AI isn't replacing SREs, it's deskilling them
SigNoz, an observability software company, published an essay in its newsletter arguing that AI is weakening site reliability engineers rather than displacing them. The authors frame the open question as whether engineers quietly lose capability before the next unfamiliar outage arrives. They point to specific habits that could fade: engineers stop reading raw log streams because a model summarizes them, stop forming their own hypotheses because a tool suggests root causes, and stop building mental models of system architecture because a tool maps dependencies. Taken one at a time, each change looks like a productivity gain.
The essay builds on Lisanne Bainbridge's 1983 paper Ironies of Automation. Its central concern is that the rare, ambiguous incidents, where human judgment matters most, are the ones teams are least practiced at handling once AI absorbs routine response. It also warns that deskilling can go unnoticed, because AI-assisted work often feels faster while dashboards stay green. A second concern, which the author calls never-skilling, is that junior engineers may never build fundamentals in the first place, which would weaken the pipeline that produces senior SREs. SigNoz says its own product follows from this view: an AI assistant that keeps humans in control, with the stated goal of supercharging SREs rather than replacing them.
Why it matters
For founders building AI tools for operations teams, the essay poses a product question: does the software keep engineers practicing, or does it take over the steps where they learn? A product that shows raw evidence, asks users to propose a cause, and keeps a human in the loop makes a different case to buyers than one that promises to resolve incidents on its own. Vendors that claim full autonomy will need evidence that customers' engineers stay capable when the tool fails.
For operators and investors, the cost may show up later and outside the usual metrics. A Bain survey of 293 respondents from April 2026 found that the expected gain in operational resilience was the smallest of every metric measured, even as release speed and cost efficiency were expected to improve more. Bain calls this a reliability paradox: AI can cut the cost of building software while raising the cost of running it, unless operations are redesigned too. Diligence on AI-driven companies should therefore ask how customers' on-call teams stay skilled.
Key points
- 1SigNoz published the essay in its newsletter; the company sells observability software covering metrics, traces and logs.
- 2The essay names three habits at risk: reading raw logs, forming incident hypotheses, and mapping system dependencies by hand.
- 3The argument builds on Lisanne Bainbridge's 1983 paper Ironies of Automation, which examines how automation erodes operator skill.
- 4The essay warns deskilling is hard to notice because AI-assisted work feels faster while dashboards stay green.
