How my team ships software with OpenClaw: Middle Management for Robots

If you took Nathen’s keynote poll yesterday afternoon, you already know your blockers: trust, security, and the fear of being the one who let AI wreck production. This talk is for people who want to go further and don't know where to start.

Six months ago, I installed OpenClaw on a spare Mac mini. I nearly quit. I didn't. Instead, my team and I figured out how to stop writing code and start supervising the agents that write it for us. We built a real production workflow: agentic coding loops that run unattended, automated reviews where multiple models debate quality, and a skills layer that wires it all into the tools we already use.

The job changed. I'm now middle management for these tools, and it turns out the hard part was never the code. It was the delegation, the review, and knowing when to trust what comes back.

In 40 minutes, you'll get the honest account: what broke, what held, and a concrete path to your first weekend with an agent runtime that actually works.

Fair warning: we're going to try some of this live. It may not all ship in 40 minutes. That's the point.