I'm really interested to know what AI tools folks are using for DevOps. Specifically, I'm looking for solutions that can automate most routine tasks, like troubleshooting assistance, and some that can help set up environments in a more dynamic way than traditional methods, such as Ansible. We're currently checking out a few options:
1. Mploi.ai - It looks promising but the free tier is limited. Has anyone tested the paid version? I appreciate that it's fully on-premise and comes with solid guard rails.
2. Resolve.ai - They claim this tool is in production use, so I'm curious about others' experiences with it.
3. Copilot4devops.com - It seems great but appears to be limited to Azure environments.
It feels like there are few tools in this space right now. We usually rely on Ansible and our own scripts, but they feel too static. It seems AI tools could be the future, and I'm on the hunt for solid commercial solutions. We're also creating some of our own tools, but I'd love to find one that includes guardrails, RAG pipelines, and a simple agent builder that can connect to local systems.
4 Answers
We've been using GitHub Copilot, and it works well for us!
I've had pretty good success using Claude for DevOps tasks. It's great for digging into issues and doing a lot of back and forth for one-off tasks, like checking failed GitHub Actions workflows and restarting them. I just keep its permissions limited to read access for safety.
Using AI to analyze logs and suggest summaries for incidents sounds pretty appealing! Personally, I use Claude heavily in development, especially if you have well-documented processes that it can reference. But I wouldn’t trust an AI tool to make changes directly to infrastructure without checks.
Could you go into more detail? Is it purely Claude that you use, or do you connect it to specific systems?
I mainly use it to sift through pod logs instead of using grep. It really saves time!
What tool are you using to accomplish that?
Thanks for sharing! Does it let you restrict what it can do or at least confirm actions before executing them?