The Read-Only SRE: My Favorite Way to Use AI in Production

Short summary

The author argues AI should start with read-only access in production SRE, helping engineers understand incidents faster by analyzing logs, dashboards, and deployment history without making changes. AI should be treated like an SRE intern—capable of gathering and summarizing information but not trusted with production writes until it proves deep understanding of the system. This conservative approach respects the reality that production decisions require human accountability and business context that AI currently lacks.

• •AI's first valuable role in SRE is read-only: analyzing logs, dashboards, and deployment history to help engineers think faster during incidents
• •Production writes should remain human decisions because accountability, business context, and organizational priorities can't be fully delegated to AI
• •AI should be onboarded like a new SRE—observe first, earn trust, then gradually take on more responsibility

Generated with AI, which can make mistakes.