Hey everyone, I'v been working on the opportunities/challenges of integrating AI into infra management. With the rise of "AI SRE" and "devops copilot" in the market I wanted to share some opinion about it, whether it's worth creating one or not. I've written a blog post : https://www.anyshift.io/blog/navigating-ai-in-your-infrastructure-dos-don-ts-and-why-it-matters the different topics : • why it's still risky to let AI (blindly) generate infrastructure as code (such as Terraform or Pulumi). • a simple look at how genAI models like gpt-4 work -> which explains why IaC code generation is not perfect yet • the security concerns of using AI-generated code in production environments • some opinion about why it matters to pair AI tools with "deterministic" practices/ like accurate infrastructure mapping would love to get your feedback on it + hear if you have some experience to share on AI/infra

Nice work, good complete overview without being too extreme on either end!! If only all articles could be so unbiased!!

Good overview, I'm just agree in part because is true that the companies save their code, but is possible using RAG in a AI internally to improve their outputs of the code. I think is the first step in order to jump at the next level

ah tks :))) Super cool to hear

pretty good article, im not a fan of the whole use big models as a wikipedia that most places are doing, i think fine tuned models calling functions using specific data in a agentic way is the way to go, its all about providing the correct tool and data for the model to use, but most companies are still at the lets use gpt as a wrapper stage

Has anyone tried any of the tools out there? e.g. k8sgpt.ai, cleric.io, https://www.vespper.com/, resolvedai.com ?

i havent tried it yet but a couple of those are on my list to get around and try it, plus im working on my own kubernetes upgrade copilot as well

Very good read , thank you @Roxane Fischer for sharing this.

@Hugo Pinheiro Can you share more about the problem you are trying to solve?

ya basically upgrading a kubernetes cluster takes research for example reading blogs of users that have done that particular update, reading release notes, running a couple apps to do api checks etc, so my idea is to automate all that using something like https://www.phidata.com/ have it output a markdown doc with details and custom commands to run and then use it in https://runme.dev/ - down the road automate it even more so that even the cluster upgrade can get kicked off by a slack bot.

Oh cool your using Runme. Did you consider foyle.io? The AI for Runme? (I'm the creator of foyle.io). You can see a demo here . Foyle automatically learns as a side effect of you getting stuff done. So each time you upgrade a cluster Foyle would make suggestions about the steps and then it would learn from your feedback so it would get smarter over time.

I did ya, its on my list of things to take a look at 🙂

Ha