census data.

6. What this data does not say

Four limits, stated plainly, because a compilation that oversells itself is worse than no compilation.

It is one country. The United States publishes the best instrumented data on graduate outcomes by major, which is why it is used here. The pattern is not unique to it, but this table is not evidence about anywhere else, and we are not going to pretend otherwise.

Major is not occupation. Somebody with a computer science degree working as an analyst is employed, and is counted as employed, and may well be doing something useful. The table measures what people studied, not what they do.

The unemployment number is a snapshot, not a trajectory. It says that this cohort is having a hard time entering. It does not say that computing is a bad field to enter, and the wage data for those who do get in says the opposite.

And we cannot tell you how much of this is machine assistance eating junior tasks. It is the obvious hypothesis, several people we respect believe it, and the honest position is that this dataset cannot separate that from an ordinary hiring correction. We wrote about what the evidence on machine assistance does and does not support separately.

7. What we do with it

We are not a university and we are not a charity, so our interest here is not sentimental. Owning the gap is the most defensible position in this industry, because whoever owns it ends up owning supply.

The way we own it is deliberately unromantic. Apprentices work in small pods alongside a senior lead on real client work that a client is actually paying for, because supervised work on something consequential is the only thing that has ever closed this gap. Progress is gated rather than time served, so the question is never how long somebody has been here. Passing is evidenced, and the evidence is a certificate a third party can verify without asking us, because a claim that only we can confirm is worth nothing to the person holding it.

That last point matters more than it sounds. The reason the gap has no owner is that training is unverifiable, so no employer will pay for someone else's, and no trainee can prove they closed it. Fix the verification and the economics of the whole thing change.

8. Method, and how to check us

Every figure here is from a public source, listed below. The table is available as a CSV from the link at the top of this page, with one row per major, the rate, the group, and the source for each row. No transformation has been applied to any number: they are reproduced as published, and where a source gives a range we have said so rather than picking the middle.

If a figure is updated by its publisher, this page goes stale. We date our papers for exactly that reason, and we will republish rather than quietly edit if a number moves.

References

1. 01National Student Clearinghouse Research Center, computing bachelor's degree completions, 2013-14 and 2022-23, studentclearinghouse.org
2. 02Federal Reserve Bank of New York, The Labor Market for Recent College Graduates, unemployment and underemployment by major, newyorkfed.org
3. 03Cengage Group, 2025 Graduate Employability Report, cengagegroup.com
4. 04Tigotek, Generation Got Cheap. Verification Did Not., on what the evidence for machine assistance does and does not support, /papers/generation-got-cheap