CHI 2027

WellVibing: A Digital Wellbeing Perspective on Software Developers' Experience with Agentic AI Coding Assistants

Riccardo Graziosi1  ·  Tommaso Calò2  ·  Alberto Monge Roffarello1
1 Politecnico di Torino    2 IT University of Copenhagen

Overview

AI coding tools read files, edit entire codebases, and run terminal commands, and a growing share of professional software development now happens in collaboration with them. We study what this way of working means for developers' digital wellbeing, a construct that research has so far examined mostly in leisure contexts such as social media and smartphone use.

We observe 15 professional developers while each works for about thirty minutes on a real task of their own, with their usual AI setup, and we follow the session with a retrospective interview that includes stimulated recall on selected episodes from the recording. The analysis uses the Digital Wellbeing Leverage Points Framework as its primary lens and Self-Determination Theory as the explanatory layer that connects mediators to wellbeing. Deductive coding on the framework's mediators is combined with an inductive pass that can surface mediators specific to AI-assisted coding, and the findings inform interventions that the research team positions along individual, team, organizational, and systemic leverage points in the discussion.

Research Questions

Session & Timeline

Each session is approximately 1.5 hours, conducted remotely via screen share. The participant confirms consent on their session page (Copy link in the roster); the Session modal drives the coding-session and interview timers.

~5 min
Setup and Consent. Welcome and rapport. The participant confirms consent on their session page. Confirm the task fits a ~30 minute slot.
~30 min
Coding Session. When the participant confirms consent, the Start coding button activates: press it as you start the screen recording, so timer time = recording position. The participant works on a real task with their own AI setup (~30 min); the observer logs episodes in the observation sheet and stars 1-3 stimulated-recall candidates. Stopping this timer concludes the participant's page.
~35 min
Retrospective Interview. Start the Interview card: the modal switches to the interview view with the protocol questions (IT/EN) and the starred episodes ready for stimulated recall (replay of recording at minute t).
~3 min
Closing. Thank-you and compensation logistics. Right after the call: session summary on the observation sheet (anomalies, spontaneous comments) and Word export if needed.
May W3 · 2026 (now)
Finalise protocol and observation rubric. Recruit pilot participants.
May W4 · 2026
Pilot study (2 to 3 participants). Refine interview script and observation sheet.
Jun W1 to Jul W2 · 2026
Data collection. 20 participants (about 3 per week). Begin open coding after the first batch.
Jul W3 to Aug W1 · 2026
Complete open coding. Axial and selective coding. Consolidate findings.
Aug W2 to Sep W1 · 2026
Write paper. Internal review with co-authors.
Sep W2 · 2026
Submit to CHI 2027.

Participants

📊 Participants Tracker ✉️ Recruitment Messages ☁️ Recordings & Materials
Name opens the participant profile Session ▸ timer & observation sheet 🔗 copies the participant's link
Loading…