You Shall Not Pass! Where and Why Developers Draw The Line on AI Autonomy

Jul 16, 2026

Upload by Patrick

AI expanding into software development shifts where work is considered human versus machine. Based on a mixed-methods study of 448 professional developers at Microsoft, the research identifies how accepted AI autonomy varies across software engineering tasks and individuals. Developers more readily accept AI producing work under their oversight, but acceptance drops for identity-defining, human-facing, and design-oriented activities. Experience with AI and risk tolerance raise acceptance, while task accountability reduces delegate autonomy.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages12

Reading Time30 min

Introduction

Related Work

You Shall Not Pass! Where and Why Developers Draw The Line on AI Autonomy Rudrajit Choudhuri 1 Christian Bird3 Carmen Badea3 Marco Gerosa2 Anita Sarma 1 1 Oregon State University, OR, USA. Email: {choudhru, [anita.sarma}@oregonstate.edu](anita.sarma}@oregonstate.edu) 2 Northern Arizona University, AZ, [USA. Email: marco.gerosa@nau.edu](USA. Email: marco.gerosa@nau.edu) 3 Microsoft Research, WA, USA. Email: {cbird, [cabadea}@microsoft.com](cabadea}@microsoft.com) arXiv :2607 .00533v2 [ cs .HC] 11 Jul 2026 Abstract As AI takes on more software work, the line between human and AI effort is shifting. Where developers draw that line around AI autonomy bears on how we design tools and roles that preserve meaningful work. Drawing on cognitive appraisal theory, work design, and automation research, we conducted a mixed-methods study of 448 professional developers at Microsoft to investigate developers’accepted levels of AI autonomy across software engineering work. Most developers accepted AI producing work under their oversight, although accepted autonomy varied substantively across tasks and individuals. Acceptance was lowest for identity-defining, humanfacing, and design-oriented work, and higher among developers with more AI experience and risk tolerance. Task accountability was associated with lower odds of allowing AI to act on developers’behalf, whereas task identity was associated with lower odds of granting AI decision-making autonomy. Task demands had the opposite effect, increasing willingness to delegate decision-making to AI. Our findings suggest that preferences for AI autonomy reflect how developers cognitively experience their work, highlighting important considerations for designing meaningful work. 1 Introduction Generative AI-powered development tools (hereafter AI tools) are transforming how software gets built. AI tools such as GitHub Copilot [42], Claude [6], and in-house assistants today write code, fix bugs, and generate tests, and their capabilities continue to expand. The question is no longer whether AI can do the work; it is should it, which parts, and how much? This quandary is not new. Long before AI, every wave of automation forced the same reckoning over what to hand to the machine [8, 30, 36] . What is different now is the pace and the reach. AI moves faster and reaches deeper into developers’work, so a single decision to defer can have downstream impacts. If we are not deliberate about this decision and let it settle based on AI tool capabilities, the result can be a set of serious compounding problems [2, 24, 34, 68, 81] . As AI absorbs tasks, developers can miss the work that builds skill and grounds judgment [24, 35, 81], leaving them to “rubberstamp” work they no longer understand well enough to evaluate. Pressure to ship leads them to offload more to keep pace [68], and as code generation grows cheaper, the same pressure plays out across the pipeline [23]: review, testing, and operations absorb rising volumes of machine-generated work. By the time defects Oregon State University, Northern Arizona University, & Microsoft Research, 2026 2026. ACM ISBN 978-x-xxxx-xxxx-x/YYYY/MM [https://doi.org/10.1145/nnnnnnn.nnnnnnn](https://doi.org/10.1145/nnnnnnn.nnnnnnn) surface, that work has passed through many hands, which makes it costlier to fix [14, 17] and accountability harder to assign [13, 74] . Prior work has examined what drives AI adoption [26, 76](e.g., trust [18, 25, 50, 88]), which tasks suit automation [52, 56], and where developers want support [22, 72], but not where they want control to remain human. Without that boundary, we cannot design work, or the tools that structure it, that keeps developers meaningfully engaged. Building on Endsley’s levels of automation [33], we distinguish two boundaries that mark it: the action boundary, where developers let AI act and produce work artifacts, and the decision-making boundary, where they let AI make decisions on their

FAQs

Which kinds of work show the lowest acceptance of AI autonomy?

Acceptance is lowest for identity-defining, human-facing, and design-oriented work, suggesting that boundaries around meaning and judgment are more strongly defended there.

Jul 16, 2026

Upload by Patrick