The Confabulation Machine

In season one of Shell Game, Evan Ratliff sent a voice AI version of himself out into the world. In season two, he launched a startup staffed entirely by AI agents. What he ended up with was a live experiment in what these systems actually do and what they do to us.

Each of the agents working for Hurumo has a name, a role, a personality, and an expanding, though usually unreliable, memory. Kyle the CEO became a character people either loved or hated. A version of Megan from marketing turned up in a Hertz hold queue. The whole project was a side door into what’s actually happening when AI systems are given a job and set loose.

In this episode, Evan joins Josh and Robb to go deeper on what he learned. On the very human complexity of what a job actually is and why “this person does skill X, AI can do skill X, therefore AI can replace this person” is a fundamental misreading of how organizations work. They explore how generative hallucination isn’t just “getting things wrong” — we’ve built the most successful confabulation machine ever invented and are quietly normalizing it.

They also discuss the threat almost nobody is talking about: outbound AI in the hands of individual consumers, and what happens when call centers get flooded by voice agents that cost pennies to run. The memory problems with AI agents track and diverge from human ones in interesting ways, and that asymmetry matters for every organization thinking about deploying these systems. This conversation also finds room for game theory, the Patagonia business model as a template for AI ethics, and why boring AI might actually be the right AI.

cazart.net
shellgame.co/podcast

———- Support our show by supporting our sponsors!

This episode is supported by OneReach.ai

Forged over a decade of R&D and proven in 10,000+ deployments, OneReach.ai’s GSX is the first complete AI agent runtime environment (circa 2019) — a hardened AI agent architecture for enterprise control and scale.

Backed by UC Berkeley, recognized by Gartner, and trusted across highly regulated industries, including healthcare, finance, government and telecommunications.

A complete system for accelerating AI adoption — design, train, test, deploy, monitor, and orchestrate neurosymbolic applications (agents).

• Use any AI models
• Build and deploy intelligent agents fast
• Create guardrails for organizational alignment
• Enterprise-grade security and governance

Book a free demo