Keynote: AI Speaks Terraform Like a Tourist
summary
AI produces Terraform that is syntactically fine but often wrong — about 30% broken — because it trains on Stack Overflow, old tutorials and vendor docs so it writes the first-apply not the millionth-apply. Concrete failures include stale S3 encryption blocks, deprecated EIP attributes and provider-version regressions. The defense is three layers: treat AI as an intern and read the plan, capture reasons with ADRs to document load-bearing oddities, and enforce policy as code guardrails.
AI-written summary · about
chapters (8)
- 0:00Introduction and job market absurdity
- 3:11Why AI speaks Terraform poorly
- 12:24Production Terraform is one-way doors
- 16:37Three-layer defense framework
- 21:12Document the shadows with ADRs
- 26:08Terraform plan is your friend
- 28:56You are the guides
- 30:56Q&A session
related talks
01h 01mKeynote: It AI-n't What You Think!
33m(No) Future of coding and scripting in AI era
58mKeynote: The past, present, and future of AI for application developers
39mBack to Basics: Cutting Through the AI-Generated Code Noise
33mKeynote: Boring Wins in AI Agents 🔥 Hard-Earned Lessons
31mMaking AI work for you: Terraform engineer blueprint
24mKeynote - The AI Playbook For Companies
29mWhy did I break production? Is AI the sickness or the cure?