Energinet

How can leaders use generative AI in a complex, critical-infrastructure organisation without weakening human judgment and responsibility? 

A master’s thesis conducted at Energinet explored Hybrid Intelligence as a leadership practice. Through three iterative design interventions, leaders worked with AI, reflected on its role in their work, and tested ways of combining machine capabilities with human judgment.

The focus was not on implementing more AI, but on developing the leadership practices needed to use it responsibly.

Energinet operates Denmark’s critical energy infrastructure. Its role creates a demanding organisational setting in which stable operations, regulation, accountability, and security must coexist with rapid technological development and the green transition. The organisation also works across complex, data-intensive workflows with long planning horizons, many dependencies, and significant consequences if decisions are delayed or poorly coordinated. This makes AI potentially valuable, but also raises the threshold for when and how it can be used responsibly.

At the same time, Energinet was already moving further into GenAI adoption. An assessment based on the 4P Hybrid Intelligence Framework showed that technological and governance foundations were relatively well established, while leadership learning, shared understanding, and practice-based AI use were still developing.

This created a clear gap between having access to AI and being able to use it well in leadership practice. The project therefore focused mainly on People and Projects: how leaders develop the ability to work critically with AI, and how concrete experiments can become learning spaces for human–AI collaboration.

The setting also made ambidexterity central. Leaders had to explore new AI possibilities while maintaining the reliability, accountability, and operational discipline required in critical infrastructure.

The Hybrid Intelligence element was not a single AI system. It was a designed leadership practice for working with AI.

The project treated human judgment and AI capability as complementary. AI could contribute structure, alternative perspectives, analysis, and suggestions. Leaders remained responsible for interpretation, context, trade-offs, and consequences.

Across the project, HI developed through three iterations.

In the first iteration, leaders worked with AI in low-risk situations. The aim was to establish a shared language and a psychologically safe learning space. AI was used as a dialogue and reflection partner, not simply as a source of answers. Participants increasingly treated AI as something they could think with.

This also strengthened what the thesis describes as a sense of partnership. One of the clearest shifts was in how leaders saw their own role in an AI-shaped future. Through interaction with AI in the HI process, participants moved from uncertainty towards greater confidence in their ability to lead in a context where AI becomes part of everyday work.

The change was not simply about becoming more comfortable with the technology. It was about feeling capable of exercising leadership alongside it — questioning AI, setting boundaries, taking responsibility, and continuing to rely on human judgment. Psychological safety, AI confidence, and a growth-oriented mindset therefore became part of the conditions for HI.

In the second iteration, leaders worked with an Interactive Workflow Assistant and an AI Opportunity Radar. The tools helped them unpack concrete workflows and discuss where AI could contribute, where human judgment was essential, and where responsibility sat.

The value of the artifacts was not the output itself. They made assumptions, decision points, dependencies, and boundaries visible.

This became a central finding: Hybrid Intelligence is created through interaction, not through AI output alone. Leaders interpreted, adjusted, questioned, and rejected AI input in relation to their own context. Human judgment remained active throughout the process.

AI also became a tool for sensemaking. Leaders used it to explore different interpretations, surface assumptions, and connect new technological possibilities to organisational goals and responsibilities. Strategic direction was therefore shaped through dialogue and reflection rather than treated as fixed in advance.

In the third iteration, leaders applied the approach to areas closer to their own responsibilities. Here, HI became most visible when participants had both ownership of the problem and decision responsibility.

Leaders challenged AI’s understanding of the context, questioned data quality, refined problems, and considered where responsibility should remain human. Skepticism became useful when it led to better questions and more careful decisions.

Across the three iterations, the development moved from:

shared understanding → safe experimentation → reflection on concrete workflows → active human–AI co-creation

The project used FERC, FEDS, qualitative coding, and the HI Radar to evaluate this development. The focus was on learning, reflection, judgment, psychological safety, and the quality of human–AI interaction rather than technical performance alone.

The case shows that responsible Hybrid Intelligence depends less on increasing AI use and more on improving the quality of the human–AI relationship.

Several outcomes stood out:

• Human judgment remained central. AI could support analysis and reflection, but leaders retained responsibility for interpretation, trade-offs, and consequences.
• Leadership became a framing practice. Leaders shaped the conditions for AI use by clarifying purpose, responsibility, acceptable risk, and room for experimentation.
• Psychological safety improved the quality of engagement. When uncertainty and critique were legitimate, leaders were more willing to challenge AI rather than accept or reject it too quickly.
• Leaders became more confident in leading with AI. Interaction with AI did not only build technical confidence. It strengthened leaders’ belief that they could continue to exercise judgment, responsibility, and leadership in an AI-shaped organisation 
• Leadership confidence supports organisational trust. In a critical-infrastructure context, responsible AI transformation also depends on the people leading it feeling capable of navigating the technology. Leaders who can work with AI critically and confidently are better positioned to create direction, stability, and trust for the organisation around them. 
• Sensemaking became part of AI use. AI helped surface alternatives and assumptions, but leaders still had to interpret what those inputs meant for the organisation.
• Interaction mattered more than output. The strongest value came from the process of questioning, refining, and contextualising AI input.
• Context and ownership affected the quality of HI. The closer the problem was to a leader’s own responsibility and decision mandate, the more active and meaningful the human–AI collaboration became.
• Skepticism could improve the process. Critical questions about data, domain understanding, and AI limitations often strengthened rather than weakened the quality of the outcome.
• Ambidexterity was practised, not just discussed. Leaders had to hold exploration and operational responsibility together in the same process.

The project also showed a clear progression across the three iterations. Leaders moved from general learning and reflection toward more context-specific and decision-near use of AI. HI therefore became more tangible as the work moved closer to real responsibilities, concrete workflows, and organisational consequences.

The broader lesson is that responsible AI transformation is not achieved through technology and governance alone.

Platform and Policies create the boundaries. People and Projects are where Hybrid Intelligence is learned, tested, and put into practice.

Hybrid Intelligence therefore develops through repeated interaction, reflection, and judgment — not simply through greater AI adoption.