AI SRE explained: how AI incident agents change on-call and whether they work

The Pragmatic Engineer interviewed Rootly, incident.io and ResolveAI about AI SRE tools, which mostly speed up incident investigation inside Slack rather than act autonomously. Deploy frequency has roughly doubled in a year at many companies, and Rootly data shows pages up with out-of-hours pages nearly doubling, while vendors rebuild investigation agents around knowledge graphs, memory and verifier agents. In-house examples include OpenAI's Sevbot, Stripe's investigation agent (about 3 minutes for a simple run, 14 for a deep one) and Google's ProdAgent in Gemini CLI, which suggests tools via MCP and does not write code.