Dutch government plans 25,000 job cuts by 2035, leaning on AI
The Netherlands plans to phase out 25,000 government roles by 2035, aiming to save €3.5 billion while using AI to absorb the extra workload.
According to TechRadar, the Dutch government has announced plans to phase out 25,000 public-sector jobs by 2035 as part of an effort to make government work more efficient. The reduction would amount to roughly 16 percent of the government's total workforce. The stated goal is to save €3.5 billion, about $3.9 billion.
TechRadar reports that the plan was laid out by Eric van der Burg, the State Secretary for Kingdom Relations and Effective Government. The outlet says €900 million of the savings would arrive by 2027. Artificial intelligence is presented as the means of handling the work that remains after the reductions, with the government expecting AI tools to raise productivity.
The package also includes other measures. TechRadar says the government intends to sell off buildings, simplify rules and move staff into smaller desk areas. Together, these steps are described as contributing to the savings target, so the cuts are part of a broader overhaul rather than a single budget decision.
The reporting does not yet spell out which roles would be affected, how the reductions would be carried out, or how AI systems would be introduced across individual agencies. Those questions are likely to shape how the plan is received in the months ahead.
Background
The Dutch civil service is large by the standards of many European countries, and successive governments have pursued efficiency programs for decades. Most such efforts combine attrition, reorganization, and the consolidation of property and back-office functions, and public employers often rely on early retirement and hiring freezes rather than direct layoffs. Automation has long been part of this picture, from digital case-handling systems to more recent experiments with machine learning for document sorting, translation and search.
The Netherlands also has a particular history with algorithmic decision-making in government. A scandal over childcare benefits, in which automated risk-scoring and rigid enforcement led many families to be wrongly accused of fraud, became a major political crisis and contributed to the resignation of the cabinet in January 2021. That episode shaped public and parliamentary expectations about how automated tools can be used in areas that affect people's income and rights.
At the European level, the EU AI Act, which entered into force in 2024, sets out a risk-based framework for artificial intelligence. Several uses common in public administration, particularly those involving access to public services, benefits or law enforcement, can fall into categories that carry stricter obligations for transparency, documentation and human oversight. Member states must also designate authorities to supervise compliance, so national plans to deploy AI in government sit within a legal structure that is still being put into practice.
Efficiency goals in government often rest on assumptions about how quickly workloads can be redistributed and how well new technology will perform. Whether those assumptions hold depends on the quality of implementation, the training of staff, and the willingness of agencies to change established ways of working.
Why it matters
The plan is significant because it links a large workforce reduction directly to the expectation that artificial intelligence will absorb the resulting workload. Few governments have framed a staffing decision this explicitly around AI productivity, and the Dutch approach may be watched by other public administrations weighing similar choices. If the model works, it could offer a template for reducing costs without relying solely on service cuts. If it does not, it could show the limits of treating automation as a substitute for staff.
The stakes are also practical for residents. Many people experience government through permits, benefits, tax matters and registration, and changes in staffing or processes can affect waiting times, accuracy and the availability of human contact. Those who depend most on public services may feel the effects first, and they have the least room to absorb delays or errors.
The mix of measures matters too. Selling buildings and shrinking office space can produce one-time or recurring savings, but they do not directly change how work gets done. Simplifying rules may reduce administrative burden, though it can also alter legal protections if not handled carefully. The relative weight of each measure will influence how much of the savings target depends on technology.
For the technology sector, the plan signals growing demand for government AI systems, along with heightened scrutiny of how those systems are procured, tested and governed.
What to watch
The first signals will be how the government identifies the roles and agencies affected, and whether it publishes timelines for the reductions. Parliamentary debate, union responses and any formal consultation will show how much support the plan has and what safeguards are being discussed.
Equally important is the AI component. Watch for details on which tools will be used, how their performance will be measured, and whether independent evaluation is planned. Clarity about human oversight, especially in decisions that affect individuals, will be central to public confidence.
Finally, track whether the €900 million target for 2027 is met and how the wider €3.5 billion goal is reported over time.
Viewpoints
Commentary from our columns: three angles on this story.
The Skeptic open questions and weak points
The headline numbers are striking, but the plan leaves a gap that deserves attention. It describes 25,000 roles phasing out and a large productivity gain from artificial intelligence, yet the connection between the two is largely assumed rather than demonstrated. Governments have a long record of announcing efficiency targets whose savings arrive later, smaller, or with costs that were not counted at the outset.
Several questions stand out. How will the work be measured before and after the change? What happens when a tool makes mistakes in a process that affects someone's benefits or permits? Who checks the results, and with what authority? Selling buildings and moving to smaller desks can produce real savings, but they are not the same as making public services work better.
None of this means the plan will fail. It means the burden of proof sits with the government. Public promises about AI tend to outpace the slower work of testing, training and maintenance. The most useful response right now is to ask for the baseline, the milestones and the independent review that would show whether the numbers hold up.
The Ledger the business angle
Seen from a budget perspective, the plan is an attempt to reshape the cost base of a large employer, and the scale is what draws attention. A target of €3.5 billion across a decade, with a sizeable share promised by 2027, sets a clear test. Savings from property sales and office changes tend to be easier to book than savings from workforce change, which carries severance, retraining and transition expenses that are not always counted in headline figures.
The technology angle is also commercially important. Public-sector AI contracts are often large, multi-year and politically sensitive, and vendors will be watching the procurement process closely. The question for buyers is not only what a system can do in a demonstration, but what it costs to run, maintain and audit over time, and who bears the risk when it underperforms.
For taxpayers and for the government itself, the most useful accounting would separate one-off gains from recurring ones and show the cost of the transition alongside the savings. Without that breakdown, it will be hard to judge whether the plan is an efficiency success or a shift of costs into other budget lines.
The Rulebook privacy, safety and regulation
When a government says artificial intelligence will absorb a large share of its workload, the rules that govern public decisions become central, not peripheral. Automated tools used in administration can affect access to benefits, permits and other services, and the consequences of error fall on individuals who often have limited means to challenge them. The Netherlands has direct experience of how damaging such errors can be.
Good governance here means clear answers to a few basic questions. Which decisions will be automated, and which will remain with a person? How can residents find out that a system was involved, and how can they appeal? What records will be kept, and who can audit them? These are not technical footnotes. They determine whether a faster process is also a fair one.
The EU AI Act provides a framework for many of these obligations, but frameworks need to be applied in practice. The plan's ambition is a reasonable subject for public debate, provided the safeguards are debated with equal seriousness. Efficiency and accountability are not opposites, but they will not align on their own.