The Netherlands will cut 25,000 civil-service jobs. Its own plan warns: don't overestimate AI

"Do not overestimate the staffing effects of digitalization and AI." That line, in bold, sits on the last page of the plan the Dutch government sent to parliament on October 26 — the plan to cut about 25,000 central-government posts.

Much of the world read the plan as a government handing civil-service jobs to AI. Bloomberg wrote that "the world is watching the Dutch." But read the 15-page letter to the end, and the government itself turns out to be the most cautious voice in the room.

It even explains why. Over the past two decades, Dutch government processes grew massive and were digitized and automated. Yet neither costs nor headcount came down. Instead, digitalization created new tasks, more staff and more outside contractors.

The plan itself is big. By 2035, about 25,000 of roughly 160,000 full-time posts in the central civil service will go — about one in six. The savings reach €3.5 billion a year from 2035.

So a government that refuses to bet on AI still plans to cut one job in six. Where do the cuts come from, then? That question leads to the most interesting part of the plan.

WHAT IS

Decisive GovernmentSlagvaardige Overheid

The Dutch government's reform to make the state simpler, smarter and smaller

1. 2025.0322% cost cut beginsThe Schoof cabinet starts cutting central-government staff costs by 22%, or €1 billion a year
2. 2026.01A coalition pillarThe D66–VVD–CDA coalition deal gets a “decisive government” chapter and a 500-rules-a-year target
3. 2026.02A state secretary namedThe Jetten cabinet takes office; Eric van der Burg leads the reform and a ministerial taskforce
4. NOW25,000 posts to goOn October 26, 2026, a letter to parliament sets out the path to 2035

"Slagvaardige Overheid" — roughly, a decisive or agile government — is the administrative reform of the Jetten cabinet, which took office in February 2026. State Secretary Eric van der Burg runs it with a taskforce of ministers, along four lines: fewer rules, joined-up public services, a more productive civil service, and breaking down the walls between ministries.

The target is the Rijksdienst, the central civil service covered by the national collective labor agreement: policy ministries, inspectorates, and executive agencies such as the Tax and Customs Administration. According to the letter's annex, it employed 160,016 full-time equivalents at the end of 2025, 70% of them in executive agencies. Another 220,000 or so people in the police, defense and the courts are not counted.

The reform's starting point was a cost cut ordered by the previous Schoof cabinet. The Jetten cabinet added more, bringing the target to €3.5 billion a year by 2035. How to keep public services running with fewer people is a question every aging society, Japan included, now faces.

Last week, DATA TIMES looked at Microsoft's offer of buyouts to 60,000 employees. When a company cuts staff, it rarely explains its reasoning in detail. A government has to. The Dutch letter lays out the numbers, the method, and even past failures.

As people who work with AI and data, what struck us was that the letter never says how many jobs AI will save. Instead it talks about rules, data sharing, meetings, evaluations and the number of government websites. We read all 15 pages.

First, how big has the Dutch central civil service become?

At the end of 2018 it had 113,500 full-time equivalents. By the end of 2025 it had 160,000 — a 41% jump in seven years. Policy and support staff in ministries grew 56% between 2019 and 2025; inspectors grew 48%.

The letter says the growth came from new and heavier tasks. Staff working on the climate and energy transition, for example, or on "Werk aan Uitvoering," a program to repair struggling executive agencies, were hired with temporary budgets. Simply going back to 2019 or 2020 levels, the letter argues, would ignore all the work that has been added since.

A savings target of €3.5 billion a year by 2035

On top of that growth, the previous Schoof cabinet and the current Jetten cabinet have stacked up cost-cutting targets. According to the letter's table, the cut rises from €393 million in 2025 to €901 million in 2027, €2.49 billion in 2030, and €3.49 billion in 2035.

Converted into people, that is about 25,000 full-time posts — a figure the letter itself calls a "rough estimate," because the same euros mean more or fewer jobs depending on whether high- or low-paid staff go. The previous cabinet had aimed for 22%, or about 35,000 posts, according to the trade publication iBestuur. The new target is about 16%.

Some units face cuts of 20% to 60%

One in six is an average. Police and defense are exempt, and so, at least in part, are the prison service, intelligence services, the public prosecutor and embassies. That pushes the burden onto everyone else. According to the letter, many organizations face a cut of about 20%. Where temporary budgets expire, it can reach about 45% — and in one case about 60%.

