AI Can Write Terraform, But It Can't Fix Your Broken Culture
Writing boilerplate Terraform, Kubernetes manifests, and Python automation scripts is no longer a competitive advantage for an SRE. AI can do it faster, and often better.
When I recently decided to build a top-down GTA clone in Go, Claude handled the game loop, the city grid, and the collision detection with only two minor corrections. At work, GitHub Copilot regularly churns out complex Terraform modules that used to take me hours of manual reading through provider documentation. The commoditization of code isn't a future prediction; it's happening right now on our laptops.
But here is the hard truth that a lot of engineering teams are about to discover: generating infrastructure code was never actually the hardest part of reliability.
The hard part is the socio-technical system. If your outages are caused by misaligned incentives, siloed teams, and a culture of blame, AI won't save you. In fact, it might just help you build broken things much faster.
The Illusion of "AI-Solved" Reliability
There is a massive temptation right now for leadership and platform teams to look at AI and see a silver bullet. The pitch is incredibly seductive: auto-remediation agents that fix alerts before you even wake up, dashboards generated instantly from natural language queries, and CI/CD pipelines written in seconds.
The misconception is that adopting these AI tools will intrinsically fix reliability problems.
The reality is that AI is an accelerator. It amplifies whatever foundation it is built upon. If your foundation is broken, AI just helps you build broken things faster or measure the wrong metrics more efficiently.
Consider the classic tension between product managers who want to ship new features quickly and SREs who need to maintain system stability. If product and engineering are not aligned on stability constraints—if there is no agreed-upon definition of acceptable failure—an AI agent that writes code ten times faster just accelerates the chaos. It means more unreviewed changes hitting production, more regressions, and ultimately, more severe outages.
It is the equivalent of putting a Ferrari engine in a car with a broken steering wheel. You'll definitely go faster, but you're just going to hit the wall that much harder.