Microsoft New Future of Work Report 2025

Source: Microsoft Research · 2025-12

A summary of recent research from Microsoft and around the world that can help us create a new and better future of work with AI.

Introduction. Welcome to the 2025 Microsoft New Future of Work Report!

As you sit down to read the 2025 New Future of Work report, it’s worth pausing to consider the thread that ties the past five years of reports together. The inaugural New Future of Work report, published in 2021, focused on new ways people could work without relying on colocation as a key productivity tool. The second, in 2022, centered on the reintroduction of physical offices and the emergence of hybrid work. In 2023, we explored how large language models could reshape everyday work, and, in 2024, how those advances moved from promise to real-world impact.

Each year, as I’ve written this introduction, I’ve found myself saying that the previous year marked a once-in-alifetime generational shift. But after five years, it’s clear that the reports aren’t capturing a series of separate revolutions. Rather, they are chapters in a single story of the digital evolution of collaboration, each representing a phase that builds on, and is enabled by, what came before.

Last year’s report highlighted research showing that AI delivers substantial gains in individual productivity. The next frontier, covered in this year’s report, is collective productivity: how teams, organizations, and communities can get better together. AI can bridge gaps of time, distance, and scale, but only if built correctly. We must design AI to support shared goals, group context, and the norms of collaboration, and this requires not just new tools but new ways of working.

This report provides research-backed insights into how AI is (or sometimes, should be) shaping work. Some of the questions it addresses include:

• Adoption and Usage: What changes are occurring in adoption and usage? What are the drivers and challenges? What are the gaps?

• Impact on Work and Labor Markets: How is AI impacting work and productivity? How are jobs evolving? Is generative AI affecting employment and wages? Where might agents reshape markets? What roles do automation and augmentation play?

• Human-AI Collaboration: How are the ways people interact with AI changing? How can human-AI collaboration be improved? How does AI use differ across modalities and time frames?

• AI for Teamwork: How can AI support teams as well as individuals? What role can AI play in team settings? What is needed to effectively integrate AI into group workflows?

• Thinking, Learning, and Psychological Influences: What are the effects of AI on cognition and thinking? Can AI be designed not just to create useful output, but to make the people who work with it smarter? How can AI serve as an effective classroom tool? Is it possible to measure psychological or well-being impacts from AI?

• Specific Roles and Industries: How is AI changing work for software engineers, program managers, researchers, and other professions?

• External Voices: What do leading scholars outside of Microsoft think are the most critical topics in this space?

Method. Key context: AI capabilities continue to advance, especially due to reinforcement learning • Long-horizon task completion capability is measurably accelerating. METR’s 50% task-completion time horizon shows frontier agents’ reliable task length has been rising exponentially with an ~7-month doubling time, turning “agent progress” into a concrete capability trend (Kwa et al., 2025). • Verifiable reinforcement learning (RL) post-training (rewarding correct, checkable outcomes) enabled strong gains on hard math/coding style tasks even when starting from a base model with no labeled reasoning traces.​ (DeepSeek-AI et al., 2025). • Scalable test-time compute frameworks gained traction, with open-weight models achieving IOI 2025 gold-level performance, showing repeatable “more compute → higher score” curves in competitive programming. (Samadi et al., 2025).

Discussion. Organizational AI adoption depends on employees as much as leaders • Across industries, the intention to use AI is influenced by social norms learned from leaders and peers (Kelly et al., 2023). • Workers can be reluctant to adopt top-down mandated AI products that prioritize efficiency above quality and creativity, undermining the traditional view of humans as the core value driver of businesses. This reluctance limits the success of AI pilot programs (Young et al., 2025; Sharma, 2025; Murire, 2024). • Leaders can facilitate AI adoption through clear communication supporting AI use, demonstrating their own learning, and setting realistic expectations about what AI can accomplish (Carter et al., 2024; Tursunbayeva & Chalutz-Ben Gal, 2024). • AI products that integrate human thinking, creativity, and expertise while amplifying their value can promote adoption without raising concerns about replacement (Ali et al., 2025; Young et al., 2025; Sharma, 2025). For example, an AI assistant can act as a thought partner, helping users explore ideas and connect concepts across their knowledge base. • Some of the best ways an organization might find to use AI “come from the edge, not the center” (Winsor, 2024). Organizations can facilitate AI adoption by creating systems and incentives for employees to share how they use AI with one another (Tursunbayeva & Chalutz-Ben Gal, 2024; Winsor, 2024). • Employees are more likely to experiment with using AI and to share those insights with others when they feel safe and trust their organizations (Tursunbayeva & Chalutz-Ben Gal, 2024; Bankins et al., 2021). • Many employees, particularly Gen X, will not adopt tools that make them conform to a way of working - they want products that are flexible enough to fit personal ways of working (Rozsa et al., 2023; Doblinger, 2023).

