SRE Deskilling: What AI Incident Automation Can Erode

AI already automates routine SRE work: grouping alerts, surfacing known-issue runbooks, flagging anomalies, and handling some remediation. The concern is not that these systems can’t help with incident response. It’s that SRE deskilling may follow when engineers stop getting practice on the work automation handles—and junior engineers may never get those reps at all.

SRE Deskilling Starts With Familiar Incidents

The source article connects this risk to research on automation in other fields. It cites a medical study in which endoscopists’ unassisted polyp detection rate fell from 28% to 22% after they used AI assistance. It also points to aviation research finding reduced situational awareness and manual flying ability among pilots who rely heavily on autopilot.

These comparisons are a warning, not a direct measurement of SRE performance. The article does not establish that automation has already caused the same decline in incident response. But the pattern is relevant: when routine decisions are delegated, people may become less practiced at making them without assistance. The article frames “when AI handles 95% of your incident response” as a risk scenario, not as a reported industry-wide measurement.

Junior Engineers Face a Different Kind of Skill Gap

There are two problems here, and they are easy to blur. Deskilling is losing a capability someone already had. Never-skilling is failing to develop a capability in the first place. An experienced engineer may get rusty at diagnosing unfamiliar failures; a junior engineer may have fewer chances to learn how to diagnose them at all.

That distinction matters because a team can look efficient while its people are getting fewer chances to build judgment. Routine incidents are not just work to clear; they can also be practice for recognizing what is unusual. If automation takes over that practice, the risk becomes visible only when the system encounters something outside the familiar patterns.

Track Readiness, Not Just Automated Response

My read is that the real question is not whether to automate routine incident work. It is whether teams can still respond effectively when automation is unavailable or the failure is unfamiliar. The article’s examples make that a credible concern, but they don’t tell us how quickly SRE skills erode, which tasks are most affected, or what level of human practice is enough to prevent it.

That leaves an important gap in the case for automation: successful handling of known incidents does not, by itself, demonstrate human readiness for novel ones. Teams should be wary of treating faster routine response as a complete measure of operational capability. The practical test is whether engineers still have opportunities to build and demonstrate unassisted judgment—not whether every familiar alert still needs a person. Automation can remove toil; it should not quietly remove the practice that helps people handle the incident nobody has seen before.