llm-kb
← Back to research
Paper

Study analyzes failure trajectories of CLI coding agents as temporal processes

A new arXiv paper presents the first large-scale empirical study of CLI coding-agent failure trajectories, treating failure as a temporal process rather than a final outcome. The study introduces a process-oriented framework to analyze how failures emerge, evolve, and become unrecoverable in LLM-based coding agents.

0 engagement·1 source·Fri, Jul 10, 2026, 03:25 PM
The paper, titled 'Failure as a Process: An Anatomy of CLI Coding Agent Trajectories,' was published on arXiv on July 10, 2026. It addresses the growing deployment of LLM coding agents in terminal-based environments and the need to understand their reliability. Unlike prior work that treats failure as a binary outcome, this study examines the onset, evolution, and irrecoverability of failures over time. The research provides a framework for analyzing failure trajectories, offering insights into how coding agents fail during software engineering tasks.

Entities

arXiv(tool)Failure as a Process: An Anatomy of CLI Coding Agent Trajectories(tool)

Related

PaperThu, Jul 9, 2026, 11:01 PM

The Patchwork Problem in LLM-Generated Code

A new arXiv paper identifies a structural failure mode in LLM-generated code: individual patches compile and pass tests but break globally due to missing configuration keys, nonexistent imports, or omitted guards. Standard CI toolchains fail to catch these issues, posing a growing risk as LLM coding tools gain adoption.

0 engagement·1 source·arxiv
arXiv
IncidentMon, Jul 13, 2026, 01:36 AM

Developer shares horror story of AI agent stuck in error loop burning API budget

A developer recounts how a background orchestration agent got stuck in an error-handling loop over a weekend, calling the LLM thousands of times sequentially and burning through weeks of API budget before daily caps kicked in. The incident highlights the need for runtime-level detection of semantic loops in AI agents.

0 engagement·1 source·reddit
PaperSun, Jul 12, 2026, 03:01 AM

Article explains how LLMs use tools and iterate to complete tasks

A technical article titled 'The Agent Loop: How AI Learns to Think, Act, and Get Things Done' describes how LLMs use tools, make decisions, learn from results, and iterate until tasks are complete. The piece provides a conceptual overview of agentic AI workflows.

0 engagement·1 source·rss
RSS
PaperFri, Jul 10, 2026, 08:10 AM

GRACE: Graph-Regularized Agentic Context Evolution for Reliable Long-Horizon LLM Agents

A new arXiv paper proposes GRACE, a method that maintains persistent system-level instructions for LLM agents as a typed semantic graph instead of flat text. This graph-regularized approach enables scoped verification and reliable context evolution over long horizons under distribution shift, addressing verification difficulties from accumulated instructions.

0 engagement·1 source·arxiv
arXiv
PaperFri, Jul 10, 2026, 03:05 PM

Researchers propose semantic framework to classify AI system failures

A new arXiv paper introduces a semantic framework for describing AI systems, distinguishing justified outputs from common failures like extrapolation, refuted assertions, and stale sources. The framework aims to help practitioners systematically evaluate correctness of AI-generated representations.

0 engagement·1 source·arxiv
arXiv