TLDR: A research roadmap, synthesized from the XP2025 workshop, addresses the integration of Generative AI into agile software development. It identifies key practitioner frustrations like fragmented tools, governance concerns, and skills gaps. The roadmap proposes short-term actions and long-term directions across five themes: tooling, human factors and AI literacy, governance and compliance, value realization, and creativity, aiming for a human-centered and trustworthy AI integration.
The intersection of Artificial Intelligence (AI) and Agile software development is a rapidly evolving field, bringing both immense opportunities and significant challenges. A recent workshop, XP2025, brought together over 30 experts from academia and industry to explore this dynamic relationship, culminating in a comprehensive research roadmap. This roadmap aims to guide the responsible and human-centered integration of Generative AI (GenAI) into agile practices.
Agile methodologies, formalized in 2001, transformed software engineering by prioritizing individuals, interactions, and responsiveness to change. Over the past two decades, its evolution was largely driven by process refinement and human collaboration. Concurrently, AI has advanced through three distinct waves: predictive analytics (2005-2015), deep learning for code and text (2016-2020) marked by tools like GitHub Copilot and ChatGPT, and the current era of foundation models and multi-agent orchestration (2021-present). This latest wave sees AI becoming an active, conversational partner in the developer’s workflow, capable of generating user stories, drafting pull requests, and even critiquing outputs.
Despite the promise of AI, practical obstacles persist. The XP2025 workshop identified several key frustrations among practitioners. These include a fragmented tooling landscape, where teams are overwhelmed by too many tools with overlapping features and inconsistent interfaces. Concerns about governance and privacy are also paramount, with uncertainties around data protection, intellectual property ownership, and compliance with regulations like GDPR and the emerging EU AI Act. Furthermore, significant skills gaps exist, particularly in prompt engineering and understanding AI’s capabilities and limitations, leading to early abandonment after poor initial experiences. Other challenges include AI integration not yielding valuable outcomes, issues with data and model quality (like hallucinations), and AI’s perceived lack of creativity.
To address these challenges, the workshop collaboratively developed a multi-thematic research roadmap. This roadmap is structured around five key themes, each directly reflecting the frustrations identified by practitioners:
Theme 1: Tooling Ecosystem & Integration
This theme tackles the “paradox of choice” in AI tools. Short-term actions include creating a systematic review and taxonomy of AI tools in agile contexts, along with an open-access tool selection guide. Long-term directions focus on developing multi-agent model selection interfaces that intelligently pick the best AI for a task, conducting longitudinal evaluations of integrated toolchains, and adapting AI models for domain-specific agile practices.
Theme 2: Human Factors, AI Literacy & Team Mindset
This theme addresses the lack of prompting skills and poor integration into team workflows. Short-term research involves assessing prompting competencies and developing role-specific training, as well as designing onboarding frameworks and team-level AI literacy workshops. Long-term goals include deploying “shadow AI” agents that observe and offer feedback, empirical studies on organizational change and mindset evolution, and defining human-AI partnership models for agile roles.
Theme 3: Governance, Compliance & Safe AI Use
This theme focuses on clarifying data protection, legal compliance, and building trust in AI tools. Short-term actions propose developing sandbox environments for safe AI evaluation and creating practitioner-focused briefs on AI regulation and intellectual property. Long-term research aims to design transparent AI audit mechanisms within agile decision-support tools and explore agent-based governance frameworks for automated policy enforcement.
Theme 4: Value Realization & Evaluation
This theme addresses the difficulty in quantifying AI’s business value and the lack of clear success criteria. Short-term efforts involve helping teams define contextual success criteria for AI adoption and developing multi-faceted evaluation frameworks. Long-term directions include building AI-driven value tracking and feedback loops, and exploring economic and organizational models to assess the costs, benefits, and long-term impacts of AI integration.
Also Read:
- Generative AI and Human Cognition: Risks, Opportunities, and the Future of Thought
- Interactive Virtual Personas: A New Frontier for User-Centered Design
Theme 5: Creativity & Multimodality in Agile AI
This theme explores how AI can move beyond routine tasks to foster creativity and innovation. Short-term research includes exploratory case studies on AI-augmented design and ideation, and experiments on multimodal AI in design activities. Long-term goals involve designing and evaluating co-creative workflows where AI acts as an active partner, and conducting longitudinal studies on creativity and innovation in AI-augmented teams.
The successful implementation of this roadmap relies on several enablers, such as establishing dedicated AI for agile software development research testbeds, ensuring the availability of high-quality, annotated datasets, developing robust evaluation frameworks, and fostering collaboration through shared infrastructure and open-source platforms. This comprehensive agenda, detailed further in the full paper available at arXiv:2508.20563, calls for an interdisciplinary approach, bringing together experts from software engineering, human-computer interaction, ethics, law, and organizational psychology to ensure that AI innovations are ethical, evidence-based, and truly meet the needs of agile teams.
