The breakfast seminar was hosted by the House of Innovation, and centered on preliminary findings from an ongoing large-scale field experiment on GenAI adoption in knowledge work, conducted in collaboration between the Stockholm School of Economics and Scania.  

It brought together leaders from retail, media, and manufacturing to examine a central dilemma facing organizations today: should leaders let a thousand experiments bloom and risk scattered pilots disconnected from real value? Or should they centralize efforts and risk moving too slowly in a rapidly evolving landscape? 

The mood was set early when the moderator, Assistant Professor Sebastian Krakowski, asked by show of hands how many people had used AI that morning. The audience exchanged an amused look as almost everyone raised their hands. Needless to say, this group did not need convincing that AI matters.  

The majority of attendees used generative AI on a daily basis.

From access to impact: What changes behavior at scale? 

Sebastian Krakowski, Assistant Professor at the House of Innovation, shared preliminary results from a randomized field experiment run with Scania as part of OpenAI's early adopter program.

One of the larger controlled studies of GenAI in the workplace to date, the study comprised three groups: a control group with no access, a group given ChatGPT access only, and a group given access plus structured training. The groups were compared over roughly six months in late 2024 and early 2025.  

The first finding was striking in itself: adoption was not a challenge. Over 80% of participants used the tools regularly, and weekly usage reached 70%, far above what many expected. Even in the control group, around half found their own way to use generative AI tools for innovation tasks, reflecting just how accessible these tools have become. 

But access alone wasn't enough to unlock value. A recurring theme in interviews was disorientation. As one director from the access-only group described it: "We could do so many things. I don't know where to start. I'm very lost, and the people around me are lost too. So, I just close it and get on with my daily tasks." 

Moderator Sebastian Krakowski, Assistant Professor at the House of Innovation.

Training changed that. Of those offered training, 50% attended, and usage patterns diverged noticeably. Power users (those using the tool a lot) increased from 7% to 12%, median usage rose, and non-users dropped from 5% to just 1%.

The training, in Krakowski's telling, worked not because it taught prompting tricks, but because it gave people a starting point.  

As one participant in the training group put it: "It was a discovery journey of a lifetime. It helped me figure out how to use it in my own work, in a very specific, focused way. Then I just went from there." 

The experiment also measured innovation outcomes, asking participants to tackle challenges around urban air quality and music composition. The results were notable: the AI-assisted groups generated around three times as many ideas and, crucially, those ideas were rated as higher quality by a panel of blind judges. More use, and better output. 

Krakowski drew a parallel to his earlier research on AI-assisted chess, where two amateur players — a snowboard coach and a school administrator — beat the world's third-ranked grandmaster in an early freestyle tournament. Their chess skill was less important than their skill in working with AI.  

"The interesting thing," he noted, "is that the chess skill became irrelevant. What mattered was your ability to use AI: when to use it, which part of the game, and how to prepare."  

The implication for workplaces was summed up in a well-known quote: “AI won't replace you, but someone using AI will.” 

Peter Muld (ICA), Jan Guhrés (Scania) and Ida Hansson Brusewitz (DI). 

Panel discussion: From experimentation to strategic value 

Krakowski then invited the panel: Ida Hansson Brusewitz (Editor, Dagens Industri / Di Digital), Peter Muld (CIO, ICA), and Jan Guhrés (Senior Manager Business Enabling Services, Scania Group). The conversation ranged across three key themes. 

Balancing experimentation and coordination 

Scania's approach to the field experiment itself was deliberately open. It gave employees access to the tools and letting the organization experiment, rather than directing them toward specific use cases.

As Guhrés explained, the goal was to see how people would find value themselves. This approach is quite unusual, as most AI rollouts in large organizations are top-down and targeted. 

Peter Muld described a hybrid approach at ICA: more structured and top-down for large-scale process transformation (where efficiency and food pricing pressures drive priorities), but more exploratory when it comes to personal productivity, where employees are given tools and the space to adapt them. 

Both agreed that while letting a thousand experiments bloom has its place, leadership plays a crucial role in focusing attention on strategically important areas and preventing fragmentation.  

One analogy that landed well was: thinking of AI adoption not as a single top-down decision, but as a portfolio of organizational experiments (testing at small scale, learning, then scaling what works). 

Sebastian Krakowski (SSE), Peter Muld (ICA) and Jan Guhrés (Scania). 

Redesigning workflows for lasting impact 

A recurring theme was that real value from GenAI comes from rethinking workflows end-to-end, rather than layering it on top of existing processes. Organizations seeing measurable gains tend to revisit task allocation, decision rights, and performance metrics alongside the technology. 

"If I save 15 minutes, I'll use them to work more, not work less. So it's hard to capture that gain directly," Muld said.  

The value is real, he argued, but diffuse. It shows up in culture, engagement, and idea generation rather than clean before-and-after metrics. 

Guhrés connected this to what Scania is learning from the experiment. He said the goal isn't just to make existing work faster, but to shift people's time and energy toward higher-value activities. To move from narrowly defined roles toward a broader "transport ecosystem" contribution.

Using AI on routine tasks, he suggested, is how you free people up to do more. 

 
Jan Guhrés (Scania) reflected on the outcomes of the study conducted in collaboration with SSE. 

From efficiency tool to strategic capability 

The panel then turned to a bigger strategic question. If everyone has access to the same tools, does AI still confer competitive advantage?  

The short answer from the room: yes and no. 

Muld pushed back gently on the purely theoretical view. In practice, early and deep adoption (especially embedding AI in core business processes rather than just personal productivity) can be a meaningful advantage, at least temporarily.  

Ida Hansson Brusewitz reflected on what this means for media and journalism – a sector already being reshaped by AI. She noted the fragmentation in adoption she sees across companies and roles.

Some teams, especially in tech, move very fast, while others use AI for marginal improvements like tidying up emails. The more transformative opportunities are still unevenly distributed. 

The panel agreed that while AI may standardize certain outputs, differentiation increasingly comes from judgment, brand, trust, and organizational capabilities that extend beyond the tool itself.  

"The tool doesn't really matter," Guhrés observed. "What we're trying to change is how people think. Once we've done that, they'll adopt anything, and keep evolving." 

The audience actively contributed to the conversation with interesting perspektives and questions. 

Collaboration as a foundation for rigorous knowledge 

This breakfast seminar was a strong example of what emerges when academia and industry come together on a deeper level.

Rather than swapping anecdotes, the collaboration with Scania allowed researchers to run a controlled experiment, measure actual performance outcomes, and generate findings that go beyond what any single company could produce on its own. 

The audience kept the conversation honest, energetic, and, at moments, genuinely surprising. Questions from the floor ranged from how to monetize AI when everyone has access, to whether organizational structures built around knowledge work will need to be fundamentally redesigned. 

As William Gibson once wrote: the future is already here, it's just not evenly distributed. On the evidence of this seminar, the same is true of GenAI at work. 

Thank you to the fantastic speakers, panelists, and participants for an engaging and thought-provoking morning at the Stockholm School of Economics.