Summary
Cole Medin explains how to run an AI software factory, a highly autonomous approach to AI coding, with the guardrails, context, and observability an engineering organization needs. His factory plans a release with a human, splits it into tickets, and implements them in a harness loop. Once many factories and applications are involved, he argues, a management layer around that loop becomes essential.
The video shows how the platform Port provides that wrapper without changing the underlying harness. It covers four core components: an agent and skill registry that manages ownership, publishing, certification, and lifecycle; a context lake that gives every factory shared business context from repositories and CI; engineering standards applied across factories; and controls to keep each factory aligned as those standards evolve.
Medin also describes how agents act as roles such as planner and backlog splitter, each with a defined platform (Claude Code, Codex, or others), model, permissions, and approval status. Enterprise runtimes like GCP Vertex AI and AWS Bedrock are discussed as well. The creator discloses a partnership with Port. The video is useful for engineering leaders and developers thinking about how to scale agentic development responsibly.
📺 Source: Cole Medin · Published October 23, 2026
🏷️ Format: Workflow Case Study
