Dev Interrupted The one where we vibe code holiday cards | Season 5 Finale
Dec 23, 2025
The hosts reflect on a transformative year, celebrating their growth and gratitude. They dive into the potential of AI in automating mundane tasks, allowing more room for creativity. A discussion on MCP agents highlights the need for safeguards and accountability for AI-generated code. They reassess outdated engineering principles and tackle the ongoing shift to remote work, questioning if return-to-office mandates are merely masking layoffs. Plus, insights from Cloudflare on internet traffic trends add a timely touch.
AI Snips
Chapters
Transcript
Episode notes
Agentic Safety Requires Decision Layers
- MCPs give agents capability but not safety without orchestration and rollback semantics.
- Builders must evolve MCP protocols to add decision management and reversibility before wide high-risk adoption.
Hold Humans Accountable For AI Code
- Engineers must remain accountable for any code they ship, even if AI generated it.
- Require rigorous testing and reviews and bundle tests with every PR to maintain safety and quality.
Dogmas Block Velocity; Triage PRs By Risk
- Rigid engineering dogmas slow teams and deserve periodic reevaluation against context.
- Not all PRs are equal; automate triage to allocate review effort where risk demands it.