Latest / AI Newspaper / The Netherlands Plans to Cut 25,000 Government Jobs as It Turns to AI

Transcript

The Netherlands Plans to Cut 25,000 Government Jobs as It Turns to AI · AI Newspaper

• 0:00
  Welcome to the AI Newspaper, your daily update
• 0:02
  on everything happening with AI. Visit AINewspaper
• 0:05
  .org to see today's stories and sign up for email
• 0:07
  alerts. So imagine a government basically firing
• 0:12
  25 ,000 of its own workers, selling off their
• 0:16
  physical desks and handing their jobs over to
• 0:18
  artificial intelligence, and doing all of this
• 0:21
  before they even know if the technology actually
• 0:23
  works at scale. Yeah, it sounds like science
• 0:25
  fiction. But the Netherlands isn't just imagining
• 0:28
  it. They are actually putting this into motion
• 0:30
  right now. Right. And today we're looking at
• 0:33
  a pretty massive stack of data, policy updates,
• 0:35
  international reports. Just to answer one critical
• 0:38
  question, can a government successfully fire
• 0:41
  a massive chunk of its human workforce and replace
• 0:43
  it with AI? Or is this, you know, a historic
• 0:47
  multi -billion euro gamble? I'd lean heavily
• 0:50
  toward gamble, to be honest. The numbers we are
• 0:53
  looking at are wild. They're projecting savings
• 0:56
  of 3 .5 billion euros a year, with up to 60 %
• 1:00
  staff cuts at some specific agencies. Wow. 60%.
• 1:04
  Yeah. And this shows exactly what you, the average
• 1:08
  citizen, are being asked to accept in the name
• 1:11
  of government efficiency. It's a huge shift in
• 1:14
  how the state operates. Let's start with the
• 1:16
  sheer scale of the bureaucratic reversal the
• 1:18
  Dutch are attempting here, because... Slashing
• 1:20
  25 ,000 full -time equivalent jobs by 2035, I
• 1:24
  mean, that is an enormous reduction. It's massive.
• 1:27
  And to really understand the shock of this, you
• 1:30
  have to look at the buildup that preceded it.
• 1:32
  This isn't just a simple trim of the fat. It
• 1:35
  is a violent reversal of, well, massive bureaucratic
• 1:39
  bloat. Right, because they've been hiring like
• 1:42
  crazy lately, haven't they? Exactly. The context
• 1:45
  here is just essential. If you look back to 2018,
• 1:49
  the Dutch central government workforce actually
• 1:51
  grew by over 40%. 40 % in just a few years? Yeah.
• 1:55
  They added 46 ,500 new jobs over that period.
• 2:00
  By the end of 2025, they are looking at a peak
• 2:03
  of around 160 ,000 full -time equivalent employees.
• 2:07
  So they essentially supercharged the size of
• 2:10
  the state over the last several years, adding
• 2:12
  all these layers of administration. And now,
• 2:14
  now they're slamming on the emergency brakes.
• 2:17
  Hard. The financial targets attached to this
• 2:20
  reversal are just incredibly aggressive. They
• 2:23
  want 900 million euros in savings by 2027. Which
• 2:26
  is right around the corner. Right. And that scales
• 2:29
  all the way up to 3 .5 billion euros annually
• 2:31
  by 2035. But here is the catch for you, the person
• 2:38
  actually relying on these services. The armed
• 2:40
  forces and the police, they are completely exempt
• 2:43
  from these cuts. Oh, wow. So the military and
• 2:46
  police are safe, which means the hammer falls
• 2:48
  disproportionately on everyone else. Exactly.
• 2:51
  The heaviest burden lands squarely on the ministries
• 2:54
  and the delivery agencies. Which, I mean, those
• 2:57
  are the departments you actually interact with,
• 2:59
  right? Yeah. The people processing your building
• 3:01
  permits, handling your tax disputes, delivering
• 3:04
  basic social services. Some of these specific
• 3:07
  departments are modeling workforce reductions
• 3:09
  of 45 to 60 percent. That level of hollowing
• 3:12
  out is just staggering for public -facing entities.
