AI Isn't Replacing SREs, It's Deskilling Them — That Gap Is a New AI-Oversight Service Business

AI isn't failing SREs. It's making them worse at the exact moment they're needed most.

That's the uncomfortable argument in a widely-read SRE newsletter post this week, and it maps directly onto a consulting offer nobody is selling yet.

The insight

The piece leans on Lisanne Bainbridge's 1983 paper "Ironies of Automation": the more you automate a process, the more critical the human operator becomes during the rare moments automation fails — and the less practice they get. The author backs that up with evidence from outside software entirely. In medicine, "endoscopists who used AI assistance for polyp detection saw their unassisted detection rates drop from 28% to 22% after a period of AI use." In aviation, "long-haul pilots who rely heavily on autopilot systems experience measurable degradation in situational awareness and manual flying ability" — which is exactly why the FAA mandates manual flying time.

The author draws a second distinction that matters for selling this: deskilling (an expert losing a skill they already had) versus never-skilling (a junior engineer who never builds the skill in the first place because AI handled it from day one). Most engineering orgs today have both problems running at once — and almost none of them are measuring either.

The money play

1. Sell a skill-retention audit, not an AI-adoption audit. Every vendor is selling "how fast can we get AI handling your incidents." Nobody is selling "how much manual diagnostic ability has your team already lost, and who on your roster could still run a severity-1 incident with the AI turned off." That's a sharper, more defensible scope — and it's uncontested territory right now.

2. Borrow the aviation fix directly: mandated manual-mode hours. Package deliberate-inefficiency game days — scheduled incidents where AI assistance is switched off and recovery time is measured — as a recurring retainer, the same way airlines mandate manual flying time instead of leaving it to pilot discretion.

3. Build the never-skilling onboarding track. For orgs hiring junior SREs into an AI-assisted environment, sell a structured ramp that forces manual-first diagnosis on a sandboxed set of real incident types before AI assist is unlocked for that engineer.

4. Turn the "human-in-the-loop, not human-on-the-side" principle into an architecture review. Audit where a client's AI incident tooling actually inserts a human decision point versus where it just auto-executes and pings a human after the fact — then redesign the escalation paths that fail the test.

5. License the two outside-industry case studies — endoscopy and aviation — as the proof points in a leadership workshop. Most engineering leaders have never seen the Bainbridge framing; it reframes "AI incident automation" from a cost-saving initiative into a skill-liability the board should be tracking.

Who buys this

• VP Engineering / Head of SRE at any company that has shipped AI-assisted on-call, alerting, or runbook automation in the last 12-18 months
• Platform teams that can feel their senior engineers getting slower at cold-start incident response but can't yet put a number on it
• Orgs actively hiring junior SRE/ops talent into an already-automated environment, worried about the never-skilling gap specifically
• Regulated industries (finance, healthcare, infra) where an auditor or regulator could eventually ask who still knows how to do this by hand

Bottom line

The pitch nobody else is making is the one buried in the aviation and medicine comparisons: automation doesn't remove the need for expert humans, it concentrates the entire cost of losing that expertise into the one day you can least afford it. Sell the audit that measures that gap before the next outage finds it for them.

Source: https://newsletter.signoz.io/p/ai-isnt-replacing-sres-its-deskilling