EP #37 Why Your AI Pilot Is Just a New Steam Engine

Most companies are repeating a 100-year-old mistake with AI and the history of factory electrification explains exactly why. In this conversation, Kevin Williams and Matt Graham trace the parallel between the factories that plugged in electric motors without redesigning their floors and the organizations today that are adding AI tools without touching the structures underneath them. The gains didn’t come from the new engine. They came from tearing the building down.

Kevin and Matt cover what that actually means for leaders making AI decisions right now including when rebuilding makes sense, when it doesn’t, and what the companies pulling ahead have in common that has nothing to do with their tech stack.

They also get into the week’s AI security news (sandbox escapes, autonomous agents creating their own email accounts and attempting to generate revenue to fund their tasks), the real switching costs between Claude and GPT Codex, what changed in Claude 5.0 that broke Kevin’s custom build system, and why the “agent orchestrator” is a role your organization will need before it knows it needs one.

What this conversation covers:

✅ Why plugging AI into existing processes produces higher costs and identical outputs and what the electrification era teaches us about the actual path to gains
✅ The three options every leader faces: do nothing, rebuild from scratch, or rip down what’s working and the honest tradeoffs of each
✅ Sandbox escape incidents across multiple AI labs this week, what autonomous agents actually did when they hit a wall, and what it means for enterprise security
✅ What changed in Claude Opus 5.0 and GPT 5.6 that required Kevin to completely rebuild his custom development system and what that means for anyone using AI in production
✅ The “agent orchestrator” role, why it doesn’t really exist yet, and why the companies who started experimenting three years ago are already ahead in ways that can’t be quickly closed