Governing Agentic Systems Development : Delegation, Coordination, and Oversight in Agentic Development Workflows

Sammanfattning

Agentic artificial intelligence (AI) is taking a more active role in systems development workflows by initiating, revising, and moving work forward, either autonomously or in interaction with human actors. This thesis examines how delegation, coordination, and oversight are enacted in agentic development workflows, and how these arrangements are embedded in workflow configurations, roles, interactions, and artefacts. The study adopts a qualitative multiple case study design based on semi-structured interviews and workflow walkthroughs. Drawing on a sociotechnical perspective, research on human-AI agency and coordination, and the concept of meaningful human control, the thesis examines how systems development work is reconfigured when AI agents and agentic AI systems become active participants in development work. The findings show that requirements, prompts, constraints, and markdown files can function as steering artefacts that embed governance logic directly into workflows under conditions of distributed execution. Governance increasingly becomes embedded in coordination structures, review practices, and oversight routines, while human roles shift from direct execution towards orchestration, evaluation, and judgement. Together, these findings show how agentic development workflows can remain governable when delegation, coordination, and oversight are continuously enacted through steering artefacts, negotiated boundaries, human judgement, and workflow practices across humans and multiple AI agents.