Robert Wright's Nonzero The Allure and Danger of Agentic AI (Robert Wright & Liron Shapira)
Mar 19, 2026
Liron Shapira, host of Doom Debates and former software engineer focused on AI safety, explains how agentic AI reshaped his coding life. He talks about job disruption, whether AI can create genuinely new knowledge, and the plausible paths from powerful agents to catastrophic risk. The conversation also covers containment, pauses in development, and the politics around safety and defense.
AI Snips
Chapters
Books
Transcript
Episode notes
Many Desk Jobs Are Mechanically Automatable
- Liron expects broad net job losses because many roles are increasingly reducible to specs and repeatable processes that agents automate.
- He cites automating tasks like YouTube thumbnail design that cut hours of work into AI prompts and a minute of human review.
Agents Are LLMs Glued To Code
- Agents are chains combining LLM cognition with tool or code connectors, enabling virtually any desk task to be automated.
- Robert Wright explains agents use LLM reasoning plus pre-existing or new code to perform tasks like scouring sites or sending emails.
Last Jobs Will Touch The Real World
- The shrinking set of human-unique tasks narrows to work touching the physical world or roles where human identity matters.
- Liron says internet and software jobs are vulnerable while oil rigs or real-world management may be last refuges.
