Authors: Xinyi Zhou, Zeinabsadat Saghi, Sadra Sabouri, Rahul Pandita, Mollie McGuire, Souti Chattopadhyay Venue: IEEE/ACM International Conference on Software Engineering (ICSE), 2026 — accepted
Summary
An observational study built on watching 36 developers actually work. It found that 48.8% of their programming actions were biased, and that more than half of those biased actions came out of an interaction with the AI rather than from the developer reasoning alone — the tool is not a neutral bystander to the bias, it is where much of it originates.
What the study ends up with is a validated taxonomy of 15 kinds of cognitive bias, covering 90 individual biases, specific to developer–LLM interaction, together with a set of concrete things developers and the people building these tools can do about them.
Links
Related
Thread: Human-AI Interaction in Knowledge Work — the research thread. Trust Dynamics in AI-Assisted Development: Definitions, Factors, and Implications — earlier study in the same line.