Terraform in the AI Agents Age | ✉️ #97

Illustration for MKDev Dispatch #97 featuring a smiling person holding a cat, with a paper airplane doodle. Text reads "Terraform in the AI Agents Age" on an orange and gray background with paper airplane pattern.

Last updated: 15/08/26 | Published: 15/08/26

Hey! 👋

So there is an ongoing debate whether Terraform - and other IaC tools - are still relevant in the world, where AI agents, equipped with the right Agent Plugins, can just do your infra for you.

Unsurprisingly, I am on the side that will tell you, that Terraform is perfect for this world, and that letting AI Agents directly configure your cloud environment via APIs is a mistake. I should probably right a bigger article on this topic, but here are some quick thoughts.

Terraform State and Locking enable better agent collaboration

I am constantly running dozens of agent sessions, and many of them within the same code base. Terraform locking allows one agent not to step on another ones feet. If multiple agents would, instead, configure things via API, they would create at least some technical debt, most likely a real mess.

Terraform Modules reduce token usage and provide better abstractions

Terraform has inputs and outputs, with variable types and many other nice things. Agents are good at reading and understanding how to use those things, and you can continuously improve your code and how agents use it to quickly build our correct, clean infrastructure.

Terraform within a mono-repo gives agents more context

And finally, if your IaC and your app code are all in the same repo, it just gives an agent better end to end understanding of how things work together. Modern harnesses - and models - are good at discovering things on their own, as long as there is something to discover, and there is nothing better for them that discovering it from some text files.

***

I feel much more comfortable with my agents building out - even imperfect, messy - infrastructure via Terraform, as I know that the changes are easier to trace, easier to maintain, and easier for many agents at once to work on.

And, as it often happens, the reasons to use Terraform in AI Agents Age are not much different from the reasons to use them in the pre-AI world.

What We've Discovered

AWS Fast Networking Is Deep Tribal Knowledge: If you are paying lots of money for running GPUs on AWS, this article will help to get the most out of them. The first tip alone will help to massively improve the network speed.
Unraveling Aurora DSQL Pricing: There are two main concerns that always stop us from using DSQL: unclear pricing and uncertainty if it will just work with regular SQL setups. This blog post helps to understand the pricing part.
AWS Answers 2026: The fourth annual Answers for AWS survey will help you with an almost impossible task of choosing the right combination of AWS services for your next project.

The 98th mkdev dispatch will arrive on Friday, August 28th. See you next time!