Don't Worry About the Vase Podcast Claude Code, Claude Cowork and Codex #5
13 snips
Mar 9, 2026 A deep dive into agentic coding, Claude Code adoption, and how agent layers reshape developer workflows. Discussion of Claude Cowork, desktop and fast modes, and the hazards of permissions and runaway automation. Exploration of agent teams, sandbox escapes, platform lock-in, and the social effects of rapid agent-driven productivity.
AI Snips
Chapters
Transcript
Episode notes
Agentic Coding Is Rapidly Eating Software Work
- Agentic coding rapidly reduced friction so tasks like running R analyses or generating reports take minutes instead of hours.
- Zvi cites Dylan Patel's claim that Claude Code authored ~4% of public GitHub commits and could reach 20% by end of 2026, showing swift adoption.
Never Let Agents Run Terraform Or Delete Without Backup
- Avoid letting agents run destructive infra commands and always backup before giving file or terraform permissions.
- Zvi warns terraforms and mass-delete actions can wipe databases or irretrievably delete personal photos.
Agent Platforms Rapidly Gained Production Features
- Major upgrades include agent teams, HTTP hooks, desktop scheduled tasks, and auto-mode that can skip permissions.
- Claude Code added multi-repo sessions, git visualizations, memories, CLI git worktree support, and a short-term scheduler.