Seven in ten central civil servants work in executive agencies that deal directly with the public. The letter admits the cuts could affect the quality and continuity of those services.

A committee of the lower house will debate the plan on December 1. Which organizations shrink, and by how much, will be spelled out in a multi-year plan due within six months and in next spring's budget update.

This is the most striking part of the letter. At the end of the annex, the government lists six lessons from 20 years of cutbacks.

Across-the-board cuts never stopped the growth

Lesson one is about method. Most past programs were a kaasschaaf — a cheese slicer. Each unit got a percentage and was left to figure out how to meet it.

The cuts happened on paper. But new programs kept being added, and the government kept growing.

So this time the government drew a hard line on what efficiency can do. The Netherlands Bureau for Economic Policy Analysis (CPB) considers 0.5% a year a realistic efficiency gain for central government. The cabinet adopted that as a ceiling.

At 0.5% a year, efficiency covers about a third of the target. The other two-thirds only appear if the state decides which work to stop, shrink or do differently. The letter admits that this kind of choice is exactly what "almost never succeeded" over the past 20 years.

"Do not overestimate the staffing effects of digitalization and AI"

Lesson two is the line we started with.

Government processes grew massive over two decades, the letter says, and were digitized and automated. But the promised savings in costs and staff never materialized. Digitalization brought new tasks, more capacity and more contracting — and several organizations are now struggling with heavy legacy systems.

Digitizing does not automatically mean fewer people. New systems need people to maintain them, people to fix the data, and people to run the new services they make possible. The government that has decided to use AI more widely is also the most cautious about its effects.

No hiring freeze

Lesson three is about blunt instruments. Past hiring freezes, the letter says, drove out good people who had other options, stopped new talent from coming in, and left young staff unable to grow. Years later, whole cohorts were missing from the workforce.

iBestuur reports that the government rejected a hiring freeze this time. The remaining lessons are equally down-to-earth: respect labor agreements and workforce planning, invest up front in change, and set clear definitions with central oversight.

So where do the two-thirds come from? The letter's answer does not start with AI.

At the very beginning, it says it is wrong to treat a "decisive government" as a synonym for budget cuts. The aim is a simpler, smarter state — one that can then also become smaller. The work runs along four lines.

Pillar 1: scrap at least 500 rules a year

Since October 1, every ministry has had its own target for rules to scrap or simplify — at least 500 a year in total, half for citizens and professionals and half for businesses.

The most interesting idea is borrowed from Germany: a reversed burden of proof. If the responsible ministry cannot show a rule is still needed, the rule goes. New rules will also be checked for a sunset clause that makes them expire automatically.

The letter names a "risk-rule reflex" — adding rules and oversight after every incident — and promises to rein it in. Its example is the Red Cross, which receives disaster-preparedness funding through 55 separate streams from ten ministries and various regions and municipalities, each with its own application and checks. The government is using it as a test case for simplifying grants to civil-society groups.

Pillar 2: never ask citizens twice for data the state already has

Today, every exchange of data between Dutch agencies has to be arranged separately, and often it isn't. So people submit the same information again and again, and miss out on benefits they are entitled to. A state commission called this part of the "broken promise of the rule of law."

The cabinet has ordered a structural solution for data sharing — used only for proactive service to citizens — within two years. What government already knows, it may no longer ask for again. Belgium and Estonia are the models, and the data will not be used for enforcement or criminal investigation.

There is a painful history here. The letter cites the Dutch childcare-benefits scandal (toeslagenaffaire), where the information demanded of citizens became so complex that it made their problems worse.

This pillar also includes the Netherlands Digital Service (NLDD), launched before the summer of 2026. Its first three projects are a working design for a sovereign government cloud, a marketplace that makes it easier for businesses to deal with government, and a method for writing legislation as executable code.

Pillar 3: one desk for every two employees — and this is where AI appears

The third pillar is productivity, and it is where AI shows up.

• Offices move to one workstation for every two employees, and surplus buildings are sold.
• Procurement gets fewer requirements, more joint buying, and a systematic search for places to use AI.
• Internal IT and facility services are standardized on a "join unless, together unless, standard unless" principle, with as few exceptions as possible.

