Working With AI
Rethinking Collaboration in the Age of Intelligent Systems
For my master’s thesis, I explored how AI could support collaboration in virtual teams without replacing the trust, agency, and human exchange that teamwork depends on. While much of the current discussion around AI in the workplace focuses on productivity and automation, I was more interested in a different question: how will AI shape the way we work together?
This project began with the observation that virtual teams face structural challenges that go beyond logistics.
Distributed collaboration often relies on asynchronous and socially “thin” communication channels, where nonverbal cues, spontaneity, and informal bonding are largely missing. As a result, misunderstandings can escalate more easily, trust develops more slowly, and social inclusion becomes harder to sustain. At the same time, AI is increasingly being introduced into collaborative systems, often without a clear understanding of what role it should actually play in relation to human teams.
My thesis set out to address this tension from a human-centered and ethical design perspective. The central challenge I defined was to design a collaboration system that restores social regulation, creates safe informal space, and supports trust-building in virtual teams without replacing human exchange. At the same time, the role of AI needed to remain clearly scoped, legible, and accountable by design, so that human agency would not be weakened but preserved as the default.
I researched the challenges of virtual teams, the role of social mediation robots, and relevant psychological theory on groups, mediation, trust, and social regulation.
The foundation of the entire project was an extensive literature analysis.
Rather than collecting sources in a purely academic way, I worked visually and structurally: I summarised key findings and translated them into large mind maps that helped me identify relationships, tensions, and recurring patterns across the research field. This process allowed me to build a mental model of the problem space and later use it as a basis for design decisions. The image here shows that early synthesis work and reflects one of the strongest aspects of my process: turning complexity into structure through research and visual analysis.
One of the most important insights from this phase was that the weaknesses of virtual collaboration are not caused by virtuality itself, but by the way collaboration channels are currently designed and used.
Reduced social presence, fewer informal moments, unclear relational signals, and a lack of trust are not inevitable. They can be addressed if systems are designed to support human interaction rather than simply optimize processes. This became the conceptual core of the thesis and the lens through which I evaluated the role of AI.
From there, I translated the research into a concrete design direction. Instead of designing AI as a neutral tool or autonomous collaborator, I explored the idea of AI as a delegated, identity-bound extension of the user: a system that supports continuity, mediation, and coordination, while leaving authorship, authority, and judgment with the human. This framing allowed me to think not just about interface features, but about responsibility, boundaries, and relationships.
Based on the findings, I developed a feature set that separated functional mediation goals from social mediation goals. This distinction helped me structure the concept while acknowledging that collaboration is always both practical and social at the same time. The functional features focused on process-oriented support, such as maintaining continuity, facilitating coordination, supporting decision-making, and managing information. The social features addressed the more sensitive but equally important layer of virtual teamwork: trust-building, psychological safety, team climate, and human connection. This step shows my conceptual design process very clearly. I did not move directly from research into screens, but first developed an underlying logic for what the system should actually do and why.
The concept eventually took shape around six core functions. “Your Agent” framed AI as a persistent, identity-bound proxy that interacts with the team space as a delegated extension of the person. “Moderator” covered structured support such as summarisation, translation, follow-ups, and subtle conflict repair. “Iterator” focused on making feedback and iterative collaboration more viable at scale. On the social side, “Social Catalyst” aimed to restore opportunities for authentic human connection, “Climate Steerer” addressed tone and psychological safety through careful nudging, and “You” remained the central point of authorship, identity, and control. Together, these features translated abstract theory into a system concept with clear ethical and interactional boundaries.
Business Model Canvas
To further ground the concept, I translated the feature set into a Business Model Canvas. This step helped me connect the interaction concept to a broader product and organizational perspective. It allowed me to think beyond isolated features and consider the system in terms of value, stakeholders, use context, and implementation logic. For me, this was an important conceptual-design step, because it ensured that the project would not remain a purely speculative vision detached from real structures of work. It also reflects my ability to use strategic design methods to give shape and coherence to early-stage concepts.
Once the conceptual foundation was clear, I moved into iterative interaction design. My aim at this stage was not to perfect visuals immediately, but to test how the concept could work as an actual system and how its logic could be expressed through interface structures. I therefore started with low-fidelity wireframing and used it as an exploratory tool to think through interactions, screen relationships, and user control.
The low-fidelity wireframes were a central part of my iteration process. I started with rough sketches and basic screen structures, then refined them through repeated feedback and comparison. At this stage, wireframing allowed me to work quickly, challenge my own ideas early, and keep the concept open enough for critical revision.
This was especially important in a project like this, where questions of role clarity, transparency, and user control had to be reflected not only in theory, but in the actual experience of interacting with the system. The image shows how I used wireframing not as a cosmetic precursor to UI, but as a thinking tool for discussion, testing, and design evolution.
Through this iterative process, I gradually translated the broader concept into clearer and more stable interaction patterns. I focused especially on making the AI’s role legible, ensuring that automated support remained distinguishable from human action, and preserving the user’s authority to approve, edit, or override outcomes. This stage was where many of the ethical principles of the thesis became concrete interface decisions.
The polished wireframes mark the point where the interaction concept had matured into a more coherent system. Compared to the earlier iterations, they show a clearer structure, more deliberate hierarchy, and a stronger sense of how the experience would work across the product. At this stage, my focus shifted toward aligning conceptual clarity with a more refined interface language. These wireframes demonstrate how I develop complexity into usable structures: not by jumping straight to visual polish, but by building up a system carefully through iteration and refinement.
The final stage of the project was a high-fidelity prototype in Figma. This was not just a presentational add-on, but an important storytelling device. Because the thesis addressed a speculative future scenario, the prototype had to make the concept tangible enough for others to understand, evaluate, and imagine in use. I therefore used prototyping as a way to stage the interaction and communicate the product vision in a vivid, experience-based form.
The high-fidelity prototype brought the concept to life and allowed me to present the system as an experience rather than only an abstract design proposal. It helped communicate how AI-mediated collaboration could feel when designed around human agency, accountability, and social sensitivity. This stage reflects two strengths that are central to my work: prototyping as a way of making ideas tangible, and visual storytelling as a way of making complex concepts understandable. To extend that storytelling beyond the live presentation, I also wrote a short narrative scenario for the written thesis, allowing the product vision to remain vivid even in a format where the prototype itself could not be embedded.
What makes this project especially representative of my design practice is the way it combines multiple layers of work into one coherent process. It began with extensive research, moved through synthesis and concept development, took strategic shape through methods like the Business Model Canvas, evolved through iterative wireframing, and culminated in a high-fidelity prototype designed for storytelling and discussion. At every stage, the project was guided by the same principle: technology should not replace human collaboration, but support it in ways that are legible, responsible, and genuinely centered on human needs.