Limitless: An AI Podcast The Good and Bad of AI Coding: Amazon Shuts Down, Autoresearch, Claude Code Review, Lovable
7 snips
Mar 12, 2026 A six-hour Amazon storefront outage caused by AI-generated code sparks debate over coding safety and managerial checks. Discussion covers AI agents outpacing human review and infrastructure risks to major cloud providers. The rise of Lovable’s no-code app platform and Anthropic’s paid code-review offering get scrutiny. Andrej Karpathy’s self-improving AI experiments and the idea of autonomous research round out the conversation.
AI Snips
Chapters
Transcript
Episode notes
Amazon Outage Caused By Junior's AI Code
- Amazon's storefront crashed for six hours after a junior developer submitted AI-generated code that brought down core systems.
- Earlier outages in December and late 2025 show a pattern of AI-generated commits causing multi-hour, costly failures.
AI Code Velocity Creates New Choke Points
- AI-generated commits are rapidly increasing on GitHub and could reach ~20% of commits by year-end, creating new systemic choke points.
- That velocity outpaces human review capacity, forcing trust in AI checks and raising failure risks.
Require Manager Approval For Junior AI Submissions
- Require managerial permission for junior developers before submitting AI-generated code to reduce blind pushes into production.
- Amazon implemented this after outages, which will slow releases but cut costly errors.