Architecture Driven AI Engineering
Agentic AI promises speed, but its nondeterminism, and unbounded agent workflows in particular, can bring risk, rework, and frustration. JD Black, Sr Director and Chief Engineer in Northrop Grumman's Chief Information & Digital Office, describes a hypothesis that the known elements of engineering success, including rigorous process, version control, deterministic feedback, and open MBSE architecture-as-code, can be automated and amplified for agents. He walks through a five-phase agentic software and IT lifecycle, from architecture and system engineering through design, planning, test-driven development, and release, in which agents execute steps with deterministic exit criteria and humans hold formal review gates. He shares what the team has observed about small-batch execution, versioning agents, skills, and workflows, and tracing requirements to SysML v2 architecture and automated acceptance tests. Attendees will leave with his requests of the community: a standardized AI package manager and deterministic agentic workflow standards.
Chapters
One membership. The whole library.
Every book and audiobook, the video library, the courses, and the MCP. The papers stay free.