Centering worker voice in AI design boosts productivity, satisfaction, and skill growth—driving both business success and worker flourishing • Worker involvement in technology design promotes sustainable productivity and job satisfaction. Historical and contemporary research consistently shows that when workers’ expertise and perspectives inform the design and deployment of workplace technologies, organizations achieve more sustainable improvements in productivity and well-being (Trist & Bamforth, 1951; Roethlisberger & Dickson, 1939; Hackman & Oldham, 1976).

• Ethnographic and HCI research demonstrates that workers adapt technology in creative ways, and that participatory design— where workers are co-designers—results in tools that better fit real workflows and foster higher adoption (Suchman, 1987; Orr, 1996; Awumey et al., 2024; Ehn, 1993; Doellgast et al., 2025).

• Combining technical and social science research methods can create AI systems that improve worker skills and satisfaction— not just accuracy—by embedding human-centric metrics, workers’ values, and skill-building into their design (Bucinca, 2025).

CEOs expect AI to transform their businesses, but leading organizational AI adoption can be challenging • A 2025 IBM survey of 2000 CEOs in 33 countries and 24 industries found that most CEOs expect AI to transform their businesses. Other industry research shows leaders believe having the most advanced generative AI is crucial to remaining competitive (de Bellefonds et al., 2024; IBM Institute for Business Value, 2025).

• However, organizational leaders have difficulties developing top-down AI strategies for many reasons, including the rapid diffusion of AI technologies, the speed with which they change, the need to communicate and reach alignment about AI, the need to prioritize AI against other concerns, and the challenge of reimagining workflows and processes (de Bellefonds et al., 2024; Leonardi, 2023).

Conclusion. • AI’s impact on work is unfolding slowly, much like other general-purpose technologies. Productivity gains rarely appear immediately; they tend to follow a J-curve, with benefits emerging only after adoption, investment, and organizational redesign (Brynjolfsson, 1993; Brynjolfsson et al., 2021). Similarly, current labor market effects remain modest overall (Chandar, 2025; Gimbel et al., 2025; Eckhardt & Goldschlag, 2025). • Yet, early signals point to pressure on entry-level roles in AI-exposed fields, especially where automation is more likely than augmentation. Payroll and other types of data show declines for junior positions, while senior roles remain stable or grow (Brynjolfsson et al., 2025; Hosseini & Lichtinger, 2025; Klein Teeselink, 2025; Humlum & Vestergaard, 2025). • This raises a deeper question about direction: prioritizing human-like AI risks the Turing Trap, a situation where substitution dominates, concentrating economic and political power and limiting broad growth. Augmentation, by contrast, expands human capabilities and creates new tasks, but requires deliberate choices and supportive systems as capabilities and adoption grow (Brynjolfsson, 2022).

Employment changes by age and exposure (Brynjolfsson et al., 2025).

• A recent model (Daley, 2025) demonstrated how increasingly capable, low-cost AI research systems could potentially negatively interact with metric-driven universities and grant funding processes.

• Using an assumption of effective research capability doubling every ~16 months (with 16 being arbitrary and any value greater than zero has the same asymptotic conclusion), this model shows demand for human research labor collapses exponentially—a dynamic consistent with automation models showing declining labor share when machines displace human tasks (Acemoglu & Restrepo, 2018). This risks less human input to the core directions of research endeavors.

• Organizations must actively intervene to build guardrails and structures around AI design, implementation and use in ways that mitigate, rather than amplify, inequality cascades of workplace AI systems (Kaurnakaran et al., 2025).