AI Shrinks the Coding Jobs Computer Science Was Built For

Only 28% of UK computer science graduates found professional work as coders or programmers in the latest survey, down from about 40% in earlier years. That is the first hard look at the class that left university in 2024, surveyed 15 months later as AI tools took on routine software work those degrees were sold on.

Matt Hiely-Rayner, director of Intelligent Metrix, compiles the 2027 university tables published on 12 September 2026. He said coding and software development were the fastest-falling graduate occupations in that data, and that the share of computer science graduates in any graduate-level job has dropped from more than 60% two years earlier to 50%.

The Fastest-Falling Graduate Occupation

The figures sit inside the Graduate Outcomes survey run by the Higher Education Statistics Agency, which is part of Jisc. HESA published the 2023/24 results on 4 June 2026, after surveying 1,002,175 graduates between 2 December 2024 and 30 November 2025. Some 353,755 people sent usable replies, a 35% response rate, which matches the pool Intelligent Metrix then sliced by subject and job.

Across all subjects, 87% of respondents were in some form of work or further study. Full-time work fell to 57% from 59%, and 7% said they were unemployed and looking for work, up from 6%. Charlie Ball, Jisc’s head of labour market intelligence, called that a cooler market rather than a collapse when the national numbers landed in June.

Computing is the bit that does not look like a gentle cooling. Among respondents whose degree sat in the computing group, 2,255 of 19,910 people with known outcomes said they were unemployed and looking for work, the highest share of graduates out of work HESA recorded for any subject, at 11%. Medicine and dentistry, and veterinary sciences, were at 2%.

COMPUTER SCIENCE GRADUATE OUTCOMES

Hiely-Rayner’s coding share is the number that will follow this cohort. A computer science degree used to move four in ten people into the job the prospectus still photographs. In this survey it moved fewer than three in ten, and only half into any graduate-level post at all.

With AI providing such responsive and cheap grunt work in this field, it’s hard not to conclude that this is behind the trend.

Matt Hiely-Rayner, director, Intelligent Metrix

Ball is more careful, and he is the person who reads these tables for a living. He said it is clear that something happened to the software developer labour market, that he is reluctant to say AI did it because there is little direct evidence, and that it is still hard to escape the conclusion that AI played a role. Pure coding jobs, he said, seemed in particularly short supply in 2025. The field has not vanished. The first job title has thinned.

Computing Courses Are Still Near a Record

The people on campus have not thinned with it. BCS, the Chartered Institute for IT, analysed UCAS placements and found 31,670 UK students accepted onto computing degrees in 2025. That was 3% lower than the previous cycle and still the second-highest total on record.

Among UK 18-year-olds, acceptances fell 9% to 16,865, the first drop in seven years, while UK 18-year-old acceptances for all degrees rose 3.5%. Computing slipped one place to the eighth most popular subject for that age group. Including international students, 36,660 people started a computing first year, about 4% lower, after international starters fell 7% to 4,990.

The mix inside those numbers is the tell. Straight computer science took 15,760 UK students of all ages, down 10% from 17,430. Software engineering fell 5% to 4,985. Computer games and animation fell 6% to 5,335. Artificial intelligence was the specialism that grew, to 1,165 UK acceptances from 820, a 42% rise.

UK COMPUTING ACCEPTANCES IN 2025

Half of UK students starting a computing degree in 2025 came from the two most disadvantaged quintiles of the Indices of Multiple Deprivation, against 41% of students overall. Computing has been one of the subjects through which teenagers from poorer postcodes were told they could buy a professional salary. They are still being placed onto those courses in huge numbers, just as the job that made the pitch is the one shrinking fastest.

Why Junior Coding Roles Dried Up

Vacancy data from before this survey already showed the junior rung giving way. In March 2025 the National Foundation for Educational Research published an analysis of Adzuna listings, funded by the Hg Foundation, and found that programming adverts fell by 68% between 2019/20 and 2024/25 on NFER’s unrounded averages. The rounded live-advert averages in that report ran from 42,000 programming roles to 14,000.

Core tech adverts were 50% lower than before the pandemic. All-job adverts were down 31%. NFER also reported a 10% fall in the number of people employed as software developers between early 2023 and 2024. Among programming ads that stated a seniority, there were four times as many senior roles as junior or entry roles in 2024/25, against twice as many in 2016/17.

HOW THE JUNIOR MARKET TIGHTENED

1. 2019/20: Average live programming adverts around 42,000 on Adzuna, per NFER.
2. Early 2023 to 2024: Employment of software developers down 10%.
3. 2024/25: Average live programming adverts around 14,000; senior ads four times junior ads.
4. 4 June 2026: HESA records computing at 11% unemployed and looking for work, the highest of any subject.
5. 12 September 2026: Intelligent Metrix puts computer science graduates in professional coding jobs at 28%.

NFER listed more than one cause: the post-pandemic hiring hangover, high interest rates, offshoring, and AI. Ball’s September reading sits on that pile rather than replacing it. The shape of the damage is still specific. Senior engineers remain hireable. The work that used to be a first job, the ticket-sized tasks a new graduate could not badly break, is the work models now finish before lunch.

That is the complaint working developers have been making in plain language for months. Firms are not deleting software as a trade. They have stopped paying someone to learn it on the clock, because a model will draft the boilerplate, the unit tests, and the first pass at a review. A London product engineer put the arithmetic brutally in August: AI does the grunt work now, so why pay a graduate to learn. The next problem is older than any model. A senior is a junior who got ten years. If far fewer people get those years, the hole shows up later, in every team at once.