• 3:15
  But what really jumps out at me is the physical
• 3:18
  reality of this plan. They are downsizing the
• 3:21
  physical footprint right now, like before the
• 3:24
  tech is even fully rolled out. That's true. They're
• 3:26
  basically burning the boats. They are forcing
• 3:29
  a new ratio in the office. One desk for every
• 3:32
  two civil servants. Wait, they're physically
• 3:34
  removing desks? Yeah. And actively selling off
• 3:37
  surplus government buildings. They are... standardizing
• 3:41
  IT systems, and they actually have a mandate
• 3:43
  to scrap at least 500 rules a year. Scrapping
• 3:47
  the rules. I mean, I guess that's the only way
• 3:49
  this even theoretically works, right? You can't
• 3:52
  really digitize a fundamentally broken, convoluted,
• 3:55
  bureaucratic process. No, you can't. You have
• 3:58
  to simplify the machinery first. If the process
• 4:00
  is a mess, the AI is just going to be a faster
• 4:03
  mess. A perfect example of this is their plan
• 4:06
  to reform the Freedom of Information Law, which
• 4:08
  is known locally as the Woo. Oh, yeah. I want
• 4:11
  to pause on the Woo for a second because this
• 4:13
  is a perfect illustration of where human labor
• 4:15
  bottlenecks the government today. Right. Think
• 4:18
  about what happens right now. When a journalist
• 4:20
  or a citizen requests documents under a Freedom
• 4:23
  of Information Act. It's a nightmare. Total nightmare.
• 4:26
  Civil servants have to manually search through
• 4:28
  physical and digital archives. They said they're
• 4:31
  reviewing thousands of pages. And don't they
• 4:33
  have to manually redact things? Exactly. They
• 4:36
  have to use a literal or digital black marker
• 4:38
  to redact sensitive personal data of citizens.
• 4:41
  Check every single line for state secrets and
• 4:44
  compile it all within a strict legal deadline.
• 4:47
  It is. agonizingly labor intensive. Right. So
• 4:51
  by reforming the law itself, the government is
• 4:54
  trying to change how much information it is legally
• 4:57
  required to actively process and publish in the
• 5:00
  first place. Yes, specifically so it takes fewer
• 5:02
  humans to run the system. But I mean, I look
• 5:06
  at this timeline, though, and I just don't buy
• 5:08
  the sequence of events. It feels like it's like
• 5:11
  they are going on a diet by throwing away all
• 5:13
  their clothes and buying smaller pants before
• 5:16
  they've actually lost the weight. That's a great
• 5:18
  way to put it. Right? The desks are being removed.
• 5:20
  The buildings are being sold off. The laws are
• 5:22
  being rewritten today, long before the AI systems
• 5:25
  meant to replace these workers are proven to
• 5:28
  work at a national scale. Yes, they are throwing
• 5:30
  away the pants and they haven't even signed up
• 5:32
  for the gym yet. They are shrinking the physical
• 5:34
  and legal footprint of the government just to
• 5:36
  force the digital transition. I mean, if the
• 5:39
  desks are gone, the humans have to leave. And
• 5:42
  if the humans leave, something else has to process
• 5:44
  the paperwork tomorrow, which is where the government
• 5:46
  is placing this massive bet on AI stepping in
• 5:49
  to fill the void. But is that a realistic roadmap?
• 5:52
  Or is it just convenient political rhetoric to
• 5:55
  justify austerity? Because the timeline here
• 5:57
  is full of massive holes. Yeah, let's talk about
• 6:00
  the timeline. The Dutch parliament passed motions
• 6:02
  calling to increase government productivity by
• 6:05
  20 % using AI. But, and this is a huge but. Those
• 6:10
  motions are entirely non -binding. Completely
• 6:12
  non -binding. And when you look for the concrete
• 6:14
  roadmap, like the detailed savings plans for
• 6:17
  how AI will actually achieve this, the government
• 6:19
  admits those won't even be presented until the
• 6:22
  first quarter of 2027. Which is wild. You are
• 6:25
  executing physical cuts and budget reductions
• 6:28
  today based on a technological master plan that
• 6:31
  won't even be drafted for another two years.
