Summary
Cole Medin gives a dev-log style tour of the mission control dashboard he built to manage his AI software factory, a system that takes an issue ticket or an entire release, plans the work, and builds it in an autonomous loop with verification at each step. He says the factory now ships hundreds of pull requests a week.
The video explains why his original fully autonomous “dark factory” approach, where a PRD went all the way to production with no human involvement, was not reliable enough beyond proofs of concept. He has since added human-in-the-loop decision points, aiming for the smallest set of decisions that gives people the most control.
The dashboard tour covers a visual map of the workflow down to individual agent skills, a log of key design decisions, version diffs between factory iterations, and views of active runs for each connected factory. It also tracks failures. Of hundreds of runs over a couple of weeks, two failures came from Claude Code rate limits and 21 stemmed from workflow issues he had to fix.
Medin also touches on the underlying infrastructure, including a Postgres database and the Archon framework, and on challenges like dependency mapping, parallel execution, app validation, and staging environments. It is a practical look at what it takes to operate agentic software delivery at scale.
📺 Source: Cole Medin · Published November 02, 2026
🏷️ Format: Workflow Case Study
