Everything About AI
NextFin News - The Dutch government is planning to eliminate 25,000 civil service jobs by 2035 in an artificial-intelligence-led overhaul aimed at annual savings of €3.5 billion ($3.9 billion), a move that would shrink the central state's workforce by roughly 16 percent and marks one of Europe's most explicit bets that AI can do more than assist bureaucrats - it can replace them.
The target was confirmed in a government statement on Monday and detailed by State Secretary Eric van der Burg of the VVD, the junior home affairs minister responsible for an "effective government." The plan does not rest on a hiring freeze. Instead, ministries will be asked whether every departing worker needs replacing, while AI systems and chatbots take over helplines, advisory bodies face the axe, and office space is cut to one workstation for every two civil servants. The savings ramp is deliberately back-loaded: €900 million a year by 2027, rising to the full €3.5 billion from 2035.
The question the market and policymakers across Europe will be asking is not whether the cuts happen, but whether they survive contact with the machinery of government. The arithmetic is seductive - €3.5 billion divided by 25,000 jobs implies roughly €140,000 in annual savings per eliminated position, a figure that only makes sense if the jobs cut are experienced, well-paid roles rather than entry-level clerks. But the coalition's own advisory committee on the future of the civil service has already warned that digitalisation and AI improve efficiency only if the state also stops adding rules, subsidies and checks. In other words, the plan only works if the Dutch state shrinks its ambitions at the same time as its headcount. That is the harder political task.
What the Plan Actually Says
The central facts, as laid out in the government statement and Van der Burg's interview with Parool, are specific enough to test. The baseline is 160,000 full-time central-government jobs. "Out of a total of 160,000 civil servants, this represents - rounding very broadly - less than 20 percent. Still, it is highly substantial," Van der Burg said. Done the precise way, 25,000 out of 160,000 is 15.6 percent - about 16 percent of the workforce disappearing over roughly nine years, or just under 2,800 positions a year on average.
The cuts will not be spread evenly. Some ministries will see staff levels fall by 20 percent, while others face reductions of up to 60 percent. Van der Burg declined to name which departments bear the deepest cuts, saying that would emerge from substantive proposals by the ministries themselves. What he did make clear is where the knife is meant to fall: managers, advisers, press officers and secretaries rather than frontline staff. The government's "effective government" program has already argued that about half of ministry jobs sit in overhead, support and management, and set a target of bringing that down to around 35 percent.
"If key personnel leave, work could grind to a halt - that's not an option. Instead, we can ask ourselves in each instance: does the person leaving actually need to be replaced?"
That framing - no freeze, but attrition plus AI - is politically deliberate. It avoids the image of mass layoffs while still delivering a workforce reduction. The government also intends to lean less on expensive independent contractors and more on digital solutions, pointing to the migration of citizen services from office-hours helplines to 24-hour online chat functions. "The advantage is that you can ask a chatbot a question 24 hours a day," Van der Burg said.
Real estate is the second leg. With hybrid work keeping office occupancy below 70 percent even on the busiest days, the state plans to end up with one workstation per two civil servants and to repurpose surplus buildings for housing, commercial space or hotels. "These are offices paid for by the taxpayer - it's simply a waste of money," Van der Burg said.
The third leg is the pruning of 176 advisory councils and bodies, including the Advisory Council on Migration, the Education Council, the Advisory Council on International Affairs and the Council for Public Administration. Van der Burg said their number had grown "out of control," though he did not yet know which would be cut.
Why This Is Different From Ordinary Austerity
Most government consolidation exercises are cyclical: a deficit widens, a finance minister tightens belts, and when growth returns the spending comes back. This is not that. The fiscal pressure is real - the Dutch budget deficit rose to €8 billion in the first half of 2026, the largest first-half gap since the pandemic year of 2021, and debt stood at €524 billion, or 43.7 percent of GDP, according to Statistics Netherlands. Fitch Ratings expects the 2026 deficit to reach 2.7 percent of GDP, lifted by a one-off €8 billion transfer of military pension liabilities, with the 2027 budget targeting 2.2 percent of GDP. But the government is not presenting these cuts as a temporary correction. It is presenting them as a permanent re-rating of how much state capacity the Netherlands needs.
The structural claim rests on technology, not the budget cycle. The National AI Delta Plan, presented in late November 2025 at the request of the Ministry of Economic Affairs, brought together more than 60 experts and produced 52 recommendations built around a stark premise: the Netherlands risks losing control over AI unless it builds its own computing capacity, energy infrastructure and adoption base. The plan calls for a National AI Impact Institute to monitor effects on work, well-being and democracy. In that framing, cutting 25,000 jobs is not austerity - it is the public-sector arm of a national industrial strategy.
The mechanism the government is betting on is straightforward: AI absorbs routine processing, document review, helpline queries and compliance checks; fewer humans are needed to supervise the work; and the remaining workforce is concentrated on frontline delivery. That is also what makes the plan structurally risky. Clerical automation is well understood. Automating policy judgment, inter-ministerial coordination and regulatory oversight is not.
