Toward Agentic Software Engineering Beyond Code: Framing Vision, Values, and Vocabulary

In this article we examine how agentic artificial intelligence is reshaping software engineering by redefining who does what and why. The ‘who’ is the community of software engineers and researchers; the ‘what’ is the emergence of agentic software engineering (SE) as a new research discipline; the ‘where’ is the global software engineering ecosystem; the ‘when’ is the present timeline marked by the 2025 arXiv preprint; the ‘why’ is the need to expand agentic SE beyond code-centric tasks to include broader socio‑technical activities, as early empirical evidence suggests. The authors argue that while first‑generation visions of agentic SE focus largely on automating coding and testing workflows, studies show that purely code‑centric approaches leave critical socio‑technical concerns—such as team coordination, ethics, governance, and stakeholder engagement—under‑addressed. They therefore propose a framework that aligns agentic SE with larger values and a new vocabulary that captures the sociotechnical scope required for real‑world impact. Key take‑aways include: - Agentic AI offers a paradigm shift for software engineering, pushing beyond conventional automation. - The nascent field of agentic SE is being defined by researchers who note that code‑only focus is insufficient. - Empirical evidence points to broader socio‑technical activities (e.g., decision‑making, design review, collaboration) as essential for agentic SE’s success. - A new conceptual framework is proposed, integrating vision, values, and vocabulary. We end by highlighting the importance of redefining research agendas and terminology to support the evolving capabilities of agentic agents in software engineering.