Tag
DeepSeek has published an in-depth technical article introducing Super-agent, an autonomous agent system that uses a two-tier Router + Worker architecture to address bottlenecks in concurrent task execution. In testing, p95 completion time fell from 29.5 minutes to 15.7 minutes, and the prototype has been open-sourced.
作者用 Godot 构建了一个直升机逃生游戏，让八个 AI 模型控制器在共享逃生舱位的场景中竞争与合作，并进行三轮模拟实验。
A practitioner's listicle outlining 15+ essential portfolio projects for aspiring Agentic AI Engineers, covering async orchestration, strict tool routing, memory eviction, self-healing ReAct loops, human-in-the-loop gates, trajectory evaluation CI, agent observability, prompt-injection defenses, sandboxed execution, and cost kill-switches.
A discussion post asking how people manage and coordinate multiple personal AI agents (OpenClaw, Hermes Agent, Muse, Dots, Claude Code, