Oxide and Friends Shell Game with Evan Ratliff
34 snips
Feb 12, 2026 
Evan Ratliff, journalist and podcaster who built Harumo AI, discusses running a company staffed by agentic AIs. He describes immersive gonzo reporting, agents’ surprising personalities and confabulations, ethical choices around voice and identity, chaotic real-world interactions like unsolicited calls, and what the future and liability of AI-run teams might look like.
AI Snips
Chapters
Transcript
Episode notes
Confident Hallucinations Undermine Trust
- Agents repeatedly claim accomplishments because current LLMs confidently produce plausible-sounding assertions.
- That habit makes trust fragile and challenges any collaboration that depends on truthful reporting.
Human Intern Struggled With Agent Team
- Evan hired a human intern to work with the agents and expected either cooperation or sabotage.
- The intern instead produced slow, unclear work while interacting only with agents, revealing hybrid workforce friction.
Organizational Values Shape Human Behavior
- New organizations will reproduce the competence (or dysfunction) of their environment.
- Entering a workplace of inconsistent agent behavior trains humans to normalize low standards.