Nune Isabekyan's Recent LinkedIn Posts

This is one of those posts where the math does the arguing for you. The combinatorial explosion of attribute combinations is something most people intuitively feel when they're debugging ("if only I had X alongside Y") but rarely see laid out this cleanly. Most AI-SRE tools right now are basically doing the same thing a junior on-call would do - open Grafana, open logs, open traces, try to mentally stitch the story together. The "AI" part is the LLM doing the stitching. But stitching fragmented data back together after you've destroyed the relationships at write time is a fundamentally lossy operation, no matter how clever your model is. The uncomfortable implication for the AI-SRE space is that the quality ceiling of your automation is set by the quality of the data going in, not by the intelligence of the model doing the reasoning. Which means: if your customers are on a classic three-pillar stack, your AI agent is operating with one hand tied behind its back. The hard question is whether you can build meaningful value anyway - through topology awareness, runbook encoding, better correlation heuristics - or whether the industry needs to fix the data layer first and the AI layer second. I suspect the honest answer is "both, in parallel." But the vendors who pretend the data problem doesn't exist are going to hit a wall.