The target came first

Start with the order of events, because the order is the story. On October 26 the cabinet published the money and the headcount: €900 million to save in 2027, €3.5 billion a year from 2035, and an estimated reduction equivalent to approximately 25,000 full-time positions across the central government. It published who is protected: police and defence sit outside the cuts. It even published the uneven shape of what is coming, with some organisations reported to face cuts near 20% and others up to 60%.

What it did not publish is the AI part of the mechanism. The cabinet has announced the savings it wants and several ways it intends to achieve them. AI-assisted productivity is one of those mechanisms, but the government has not yet published the tasks, baselines or savings attributable specifically to AI. The section of the government's own announcement headed "a more productive civil service" contains the AI commitment in a single sentence: in the first quarter of 2027, the cabinet will present goals and savings plans for using AI to work more productively. Until then, there is no list of tasks AI will take over, no productivity baseline, and no AI share of the €3.5 billion. A reader who saw only the wire headline about an AI-led overhaul would assume the technology plan came first. The letter runs the other way.

None of this makes the target imaginary. A savings line for 2027 is a budget instruction, and ministries will have to meet it with or without working AI tools. That is the risk in the sequence. The first annual savings target applies to 2027, the same year in which the cabinet plans to publish its AI productivity goals. That leaves limited time to explain how any AI-related savings would contribute to the initial target.

The growth it is reversing

The cuts also need their denominator. Reporting put the central government at about 160,000 full-time jobs, a record reached after years of growth. Against a reported central-government workforce of roughly 160,000 FTE, the estimated reduction equals about 16%. Reuters characterized the proposed cuts as around 20%, but the source material reviewed for this article does not establish the denominator behind that higher percentage. Reuters put the increase since 2018 at about 46,500 jobs, roughly 40%, accumulated while successive governments said they wanted a smaller state.

The estimated 25,000-FTE reduction is equivalent to roughly 54% of the 46,500 positions Reuters reports were added since 2018. The cabinet says the reduction will be spread over several years but does not give a final workforce-reduction date in its announcement. It is still one job in six, and the exemption list concentrates it: if police and defence keep their staff, the deepest reported cuts, up to 60% at some organisations, fall on the rest of the service. Dutch reporting says the tax administration, UWV and IND have not been granted advance exemptions. That does not establish how much each organization will ultimately be required to reduce. There is reportedly no general hiring freeze: vacancies are judged one by one, so the reduction depends on thousands of replacement decisions continuing to go the same way.

The part AI cannot do

To its credit, the letter does not claim software will find €3.5 billion on its own. It says plainly that alongside general efficiency measures, strategic choices are necessary about which tasks will and will not be done. Around the headcount sit the less glamorous items that may matter more than any model: at least 500 rules a year to be scrapped or simplified, a brake on new rules, ICT systems standardised across ministries, one workplace for every two employees, and a sharp reduction in the advisory and evaluation machinery, currently 28 permanent colleges, about 150 committees and roughly 400 policy evaluations a year.

That list reframes what AI is being asked to do here. Deleting a rule deletes the work of administering it, permanently and at no licence cost. AI, at best, speeds up work that still exists. A government that cuts tasks first and automates what remains has a coherent programme. One that automates first has bought tools for a workforce it is also removing.

The test arrives in the first quarter of 2027, when the AI document itself is due. Judge it by three things: whether it names specific tasks rather than product categories, whether it states the productivity baseline those tasks are measured against, and whether service levels for citizens are written down before the staff leave. Every other government announcing AI savings this year should be read against the same three questions.

Sources

  1. [1] Rijksoverheid.nl, "Kabinet werkt aan slagvaardige overheid: eenvoudiger, slimmer en kleiner" (government news release summarising the letter to parliament, October 26, 2026)Read source
  2. [2] Reuters, "Netherlands aims to cut around 20% of government jobs over next decade" (October 26, 2026; workforce growth since 2018, savings in euros and dollars)Read source
  3. [3] Bloomberg Law, "Dutch Government Plans to Cut 25,000 Jobs in AI-Led Overhaul" (October 26, 2026)Read source
  4. [4] Engadget, "Dutch government to cut 16 percent of workforce as part of AI efficiency push" (October 26, 2026)Read source
  5. [5] HBP Media, "Kabinet wil 25.000 ambtenaren schrappen bij rijksoverheid" (October 26, 2026; tax administration, UWV and IND not spared in advance)Read source