More or fewer programmers needed in 2027? More developers, a narrower way in

Anyone who started a computing degree at a Dutch university of applied sciences in September 2022 picked the right moment. That spring, the Dutch employment agency UWV counted more than 37,700 open ICT vacancies, a peak. One quarter later, ChatGPT came out. Students who finished in four years graduated this summer, into a market where the twelve largest tech companies hire about 65% fewer new graduates than in 2019. Meanwhile, Google says 75% of its new code is written by AI.

So will we need more or fewer programmers in 2027? I put the latest figures from Statistics Netherlands (CBS), UWV, Eurostat, the US Bureau of Labor Statistics and Indeed next to the best research on AI and programming. The answer is less dramatic than the headlines and less comfortable than the reassuring blog posts. More developers will be needed, but the way in for beginners is getting narrower. Those two things are connected.

Figures checked on 29 October 2026, at the original source where possible. Updated on 30 October with the CBS vacancy figures for the third quarter of 2026. Forecasts and my own expectations are marked as such.

More or fewer programmers needed in 2027? The short answer

More software developers, fewer programmers in the narrow sense, and less room for beginners. No official body publishes a forecast for 2027 alone. The forecasts that exist run to 2030 or 2035, and all of them point to growth. Each new round is just lower than the last. In the Netherlands, UWV and the research institute ROA expect the market for software developers to become less tight up to 2030, but not slack. Based on current figures, a fall in the total number of jobs in 2027 is unlikely.

That sounds like a contradictory answer, but it becomes clear once you see that "programmer" is really two occupations.

US statistics measure two occupations, and they are moving in opposite directions

The Bureau of Labor Statistics (BLS) tracks two occupations separately. The software developer works out what a system should do, designs it, connects it to other systems and is responsible when it breaks. The computer programmer writes and tests code to a specification someone else made. In the projections the BLS updated on 27 August 2026, the two go separate ways:

For the decline in programmers, the BLS states the reason plainly: companies use technology, "including artificial intelligence (AI), to automate repetitive programming tasks". Part of that work moves to developers. For the growth in developers, the BLS names, among other drivers, "continued expansion of software development for artificial intelligence". AI removes work on one side and creates it on the other.

One thing stands out when you put the older rounds next to this one. For 2023–2033 the BLS projected +17.9% for developers, for 2024–2034 about +15%, and now +10%. Still faster than average, but a little less each year. That is the honest picture: the profession is growing, just more slowly than predicted.

Most of the drop in job postings came before today's AI coding tools

Indeed measures daily how many software development jobs are posted in the US, with February 2020 set to 100. The series is public at the Federal Reserve Bank of St. Louis, and the timeline is revealing.

In February 2022 the index stood at 229: more than twice as many postings as just before COVID. A month later, the US Federal Reserve started raising interest rates. By the time ChatGPT came out on 30 November 2022, the index had already fallen to 147. At the end of 2023 it was at 72, at a time when AI assistants mostly completed code and answered questions. The bottom came in May 2025, at 61.1. After that it went up, in the same period in which coding agents such as Claude Code took off. On 23 October 2026 the index stood at 80.8.

That does not prove AI is good for job postings. It does show that the story "AI wiped out developer jobs" does not match the order of events. The timing fits better with cheap money becoming expensive and with companies that hired too many people during COVID. According to Indeed Hiring Lab, US software development postings grew by almost 15% since Claude Code launched, while all postings combined fell by 7%.

But look at who is wanted. Of the growth between May 2025 and May 2026, 71% came from senior roles, and 37% of the growth came from postings with "AI" in the job title. At the same time, tech layoffs continue: according to layoffs.fyi, more than 134,000 people had lost their jobs this year by 21 October, more than in all of 2025. Some of the people leaving are not the profile that is being hired.

In the Netherlands the shortage is easing, and AI is not the main reason

The Netherlands is going through a calmer version of the same movement. At the end of September 2026, CBS counted 375,000 open vacancies in total, 92 for every 100 unemployed people. In the information and communication sector, which covers most software companies, there were 48 vacancies per 1,000 jobs. In early 2022 that figure was 91, a year ago 49. It has not been this low since the summer of 2020. That is cooling, not collapse. In the first quarter of 2026, UWV still rated ICT occupations as tight to very tight in 34 of the 35 Dutch labour market regions.

The forecast to 2030 is clear. Based on ROA's projections of November 2025, UWV writes: "Bij ict beroepen wordt de arbeidsmarkt tot 2030 minder krap" (for ICT occupations, the labour market will become less tight up to 2030). For the 371,500 software and application developers in the Netherlands, UWV expects a market that becomes "(slightly) less tight". The main cause UWV gives is not AI, but "the (very) low replacement demand". Software developers are a relatively young occupation. Few of them retire, so few positions need to be filled again.

That is the detail missing from almost every article on this subject. In occupations with many older workers, the wave of retirements keeps shortages high. In ICT that wave is missing. The shortage is easing mainly because there is less to replace, not because a language model is taking over the work.

