119 developers on what the habit has done to them (external link)
De Souza Santos, Santos, Leça, Magalhaes and Wessel surveyed 119 practitioners about dependence, overreliance and addiction-related behaviour around LLM use, and the notable thing about the paper is how carefully it declines to be the paper everyone wanted.
What they found is mostly functional dependence: the tools have become part of the routine, the way an IDE or a search engine is part of the routine, and that is not a pathology. Overreliance shows up in a specific and recognisable form — reaching for the model instead of the documentation or a colleague. Reports of anything genuinely addiction-shaped were the minority, and were mostly about difficulty moderating use or a bit of emotional attachment.
I want to sit on the overreliance finding, because it is the one with teeth. Asking the model instead of reading the docs is not laziness; it is usually faster and usually right, which is exactly what makes it corrosive. The documentation tells you how the thing works. The model tells you what to type. Those feel like the same answer right up until the moment they are not, and the gap between them is where your mental model of the system was supposed to come from. Skip it often enough and you keep shipping while quietly losing the ability to tell when the answer is wrong — which, per every other study I have linked this week, is now the job.
Same for asking the model instead of a colleague. That is not just your understanding; it is the team's. Nobody else learns what you learned, and nothing gets corrected.
The authors' recommendation is trust calibration, professional judgement and verification throughout. Fine, though it has the flavour of a conclusion written to be unobjectionable. 119 self-selected practitioners describing their own habits is a weak instrument, and they say so.
Credit where it is due: they had every opportunity to write "developers are addicted to AI", which would have been covered everywhere, and instead published "mostly it's just habit, but watch this one specific thing". That is the less shareable and more useful paper.