LessWrong (30+ Karma) “AI #149: 3” by Zvi
Jan 4, 2026
In a world transformed by AI, 2026 is seen as the year of the coding agent, where Claude Code drastically reshapes software development. The conversation dives into the practical utilities and potential risks of language models, including deepfake scams and legal implications of AI in violent incidents. Zvi discusses the debate on job displacement, patient preferences for human doctors, and the paradox of efficiency in knowledge work. The podcast explores public attitudes towards AI regulation, making a compelling case for caution amidst rapid technological advancements.
AI Snips
Chapters
Books
Transcript
Episode notes
Democratized Access To Powerful LLMs
- For about $20/month you get near-universal access to powerful LLMs that rapidly improve.
- Zvi calls this an egalitarian, democratizing force that most people should start using.
Developer Replaced By Claude Code
- Boris Cherny stopped writing code after Claude Code matured and now all his recent commits were generated by Claude.
- He reports hundreds of PRs and tens of thousands of lines added or removed authored by the agent.
A New Abstraction Layer In Programming
- Programmers now need to master a new programmable abstraction layer around agents, prompts, modes, and integrations.
- Zvi compares the transition to handling an alien tool with no manual, radically refactoring software engineering.
