A 16-minute video from Cole Medin that works like a dev log. He tours the mission control dashboard he built to manage his AI software factory, the system where an issue ticket, or an entire release, is planned and then built in an autonomous loop with verification and human review where he wants it.
What it is. Medin's factory takes an issue ticket or a whole release, plans it, and builds it in an autonomous loop. The dashboard is the control plane he put on top of it.
Why it matters. He says the companies building software factories at scale all end up with a dashboard like this, because the factory is a living and breathing machine that must keep evolving. He initially disliked the idea, fearing less autonomy and more to manage, but found it the only way to manage the complexity: dependency mapping, parallel runs, app validation, staging, and every human-in-the-loop point.
How to apply. Aim for the minimal set of decisions that bubbles back up to a human while giving the most control. His early dark factory split a PRD into issues, reviewed each, and merged straight to production. He found that works for proofs of concept but is not enough, so he dialed back the autonomy and renamed it the software factory.
What it is. The dashboard is itself built by the factory, which he calls a flywheel: a better factory makes a better dashboard, which makes the factory easier to manage. Before it, he managed everything in GitHub.
Why it matters. The homepage visualizes the whole factory, every decision point and skill. He says it was hard to picture the factory's shape before, and so hard to improve it. Other tabs track key decisions, version diffs, health, and runs.
How to apply. Document each improvement so you do not retry the same ideas. Compare any two factory versions (workflows, steps, prompts, skills). Track failures by cause: in his case, two came from hitting Claude Code rate limits and 21 from workflow issues he had to fix. The factory is dynamic, doing more steps for complex issues. A scorecard (runs, merges, duration) and test scenarios show whether to change a version, and a rewind feature uses Neon branching to return to older versions.
What it is. Archon is his open source harness builder. It turns any AI coding process into a reusable workflow and handles hard parts of agentic engineering such as work trees, so work runs in parallel.
Why it matters. Not a single coding agent in his factory runs outside an Archon workflow. The Archon UI shows every run down to the logs, and the dashboard links to it for any step. He credits his Archon co-author, Rasmus, for the workflows in the SDLC pack that ships by default.
How to apply. Store every agent turn, tool call, and decision in Postgres. He hooked Archon to Neon, which gives full audit trails and lets him dig into what went wrong later. In his words, without that record he has no way to know why things failed when he wants to evolve the factory.
What it is. Owner decisions are the questions and blockers that need his attention, since the factory is not dark and the agent asks for a human when it deems it necessary.
Why it matters. Before the dashboard, the agent mentioned him in an issue or pull request, and he had to watch the repository, email notifications, and Slack alerts. He wanted one place for everything in his repository, both for action and for observability.
How to apply. A GitHub webhook triggers a Neon function whenever a new issue or pull request appears. Because he manages the factory with issues and PRs, that gives a single entry point, stored in the same Postgres database as the Archon runs. He describes this as a central place for all observability and for diving in with his agent to improve the system.
What it is. Neon's new real-time feature, in public beta at the time of the video, makes any SQL query live data.
Why it matters. Many things happen every minute as runs proceed in parallel. With live queries the dashboard receives changes quickly, so he can watch new pull requests, issues, and comments arrive as they happen instead of only keeping a record.
How to apply. Wire the dashboard's SQL queries to real-time subscriptions. He shows a query on screen and new records appearing right away. This is especially helpful while he works on improving the factory itself, because he can see the effect of a change live.
Medin says the software factory is the big thing he is pouring time into, and he is open sourcing the whole thing so people can use it to build software while he makes it more reliable using the dashboard. If you build your own factory, he encourages using his resources as a starting point, and recommends Neon because you need a backend to store and process everything and it is not enough for it all to live in GitHub. He notes a partnership with Neon and asks viewers to like and subscribe.