AI isn't replacing SREs, it's deskilling them
Points and comments are a snapshot, not live.
AI is deskilling SREs by automating routine work, leaving them unprepared for rare, novel outages.
The article draws on Bainbridge's 1983 "Ironies of Automation" paper to argue that as AI handles more incident response (alert reduction, runbook execution, log analysis), human SREs lose the hands-on practice needed for the 5% of novel, complex failures. Research cited shows similar deskilling in medicine (endoscopists' detection rates dropping from 28% to 22% with AI) and aviation (pilot skill degradation). The author warns of a "never-skilling" problem for junior engineers entering the field. Recommendations include deliberate inefficiency (designating human-practice incidents) and building for human-in-the-loop rather than human-on-the-side.
What commenters are saying
Commenters broadly agree that AI deskilling is real and concerning. Several share experiences: one describes colleagues spending hours using Claude on blindingly obvious issues during outages, another notes two devs blindly sharing AI output back and forth instead of thinking. A common prescription is writing detailed specs and tests before using AI, treating it as a syntax generator rather than a partner. Some express relief that AI lets them focus on system-level design rather than day-to-day coding, dismissing LeetCode-style practice. Others worry about the risk of LLM companies going under or removing access.