• 6:33
  The friction there is just obvious. And this
• 6:35
  is causing massive skepticism. Especially when
• 6:38
  you overlay this plan with the broader landscape
• 6:40
  of government AI adoption. Right, because governments
• 6:43
  aren't exactly known for being tech pioneers.
• 6:45
  Not at all. An OECD report from September 2025
• 6:48
  painted a very stark picture. It showed that
• 6:51
  government AI adoption severely trails the private
• 6:54
  sector. Private companies are deploying AI at
• 6:57
  scale to cut costs, sure, but most government
• 7:00
  initiatives remain completely stuck in pilot
• 7:02
  phases. Which makes this whole thing feel so
• 7:05
  premature. And we are seeing critics call this
• 7:07
  out openly. Labor MP Annalise Dodds over in the
• 7:10
  UK warned about what she called performative
• 7:13
  mentions of AI. Performative mentions. That phrase
• 7:17
  captures the anxiety perfectly, doesn't it? It
• 7:19
  really does. The implication is that AI is functioning
• 7:22
  as a shiny buzzword. It's just a way to make
• 7:25
  preplanned redundancies and brutal budget cuts
• 7:27
  look like a sophisticated, forward -thinking
• 7:29
  technological strategy. Yeah, I have to push
• 7:32
  back on the core assumption of technological
• 7:34
  readiness here, because when you look at other
• 7:37
  advanced economies, the idea of a seamless leap
• 7:40
  to a hyper -efficient AI -driven state just seems
• 7:42
  completely absurd. Oh, totally. Like, look at
• 7:44
  the UK. As recently as early 2025, nearly 10
• 7:48
  % of English health trusts still lacked basic
• 7:51
  electronic patient records. That's crazy. Right.
• 7:53
  You have modern hospitals dealing in physical
• 7:55
  paper files. And at the same time, the UK's Environment
• 7:58
  Department had AI guidance that was still telling
• 8:01
  staff to use Google Bard. Oh man, Bard. Yes,
• 8:04
  a brand like Google retired years ago in favor
• 8:06
  of Gemini. So how can a government mandate a
• 8:09
  leap to an AI workforce when their basic IT infrastructure
• 8:11
  is lagging that far behind? You have hit on the
• 8:14
  absolute foundational problem. AI is not magic.
• 8:18
  It requires a modern, clean, unified data infrastructure
• 8:22
  to function at all. It needs good data. Exactly.
• 8:25
  If your data is sitting in filing cabinets or
• 8:28
  if it's trapped in outdated proprietary legacy
• 8:30
  systems that literally cannot communicate with
• 8:33
  each other, an AI model is completely blind.
• 8:36
  You cannot automate a process that hasn't been
• 8:39
  digitized in the first place. And you certainly
• 8:41
  can't automate a process where the underlying
• 8:43
  data is fragmented across a dozen different incompatible
• 8:46
  servers. Right. But let's play devil's advocate
• 8:49
  for a second here. Okay. Let's assume the Dutch
• 8:51
  government somehow manages to modernize its underlying
• 8:54
  IT infrastructure overnight. The immediate next
• 8:58
  hurdle is a massive political minefield. Whose
• 9:01
  AI are they actually going to use? Oh, right.
• 9:05
  Because in Europe right now, this touches on
• 9:07
  the deeply sensitive issue of digital sovereignty.
• 9:10
  It's a huge issue. The Dutch are making a pointed
• 9:13
  effort to favor domestic and European tech suppliers.
