AI SRE tools cut routine MTTR while Bainbridge 1983 predicts longer resolution for novel failures

Rootly and similar platforms deploy agents that query telemetry, form hypotheses and close capacity incidents without paging engineers. The blog post from Sylvain Kalache, former LinkedIn SRE, states these tools handle night-time events reliably. Data from internal deployments show median MTTR falling on known patterns while complex cases remain unmeasured. Bainbridge's Ironies of Automation documented that operators lose skill on normal tasks yet retain responsibility for rare anomalies. Aviation regulators responded with mandatory six-month simulator checks under FAA rules for engine failures occurring less than once per 100,000 flight hours. Software lacks equivalent recurrent training mandates despite comparable automation density. Uptime Labs simulations at Rootly place engineers in Slack-based e-commerce outages against LLM stakeholders. Participants practice hypothesis formation and coordination on incomplete data. This approach directly addresses the skill erosion Bainbridge identified. Without mandated simulation hours or telemetry drill requirements, mean time to resolution for ambiguous, high-severity incidents is expected to rise as fewer responders accumulate live system intuition.