Cloud networks didn't get self-healing overnight. A new paper traces exactly how they got there, one painful generation at a time.
Drawing on production experience running network infrastructure at hyperscale, researchers lay out a five-generation maturity model for AI operations, or AIOps, in cloud network infrastructure. The path runs from manual, human-driven troubleshooting through scripted automation and rule-based systems, then AI-assisted operations, and finally fully autonomous incident resolution. The paper identifies the specific architectural patterns and organizational obstacles that gate each transition, and it documents the metrics teams actually use to decide when a system is ready for more autonomy. Crucially, the authors argue the jump between generations is not a pure engineering problem.
That last point is the real contribution here. Most autonomous-ops pitches focus on model capability, as if better anomaly detection alone gets you to a system that can act without a human in the loop. This paper instead treats trust frameworks, knowledge management, and operational culture as co-equal blockers, which matches what infrastructure teams have quietly learned the hard way over the past decade.
It is a useful corrective to vendor demos that show an AI agent fixing an outage in thirty seconds and skip the harder question of who is accountable when it doesn't.