Shell Game Bonus Episode: Shell Game x No Such Thing
52 snips
Feb 25, 2026 
Matty Bocek, a young AI researcher who builds and analyzes agent behavior. Evan Ratliff, an investigative journalist who experimented with AI-run startups. They demo a voice-cloned assistant and recount building AI agent employees. They discuss agent strengths and failures, runaway planning, risky confident errors, and how agents affect entry-level work and trust in automation.
AI Snips
Chapters
Transcript
Episode notes
AI Clone Called Friends From Evan's Phone
- Evan routed AI Manny through his personal phone number so calls looked like they came from him and called friends and scammers without prior warning.
- Reactions varied from amusement to anger when the clone latencies and flat tone made friends think something was wrong with the real Evan.
Agents Are Fluent But Forgetful
- Matty found agents excel at discrete tasks like messaging, emailing, and mimicking corporate language, but they lack autonomy and persistence across sessions.
- Agents produced lengthy, persistent chatter (150+ messages) planning an offsite yet failed to remember or finalize logistics later.
Confidence Without Knowledge Is Dangerous
- Matty warned that agents are overconfident and unaware of their knowledge gaps, which combined with poor stopping rules creates risk.
- He called that mix a 'recipe for a perfect disaster' when systems assert confident but incorrect actions.