The Challenge
Organizations increasingly mandate generative AI tools to accelerate software engineering productivity, yet adoption remains starkly uneven across teams. Existing literature predominantly relies on individual-level models like TAM and UTAUT, assuming adoption is merely an individual choice while overlooking environmental confounds such as team norms, codebase complexity, and management behavior. Crucially, companies deploy GenAI with high expectations but fail to provide structured organizational learning support. This disconnect creates a 'deploy and figure it out' dynamic that often negates anticipated efficiency gains and introduces significant workflow friction for practitioners navigating uncalibrated managerial demands.
Core Findings
Conducting sequential paired semi-structured interviews with 54 developers across 27 matched teams (pairing one frequent and one infrequent user sharing identical codebases, roles, and managers) at a multinational firm, the authors isolated sociotechnical drivers of tool usage. Frequent users conceptualized GenAI as a collaborative partner, embraced experimental integration, and demonstrated adaptive persistence during failures. Conversely, infrequent users framed tools as rigid utility features, used conservative integration, and quickly abandoned tools upon output degradation. Crucially, the authors uncovered the 'Productivity Pressure Paradox': top-down managerial pressure for immediate output gains without dedicated learning time starves developers of the temporal slack needed to build AI fluency, ultimately depressing long-term adoption and overall productivity.
Strategic Takeaway
For technology executives and digital leaders, this study demonstrates that treating GenAI integration as an individual workflow optimization rather than a systemic sociotechnical transformation guarantees suboptimal adoption. Mandating productivity gains while maintaining identical delivery deadlines triggers the Productivity Pressure Paradox, deterring developers from engaging in necessary experimentation. Strategic leaders must move past the 'deploy and figure it out' model by establishing contextual social learning structures, formalizing time-slack for capability development, and facilitating peer demonstrations tailored to local codebases to transition conservative users into adaptive collaborators.