LLMs are good at turning well defined requirements to code.
But even now it’s struggling on a project to understand the correlation between “It is creating Lambda code to do $x meaning it needs to change the corresponding IAM role in CloudFormation to give it permission it needs”
The LLMs are fantastic at writing terraform when you tell it what to do which is a huge timesaver, but good heavens is it terrible at actually knowing what pieces need to be wired up for anything but the simplest cases. Job security for now I guess?
I was able to one shot CDK, Terraform and CloudFormation on my last three projects respectively (different clients - different IAC). But I was really detailed about everything I needed and I fed ChatGPT the diagram.
I guess I could be more detailed in the prompt/md files about every time it changes lambda code, check the permissions in the corresponding IAC and check to see if a new VPC endpoint is needed.
But even now it’s struggling on a project to understand the correlation between “It is creating Lambda code to do $x meaning it needs to change the corresponding IAM role in CloudFormation to give it permission it needs”