The Productivity-Experience Paradox
Article: The Productivity-Experience Paradox by Annie Vella
Summary
Vella presents the second set of findings from her Masters research at the University of Auckland on how AI changes software engineering. The study followed engineers from October 2024 to April 2025 and asked two things at two points in time: how AI coding assistants affected their perceived productivity, and how they affected developer experience, measured with the DevEx framework’s three dimensions — feedback loops, cognitive load, and flow state.
Perceived productivity went up and stayed up: 84% said AI improved their productivity at both timepoints. Developer experience drifted the other way. The share of participants for whom at least one DevEx dimension got worse nearly doubled, from 14% to 27%, and only 37% of those who started out positive on all three were still fully positive at the end. Nobody who turned negative came back to fully positive.
The dimensions pulled in different directions. Flow state declined most steeply (7% to 20% rating it worse), cognitive load worsened slightly (6% to 9%), and feedback loops were the only dimension that improved. Vella suggests these are linked: every fast AI response has to be read, judged, and re-prompted, so the quicker feedback is paid for with constant context switching. Within individuals over time, the correlation between change in flow and change in productivity was 0.02 — the coupling between experience and productivity that DevEx takes for granted appears to break down with AI.
She frames the risk with two ideas. Csikszentmihalyi’s “junk flow” describes the rewarding but shallow pull of orchestrating agents, closer to a slot machine than to a challenge that stretches your skills. MacIntyre’s distinction between internal goods (skill, mastery, satisfaction) and external goods (output, money, status) describes what is being traded: output holds or rises while the craft erodes. Strain that doesn’t show on a dashboard still surfaces later as burnout and turnover.
Her conclusion is that this is a system-design problem rather than a personal failing. Frameworks like DevEx were built for developers who write code, and need rethinking for work centred on specification, judgement, and verification — the task is to design new roles that are worth doing.