The Ruby AI Podcast Minerva Magic: OpenClaw, Agent Status Pages, and Training an AI Coworker in Ruby on Rails
Apr 28, 2026
They test treating an AI agent like a co-founder, giving it domains, goals, accounts, and room to self-train. They walk through building and deploying a Rails product with agent-written code, PR review role-play, and headless QA. They tackle memory systems, status pages for agents, cost controls, security limits, and what autonomy looks like in real-world agent workflows.
AI Snips
Chapters
Books
Transcript
Episode notes
Shark Tank Simulation Led To Ups.dev
- The agent proposed ups.dev as a niche product: status pages for agents, found via a Shark Tank simulation and critical review workflow.
- It used a private GitHub PR process as simulated sharks to reject risky ideas like AIG.dev over trademark concerns.
Train Agents On Domain Best Practices
- Out‑of‑the‑box autonomous agents make shaky engineering choices; focused domain training improves results significantly.
- Valentino trained the agent on Rails 8 best practices and core Ruby books to guide it toward simpler MVP decisions.
Lock Down Agent Production Access
- Limit agent privileges and keep human control over production credentials and deployment keys.
- Valentino gave the agent an SSH key to set up servers, then rotated keys and removed access so only he could deploy to production.
