No. 119
2026-11-15
the AI pulse — brewed, not raw
 AI CRITIQUE

AI isn’t firing SREs, it’s dulling their skills

Read the article    English (automatic translation)

Summary

As the article argues, as AI automates routine SRE tasks (alert classification, resolving known issues, daily summarization), humans are losing the practice needed to handle the rare and complex 5% of incidents. The author refers to this as 'skill atrophy.' A more concerning variant, termed 'never-skill-acquisition,' describes how junior engineers may fail to acquire fundamental skills entirely.

Why it matters

SRE is a domain where a single wrong decision during cascading production failures can cost millions of dollars. However, the article shows that tools designed to reduce human error are silently eroding the very judgment they rely on. The parallels drawn with aviation autopilots and AI-assisted endoscopy suggest this is not a hypothetical risk, but a pattern observed in safety-critical professions. If the next generation of SREs enters the industry without reading raw logs or building mental models of system dependencies, organizations will lose their last line of defense when AI recommendations go wrong.

Highlights

  • The 1983 'irony of automation' paper is applied to SRE: Increased automation leads to fewer rare failure interventions, which in turn weakens human competency for those rare failures.
  • 'Never-skill-acquisition' is introduced as a distinct risk: Junior SREs starting their careers in an AI-heavy environment may never develop the intuition and system mental models that senior engineers built over years of manual incident response.
  • Proposed measures include 'intentional inefficiency' (mandating human handling of incidents even if AI could solve them, similar to FAA manual flight requirements) and keeping humans as active problem solvers rather than passive approvers of AI suggestions.

Critical take

The text is a summary of an opinion piece rather than original research. It borrows its strongest evidence from medicine and aviation but generalizes to SRE without citing any empirical data on SRE teams' incident management performance over time. The 'intentional inefficiency' prescription, while intuitively logical, glosses over a clear tension: forcing humans to handle incidents that AI could solve is a cost that most engineering organizations, under pressure to reduce MTTR and headcount, will not sustain.

This critique was written by the local AI model Qwen3.8-27B. I chose the story; I do not review each critique before it is published.
Back to the issue