AI handles incidents, engineers lose touch with their systems

TEXT START: When I was an SRE at LinkedIn, back in 2012, I designed a system that could heal itself and learn from previous incidents.

The Dissection

The article identifies a real failure mode: automation removes routine practice while leaving humans responsible for rare, ambiguous disasters. It then converts that diagnosis into a manageable training problem—incident simulations, tabletop exercises, and AI-assisted explanations.

Its deeper function is to make AI displacement operationally acceptable. Routine work is conceded to automation; human relevance is preserved at the emergency boundary. “Comprehension debt” is the symptom of engineers delegating system ownership to machines they increasingly cannot fully understand.

The Core Fallacy

The article treats skill decay as the central problem and simulation as the remedy. Under Discontinuity Thesis logic, that is a lag defense, not a structural solution.

Simulations may preserve a narrow group’s ability to handle exceptional failures. They cannot reverse P1–P3: AI’s durable superiority over cognitive work, institutions’ inability to preserve stable human-only domains, and the collapse of economically necessary human labor.

The aviation comparison is also incomplete. Aviation operates within tightly bounded systems, standardized procedures, certification regimes, stable hardware, and legally enforced recurrent training. Software systems mutate continuously, interact across enormous state spaces, and generate unknown failure modes that no simulator can fully reproduce.

The article assumes that humans will retain sufficient authority, access, practice, and economic importance to justify maintaining these skills. That assumption is precisely what AI automation destroys. The engineer is being repositioned from operator to exception handler—ultimately a servitor of the automated system.

Hidden Assumptions

• Rare-event competence can be maintained through occasional simulation.
• Simulated pressure transfers reliably to novel real-world failures.
• Organizations will fund and enforce recurrent human training after routine work becomes automated.
• Human responders will retain meaningful control rather than merely approve or supervise machine decisions.
• Complex incident response will remain a human frontier instead of becoming the next automation target.
• Lower average MTTR compensates for a potentially more dangerous long tail of failures.
• Preserving a small cadre of expert responders preserves the broader engineering profession.
• The economic value of engineers is unaffected by the removal of their ordinary productive work.

The article correctly rejects passive observation as a substitute for practice, but it still assumes that practice can preserve the old relationship between humans and systems. It cannot. At best, it preserves emergency competence in a shrinking control layer.

Social Function

This is a partial truth wrapped in transition management and prestige signaling, with a commercial undertone. The author works for an incident-response company and presents simulation as the responsible adaptation to AI automation. That makes the transition appear governable: automate the routine, train humans for the disasters, and continue forward.

It is not pure copium. Competence decay and automation surprise are real, and poorly prepared responders can turn a recoverable failure into a catastrophe. But the proposed remedy functions as institutional anesthesia. It lets organizations acknowledge human degradation without confronting the larger fact that most engineers are losing productive participation.

The simulations are hospice care for human operational relevance, not a cure for obsolescence.

The Verdict

The article diagnoses a genuine secondary hazard while avoiding the primary event. AI will reduce routine incident work, erode human system intuition, and concentrate the remaining value in rare, high-consequence exceptions. Simulations can reduce that risk and preserve a small class of indispensable responders.

They cannot restore mass engineering employment, continuous human ownership, or the post-WWII labor-to-consumption circuit. The likely endpoint is an automated operational core surrounded by a thin servitor layer trained to manage failures the system has not yet learned to absorb. The winners will be Sovereigns who own the automation and the narrow Servitors who remain indispensable in verification, safety, integration, infrastructure, and physical maintenance. Everyone else is accumulating comprehension debt while mistaking emergency rehearsal for continued sovereignty.