[Paper] Preparing Students for AI-Driven Agile Development: A Project-Based AI Engineering Curriculum

Source: arXiv - 2603.09599v1

Overview

The paper introduces a project‑based AI Engineering curriculum that blends agile software development with generative‑AI tools. By embedding AI assistance directly into the sprint workflow, the authors show how students can acquire both agile and AI‑augmented engineering skills in a realistic, hands‑on setting.

Key Contributions

• Curriculum Blueprint: A set of guiding principles and a detailed semester‑long course structure that intertwines agile practices with AI‑enabled development tasks.
• Interdisciplinary Case Study: Real‑world student projects spanning seven two‑week sprints, where AI tools support everything from requirements clarification to automated testing and documentation.
• Mixed‑Methods Evaluation: Early quantitative and qualitative evidence that the integrated approach improves hands‑on competence in AI‑assisted engineering.
• Pedagogical Insights: Identification of necessary adaptations to traditional agile teaching (e.g., frequent tool updates, oral verification of concepts) and recommendations for educators.

Methodology

1. Curriculum Design: The authors defined six “AI‑infused” touchpoints within a standard Scrum sprint—requirements, backlog grooming, architecture, coding, testing, and documentation. Each touchpoint pairs a generative‑AI or agentic tool with a reflective activity on human responsibility and quality.
2. Project Execution: 2nd‑semester bachelor students formed cross‑disciplinary teams (software engineering + AI basics) and delivered a realistic software product over seven 2‑week sprints. In each sprint, students used tools such as GitHub Copilot, ChatGPT, and automated test generators, while instructors facilitated brief reflection sessions.
3. Evaluation: A mixed‑methods approach combined (a) pre‑/post‑course surveys measuring self‑reported AI‑assisted development confidence, (b) analysis of sprint artefacts (backlog items, code quality metrics), and (c) semi‑structured interviews to capture student perceptions and challenges.

Results & Findings

• Skill Growth: Survey scores for “confidence in using AI for coding and testing” rose by ≈30 % from the start to the end of the semester.
• Productivity Boost: Teams that leveraged AI for code snippets and test generation completed ~15 % more backlog items per sprint than a control group using only traditional tools.
• Quality Trade‑offs: Automated code suggestions sometimes introduced subtle bugs; students who performed oral verification (explaining the generated code to peers/instructors) showed significantly fewer defects.
• Tool Evolution Pressure: Rapid updates to AI services forced instructors to re‑calibrate sprint activities mid‑course, highlighting the need for flexible curriculum scaffolding.

Practical Implications

• For Developers: The study validates that integrating generative AI into daily agile tasks can accelerate backlog completion without sacrificing quality—provided teams maintain a “human‑in‑the‑loop” verification step.
• For Tech Leaders: Curriculum insights can be repurposed for on‑the‑job training programs, enabling existing engineers to adopt AI‑augmented workflows safely.
• For Tool Vendors: The need for stable, versioned APIs and clear documentation becomes evident; developers will favor tools that support transparent hand‑off to human reviewers.
• For Educators & Bootcamps: The modular sprint‑based design offers a ready‑to‑use template for short‑term intensive courses that aim to produce AI‑savvy agile practitioners.

Limitations & Future Work

• Early‑Stage Evidence: The evaluation is limited to a single cohort and a short (one‑semester) timeframe, so long‑term retention of AI‑augmented agile skills remains untested.
• Tool‑Specific Bias: Results may be tied to the particular AI services used (e.g., Copilot, ChatGPT); broader tool diversity could affect outcomes.
• Scalability Concerns: Managing frequent AI tool updates in larger classes could strain instructor resources.

Future research directions include longitudinal studies tracking graduates into industry, expanding the curriculum to cover AI ethics and governance in agile contexts, and developing automated scaffolding to keep course material in sync with rapidly evolving AI APIs.

Authors

• Andreas Rausch
• Stefan Wittek
• Tobias Geger
• David Inkermann

Paper Information

• arXiv ID: 2603.09599v1
• Categories: cs.SE
• Published: March 10, 2026
• PDF: Download PDF