Intel / article

Helpful but Fallible: Developer Experiences of AI Tools Under a Coordinated Industrial Roll-out

Source: arxiv.org
Author: Andreas Bexell, Rushali Gupta, Lo Gullstrand Heander, Emma S\"oderberg, Per Runeson, Sigrid Eldh, Wei Shi, Konstantin Malysh
Date: Published Sep 21, 2026

Context

An organization's expectations can shape how developers experience AI coding tools. In a case study at a Swedish telecommunications company, researchers interviewed twelve software professionals across three sites during a coordinated rollout. The interviews surfaced differences between management and developer expectations, alongside concerns about the risks of using the tools.

The authors describe AI tools as useful but fallible, with their value depending on developers' skills, the scale of the systems they work on and organizational expectations. This is evidence about adoption as experienced by staff, rather than a measured productivity increase. It gives managers a reason to examine how expectations and perceived risks affect usage alongside technical performance.

Key takeaways · AI-distilled

Interviews spanned 12 software professionals across three sites at a large Swedish telecom, analyzed with process coding and thematic analysis, then interpreted through the extended Technology Acceptance Model (TAM2).
The interviews surface a management-developer expectation gap that maps onto TAM2's constructs of subjective norm and voluntariness - how much pressure and choice developers feel around adoption.
Participants weighed perceived risk heavily, a factor TAM2 and similar acceptance models don't represent at all, meaning standard adoption models may be missing a real driver of developer resistance.
The paper's framing: AI-devtools are helpful but fallible assistants whose value is shaped by three factors together - organizational expectations, system scale, and developers' own skills - not the tool in isolation.

Read the source arxiv.org