• 9:17
  over American or Chinese products. They really
• 9:19
  want to avoid being locked into the ecosystems
• 9:22
  of foreign tech giants. And we are already seeing
• 9:25
  them take concrete and honestly painful steps
• 9:28
  away from big tech. The Dutch tax office is currently
• 9:31
  abandoning the Microsoft 365 cloud environment.
• 9:34
  Which is huge. It is. They are migrating back
• 9:37
  to their own servers and utilizing open source
• 9:39
  software. The government is even building its
• 9:41
  own customized Linux desktop operating system
• 9:44
  for its workers. I just... Transitioning a massive
• 9:48
  bureaucracy off a familiar system like Windows
• 9:50
  onto a custom Linux desktop is notoriously difficult.
• 9:54
  I mean, have you ever tried to switch a non -technical
• 9:56
  worker to a brand new operating system? It's
• 9:58
  a nightmare. People hate changing their workflow.
• 10:01
  It is a massive shock to the daily workflow of
• 10:03
  a civil servant. Usually the path of least resistance
• 10:06
  is to just buy an off -the -shelf enterprise
• 10:07
  solution from a U .S. tech giant. But this pivot
• 10:10
  reflects a very real anxiety across Europe regarding
• 10:13
  data security. It's about who holds the keys
• 10:16
  to the kingdom. If a U .S. company runs the AI
• 10:19
  infrastructure, there is an underlying fear about
• 10:21
  the U .S. government legally demanding access
• 10:23
  to Dutch citizens' data under laws like the CLOUD
• 10:26
  Act. And that fear is driving policy everywhere.
• 10:29
  We saw this explode in the U .K. over health
• 10:31
  care data recently. Right, the Palantir contract.
• 10:34
  Exactly. The U .K. government signed a 330 million
• 10:38
  pound contract with the U .S. firm Palantir to
• 10:40
  manage health data. And the public backlash was
• 10:43
  immediate. And intense. Because it's medical
• 10:46
  records. Yes. UK AI Minister Kanishka Narayan
• 10:50
  actually addressed it head on, noting the deep
• 10:53
  public discomfort with sensitive citizen data,
• 10:55
  especially personal medical records, ending up
• 10:58
  in the hands of, quote, foreign billionaires.
• 11:00
  I look at this buy local mandate, though, and
• 11:03
  it seems like it could cripple this massive efficiency
• 11:06
  drive before it even starts. How so? Well, the
• 11:09
  reality of the global AI race right now is that
• 11:11
  very few cutting edge frontier models are actually
• 11:14
  built in Europe. So if a government restricts
• 11:17
  itself to only using domestic or European suppliers,
• 11:20
  aren't they basically forcing themselves to use
• 11:22
  inferior tools to solve a massive productivity
• 11:26
  crisis? That is the ultimate tradeoff they are
• 11:28
  navigating right there. It is a constant friction
• 11:31
  between optimal efficiency and national sovereignty.
• 11:35
  I mean, relying solely on domestic tech means
• 11:37
  you probably won't get the absolute fastest,
• 11:40
  most capable reasoning model available on the
• 11:42
  global market. But when you were dealing with
• 11:45
  the nation's tax records, legal judgments, and
• 11:48
  the core identities of your citizens, control
• 11:50
  and security just have to supersede raw processing
• 11:53
  power. A sovereign state cannot outsource its
• 11:55
  core functions to a server farm in another jurisdiction.
• 11:58
  Okay, so... Let's say they build the infrastructure,
• 12:01
  they source secure European AI, and they launch
• 12:05
  these automated systems to handle the workload
• 12:07
  of the 25 ,000 human workers they just cut. Here
• 12:11
  is where the plan might actively self -destruct.
• 12:14
  Oh, the flooding issue. Exactly. Deploying AI
• 12:17
  to save money might actually trigger a paradox
• 12:20
  where the technology creates infinitely more
• 12:22
  work for the government. This is honestly one
• 12:25
  of the most fascinating unintended consequences
• 12:27
  emerging in this whole space. There's a recent
• 12:30
  study from the Herdy School in Berlin that documented
• 12:32
  84 separate cases across 11 different countries
• 12:35
  of a phenomenon they call agentic flooding. Agentic
• 12:39
  flooding. So think about what this looks like
• 12:42
  for you as a citizen, okay? So you realize you
• 12:45
  can use an AI chatbot to instantly draft a highly
• 12:48
  polished, perfectly formatted, legal -sounding
• 12:51
  complaint or administrative request. Yep. Yeah.
