Large language models are changing the relationship between people, software, and computers. Rather than only selecting functionality that has already been implemented in applications and interfaces, people can increasingly describe what they want to accomplish and have computational functionality created in response.

I call this LLM-mediated computing. I am interested in what happens when the capabilities of a computer are no longer entirely predefined, but instead emerge through interaction between a person, a language model, and the underlying computational system.

My work in this area spans several connected questions.

One concerns computing beyond predefined applications. If executable code, interfaces, and functionality can be generated in response to people's intentions, the computer becomes a more malleable medium. My work on reflective conversation, proto-interpretation, and generative user interfaces examines how such systems interpret instructions, produce computational possibilities, and allow people to iteratively shape what the computer does.

A second concerns the future of programming. Together with Jakob Tholander, Martin Jonsson, and others, I study how programming changes when people increasingly work with language models that can generate and modify code. Rather than treating AI simply as another programming tool, we are interested in whether this changes the role of the programmer, the status of source code, and the relationship between intention, specification, implementation, and execution. This includes our work on co-coding with AI and the possibility that programming increasingly becomes something performed through interaction rather than through directly writing every instruction.

A third concerns programming education. If producing software increasingly involves describing intentions, evaluating generated solutions, debugging through dialogue, and understanding code that one did not write directly, then learning to program may also need to change. In my teaching and research, I am exploring how large language models can be incorporated into programming education while maintaining an emphasis on computational understanding, critical evaluation, and the ability to reason about what programs actually do.

Across these strands, the broader question is what happens to computing when programming becomes increasingly intertwined with interpretation. Who can shape a computer, what knowledge is required to do so, and what does it mean to understand and take responsibility for computational behaviour when some of the translation from intention to execution is performed by a language model?

Related work includes Reclaiming the Computer through LLM-Mediated Computing, Co-Disclosing the Computer: LLM-Mediated Computing through Reflective Conversation, Proto-Interpretation: The Temporality of Large Language Model Inference, Beyond Interface Generation: GenUI as Interpretive Runtime, and The End of Programming as We Know It: Envisioning Radical Re-Conceptualizations of Co-Coding with AI.

A longstanding part of my research concerns what happens when digital technologies become embedded in everyday life.

Smartphones are particularly interesting because they are not simply devices that we occasionally use. They are carried with us, provide access to many different services and infrastructures, mediate communication with other people, record aspects of our behaviour, and increasingly participate in how we navigate and understand everyday situations.

My work in this area has examined mobile communication, social media, self-tracking, digital wellbeing, patterns of smartphone use, and people's experiences of living with digital technologies. More recently, I have been interested in understanding smartphones as layered forms of mediation: how the device, operating system, applications, algorithms, notifications, and wider infrastructures become entangled in people's experiences.

This research has also contributed to work outside academia. I have been part of an expert group supporting the Swedish Public Health Agency in its scientific evaluation of research on screen time and digital media use among children and young people. This work connects questions about everyday technology use to the broader challenge of translating a complex and sometimes contradictory evidence base into guidance and public policy.

Across this work, I am interested not simply in whether technologies have positive or negative effects, but in understanding how relationships between people and technologies develop, how technologies come to structure everyday possibilities, and how we can make sense of their consequences without reducing complex patterns of use to simple measures of exposure.

Examples of related work include Personal Tracking as Lived Informatics (2014), Personal Tracking of Screen Time on Digital Devices (2016), Uncovering smartphone usage patterns with multi-view mixed membership models (2016), and “In our world, calories are very important” (2025), a study of wheelchair users' experiences with commercial self-tracking technology.

The Image Explorer is an ongoing collaboration with GPS400: Centre for Collaborative Visual Research at the University of Gothenburg.

The project explores how AI models can support new ways of navigating large visual collections. Rather than treating AI primarily as a way of making image search faster, we are interested in whether computational interpretations can reveal similarities, contrasts, clusters, and unexpected relationships that invite further human interpretation.

We are developing tools for exploring thousands of images through representations produced by contemporary AI models. This makes it possible to move through a visual collection according to relationships that are difficult to capture through conventional metadata or predefined categories.

The project raises questions about the relationship between computational and human interpretation: the model can suggest structures and connections, but what those connections mean remains something for researchers and users to investigate.

The Image Explorer is developed together with researchers at GPS400 through Appademin at the Department of Applied IT.

This strand of my research explores how smartphones and activity-tracking technologies can help us understand everyday movement, and how people engage with the data they collect.

Earlier work includes Pass the Ball: Enforced Turn Taking in Activity Tracking (2015), which explored the social dimensions of activity tracking. More recent work extends this interest to the long-term movement histories already recorded by everyday devices.

During the COVID-19 pandemic, we developed the WFH Movement app, which allowed participants to contribute historical step-count data from services such as Apple Health and Garmin. Because these records extended back to the period before widespread working from home, they allowed us to study how everyday movement changed as people's routines changed.

This resulted in a longitudinal study of movement patterns in Sweden and has led to further work using smartphone-based movement histories in health-related settings, including studying patterns of recovery before and after orthopaedic procedures.

I am particularly interested in the methodological possibilities here. Smartphones continuously collect traces of everyday life, often over long periods of time. With appropriate consent and careful design, these traces can make it possible to study changes that would be difficult to capture using traditional questionnaires or short-term measurement studies.

I have been building experimental mobile and web systems as part of my research since the mid-2000s.

Earlier work includes Populations, which explored alternative approaches to constructing and deploying mobile software; Research in the Large, which investigated the methodological possibilities and challenges of conducting research through widely deployed systems and large-scale usage data; and the Mobile 2.0 project at the Mobile Life Centre.

I also built a number of experimental mobile services and web applications, including Spotisquare, which connected Spotify playlists to Foursquare locations, and Buzzy, an interactive conference application developed for MobileHCI.

These projects belong to an earlier technological period, but many of the questions behind them remain central to my work: what happens when new computational capabilities become available, what kinds of interactions they make possible, and how we can learn about them by actually building and using things.

Old Videos

Videos of some of my earlier work:

• Geo Chat
• Columbus
• Push!Photo
• Push!Music
• Context Photography
• Portrait Catalogue