Even here, the language on AI is careful. AI will be used more widely, but within the conditions of security, data quality and sovereignty — the state's control over its own data and systems. That requires changes to infrastructure and a fundamental redesign of how work is done. Concrete AI goals and savings plans are due in the first quarter of 2027.

Pillar 4: from 400 evaluations a year to the 10 or 20 that matter

The fourth pillar targets what the letter calls the "Hague walls": silos between ministries, between policy and delivery, and between layers of government, buried under coordination, reporting and oversight.

The numbers show how thick the walls are.

Source: DATA TIMES, based on the Dutch government's letter to parliament (October 26, 2026)

From 400 evaluations a year down to 10 or 20. Evaluations exist so government can learn, the letter says, but there are so many that they cost a lot and change little.

Progress reported twice a year, in numbers

The plan comes with a monitor: the cost of regulation for businesses, how citizens and companies rate public services, and the size of the civil service (running costs, staff and contractors). It draws on Sir Michael Barber's delivery method from the UK and on Rotterdam's city targets, and will be reported to parliament every spring and autumn.

The Netherlands was not the only place where AI and headcount made news in October. Put side by side, the choices look very different.

Barclays: about 5,000 jobs by the end of the year

On October 22, British bank Barclays said it would cut about 5,000 jobs, roughly 5% of its workforce, by the end of 2026. Reuters reported that about 1,000 had already gone and that around 60% of the cuts fall on support functions such as software developers. The same day, the bank reported third-quarter pre-tax profit of £2.3 billion, up 12% from a year earlier.

Asked about AI's role, CEO C.S. Venkatakrishnan told reporters it was "the icing on the cake, not the cake." Like Microsoft's buyouts, it is a profitable company trimming headcount first.

Denmark: free up time, not cut jobs

Denmark has made a different choice. According to the Danish political outlet Altinget, the government, Local Government Denmark (KL) and Danish Regions set a goal in June 2025 to free up 50 million hours of public-sector work — about 30,000 full-time jobs' worth — with AI by 2035.

The freed-up time is meant to move into work with citizens. It is not a promise to cut jobs. Even so, unions called the target unrealistic.

Japan: no more posts, but digital staff first

Japan, where DATA TIMES is based, faces a shrinking workforce too. Its cabinet policy on managing the national civil service aims to "keep public services sustainable without increasing the number of national civil servants, using the power of digital technology."

Under that policy, a Cabinet Secretariat summary shows Japan added 396 posts for digital transformation in fiscal 2026 (782 over two years), including 89 to prepare government for using AI and 45 for project management offices that steer each ministry's digital work. At the same time, it postponed 116 planned post reductions government-wide to keep the people it needs "until the effects of digital transformation appear."

The Netherlands cuts, Denmark frees up time, Japan holds the line. All three judge AI's effect over a long horizon, around 2035. And the Netherlands and Japan share one more thing: they put people and money in before expecting AI to pay off. Japan is adding digital staff first; the Dutch wrote "investment up front" into their lessons. Neither expects digitalization to pay off quickly.

The letter mentions AI concretely in only two places, both under the third pillar: procurement, and the AI section itself. Most of the shrinking is designed to come from scrapping rules, not asking twice for data, and cutting evaluations, meetings and advisory bodies. The government even caps efficiency at a third of the savings.

Headlines still said "AI." It is an easier story. But what actually makes a state smaller is not the tool, but the political decision about what to stop doing. The Dutch letter is worth reading because it doesn't dodge that.

Lesson two applies to companies just as much. Put AI on top of today's workflows and you need people to check its output, clean the data and run the new things it makes possible. That is exactly the road Dutch government took through 20 years of digitalization.

In our view, the organizations that really get lighter with AI are the ones that decide first what work to stop. Drop a meeting, kill a report, skip a sign-off — then put AI in the space you freed. Do it the other way around, and the work grows.

Two ideas here are ready to use, in government or in business. One: never make people hand over information an organization already has. Anyone who has filled in the same form again and again after moving house knows the pain. Two: stop trying to evaluate and report on everything, and pick each year what truly needs it.

For students thinking about careers, it is striking that a government now talks about scrapping 500 rules a year and writing laws as code. People who understand data and software can help rebuild how the state itself works. The Dutch letter reads like the opening of that era.