Where AI does play a role is exactly where you would expect. In its regional analysis for Rijnmond (the Rotterdam area), UWV writes that language models are already taking over simple coding work and that programmers are vulnerable, "especially where starter jobs are concerned". At the same time, fewer new people are coming in. Intake in technical and ICT programmes at Dutch universities of applied sciences fell by 7.5% in 2025, according to provisional figures from the Netherlands Association of Universities of Applied Sciences. The universities blame fewer pupils in the relevant school tracks and fewer choosing a technical profile, not AI. Fewer beginners applying and fewer starter jobs: for now, the two movements offset each other.

Europe is missing its own target for ICT specialists

According to Eurostat, 10.45 million ICT specialists worked in the EU in 2025, 5% of all employment. Growth fell from 7.1% in 2020 to 2.6% in 2025. The European Commission wants 20 million ICT specialists by 2030 and wrote in its June 2026 report that progress "seems to be stagnating". That "shortage" of almost ten million is a policy target, not a count of vacancies nobody can fill. Anyone using it as proof of huge demand for programmers is mixing up two different things.

For 2027 itself, there is one fixed European date that creates work: 11 December 2027. That is when the main obligations of the Cyber Resilience Act apply. Anyone selling software, or devices containing software, in the EU must then manage vulnerabilities, keep a software bill of materials (SBOM) and deliver security updates for years. That is not work you leave to a language model without an experienced developer checking it. Nobody has calculated how many jobs it will create.

How much faster does AI really make a programmer?

This is where claims and measurements are furthest apart. Sundar Pichai wrote on 22 April 2026: "Today, 75% of all new code at Google is now AI-generated and approved by engineers, up from 50% last fall." That number says where the lines of code come from. It does not say how much faster software gets finished, nor how many fewer engineers Google needs because of it.

The best controlled research shows more modest effects. In July 2025, the research institute METR published a randomized study of sixteen experienced open-source developers working on their own projects. With AI, they took 19% longer on their tasks. Beforehand, they expected to be 24% faster. Afterwards, having been slowed down, they still believed AI had made them 20% faster.

That last point may be the most important finding in the whole AI debate. Feeling faster is not a measurement. An agent that writes two hundred lines in thirty seconds feels like a win, even if you then need an hour to understand why test 14 fails.

The tools have improved since then. In February 2026, METR reported that returning participants were about 18% faster with newer tools. The confidence interval ran from 38% faster to 9% slower, and METR itself called this "only very weak evidence". At Microsoft, engineers who started using Claude Code and Copilot CLI merged about 24% more pull requests. That is an observational study, not an experiment, and the authors warn that a merged pull request is not the same as delivered value.

So a realistic estimate for 2026 is somewhere between zero and a quarter more output per developer. That is a lot, but it is not five times as much. With a quarter more output, you do not cut half your team. Companies that do so anyway will run into trouble sooner or later.

Why beginners have it hardest

If AI makes experienced people only somewhat faster, why are so many fewer beginners being hired? The June 2026 report by investor SignalFire gives the sharpest answer: "Junior engineers can slow a team down because the operational bottleneck has moved from writing code to reviewing it."

That is the core of it. A junior's job used to be: write this function, then a senior will look at it. Writing was slow, reviewing was quick. Now an agent writes that function in a minute, and reviewing is the bottleneck. A junior who hands in AI code without fully understanding it makes the senior's work harder. SignalFire counts about 65% fewer new graduates hired at the twelve largest tech companies than in 2019, and about 76% fewer at early-stage startups. The total number of engineers those companies hire fell by only 11%.

There is a second problem underneath. In a randomized study by Anthropic, published in January 2026, 52 mostly junior engineers learned an unfamiliar Python library, with or without AI. On the comprehension quiz afterwards, the AI group scored 50%, against 67% without AI. The gap was largest on debugging questions. If you learn with an AI that does the work, you learn less. That hits exactly the skill you need as a senior: seeing why code does not work.

The Stanford study "Canaries in the Coal Mine" confirms the pattern at scale, using payroll data on millions of American workers from the payroll processor ADP. In the version of 12 August 2026, employment of 22- to 25-year-olds in AI-exposed occupations is 19% lower than if it had kept pace with less-exposed peers of the same age. For experienced workers the researchers see no comparable gap. That figure covers all AI-exposed occupations, not just developers, and the authors describe their results as descriptive, not as proof of cause and effect.

That caveat is fair. Yale Budget Lab and Brookings saw the shift in the occupational mix begin in 2021, before ChatGPT. The Economic Policy Institute sees unemployment rising among young people both with and without degrees, which points more to employers hiring less in general. The most honest picture: beginners really do have it harder, AI is a growing cause of that, but not the only one.

And not every employer makes the same choice. According to Bloomberg (February 2026), IBM will triple its entry-level hiring in the US, in rewritten roles with more customer contact and oversight of AI. HR chief Nickle LaMoreaux: "And yes, it's for all these jobs that we're being told AI can do." AWS CEO Matt Garman called replacing juniors with AI "one of the dumbest things I've ever heard" back in 2025, and said in June 2026 that Amazon is hiring "11,000 interns and new college grads" this year. Their reasoning comes down to this: if you do not train juniors, you will have no seniors in five years.

