3 ms·
Short answer: you can, but it breaks down fast. LLMs are decent at writing Terraform, but Terraform wasn’t designed for agents. A tiny app change often explode
by nodar-d 10mo ago
Short answer: you can, but it breaks down fast.
LLMs are decent at writing Terraform, but Terraform wasn’t designed for agents. A tiny app change often explodes into hundreds of lines of diff across modules, IAM, networking, state.
> use AI to write terraform for me
The real problem isn’t generating config, it’s control. With vibecoded Terraform you get:
* no guardrails on what can change
* hard to review whether a diff is safe
* easy to accidentally destroy or over-permission infra
* state + implicit dependencies are brittle for autonomous edits
We built Neptune to solve those issues and eliminate as many footguns as possible. We are spending a lot of our time thinking about guardrails so you can deploy code with AI safely
- JojoFatsani 10mo agoI write make Terraform PR’s with Cursor daily. The Terraform Registry is easy for the agent to comprehend, HCL has been stable for years now. Good terraform linters and plugins exist. and you can pretty easily tell your agent to run and analyze terraform plans without applying. If Cursor/CC can’t be trusted with my TF code base why should I trust your tool? The same guardrails are needed for any of those not to mention the pesky meat-based code contributors.