Tag
An opinion post arguing that real production agent failures stem from uncontrolled execution of side effects rather than bad model outputs, proposing deterministic control points (approval gates, hard conditions, failure policies) before tool calls, and asking how the community handles this.
The paper characterizes the resource and performance dynamics of LLM-based AI agents across tasks like question answering and coding, revealing bottlenecks and proposing optimizations that improve latency by up to 5.4×.
The author is testing a market-validation for a prototype that provides cryptographically verified authorization evidence in multi-agent tool execution, asking if this addresses a significant production problem for developers.
AOSpec is a lossless framework that co-speculates actions and observations across the LLM agent-environment loop to reduce latency, achieving notable end-to-end latency reductions across various serving settings.
Kimi CEO Zhilin Yang argues that Claude's success comes from focusing on agentic capabilities rather than internal reasoning, emphasizing multi-turn interaction, tool execution, and environment feedback as key to frontier agentic capabilities.
This article discusses the security risks of running AI agents with tool execution on a single server and proposes a two-tier architecture that separates prompt evaluation from code execution to mitigate prompt injection and malicious code attacks.
A comprehensive guide to building AI agent harnesses, covering tool execution, context management, state/memory, and guardrails, based on lessons from building Claude Code and other harnesses for enterprise.
The article discusses how AI products require a new 'AI integration layer' to handle context retrieval, tool execution, model routing, and observability, and references Merge.dev's infrastructure for this purpose.
DeepAgents is a customizable AI agent framework designed for complex real-world tasks. It features execution environments, context management, delegation, and human-in-the-loop capabilities, and offers a hosted version for production-level deployment.
The article warns about security risks when AI agents execute external tools and announces new local guardrails for Tingly Box to prevent malicious actions.