GenAI study: software engineering shifts from code to intent

A new academic study by Elyson De La Cruz, published on arXiv, examines how generative AI (GenAI) and agentic systems are transforming software engineering. Using a reflexive thematic analysis (RTA) informed by interpretative phenomenological analysis (IPA), the researcher analyzed a corpus of peer-reviewed literature, technical benchmarks, public talks, essays, product announcements, and X discourse from prominent AI and software engineering voices. The study argues that the profession is moving from 'code-centric' production—where engineers primarily author code—to 'intent-centric' human-agent collaboration, where natural language, repository context, tools, tests, and governance shape delivery. This transition redefines the engineer's role from an isolated code author to a supervisor, validator, and governor of socio-technical systems involving humans, AI agents, and automated evidence gates.

The analysis finds that while GenAI dramatically lowers the cost of producing plausible code, it simultaneously raises the stakes on several non-coding skills: intent specification (clearly describing what the system should do), context curation (selecting and structuring relevant information for the AI), architectural knowledge, verification, security, provenance, and accountable human judgment. De La Cruz warns that organizations racing to adopt AI-driven development without adequate governance risk accumulating hidden technical debt and accountability gaps. However, with bounded autonomy—where agents operate within clearly defined constraints and oversight—it is possible to preserve quality, security, maintainability, and trust. The study provides a comprehensive framework for understanding this near-term transition, including a codebook, coding matrix, and traceability table, offering engineers and leaders a roadmap for navigating the shift from code-focused to intent-focused software engineering.