Cheaper software means more software, but not automatically more working hours

The strongest argument for more developers is an old one. When something gets cheaper, we often use so much more of it that total demand rises. Economists call this the Jevons paradox. The best-known labour market example: in the US, the number of bank tellers rose from about 500,000 in 1980 to about 550,000 in 2010, while ATMs appeared everywhere. Each branch needed fewer people, so banks opened more branches, and tellers moved from counting cash to advising customers.

For software, that sounds logical. Every company has a list of tools and integrations that were never built because they were too expensive. If they cost half as much, they get built, and someone then has to maintain them, secure them and make them comply with the Cyber Resilience Act. But nobody has reliably measured how strongly demand for software responds to a lower price. The argument is plausible, not a number.

The money flows for 2027 show where the slack is. Gartner expects worldwide IT spending to rise 15% in 2027 to $7.35 trillion. According to coverage of that forecast, software spending grows by almost 15% and data centre systems by more than 50%, but IT services by only 4.9%. IT services is the part closest to paid human hours. My reading, not Gartner's: the extra money goes mainly to AI infrastructure and software, and only a small part goes to extra developers.

And the predictions of the AI companies themselves? Dario Amodei, CEO of Anthropic, said in March 2025 that within three to six months AI would write 90% of code. Anthropic now says through a spokesperson that for them it is 70 to 90%. Yet developer job postings in the US have been rising since May 2025. Writing code was apparently never the whole job.

My expectation for 2027

What follows is my estimate based on the figures above, not a measured fact.

By the end of 2027, slightly more software developers will be working in the Netherlands than now, in a market less tight than in 2022 but not slack. Job postings will more often ask for someone who can design systems, judge AI-written code and explain why something is secure, and less often for someone who turns a ticket into code. The job title "programmer" will fade further, just as in the US projections. Beginners will keep having a harder time than in 2021 and 2022, but employers who hire no juniors now will notice around 2030 that they lack mid-level people. Some of them will therefore restart traineeships in 2027, in a different form: less typing code, more reviewing, testing and customer contact.

Two indicators will show whether I am wrong. If the US Indeed index stays above 80, the recovery continues. If it drops back to its 2025 level, around 62, this picture no longer holds. In the Netherlands, the CBS vacancy rate for the information and communication sector gives a new signal every quarter.

What to do with this now

If you are studying or just starting out, use AI, but not to skip the learning. The Anthropic study shows the cost: people who let AI do the work understand less of the code and debug worse. Write the hard parts yourself, let AI check your work, and make sure you can explain every line in a job interview. That last point is exactly what employers want to know.

If you are experienced, your value shifts to the part AI does badly: deciding what should be built, connecting systems, security and proving that code works. I wrote earlier about what that means for PHP developers in 2027, and about how to build an AI agent yourself in which a human says yes before anything happens.

If you hire people, plan with measured gains, not with what vendors promise. A quarter more output per developer is a good result. Stopping your intake of juniors on that basis turns a saving today into a problem five years from now.

The student who started in September 2022 has it harder than expected. But demand for people who can conceive, build and take responsibility for software will not disappear in 2027. Only the easy way in is disappearing.

Further reading on this site

• The 20 best AI sources for 2026, according to my colleague

Sources

• U.S. Bureau of Labor Statistics, Occupational Outlook Handbook: Software Developers and Computer Programmers (updated 27 August 2026); AI impacts in BLS employment projections
• Indeed via FRED, Software Development Job Postings on Indeed in the United States; Indeed Hiring Lab, AI and Job Postings: From Destruction to Creation? (8 July 2026)
• CBS (Statistics Netherlands), Labour market tightness eases slightly (Dutch, 30 October 2026) and vacancy rate by industry (StatLine 80567NED)
• UWV, Ontwikkeling krapte naar beroep 2024-2030 (18 December 2025), ICT-beroepen in beeld, labour market tension indicator and Regio in Beeld Rijnmond 2025-2026 (all Dutch)
• Vereniging Hogescholen, intake at universities of applied sciences (12 November 2025, Dutch)
• Eurostat, Number of ICT specialists in the EU continues to grow (27 May 2026); European Commission, State of the Digital Decade 2026
• European Commission, Cyber Resilience Act
• METR, Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity and update of 24 February 2026
• Microsoft study of Claude Code and Copilot CLI, arXiv 2607.01418
• Anthropic, How AI assistance impacts the formation of coding skills (29 January 2026)
• Google, Sundar Pichai at Cloud Next 2026 (22 April 2026)
• SignalFire, State of Talent Report 2026 (22 June 2026)
• Brynjolfsson, Chandar and Chen, Stanford Digital Economy Lab, Canaries in the Coal Mine (revised 12 August 2026)
• Economic Policy Institute, Class of 2026: what occupation data show about AI; TechTarget on the Yale Budget Lab study
• IBM: PYMNTS on IBM's entry-level hiring; AWS: Platformer interview with Matt Garman (23 June 2026)
• Gartner, IT spending forecast 2027 (21 October 2026); Network World on the breakdown by segment
• Bank tellers and ATMs: Conversable Economist, after James Bessen's work
• layoffs.fyi, Tech Layoff Tracker