What SREs Should Automate — and Never Automate — with AI

Five key takeaways:

1. Automate based on impact and recoverability, not on whether the AI is technically capable of doing the task.
2. Alert triage, anomaly detection, incident summaries, capacity forecasting — these are the easy wins. Low risk, high value.
3. Production changes, incident command, security response, severity calls — keep a human's name on these. Always.
4. Reversibility and blast radius are better questions than "can the AI do this."
5. The goal isn't AI replacing engineers. It's AI clearing enough noise that engineers can actually think.

I've sat through the version of this conversation that sounds like a vendor pitch — AI triages everything, drafts your runbooks, predicts outages before they happen, and nobody gets paged at 2 a.m. anymore. I've also watched the other version happen in real time: an automated remediation script restarts the wrong service, confidently, at 11 p.m., and a 20-minute blip turns into a four-hour outage while...

Copyright of this story solely belongs to hackernoon.com. To see the full text click HERE