Harnesses: How an AI-Native Company Actually Runs

​Harnesses: How an AI-Native Company Actually Runs
Thursday, October 15, 5:30 to 8:00 PM

Most of what goes wrong when you build with AI happens around the model, in the process nobody wrote down.

Pat Santora runs Sanvio Labs, a studio where one person and a team of AI agents take products from idea to production. What makes that work is a harness: the rules, procedures, gates and hooks that tell agents what to do, check what they did, and stop them before a mistake ships.

In this two-hour session, hosted by Humanbulb, you'll watch a harness get built live from an empty repo, see where agents belong and where they don't, and hear what broke along the way. You'll also learn how to borrow other people's skills without handing a stranger the keys. Then you build your own. You leave with a working starter harness: an instructions file, a skill, a gate that blocks, a hook that enforces it, and your first agent.

Before you come: bring a laptop with a coding agent installed and working on your own account (Claude Code, Kiro, or Codex), and clone the starter repo: https://www.google.com/url?q=https://github.com/SanvioLabs/sanvio-harness-starter&source=gmail&ust=1790290811498000&sa=E. The hands-on runs in pairs, so if you can't get set up, pair with someone who has. Doors open at 5:30 for food and setup, and we start at 6:00.

What would you trust an agent to do while you sleep?