AI and the Future of Work: Uneven, Uncertain, and Unresolved

AI is changing quickly, and the big question is how it will affect work. A recent OECD conference highlighted the technology’s real-time transformation—becoming both an asset and a liability for workers—and pointed to a promising future that will, nonetheless, require guardrails. Here are five of our major takeaways from the discussion.

1. AI’s effects on the labor market are uneven.

Jobs aren’t disappearing yet, but AI is transforming how people work and putting some jobs at greater risk, especially office and administrative support workers, whose tasks are more easily automated. Health care practitioners face lower risk—their roles demand reasoning, physical, and social capabilities AI cannot yet match. A more likely scenario in health care is that AI handles administrative work, freeing up providers for more direct patient care.

Roles requiring physical activity are also unlikely to be replaced soon, since physical machinery and robots are more expensive than AI. Agriculture, personal services, and food preparation could be buffered from AI-related job losses for this reason. Wide-scale job loss is unlikely, experts say, but some roles will disappear, prompting workers to pivot.

Where workers live will also shape their experiences. In Germany, government investment is helping small- and medium-sized enterprises experiment with AI to compete with larger organizations who have invested heavily in the technology. Bosch, a major German technology and services company, has already trained 100,000 employees in its own AI academy. Germany’s approach aims to empower businesses of all sizes, though it remains unclear whether other countries will make similar investments in smaller firms.

2. Generative AI has dominated attention so far, but agentic AI may prove more disruptive.

Generative AI creates content from existing data; agentic AI runs tasks more autonomously. While the former has been the focus of most existing labor research, the latter could be more important to the future of work. For example, while generative AI could power a bank’s chatbot, agentic AI could go further—identifying a potential customer, targeting their interests, scheduling a call, and briefing the employee beforehand. It handles entire processes, not just tasks. With agentic AI evolving rapidly, experts acknowledge a lack of collective clarity around which tools to invest in and how to plan.

How should workers prepare? Experts advise honing critical thinking and problem-solving skills, which are harder for AI to replicate, and emphasize that human judgment and empathy remain essential in fields like health care and teaching. The biggest productivity gains come when AI replaces a process rather than a single task—expect a growing trend toward AI agents handling more elements of a work stream (like queuing up a potential customer for a salesperson) rather than simple tasks (like answering sales-related questions). Policymakers will increasingly focus on AI governance and will seek evidence of best practices to shape future regulations.

3. AI can be a positive force for efficiency and inclusion.

A consistent theme was that AI will continue to drive efficiencies at scale and low cost. Government agencies are beginning to use AI to administer social protections programs, such as unemployment benefits, to reduce paperwork, improve eligibility identification, and free up frontline staff to focus on more complex cases. Automation without human oversight will not succeed, but using AI for rote and time-consuming tasks can make these systems more efficient and responsive.

AI also shows promise for workforce training and inclusion. Vocational, education, and training (VET) programs are seeing early success using AI to train neurodivergent people for  work. Tailoring training to individual needs was previously seen as too costly; AI makes customized learning pathways far more feasible. AI has also made high-touch skills development—such as coaching people on social interactions—more accessible and affordable.

Many instructors currently lack the resources to learn and apply these tools, but the demand will grow. As the workforce in OECD countries continues to age rapidly, definitions of who works will broaden, creating more opportunities for those previously overlooked. More inclusive training will benefit all workers—particularly as AI adoption pushes more people to change careers.

4. Yet, inequities in AI use are already forming.

Larger firms have invested more heavily in AI, consolidating early advantages. Within organizations, AI implementation decisions  frequently bypass frontline workers, creating tensions between outsized expectations and on-the-ground realities. Among workers, adoption is uneven—some become power users  while others resist—and the roles most at risk are clerical jobs, which are disproportionately held by women. Future job loss may not be extensive, but it could be concentrated.

What can be done? OECD countries are developing policies and programs to level the playing field, including sandboxes where smaller firms can experiment without major investment. Labor advocates are working to give workers a voice in AI decision-making. Early adopters are being encouraged to share their experiences with reluctant colleagues. But without better data—ideally collected by trusted government entities—it will be difficult to assess what’s working and where gaps exist. The task ahead is not simply be mindful of growing inequities, but to measure them and find ways to close them.

5. Worker voice and social dialogue are essential to equitable implementation.

Workplaces of the future will have AI embedded in them, but human oversight will remain critical. The more pressing question isn’t whether AI will replace workers—it’s how much power workers will have to shape its implementation.

Early evidence is instructive: Australia’s “robodebt” automated benefits system was error-ridden and later deemed illegal, in part because frontline workers were never consulted. In contrast, Danish organizations have achieved successful AI integration in the workplace through robust processes that mandate dialogue between workers and managers. Whether Denmark’s conditions could be replicated elsewhere remains an open question.

Those who lack power and voice today will likely also have less representation as AI becomes more embedded. But the future is still being written. Ensuring workers have a seat at the table when AI decisions are made will improve the technology’s success and build the trust needed as these tools move from novelty to necessity.