A Reference Architecture for Governed Agentic Software Delivery : AI Delivery Control Plane
Enterprises now run software delivery in two disconnected worlds. In one, enterprise architecture, portfolios, standards, SDLC gates and budgets decide what is built and how. In the other, AI coding agents, multi-provider large language models (LLMs), Model Context Protocol (MCP) servers and agentic workflows increasingly do the building. Because the two do not share context, decisions or evidence, organisations get local productivity gains alongside rework, unmanaged risk and AI spend that cannot be tied to value. This paper proposes the AI delivery control plane, a reference architecture that connects the two worlds in both directions. Downstream, organisational knowledge, standards, decision rights and budgets flow into every AI action as context and guardrails. Upstream, every result, decision, cost and piece of evidence flows back into a versioned enterprise knowledge graph.
Authors
- Vengala (ORCID: https://orcid.org/0009-0000-4352-4080)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-20
- DOI
- https://doi.org/10.5281/zenodo.24401826
- Primary Topic
- Software Engineering Techniques and Practices
- Type
- article
- Field-Weighted Citation Impact
- 0.00