Why the retirement wave could be government’s AI moment

The debate about artificial intelligence (AI) and work tends to focus on a familiar question: which jobs will AI replace? However, when it comes to government and the public sector, that may be the wrong starting point.

A more useful question is emerging from a demographic reality already visible across many countries: as public servants retire, should governments automatically replace every position, or should they first redesign the work itself?

According to the Organisation for Economic Co-operation and Development’s (OECD) Government at a Glance 2025, 27.1% of employees in central administrations across OECD countries were already aged 55 or older in 2023, compared with 19.1% aged 18 to 34.

In Greece, Italy, Portugal and Spain, more than 40% of the central government workforce was 55 or older. The risk is not only the number of people who may leave over the coming years. Workers aged 55 and above hold, on average, 42% of senior management positions in OECD central administrations, meaning that retirement can also take decades of institutional knowledge with it.

This workforce transition is happening while the populations that governments serve are ageing as well. The International Monetary Fund (IMF) projects that the old-age dependency ratio across OECD countries will rise from about 33 older people for every 100 working-age people in 2025 to 52 by 2050, and estimates that ageing could increase public spending on pensions and healthcare by around 3% of gross domestic product (GDP) over the next 25 years.

Governments will therefore be asked to deliver more services to ageing societies while a significant share of their own workforce approaches retirement and fiscal space becomes harder to find.

This is where AI changes the equation, not as a pretext for mass layoffs, but to use natural workforce turnover as an opportunity to redesign organizations before simply reproducing the same structures and roles.

Evidence is beginning to show what this could mean in practice. In 2025, the UK Government Digital Service published results from a trial involving 20,000 civil servants across 12 organizations using Microsoft 365 Copilot.

Participants reported saving an average of 26 minutes per day, equivalent to roughly 13 working days a year, and more than 70% said the tool reduced time spent searching for information and performing mundane tasks while increasing the time available for more strategic activities.

A later evaluation by the UK Department for Work and Pensions, involving more than 3,500 employees and using a comparison group and econometric analysis, estimated an average saving of 19 minutes per day across routine tasks, with some of the largest gains in information search and email writing. These findings are not measures of job replacement, but they do show that parts of public sector work can already be performed differently.

Research from the Alan Turing Institute points in the same direction. Using time-use data from Britain’s public sector, researchers estimated that about 41% of public-sector working time is spent on activities that could be supported by generative AI, with the share rising to 47% among non-frontline workers and falling to 38% among frontline workers.

The distinction matters because being suitable for AI support does not mean that the person performing the task is redundant. It means governments can start separating work that requires judgement, empathy, discretion or accountability from work that consists largely of searching, checking, drafting, classifying and moving information through a process.

That shift in perspective matters because the relevant unit of automation may increasingly be the task rather than the job.

I have seen this transition firsthand in Niterói, Brazil, where the municipality uses Colab’s Government AI Layer in its school pre-enrolment process. Across 14,985 applications processed through services using AI, the system performed 11,733 document analyses, including proof-of-residence documents and medical reports.

In the three services with standardized outputs, the AI produced a conclusive classification in 8,759 of 10,908 analyses, or 80.3%, determining whether documents met the requirements configured for the service.

Human review and administrative steps remain part of the workflow, but the significance is already clear: AI is not only helping public servants work faster, it is beginning to perform parts of the administrative work itself. At an estimated four minutes of manual work for each conclusive classification, this volume corresponds to roughly 584 hours of staff time and offers a concrete example of how task-level automation can absorb part of the workload that governments would otherwise need to reproduce as employees retire.

This suggests a different model for public sector automation, one that could be described as attrition-led automation. Imagine an agency where 100 people perform a mix of case processing, document review, database searches, drafting, verification and citizen communication.

If 20 employees retire over several years, the traditional assumption is that 20 replacements are required. A different approach would first examine the work those employees performed, identify which components can be automated or augmented, preserve human review where consequences are significant, and only then determine which capabilities actually need to be hired back.

The answer may still be 20 people, but it may be 15 or 10, or it may be that the same budget is better redirected towards nurses, teachers, inspectors, social workers or other roles where human capacity remains scarce.The objective should not be headcount reduction as an end in itself, but a more deliberate allocation of human effort.

The same transition could also help governments address the loss of institutional memory that comes with retirement. Experienced public servants often carry knowledge about processes, exceptions, rules and informal practices that was never fully documented.

Governments can use the period before these employees leave to capture that knowledge, redesign workflows and build systems that make it available to the next generation of public servants. In this sense, AI can become not only a productivity tool, but part of a deliberate succession strategy that helps institutions retain capability even as people change.

None of this removes the need for safeguards. Public sector AI requires clear accountability, auditability, security, contestability and human oversight, especially where decisions affect rights, benefits or access to services.

Automating a poorly designed process can simply make bad bureaucracy faster, so the goal should be to redesign work before automating it and to be explicit about where human judgement must remain central.

The retirement wave is coming whether governments are technologically prepared for it or not. Instead of recreating today’s bureaucracies one retiring employee at a time, governments can use natural workforce turnover to rethink how administrative work is organized, which tasks can be performed differently and where human judgement adds the most value.

The most important workforce transition that AI brings to government may therefore begin not with a layoff, but with a retirement.