AI Isn't Replacing SREs. It's Deskilling Them

AI is unlikely to eliminate SRE roles outright, but automating routine incident response could weaken the expertise needed for rare, high-stakes failures. Drawing on Lisanne Bainbridge’s 1983 “irony of automation,” the argument is that automation leaves people monitoring passively until the system fails, then asks them to handle emergencies they have had little chance to practise. In SRE, AI already groups alerts, detects log patterns, runs known fixes, and suggests causes; more autonomous diagnosis, prediction, and rollback are approaching. If humans mainly handle novel, cascading incidents, they may lose the hands-on repetition that builds intuition for those very events.

The risk includes both deskilling - losing abilities once practised - and “never-skilling,” when junior engineers never get enough experience to develop them. The author points to research on reduced polyp-detection performance among doctors using AI and declining manual skills and situational awareness among pilots relying on autopilot. SREs may similarly stop reading raw logs, forming hypotheses, or learning system architecture, while AI’s apparent productivity gains conceal the loss. The proposed response is not to reject AI, but to preserve deliberate practice: sometimes let engineers resolve incidents AI could handle, and design tools that involve people in reasoning rather than merely asking them to approve recommendations. AI should assist under guardrails, with autonomy increasing as trust is earned, while humans remain engaged and capable when unfamiliar production failures arise.