Bridging the Gap? An Analysis of Competency Misalignment Between Industry AI Practice and Intended Learning Outcomes in Swedish Higher Education

Sammanfattning

Introduction: The integration of generative artificial intelligence into professional software development has produced a new generation of agentic coding tools that have shifted the role of the developer from manual code production toward the supervision, evaluation, and orchestration of machine-generated implementation. This shift raises a question about computer science education: whether the documented intended learning outcomes of contemporary university programs reflect the competencies that practitioners now exercise in AI-assisted development. The present study addresses this question through an empirical audit of computer science programs at four major Swedish universities. Research Question: The primary research question is: To what extent do the intended learning outcomes of computer science programs at Swedish universities align with the emerging competencies utilized in professional AI-assisted development? A sub-research question complements this main inquiry by asking what the primary areas of misalignment are between competencies practiced in industry and the intended learning outcomes documented in Swedish academic curricula. Method: A three-phase qualitative comparative research design was employed. The first phase applied inductive thematic analysis to 71 practitioner-led video tutorials demonstrating end-to-end AI-assisted software development, producing a competency framework of 97 distinct labels organized into eight themes. The second phase applied this framework as a coding frame to the deductive qualitative content analysis of 133 official course syllabi from KTH Royal Institute of Technology, Stockholm University, Uppsala University, and Chalmers University of Technology, with each course evaluated through its full documentation including intended learning outcomes, course content descriptions, and examination formats. The third phase performed a programmatic cross-case synthesis to identify consistent patterns across the four institutions. Results: The audit found that 39 of the 97 practitioner competencies (40.2%) had a conceptual equivalent in at least one analyzed syllabus, while 58 competencies (59.8%) were absent from every curriculum analyzed. Per-institution coverage ranged from 22.7% at Stockholm University and Chalmers, to 28.9% at KTH, to 32.0% at Uppsala. The alignment that does exist is concentrated in foundational software engineering competencies that predate AI-assisted development, including system architecture, database design, traditional testing, and version control. The misalignment is composed of two analytically distinguishable layers: a layer of new operational competencies that emerged from practitioner adaptation to agentic coding tools (including prompt-based feature specification, review of AI-generated implementation plans, evaluation of AI-generated outputs, configuration of agent autonomy and runtime infrastructure, and lifecycle governance of AI agents across sessions), and a broader layer of modern applied software competencies (including design system workflows, production deployment and platform operations, and applied commerce integration). The two-layer pattern is consistent across all four institutions, indicating a discipline-wide rather than institution-specific gap. Discussion: The findings indicate that the curricula examined describe the durable foundations on which AI-assisted practice rests but do not engage with the operational layer at which contemporary practitioners conduct AI-assisted work. The gap is structural rather than localized: no institution exceeds approximately one-third coverage of the practitioner framework, and no institution exceeds 28.6% coverage on any of the four themes most directly tied to AI-assisted operational practice. The audit is descriptive; it surfaces points of consideration for educators, students, and industry, and it raises ethical and societal implications including vendor dependency, equity of access to commercial AI tools, and labor market pressure on entry-level software work. The study is limited by the temporal validity of the practitioner framework in a fast-moving field, the constraints of the YouTube tutorial corpus, and the four-institution sample. Future research directions include temporal replication of the framework, institutional and national replication of the curriculum audit, methodological triangulation against industry signals beyond tutorials, and empirical investigation of the gap from the student side.