AI TrackerFrontier desk · AEST
Back to news
normalaitech

Free to Look, Free to Reason: Simple LLM Harnesses Outdiagnose Hand-Crafted Root Cause Analysis Agents

arXiv cs.AI · 27 Oct 2026, 03:12 AEST

A study compares simple harnesses for root cause analysis of microservice failures against hand-crafted LLM agents. The findings indicate that the simple harnesses are significantly more accurate than the published agents and non-LLM methods.

Key facts

Read aloud

A study compares simple harnesses for root cause analysis of microservice failures against hand-crafted L L M agents. The findings indicate that the simple harnesses are significantly more accurate than the published agents and non-L L M methods. Root cause analysis RCA of microservice failures has attracted many L L M agents whose investigation is hand-crafted, through standard operating procedures, fixed stages and role agents. The simple harnesses are the most accurate PR@one sixty five-seventy seven percent on three mid-tier models, ahead of the published agents sixteen-fifty three percent and the non-L L M methods six-thirty nine percent, on every model on which both families were run. A stronger model mainly improves reliability, a multi-turn harness mainly improves aptitude, and more reasoning effort changes accuracy by a few points at most while multiplying the time by up to seven.

Sources in this cluster