Van der Burg himself supplied the diagnosis that makes the structural case. Since 2018, he said, the number of civil servants implementing laws and regulations rose 28 percent, while policy-making staff rose 56 percent. "We have made policy incredibly complex," he said. "This steadily increases the risk of errors, both by citizens and civil servants." The implication is that complexity, not the people, is the cost driver - which is exactly why the advisory committee insists that technology alone will not fix it.
The Counter-Thesis: AI Cannot Cut Through Red Tape
The strongest objection to the plan comes not from unions but from the government's own advisers. The advisory committee on the future of the civil service, whose conclusions are due later this year and which Van der Burg is partly following, has argued that digitalisation and AI can make government more efficient "only if the build-up of rules and checks is also tackled." Without deregulation, the committee's logic runs, AI does not remove work - it digitises the same work at higher speed and adds a new layer of technology governance on top.
That view has outside backing. Government economist Jos Blank has written that the Dutch state has been underperforming for decades, and that a constant stream of new rules, subsidies and mergers is making public services worse rather than better. Former senior civil servant Roel Bekker has criticised growing "policy overload" and called for cutting the number of policy officials by 20 to 30 percent. Policy officials have almost doubled to 20,000 in the past decade, according to experts cited in Dutch reporting.
The counter-thesis, sharpened, is this: the Dutch state's cost problem is not headcount, it is scope. Every new regulation creates compliance work, every subsidy creates an application-and-audit pipeline, every check creates a reviewer. If the coalition continues to legislate at its current pace while cutting staff, the remaining civil servants will face higher caseloads, error rates will rise, and the promised €3.5 billion will be consumed by remediation and litigation rather than banked as savings. AI chatbots can answer a citizen's question at 2 a.m.; they cannot decide whether a subsidy application meets a rule that changes every year.
There is also an execution-timing risk baked into the numbers. Only €900 million of the €3.5 billion is scheduled to be saved by 2027 - about 26 percent of the steady-state target. The steepest cuts arrive in the second half of the decade, which means the plan depends on political continuity through 2035 - and Dutch cabinets have a habit of fracturing over exactly this kind of structural reform.
Second-Order Effects: Who Benefits, Who Is Exposed
Look past the first-order budget math and the plan creates clear winners and losers across the Dutch economy. The direct beneficiaries are the firms positioned to sell AI systems, automation platforms and digital-identity infrastructure to the public sector - and the real-estate dynamics cut both ways, with the state exiting office space while the freed buildings are proposed for housing conversion. The Dutch tech ecosystem, already buoyed by the AI Delta Plan's emphasis on domestic computing capacity and talent, stands to gain procurement tailwinds if the government buys locally rather than from American hyperscalers.
The exposure sits with the service layer that has grown around the Dutch state. Independent contractors, consultancy firms and the advisory bodies themselves face a shrinking market. The government's stated intent to reduce reliance on "pricey independent contractors" signals a shift of spend from external consultants to internal platforms - a redistribution within the same budget line rather than pure savings, which is one reason the €3.5 billion figure should be read as a target, not a forecast.
For the rest of Europe, the Netherlands is a test case. Several EU governments are managing fiscal consolidations under the bloc's reactivated deficit rules, and most are eyeing AI as a productivity lever. If the Dutch experiment delivers cleaner services at lower cost without a visible drop in quality, it becomes a template. If error rates climb and citizens face slower resolutions, it becomes a cautionary tale that AI in government is harder than AI in commerce - because the citizen cannot switch provider.
What Would Prove the Thesis Wrong
The plan should be judged on a specific, observable test, not on rhetoric. The falsifying signal is this: if the number of advisory bodies does not fall materially from 176, and if policy-making staff does not decline alongside total headcount, then the state is cutting frontline capacity while preserving the complexity machine - and the €3.5 billion savings target will not materialise. A secondary signal is service quality: if citizen-facing resolution times or error rates in permits, benefits and tax filings deteriorate over the next two years, the productivity claim is failing regardless of the headcount number.
The scenarios break down by time horizon. In the short term, through 2027, expect mostly attrition, hiring restraint and the first wave of chatbot and back-office automation - savings of roughly €900 million a year are plausible because they rely on vacancies rather than dismissals. In the medium term, to 2030, the 20-to-60 percent departmental cuts bite, and the real test begins: whether AI can absorb the work of experienced policy staff without degrading output. In the long term, to 2035, the question is whether the Netherlands has structurally reset the size of the state or merely deferred the cost into slower service delivery and higher contractor bills.
The base case is that the headcount target is largely met - attrition-based cuts are hard to reverse once made - but the savings fall short of €3.5 billion because complexity was never cut at the source. The upside case is that AI-driven automation does lift public-sector productivity enough to offset the headcount reduction, making the Dutch state a rare example of a government that got smaller and worked better. The downside case is a slow-motion degradation: fewer civil servants, the same volume of rules, rising backlogs, and a political backlash that restores spending without restoring capacity.
The Dutch government has correctly identified that it cannot cut its way to efficiency with incremental trims. But the hard truth of public-sector AI is that technology removes tasks, not mandates. Unless the coalition is willing to repeal the rules that generate the work, the 25,000 jobs may disappear - and the work they did will simply wait, unpaid, in a queue.
Explore more exclusive insights at nextfin.ai.
