Essay: AI handles routine incidents, but SREs may lose skills for the rare ones

A February essay from observability company SigNoz, now trending on Hacker News, argues that as AI takes over alert triage, log reading and known fixes, site reliability engineers get less practice for the novel outages that still need humans. It builds on a 1983 paper, Ironies of Automation, which found that automating a job leaves operators worse at the rare failures they must step in for. The author suggests deliberately handing some incidents to humans even when AI could solve them, and keeping people actively working the problem rather than just approving AI suggestions. SigNoz sells AI features in its own product, so read it as informed opinion, not research. On Hacker News (491 points, 369 comments), the thread leans towards agreeing. Of 15 top-level comments we read, 6 back the worry with their own experience, 2 push back (one says losing mundane skills is fine, one says the tiers described don't match real on-call work), and 7 are side points. No consensus that it is a crisis. Sorted take: pick one routine task a month that your team does by hand on purpose.