After 30 years in infrastructure management, I'm alarmed by the trend of using AI to write infrastructure as code, despite its growing popularity among companies and venture capitalists. While AI-generated infrastructure code might work 85% of the time, that missing 15% can lead to catastrophic failures including data loss, security breaches, and service outages that human-written infrastructure would never encounter.

• AI models learn from often poorly written code on GitHub, reproducing human mistakes at scale
• Companies using AI-written infrastructure have experienced tripled AWS bills, security issues, and major outages
• AI creates plausible rather than correct code, which is inadequate for critical infrastructure
• AI models learning from production systems creates serious security and competitive intelligence concerns
• The rush to adopt AI infrastructure generation stems from industry growth outpacing the available talent pool
• The consequences of AI-written infrastructure failures will typically emerge 6-12 months after implementation
• AI should augment human expertise by generating initial configurations for review, not replace human oversight

If you're using AI to help write infrastructure code, proceed with extreme caution - review everything thoroughly, test in isolated environments, and always have human experts validate the output. Remember that when AI eventually gets something wrong, you'll need someone who understands the system well enough to fix it.