• 12:54
  So it used to take you like hours of frustrating
• 12:56
  paperwork to file a grievance with the city.
• 12:58
  And now it takes three seconds to generate a
• 13:01
  10 page legal claim. Exactly. And suddenly government
• 13:04
  institutions are being like absolutely bombarded
• 13:07
  with perfectly formatted demands. It is literally
• 13:10
  an automated arms race. So citizens bring a machine
• 13:13
  gun of AI generated complaints, like firing them
• 13:16
  off by the thousands. And the government is forced
• 13:19
  to build a massive AI shield just to process
• 13:21
  and read and deflect the paperwork. Right. Nobody
• 13:25
  is actually solving problems. So the machines
• 13:27
  are just talking to machines. And the bureaucracy
• 13:29
  is certainly losing that race. Marie Keblom,
• 13:32
  the chief economist at ING, pointed out a crucial
• 13:35
  historical pattern here. Previous cuts to the
• 13:38
  civil service have often resulted in increased
• 13:41
  hiring later on just to deal with the resulting
• 13:43
  backlogs. Because AI doesn't just process existing
• 13:47
  needs. Exactly. By lowering the friction for
• 13:50
  citizens to interact with the government, the
• 13:52
  technology creates entirely new needs. The state
• 13:55
  might find itself forced to hire more people
• 13:57
  just to manage the influx of automated requests.
• 13:59
  And beyond the sheer volume of work, I look at
• 14:01
  the financial modeling here and see a massive
• 14:04
  trap regarding vendor lock -in. Oh, for sure.
• 14:07
  New Zealand is actually undergoing something
• 14:09
  similar right now. They are cutting 8 ,700 public
• 14:12
  service jobs and taking this really... quote
• 14:15
  -unquote aggressive approach to AI to bridge
• 14:18
  the gap. And internal documents from New Zealand
• 14:20
  revealed a shocking blind spot. Officials entirely
• 14:24
  failed to model the actual long -term costs of
• 14:26
  rolling out this AI infrastructure. Because once
• 14:29
  all the human workers are gone, you are completely
• 14:31
  at the mercy of the AI vendor's pricing strategy.
• 14:34
  You are essentially trapped. Right. To understand
• 14:37
  the danger here, you really have to look at how
• 14:39
  AI pricing works. In the AI world, you are billed
• 14:43
  by the token. which is essentially a fragment
• 14:45
  of a word. So every word costs money. Yeah. Every
• 14:48
  single time the AI reads a citizen's complaint
• 14:50
  or writes a government response, the meter is
• 14:53
  running. The government is paying for every single
• 14:55
  token. Once competition thins out and the state
• 14:59
  is completely dependent on a specific AI model
• 15:01
  to keep the lights on, the providers can easily
• 15:04
  jack up token prices and licensing fees. So the
• 15:07
  3 .5 billion euros the Dutch hope to save on
• 15:10
  human salaries could just evaporate overnight.
• 15:13
  eaten up by spiraling software licensing costs.
• 15:15
  Exactly. And shrinking a government by 25 ,000
• 15:18
  jobs is not just a software update. There is
• 15:21
  a profound human element here, both for the workers
• 15:24
  being displaced and for you, the citizen, relying
• 15:26
  on the state. The workforce dynamics alone are
• 15:29
  a massive hurdle. Civil servants are not a standard
• 15:33
  demographic. They are heavily unionized and uniquely
• 15:36
  positioned to resist sweeping cuts. Right. They
• 15:39
  have a lot of power. To give you some context
• 15:41
  on the density, in the U .S., a quarter of federal
• 15:44
  employees belong to a union compared to just
• 15:46
  10 % of the general workforce. In South Korea,
• 15:49
  a massive 66 % of civil servants are unionized.
