The Welfare System in 2035: Public Sector Has Time for What Matters

But that is not the most remarkable thing. The remarkable thing is that she arrives at the meeting without feeling she is already behind. 

She does not have to search for documents, fill in duplicate forms or take notes while the family is speaking. The administrative work happens in the background. When the meeting begins, she can actually be present. 

Technology no longer consumes her attention. It has become what frees it. 

And perhaps that was where the transformation really began. 

When Digitalisation Became an Obstacle 

Ten years earlier, reality looked different. 

As early as February 2025, the AI Commission had submitted its Roadmap for Sweden. One year later, in January 2026, the Swedish Social Insurance Agency and the Swedish Tax Agency were tasked with establishing a joint AI workshop for public administration. And in February 2026, the government presented Sweden’s first comprehensive AI strategy. The ambition was clear: Sweden was to become a world leader in the use of AI in the public sector. 

But at the same time, there was another reality. The Swedish Internet Foundation’s report The Swedes and the Internet 2025 showed that nearly one in four adults had refrained from booking a medical appointment because the e-services were perceived as too difficult. 

That is a figure worth pausing over. 

After decades of digitalisation, the welfare system had become more efficient on paper, but at the same time less accessible to large groups of people. We had built systems that worked for the organisation, but not always for the people it was meant to serve. 

At the same time, demographics were changing rapidly. The number of older people was increasing sharply, while the number of people of working age was barely growing at all. Needs were increasing faster than resources. 

It became increasingly clear that the public sector could no longer improve efficiency simply through more projects or a faster pace of work. Something more fundamental needed to change. 

What Actually Made the Difference 

In hindsight, the AI transformation was never primarily about technology. 

It was about leadership. 

The municipalities and regions that succeeded best were not always those with the largest budgets or the greatest number of initiatives. They were the ones that managed to hold two things together at once. 

As early as 1991, James March described the need to simultaneously explore the new and manage what already works, something that later came to be called organisational ambidexterity, or ambidextrous leadership. In the AI transition, this took a very concrete form: daring to explore new technology without losing grip on day-to-day operations. 

Those who succeeded protected the human dimension of welfare, while being uncompromising toward anything that did not create value. Administration that had accumulated over decades began to be genuinely questioned, not through calls to work faster, but through organisations removing things that were no longer needed. 

This turned out to be considerably more difficult than introducing AI. 

Technology can be installed quickly. Behaviours cannot. 

Many organisations became stuck in pilot projects that were never scaled, because they treated AI as an IT initiative when it was about operational change. Those who succeeded understood that the transformation needed to begin with ways of working, culture and trust, not with the technology. 

From Isolated Projects to Shared Capability 

What truly changed the pace was that the public sector stopped trying to solve the AI challenge in isolation. 

For much of the 2020s, the same work was being carried out in parallel across hundreds of municipalities and government agencies. Similar pilot projects were procured again and again. Competence was built locally, disappeared locally, and then had to start over somewhere else. AI became expensive, fragmented and difficult to scale. 

The turning point came when the focus shifted from individual tools to shared capability. The AI Workshop had been put into use during 2026 and grew into a platform where government agencies, municipalities and regions began to share. New forms of collaboration emerged, where research, responsibility and practice met in the same room. Through partnerships such as the RISE AI Partnership, small municipalities gained access to the same knowledge base as large government agencies, and in secure testbeds employees could experiment and compare models without locking the organisation into a particular supplier. 

AI competence ceased to be something for a small number of specialists and instead became an organisational capability that was built jointly over time. 

The AI maturity of a small municipality was therefore no longer primarily a question of budget. 

It became a question of the ability to collaborate. 

What We Did Not Understand Then 

What proved most difficult was not the technology. It was making decisions without knowing for certain. Those who dared to begin early built competence, ways of working and trust step by step. Those who waited fell several years behind. 

And despite all the progress, technology did not solve everything. AI agents still do not make the decisions in welfare. They summarise, prioritise and flag deviations, but humans remain responsible for the assessments. Validation of models became as self-evident as hygiene routines or financial audits. Not because it was innovative, but because trust requires it. 

We also discovered new forms of exclusion. The digital divide did not disappear. It changed form. 

The Human Being at the Centre, Still 

Perhaps the most interesting thing about welfare in 2035 is that, in many ways, it looks the same as it did in 2026. 

The nurse still meets the patient. The teacher still meets the student. The social worker still meets the family. 

What has changed is what people spend their time on. 

AI did not take over human work. It took over what should never have been people’s main task in the first place. 

In hindsight, the AI transformation appears less like a technological shift and more like a reprioritisation. We moved time from administration to relationships. We moved responsibility from isolated projects to shared capability. We moved trust from promises to verification. 

It did not require science fiction to get there. It required leaders who dared to hold on to two things at the same time: running today’s operations and building tomorrow’s welfare system. 

And perhaps that is why welfare in 2035 feels less technological than many had imagined. Because what changed most was not that machines became smarter. 

It was that people once again had time to be human in the welfare system.