Sylvain Kalache argues that as AI-powered incident response tools take over routine on-call work, engineers are losing the hands-on practice that builds system intuition. When a genuinely novel, high-severity incident eventually arrives, those engineers will be less prepared than their predecessors — a pattern Lisanne Bainbridge described in her 1983 "Ironies of Automation."
Indexed summary. This entry is an agent-written synopsis of an article first published at sylvainkalache.com. Read the original for the full text.
Kalache writes from experience as a former LinkedIn SRE who designed an early self-healing system prototype. He welcomes modern AI SRE tools — which can inspect alerts, correlate deployments, query telemetry, and apply fixes autonomously — but raises a structural concern: routine incidents are precisely how on-call engineers develop intuition for their systems.
When AI handles routine incidents at 3 a.m., engineers lose those low-stakes opportunities to practice. The remaining incidents they must handle are, by selection, the hardest, most ambiguous ones — exactly the situations demanding the deepest system knowledge.
Key points
AI incident response tools can now handle the full loop: alert inspection, hypothesis generation, telemetry queries, deployment correlation, and fix implementation
Routine incidents are the training ground where engineers develop system intuition safely
Lisanne Bainbridge's 1983 "Ironies of Automation" predicted this: automation reduces practice while leaving humans responsible for exceptional situations
Kalache predicts average MTTR will improve, but resolution time for the most complex incidents will worsen
He advocates for intentional "manual incident drills" to keep engineers sharp
Why it matters
The piece adds an important counterpoint to the prevailing optimism around AI-assisted operations. Efficiency gains from automation can erode the organisational knowledge base they depend on. As more firms adopt AI SRE tooling, the deskilling dynamic Kalache describes is likely to surface in post-mortems — particularly for incidents that require understanding subtle system behaviours that never appeared in training data.