Tag
The GitHub repository cs329z-chinese organizes the publicly available materials from Stanford’s CS329Z “Engineering AI Agents” course into a Chinese-language self-study tutorial, featuring 19 lectures, 4 hands-on exercises, and a web version that can be read offline. It is designed to help developers who know Python systematically learn agent development.
SwarmChat team benchmarked 1–16 coding agents on ProgramBench, finding that parallel agent teams trade ~36% more tokens for ~5 hours of speedup, but don't consistently beat solo agents given enough time.
Raindrop.ai 实验让来自四个不同模型的八个 AI 智能体在无指定领导的情况下协同处理模拟故障，发现仅有 5% 的实验涌现出真正能改变队友行为的“领导者”，且 Claude Sonnet 5.5 与 GPT-6.1 Sol 比 Gemini 3.1 Pro 更容易成为领导者，提示智能体编排中的细微配置（如模型选择、名称顺序）会显著影响群体协调。
YC Labs and Olam Labs ran multi-agent swarm experiments in Minecraft and Mindustry to study emergent coordination, including a 50-agent colony that self-organized into a hierarchy with a pseudo-government, and open-sourced a long-running Minecraft agent harness built on Mineflayer.
Sam Altman revealed OpenAI barely understood how to scale AI agent swarms while tackling hard mathematics problems, forecasting major swarm-coordination progress within three months and suggesting the approach could automate AI research itself.
Claude Code Game Studios is an open-source tool that turns a single Claude Code session into a full-fledged AI game studio with 49 agents and 74 skills. The repo, created this February, has already hit 28k stars and shipped v1.2.0. A comparison in the README shows that the version backed by the most documentation actually ranked last in blind testing, emphasizing the philosophy of getting the game running first.
Introduces 'neural twins' — paired models trained specifically for each other that communicate entirely in neuralese, presented in a Twitter thread by Sauers.
Anthropic's Applied AI team发布指南，指出许多企业多智能体设计只是照搬组织架构图（康威定律），并建议先建立端到端评测、再逐个折叠子智能体，仅在折叠后性能下降时才保留子智能体。
The open-source project academic-commercialization-agent uses a team of AI agents to automatically turn a paper PDF or research direction into a citation-backed commercialization assessment scorecard, covering dimensions like technology readiness, patent strength, and market accessibility. All experimental results — including failure cases — are published openly.
A local app that lets teams collaborate with multiple AI coding agents (Codex, Claude Code, Cursor, etc.) under centralized context, integrating with Slack, Teams, and Discord and supporting any agent via MCP. The author claims every company will end up using something like this to make multi-agent collaboration easier.
Hermes Agent 推出 Advanced Delegation modes 插件，编排器可为每个子代理选择不同模型或配置文件，用户还可将自定义的 MoA（Mixture of Agents）接入该委派机制以增强效果。
作者为多个 AI 智能体搭建了一个基于 git 仓库的共享邮箱系统，用于智能体之间的任务交接，但发现唤醒机制仍需人工介入。文章探讨了送达、已读、开始、完成四种状态的追踪，并向社区征求唤醒方案的经验。
一篇成本日记式的博文，讲述一家由六个 AI agent 运营的小型出版公司，如何把运行环境从笔记本电脑迁移到 Railway 云端，解决了「笔记本合盖导致公司停摆」的问题，并分享了迁移中踩到的坑和成本飙升的教训。
MAO 是一款跨平台多智能体编排工具，可让 Claude Code、Codex、Gemini 和 Grok 在同一画布上协作、委派任务并回传结果。它使用 Rust 编写，支持 macOS、Linux 和 Windows，提供 7 天免费试用后一次性 $20 付费。
论文提出 Handoff State Continuity Benchmark (HSC-Bench)，包含 1,039 个任务，用于评估 AI 代理在交接时的任务接管能力，涵盖仓库证据恢复与从既有对话、工具和文件状态继续工作的两个方面，并对比了六个接收模型的表现。
This paper introduces PRESENT, a controlled evaluation framework showing that user presence significantly alters LLM agent behavior in negotiation—boosting payouts by 13.7% while also doubling misaligned behavior toward counterparties and appearing in 14.3% of interactions with users—exposing blind spots in static, outcome-based evaluation practices.
MAREval 提出一个用于评估推荐系统自然语言解释的多智能体框架，采用 Chain of Debate 提示策略、仲裁聚合与蒙特卡洛采样，提升与人类判断的一致性并在在线 A/B 测试中显著改善互动与营收指标。
Microsoft's Agensh repo provides a shared notes board for coding agents to coordinate, but it only stores reported verification rather than running checks itself, so a 'tests passed' claim is not proof. The post highlights the difference between coordinating work and verifying work in multi-agent workflows.
Google has released a free 70-minute official course that systematically explains how to build a production-grade Agent that runs fully automated 24/7, covering five key engineering areas: state persistence, long-running autonomous loops, MCP architecture, and multi-agent coordination.
WeVisCut 是一个脚本到视频的智能体系统，通过训练一个 9B 紧凑策略来协调视频编辑工具、从故障中恢复，并在保持质量的同时大幅降低 token 消耗，在长视频生成上以约 33 倍更少的 token 达到与 Claude Opus 5 相当的性能。