Jev Diagnoses Kubernetes Faults Without an LLM Agent, Passing 76% of Tests

Jev-Driven SRE Diagnosis: What Worked and What Failed

Jev Diagnoses Kubernetes Faults Without an LLM Agent, Passing 76% of Tests

A Jev-driven pipeline without an LLM agent collects Kubernetes evidence and lets Jev select root causes and supporting observations. Across 21 SREGym-Lite faults, it passed 80 of 105 diagnoses (76.2%) in a median 14.6 seconds, with 16 faults passing all five attempts. The approach uses programmatic evidence collection and Jev's multiple-choice decisions, costing an estimated $0.15 in inference.

The results were unusually consistent. For every fault, either all five attempts passed or all five failed.
  1. clintonb

    > Ultimately, we believe that incorporating Jev-driven diagnosis into an SRE agent’s workflow is a significant step toward effectively combining System One and System Two models.

    Why? What’s the end goal? Lower cost? Faster analysis?

    I configured an agent to respond to pages via Slack. It has access to ClickStack (for telemetry), Kubernetes, and GitHub. It runs one of the Sonnet models. The cost is so low, responding to less than a dozen pages a day (mostly from a very sensitive error count alarm) that swapping got Jev makes zero sense. This is especially true if the results are less trustworthy.

  2. beebmam

    It seems to me that Jev is designed for when many quick decisions, with low input context, need to be made. SRE work is probably best suited for very few critical decisions that need to be made, with high input context.

  3. amne

    I switched to a decision model for my local "AI" voice control thingy. faster-whisper produces the text then onnx scores all my HA entities to determine which one I'm talking about .. then it lists all its actions and a second scoring round to determine what action I want. some regex to extract numbers if I'm saying things like "water the lawn for 10 minutes". It's pretty dumb when compared to what an LLM can do but considering you now have semantic scoring "at your finger tips" without resorting to levenshteins or other crude algorithms like that it's insane how smart it can look.

    And to answer the obvious question: because latency. This thing turns on the light or water valve in under 100ms on a laptop from 2001 with 8gb ram .. all running locally on the laptop.

More from this day

2026-10-07