Year: 2025
Type: conference-paper
Study design: Observational Study
Abstract: AI coding tools can boost software-development productivity, yet their English-dominated training data pose challenges for non-English-speaking developers. In Ecuador's bilingual context—Spanish drives workflows while English is essential for APIs and documentation—these tools fail to parse non-English code elements, including Spanish identifiers, causing inefficiency and mistrust. This study investigates perceptions and challenges among 65 developers—45 Spanish-survey (ES) and 20 English-survey (EN) respondents, all native Spanish speakers with varying English skills. Using mixed methods we surveyed tool usage, linguistic barriers, and desired enhancements. Participants use AI assistants to automate routine tasks but report issues when suggestions span multiple files or ignore local naming conventions. ES respondents abandon recommendations more often than EN peers (25% vs 17.9%), indicating linguistic inequities rather than programming expertise. Qualitative analyzes highlight three needs: (i) multilingual NLP that understands hybrid Spanish-English code, (ii) context-aware generation that adapts to regional API norms, and (iii) transparent feedback that exposes provenance and confidence for verification. We argue that balanced Spanish-English corpora and bilingual tokenisers are pragmatic steps toward these goals, enabling automation while preserving developers' reasoning skills. Though limited by a modest convenience sample and self-reported practices, this work provides the first empirical snapshot of AI coding-tool adoption in Latin America and underscores the need for inclusive, context-sensitive tools to foster equitable software-engineering advancement globally.
Institutions: University of Guayaquil, Eindhoven University of Technology
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
SDG: Reduced inequalities
Open Access status: gold
OpenAlex ID: W4412703908