{"id":357,"kind":"story","title":"Why LLM Agents Fail in Production: Lack of Design Patterns","url":"https://arxiv.org/abs/2605.13850","host":"arxiv.org","by":"mrlatte","score":1,"comments":0,"time":"2026-05-15T09:00:00.000Z","origin":"archive","language":{"source":"ko","reading":"en","translated":true},"summary":["Two papers summarizing agent design have been published side by side on arXiv.","One divides 27 patterns into two axes: cognitive functions and execution topology, while GraphBit claims to have eliminated hallucination routing using a DAG and Rust engine.","If you've experienced an agent working fine in prototype but entering an infinite loop in production, the classification chart is useful.","However, the chart doesn't replace judgment for domain-specific failures."],"discussion":"https://startupxo.com/en/357/","markdown":"https://startupxo.com/en/357.md","api":"https://startupxo.com/api/v1/item.json?id=357&lang=en","text":null,"edited":null,"sources":[],"thread":[]}