Why did I break production? Is AI the sickness or the cure?
summary
AI speeds prototyping but still breaks things: models are 95% syntactically correct yet only 55% secure, big-company claims of 75% AI‑generated code are dubious, and Amazon forces senior review of AI code after outages. Examples include Diablo 4 login collapse, an AWS DNS race condition, and the Mars Climate Orbiter unit mismatch; automated stacks like self-healing Cloud Code run static analysis, intent tests and iterate pre‑PR, but results are mixed — productivity gains plus repeated vulnerabilities mean SAST/DAST, tests and human reviews remain necessary.
AI-written summary · about
chapters (16)
- 0:00Introduction and context
- 1:12Why breaking production happens
- 5:06Historical outages and AI role
- 8:41Infrastructure challenges today
- 12:14AI complexity and costs
- 13:38Code quality with AI
- 16:47Self-healing workflows
- 17:50Automated tools limits
- 18:31Code reviews and AI burden
- 21:00AI-assisted code review
- 21:41Testing strategies
- 22:24Supply chain attacks
- 23:05Fun outage examples
- 24:30Q&A: AI curse or solution?
- 25:55Q&A: Static code analysis future
- 27:00Q&A: AI in system design
related talks
19mIf AI Writes the Code, Who Runs Production?
39mBack to Basics: Cutting Through the AI-Generated Code Noise
39mAI-assisted development: Common pitfalls and smarter practices
19mThe AI Code Review Bottleneck: Escaping the Velocity Trap
11mAI took over dev. Prod is next. Here's why it has to be
23mVibes Not Vulns: Securing the Era of AI-Written Software
20mThe AI bugpocalypse is here. Now what?
59mUsing AI for Secure Code Creation: Enhancing Software Security