WHERE EXPERTS DISAGREE

• Intelligent Metrix: Hiely-Rayner treats cheap, fast AI grunt work as the force behind the coding drop.
• Jisc: Ball says there is little direct evidence AI did it, then adds that it is still hard to escape that AI played a role in a wider cooling of graduate hiring.

Both men are describing the same tables. One names the tool. The other refuses to pretend the tool is the only thing in the room, then names it anyway.

A Pivot Into Security That Barely Shows

Ball said there are signs computer science graduates are moving into neighbouring work such as cybersecurity and network engineering. That would be the healthy version of this story: the degree still pays, the job title on the LinkedIn line just changes. The official record of the last detailed occupation split does not yet show a rush through that door.

In the government’s Cyber Security Skills in the UK Labour Market 2025 report, which used Graduate Outcomes data for the 2021/22 academic year, 38% of computer science graduates in work were already coded as programmers and software development professionals, down from 42% for 2019/20. Only one percent entered cyber security roles from computer science. Among people who had actually studied cyber, 31% landed in cyber professional jobs.

WHERE THE OVERFLOW WAS SUPPOSED TO GO

• Cybersecurity: Ball sees graduates diversifying into it, but the 2021/22 occupation split still put only 1% of computer science graduates in cyber professional jobs.
• Network engineering: Named by Ball as a second related destination, not as a mass employer of the 2024 class.
• Graduate-level work of any kind: Down to 50% for computer science, so a large minority never make the pivot at all inside 15 months.

The 1% figure is older than this survey, so it cannot contradict a 2025 shift on its own. It can show how small that door was when coding still took 38% of the employed CS class. If cybersecurity is now the shock absorber, it is starting from almost nothing.

How Birmingham Is Rewiring Its Degrees

Universities can see the same labour market their students are about to enter. Their first institutional answer is not to shrink computing. It is to put AI on more transcripts.

The University of Birmingham rose to 14th in the 2027 Intelligent Metrix tables, its highest place, after strong student satisfaction. On 8 July 2026 the university said it had ranked first in the Russell Group in the 2026 National Student Survey, with overall satisfaction of 87.4%, and that Artificial Intelligence and Computer Science was among 16 subjects placed first in that group.

Adam Tickell, Birmingham’s vice-chancellor, said the university was weaving employer-demanded skills into courses and offering undergraduates an extra year, an intercalated year, in subjects such as AI and data science without the usual prerequisites. The School of Computer Science already runs a Year in Artificial Intelligence and Data Science for students on any subject except computer science, taken in the penultimate year, with a 55% second-year average as the academic bar.

The year is 120 credits of compulsory modules: artificial intelligence and machine learning, programming for AI and data science, storing and managing data, current topics, data analytics, and visualisation. It is assessed pass/fail, it does not feed the home-degree classification, and it costs an extra year of tuition, with a Student Loans Company year of support for those who qualify. Successful students graduate in their original subject “with a Year in Artificial Intelligence and Data Science.” Applications for 2026 entry opened in March 2026.

This is about equipping people for the world they are in, whether or not we like the world we are becoming.

Adam Tickell, vice-chancellor, University of Birmingham

The design is revealing. Computer science students are not the target. Everyone else is being offered a crash year in the tools that are eating the first CS job, while CS itself remains a near-record product. Birmingham is not wrong to teach people how to use the models. It is answering a collapse in junior coding demand by spreading AI literacy, not by promising another 40% of CS graduates a programmer title.

The Same Squeeze Hits Economics

Intelligent Metrix also found weaker employer demand for well-paid graduate roles in financial categories, including economists and management consultants. That is a second subject that used to be treated as a safe professional ticket, now meeting the same class of tools: models that draft analysis fast and cheap.

Ball has been asking finance employers what they are actually doing with those tools. Most, he said, told him they are not replacing jobs with AI. They are changing the jobs, and they want new hires who already know the tools. They are not unanimous. Some firms with an appetite for risk are using AI and laying people off, hoping it gives them an edge. Ball’s warning is that the experiment is still too new to treat as a settled headcount plan.

It shows the signs of an industry that is changing, and graduates are feeling the sharp end of that because they are the largest group of new employees. It’s not collapsed entirely, there are still roles available, but pure coding jobs seem to have been in particularly short supply in 2025.

Charlie Ball, head of labour market intelligence, Jisc

The finance version of the same sentence would swap “pure coding” for the junior analyst tasks that used to train economists. The common mechanism is the bottom rung. The tasks a new graduate could be trusted with, because they were bounded and checkable, are the tasks a model can be trusted with for the same reason.

Seniors Have to Come From Somewhere

None of this makes a computer science degree worthless. Median full-time pay 15 months after graduation, across all subjects, was £30,000 in the HESA release, up from £28,500, and 75% of working graduates were in high-skilled jobs, down one point. Ball’s June line still holds for the average graduate: this is a softer market, not a structural collapse.

Computer science is where the average stops hiding the change. The degree was marketed as a machine for producing programmers. The class of 2024 produced far fewer of them. Universities still placed 31,670 UK students onto computing courses in 2025, and the specialism that grew was AI. Those students will walk into this labour market as it currently exists, not the one on the older prospectuses.

The quieter cost sits a few years out. If junior coding work is the training loop, and models now occupy that loop, firms can look fully staffed on senior organograms while hiring almost no one who will be senior in 2032. Ball is right that the occupation has not collapsed. The first job that used to feed it has, and the courses that feed that job are still full.