Every government in the rich world is facing the same quiet problem: its workforce is retiring faster than anyone is planning for.
In OECD countries, 27.1 percent of central government employees were 55 or older as of 2023, compared with just 19.1 percent between 18 and 34. In Greece, Italy, Portugal and Spain, more than 4 in 10 central government workers fall into that older bracket. These are not junior staff either, since workers 55 and up hold 42 percent of senior management jobs in OECD administrations on average, meaning the people leaving often carry institutional memory that was never written down.
The usual instinct when someone retires is to post the same job and hire a replacement. A growing body of research, including this piece, argues that instinct is a mistake. Before refilling a role, governments should ask what the departing employee actually spent their time on, and how much of it was searching for documents, checking forms, or drafting routine replies rather than using judgment that only a person can provide.
The evidence that this split is real is piling up. A trial across 20,000 UK civil servants using Microsoft's AI assistant found average time savings of 26 minutes a day, with most of the gain coming from cutting time spent hunting for information. A separate study estimated that 41 percent of public sector working hours involve tasks that AI tools could support, rising to 47 percent for staff who do not deal directly with the public.
A city government in Niterói, Brazil, already runs this model for school enrollment. An AI system reviews residency and medical documents submitted by families, and in cases with standard paperwork it reaches a clear decision 80 percent of the time without a person needing to do the first pass. That is hundreds of hours of staff time freed up, not by cutting a job, but by changing what the job involves.
Here is where it gets interesting, and where the United States offers a warning rather than a model. Washington did not take the slow, careful route: mass federal layoffs and a deferred resignation program pushed out well over 150,000 workers in 2025, hitting agencies like the IRS hard enough to gut large parts of its technical staff. The fallout has been expensive and messy, and several reports since have found that the fear and distrust left among remaining staff is now one of the biggest obstacles to getting federal workers to actually use AI tools.
That is the real lesson for any organization, public or private, watching this play out. Replacing headcount with technology works when it follows a plan built around what the work actually requires, done gradually as people leave on their own terms. It backfires badly when it is done as a shock, because the people left behind stop trusting the tools meant to help them.
Governments also need to be honest that automation of a messy process just makes a messy process faster. The real design question is not whether a task can be handed to AI, but whether a human still needs to catch the cases where the stakes are high, like benefit decisions or legal rights. Skipping that step is how automated systems end up producing fast, confident, and wrong decisions at scale.
The retirements are coming regardless. The choice left to government leaders, and to any manager watching their own workforce age, is whether to use that departure as a chance to rethink the work, or just to keep reproducing the same job descriptions on autopilot.