← Back to context

Comment by Rodeoclash

3 days ago

I find it so hard to reconcile using Chat-GPT to write some Terraform code which hallucinates the way of using the timer provider, tells me that it's right because it "simulated" the rest on an AWS staging environment that it has access to (can I see it? No.) then when confronted with the evidence that this doesn't work in a GitHub issue tells me I've found the "smoking gun" of evidence with what I read in the news articles and my LinkedIn feed about how programmers are out of a job in six months.

I feel like I'm taking crazy pills here, is anyone else getting this?

As an SRE, I completely agree. I've had some more success with claude code recently. I use it as a fancy find and replace, if I explain it in detail what I did in one file it can repeat it in a dozen other files. Which is super handy, like once every 2 days, when I have to work in a legacy monorepo.

Whenever I try to use it to generate code from nothing I end up losing more time than winning.

Have you tried using cursor?

Chat GPT seriously is outdated nowadays, and not the preferred choice for writing code.