• 15:53
  And the Dutch public sector unions are incredibly
• 15:56
  strong. They are expected to fight these cuts
• 15:58
  fiercely. And it isn't just about preserving
• 16:01
  jobs. It is about protecting the fundamental
• 16:03
  nature of public service. Look at France for
• 16:07
  a totally different approach to this. France
• 16:09
  is currently rolling out an AI assistant to all
• 16:12
  of its public sector workers. And early trials
• 16:14
  showed it cut the time spent summarizing documents
• 16:17
  by 16%. Which is a solid gain. It is. But the
• 16:20
  French government explicitly promised that this
• 16:22
  technology will not be used to replace civil
• 16:25
  servants. More importantly, they stated it will
• 16:27
  not be used to dehumanize relations between the
• 16:30
  state and the citizen. The AI is an assistant
• 16:32
  to the human, not a replacement. Contrast that
• 16:35
  with the Dutch plan, where those delivery agencies,
• 16:38
  the exact departments where you go when you need
• 16:40
  help, are facing up to 60 percent cuts. It's
• 16:43
  completely different. It fundamentally alters
• 16:46
  how you interact with your government. And this
• 16:48
  brings me back to the danger highlighted in that
• 16:50
  Hurdy School study on agentic flooding. They
• 16:53
  noted that the quickest fix for governments facing
• 16:55
  an avalanche of AI complaints is to simply impose
• 16:58
  financial fees on filing a claim just to slow
• 17:01
  down the volume. Which creates a massive immediate
• 17:04
  equity issue. Right. If the government erects
• 17:08
  a paywall to stop the chatbots, doesn't that
• 17:10
  just shut out real people with genuine non -automated
• 17:13
  grievances? Absolutely, it does. If you have
• 17:16
  been wronged, but you can't afford the fee designed
• 17:18
  to stop the AI spam, you are entirely locked
• 17:22
  out of the system. How much efficiency will you
• 17:25
  actually tolerate before you feel completely
• 17:27
  abandoned by a faceless automated state? That
• 17:30
  is the line every government has to walk in the
• 17:33
  next decade. When you have a complex problem,
• 17:35
  something involving your health care, your taxes,
• 17:38
  your housing, you want to talk to a human being
• 17:41
  who has the discretion and empathy to understand
• 17:43
  your unique context. Because life is complicated.
• 17:46
  Exactly. And AI operates on parameters and probabilities.
• 17:50
  It does not have discretion. If you hollow out
• 17:53
  the delivery agencies by 60%, you are removing
• 17:56
  the human safety net that catches the edge cases.
• 17:59
  And honestly, life is nothing but edge cases.
• 18:01
  When those edge cases inevitably get mishandled
• 18:04
  by an automated system, the erosion of public
• 18:06
  trust could be catastrophic. The efficiency gains
• 18:09
  might look incredible in a spreadsheet in 2027,
• 18:12
  but the social cost of removing humans from the
• 18:15
  loop could be immense. It really forces us to
• 18:17
  rethink what a government is actually for. Is
• 18:20
  it just a processing engine for rules or is it
• 18:23
  a human institution designed to serve society?
• 18:26
  As we watch governments turn to AI to cut human
• 18:28
  jobs, it leaves you wondering about the future.
• 18:31
  Yeah. Will this looming bureaucratic battle between
• 18:34
  citizen AI generating endless requests and government
• 18:37
  AI trying to process them eventually require
• 18:40
  hiring a whole new class of human civil servants
• 18:43
  whose only job is to act as referees between
• 18:46
  the machines? It's a crazy thought, but we might
• 18:48
  just need them. Thanks for listening to the AI
• 18:50
  newspaper. Visit AINewspaper .org to catch up
• 18:54
  on the latest updates and sign up for free email
• 18:56
  alerts.