Learning Python as a Blind Programmer in an AI-Mediated Environment: An Autoethnographic Study

Programming education is increasingly mediated by AI tools, yet little is known about how blind learners—particularly non-native English speakers—integrate these tools into learning, or where AI fails to address accessibility barriers. This paper presents an eight-week autoethnographic study of a blind, Chinese-speaking doctoral student learning introductory Python through Khan Academy’s curriculum with screen readers, a Braille display, and AI assistants (ChatGPT, Gemini). Drawing on detailed learning diaries and human–AI interaction logs analyzed via reflexive thematic analysis, we examine how AI mediates accessibility, cognition, and language during programming learning. Findings show that AI simultaneously functions as an accessibility interpreter, cognitive scaffold, and linguistic mediator, yet also introduces epistemic uncertainty, accessibility hallucinations, inconsistent cross-lingual terminology, and clear limits with non-textual cues. We argue that AI reconfigures rather than solves accessibility, and derive design implications for AI-assisted programming tools prioritizing explicit code structure, screen-reader integration, calibrated trust, and terminology-aware multilingual support.