IaC in the AI Era

When I joined my last company, they were on a multi-year journey of ClickOps: using the web interface to create and modify cloud infrastructure. Fortunately, I think we’re moving away from cloud UIs as an industry, because AI makes it so much easier to use the cloud through other means (e.g. the CLI). This means the value of Infrastructure as Code (IaC) has become even more important.

At this company, codifying the infrastructure was top of mind. It would make it easy to detect unused and misconfigured resources, where misconfigurations include security gaps and areas of optimisation.

I’ve done this type of work before, years ago, but this time it was going to be very different.

Typically the cloud-specific CLI is difficult to use because it’s hard to remember all the specific flags and arguments. This problem vanishes with AI. The infrastructure can be pulled into code and imported without much work, besides the normal guidance required to stop AI writing slop.

With all the infrastructure in code, it’s easy to spot misconfigurations between resources and ensure everything behaves the same. It’s also easy to load smaller, simpler context into AI, instead of requiring AI to make many requests to the cloud to understand what’s happening.

It’s also now easier to create and modify infrastructure, or even generate architectural diagrams if need be. And we mustn’t forget that we have an audit log of changes now too.

Following all this, some of the lowest hanging fruit I grabbed was a cost analysis of the entire cloud usage. This resulted in thousands of dollars of savings per month.

When AI is doing most of the work, what stands out as important is how easily observable your software and infrastructure is.