AI handles incidents, engineers lose touch with their systems

Sylvain Kalache argues that as AI SRE tools get better at routine incidents, engineers get fewer chances to build real troubleshooting intuition. The paradox: mean time to resolve drops, but when automation hits a novel severe incident, responders are less prepared. He draws on Bainbridge's 1983 ironies of automation and aviation's mandatory six-month simulator checks, then advocates for incident simulators in software. Rootly and Uptime Labs built a simulated e-commerce outage where engineers investigate while coordinating with LLM-powered stakeholders in Slack.