AI-Agent Use and Job Satisfaction among Software Developers: Evidence on Coordination and Implementation Friction

Abstract

AI agents can plan and execute multi-step work, yet little is known about how their use is associated with employees’ job satisfaction. Drawing on research on agentic information systems and job demands–resources theory, this study distinguishes adoption frequency from the resources and demands reported by current users. Data come from the publicly released 2025 Stack Overflow Developer Survey. The adoption analysis includes 22,289 employed professional developers across 165 countries, and the within-user analysis includes 5,988 current agent users across 141 countries. We estimate ordinary least-squares models with standard errors clustered by country and assess robustness using a cross-fitted doubly robust adjusted contrast, inverse-probability weighting for complete-case selection, a binary satisfaction outcome, and false-discovery-rate correction. Compared with respondents who neither used nor planned to use agents, weekly and daily use are associated with job-satisfaction scores that are 0.175 and 0.239 points higher, respectively, on the 0–10 scale; lighter use and copilot-only use are not statistically distinguishable from the reference group. The doubly robust adjusted contrast is 0.145 points (95% CI [0.071, 0.219]). Among current users, perceived improvement in team coordination is the most consistent positive correlate of satisfaction. Implementation friction is negatively associated with satisfaction, whereas the negative association for organizational restrictions is less stable across robustness analyses. Efficiency is no longer independently associated with satisfaction when demands are modeled jointly; capability perceptions and use intensity are also null. Taken together, the results point to implementation quality rather than use intensity alone. Because the survey is cross-sectional and agent use is self-selected, all estimates are associational.