Netherlands plans 25,000 fewer government jobs by 2035, but the AI savings are not counted yet

The Netherlands has put a number on a shrinking civil service, and it has put AI in the story. The Dutch government's plan, announced October 26, 2026, aims to cut about 25,000 full-time central government posts over several years. International headlines called it an AI-led overhaul. The cabinet's own release is more careful: it says AI plans are coming in 2027 and does not say how many of the jobs AI is expected to replace.

What the cabinet actually announced

The release from the Dutch government, titled around building a government that is simpler, smarter and smaller, sets the staff target at an estimated 25,000 full-time equivalents, spread over several years. Savings begin at €900 million in 2027 and rise to €3.5 billion a year from 2035. Police and defence are excluded. State Secretary Eric van der Burg, who is in charge of the program, is quoted: "We kunnen niet blijven doen wat we deden," which translates as "We cannot keep doing what we have been doing."

Bloomberg's report, republished by The Star, converts the €3.5 billion to about $3.9 billion a year and says the cuts will fall hardest on ministries, inspectorates and implementing agencies, with some departments facing reductions as high as 45% or even 60%. Those department-level figures come from Bloomberg's reading of the letter to Parliament, and we did not find them in the government release we opened.

Where AI comes in

The release says the government will present goals and savings plans for using AI to make civil servants more productive in the first quarter of 2027. That is all it says about AI's role. Bloomberg reports that parliament has passed non-binding motions calling for a 20% productivity gain through AI, which van der Burg cited as a reason for the cuts.

A Dutch analysis on De Belegger reaches a plainer conclusion: AI savings are not yet quantified, and there is no proven yield that could pay for the full reduction. It adds a piece of arithmetic worth keeping. If workload stays the same, the roughly 135,000 remaining FTE would each need to deliver about 18.5% more work. The author presents that as simple division, not a measured productivity forecast.

What else does the work

Most of the plan has nothing to do with AI. The release lists at least 500 rules a year to be scrapped or simplified, a yearly simplification law, and a new brake on rules, including a study of the German principle under which a rule lapses unless ministries justify it. The 28 standing advisory councils and about 150 advisory committees are to be cut, with targets set this year, and policy evaluations, now about 400 a year, are to fall sharply. Facilities services at ministries shrink, ICT is standardized, surplus buildings are sold and the target is one workplace per two employees. A legal framework for agencies to share data is due within two years, so residents do not resubmit information the government already holds.

That list matters because it shows where the money is. A smaller rulebook and fewer committees reduce workload for reasons that have nothing to do with a model. Whether AI adds to those savings is a separate, unproven question.

The growth the cuts reverse

The context is a civil service that grew despite earlier promises. According to the De Belegger analysis, central government FTE rose from about 113,500 in 2018 to 160,000 at the end of 2025, roughly 41%. A TrendingTopics report puts the base at about 170,000 civil servants, and Bloomberg also cites 160,000, so the baseline depends on what is counted. The same TrendingTopics piece says the governing coalition holds 66 of 150 seats, so the plan needs opposition votes. De Belegger adds one caution: 25,000 FTE is not 25,000 people receiving dismissal letters, because retirement, unfilled vacancies and reorganizations all count.

What the money implies

Dutch analysts have already done some arithmetic on the release. Dividing €3.5 billion by 25,000 posts gives about €140,000 per FTE, which De Belegger says is not an average civil servant's salary, since the savings also cover external hiring, housing, ICT and other costs. The cabinet assumes at most 0.5% general efficiency savings per year, which the letter says covers about one third of the financial task. In other words, the plan leans on efficiency in general before AI is counted at all.

Next steps are dated. A multi-year plan for the future civil service is due within about six months, based on the Commissie Toekomst Rijksdienst, and details of the cuts follow in the spring memorandum, the voorjaarsnota.

Why it matters beyond the Netherlands

Governments elsewhere are watching. Semafor notes that New Zealand is considering a similar proportional cut and that the United Arab Emirates is pushing to become the first "AI-native" government. The Dutch plan is an early test of whether a state can book AI savings before it has them. For the race to powerful AI, that is a small but real signal: the political system is already spending expected productivity gains that no one has measured.

Our report on Anthropic's employment data shows the private-sector version of that question, where hiring of junior technical staff is already shifting.

What we would watch

Three documents will tell us whether AI is doing real work here. The first-quarter 2027 AI goals and savings plans, which should carry a number. The spring budget memorandum, where concrete cuts are due to return. And the multi-year civil service plan the release promises within six months. Until a savings figure for AI appears in one of them, this is a staffing target with an AI hope attached.