The short version

AI can now write software. Good software, often. Fast software, always.

But writing code and shipping code are two different things. Who checks the AI’s work? Who makes sure it didn’t quietly change something nobody asked it to change? Who tracks the cost? Who says “yes, this is ready to go live”?

That’s what UPDSS does.

What it actually is

UPDSS is a system I built that manages AI programmers. Think of it as a supervisor for AI agents.

I run nine AI agents from my computer. Each one works on a different part of a software product. They write code, fix bugs, build features. But none of them can ship anything without going through three checkpoints. At each checkpoint, I review what they’ve done, check the evidence, and decide whether it moves forward.

The AI does the work. The system makes sure the work is trustworthy. I make the final calls.

Why it matters

Most people using AI for coding right now are doing it freestyle. Copy-paste from ChatGPT, hope it works, fix it when it breaks. That’s fine for small experiments. For serious software that people depend on, you need structure.

UPDSS is that structure. It tracks every piece of work, enforces quality checks automatically (the AI literally cannot skip them), and keeps a complete record of what was done, why, and at what cost.

The numbers

These are real, from eight continuous weeks of operation:

  • 50+ software releases shipped
  • ~$500 total cost for all AI usage across all products
  • 6 different products managed through one system
  • 9 AI agents directed by one person (me) through a terminal
  • Zero rule violations. No AI agent approved its own work, shipped without permission, or worked outside its assigned scope.

What’s on this site

If you want to go deeper:

  • About explains why UPDSS exists and what makes it different
  • How It Works shows the three layers of the architecture (with diagrams)
  • For Product Leaders is written for managers and executives who make governance decisions
  • For AI Agents shows what the AI agents themselves operate under (the page nobody else has)
  • The full architecture article is the deep technical walkthrough (4,000 words, for engineers)

Start wherever feels right. If you have questions, DM me on LinkedIn.

Want to support this?

Everything on this site was built by one person directing AI agents for roughly $500. The project is being open-sourced. If you’d like to help, visit the Support page.