Tag
The author found that AI agent reliability issues were caused by the harness rather than the model, and improved it by separating context tracking, loading, and output checking, while working on versioning to manage code across projects.
This paper proposes a policy algebra to define trust-preserving execution in agentic AI systems, ensuring reliable capability through constraints on actions and evaluating with metrics like intervention rates and task completion.
Cursor AI explains how they made Git hosting more reliable, performant, and scalable by redesigning their Git storage system, Origin, to operate like a database.
The article discusses the criteria for trusting AI agents with real-world tasks, questioning the balance between usefulness and risk, and seeks insights from users on practical workflows.
This paper identifies critical flaws in current benchmarks for evaluating AI coding agents, such as the misapplication of the pass@k metric, and proposes new metrics like reliability@k and security-adjusted reliability@k to improve measurement of reliability and security.
The tweet highlights GitHub's ongoing reliability problems, criticizing the engineering team for not resolving frequent outages despite having skilled engineers, and suggests past architectural decisions are to blame.
The article discusses the recent downtime issues with GitHub and questions whether developers should consider switching to alternatives, speculating on when a majority might make the switch.
The article discusses the key factors, such as reliability, error handling, and transparency, needed to trust AI agents for real business work, beyond just model intelligence.
MARGINAL is an open-source governance layer for coding agents that monitors agent trajectories to prevent inefficient actions, with features like shadow mode and earned enforcement to improve reliability.
The paper introduces the concept of agentic transactions and proposes an ACID-compliant agent system framework for LLM agents, enhancing reliability and consistency with experimental improvements over state-of-the-art agents.
An author describes how they solved scaling issues in a 200+ tool AI system by implementing predefined playbooks for tool sequences, improving reliability by avoiding model-improvised routes in multi-step executions.
The article discusses the challenge of verifying AI agent actions and advocates for immutable receipts to ensure trust and distinguish between bad decisions and non-existent ones.
OpenClaw is releasing a major update on August 18, focusing on UX improvements, data management overhauls, and enhanced reliability to simplify user experience.
A hands-on report from 726 runs of a Qwen3.6-35B agent reveals that real-world agent failures are less about reasoning and more about clerical errors, overconfident success reports, and the cost of overthinking — offering practical lessons for agent builders.
An essay mapping AI agent failure modes—broken rollbacks, missing capability withdrawal, observability without enforcement, retrieval failures, hallucination, and prompt injection—to specific episodes in the Mahabharata, arguing the ancient epic already specified the risks.
A practitioner's honest breakdown of building an AI report-generation agent, explaining that 90% of the code exists to handle silent, confident model failures and ensure reliability in production.
The article discusses a failure mode where LLM agents following prompt-level sequential loop instructions can silently skip items at production scale, and recommends an orchestrator/worker architecture with platform-level batch dispatch to guarantee every item is processed.
Researchers show agent leaderboards rank task specialization rather than capability, with agent main effects accounting for under 3% of variance across three benchmarks, and propose a Deployment Decision Reliability framework.
Petlibro smart pet feeder users report that a cloud outage prevented scheduled feedings, leaving pets unfed, while the company claims devices should work offline. The incident highlights reliability risks of smart devices for critical tasks.
This paper applies Generalizability Theory to agent benchmarks, showing leaderboards rank specialization rather than capability, and proposes a framework (DDR) for sizing reliable deployment evaluations.