The Dutch cabinet wants 25,000 fewer full-time jobs in its central government by 2035, out of about 160,000, with police and defence left out. From then on it expects to save €3.5 billion a year. The plan for how AI will do part of the work does not exist yet, because the cabinet has promised goals and savings plans for AI only in the first quarter of 2027. The job target comes first and the AI plan second, so the useful question is what would make the target reachable.
Faster work brings more work
A government is a machine for producing cases: applications, objections, permits, checks. When a case gets cheaper to handle, more of them get started. Make filing easier and more people file. Add an exception for one group and someone has to write the rule, train staff in it and later check whether it worked.
That is part of how the central government reached an all-time high of 160,000 people at the end of 2025. Since 2018, staff who write policy grew 56% and staff who carry out the laws grew 28%. The state secretary in charge, Eric van der Burg, puts it down to making policy incredibly complex.
Now put AI into that machine. British civil servants given an AI assistant for routine office tasks reported saving 26 minutes a day. That is about 6% of a working day. Even if every saved minute became a smaller team, which never happens cleanly, speed alone would cover about a third of a 16% cut. The rest has to come from work that no longer exists.
The cabinet is cutting work, not only hours
The letter pairs AI with a list of things that create less work. At least 500 rules a year are to be scrapped or simplified. The cabinet is studying a German idea: a ministry must show that a rule is still needed, and if it cannot, the rule lapses. The 28 standing advisory councils and about 150 advisory committees are to be cut back, and so are the roughly 400 policy evaluations a year. Offices are to hold one desk for every two staff.
Residents may notice two things. The cabinet wants chatbots to answer questions at any hour, and it is working on a law that would let agencies such as the tax office and the chamber of commerce share data, so that people are asked less often for information the government already holds.
There is no hiring freeze. Each time someone leaves, the question is whether the job needs filling at all, which is the logic behind governments using AI to avoid refilling retiring staff. Some departments face cuts of 20% and others of up to 60%, and the deepest ones fall where work can be removed, not only sped up.
My expectation is that in the first years most of the €3.5 billion will come from rules, committees and buildings, and AI will get the credit. The plan succeeds if the amount of work falls as fast as the headcount. If new rules keep arriving, fewer people will be handling a growing pile, and the errors will show up in residents' lives.
The same rule applies to a team of ten
Take a finance team that spends two days, or 16 hours, on a monthly report. AI cuts it to two hours. At first that looks like 14 hours back every month. Then the board asks for the report every week, and the team spends 104 hours a year on it instead of 192, which is still a saving. Then someone asks for it every working day, and the team spends about 500 hours a year, more than before AI.
Nothing went wrong in that story. The work simply got cheap, so more of it was ordered. Before you give AI a task, decide what stops: which report, which approval, which meeting. Without that decision, the time you saved goes to whoever noticed the work had become cheap.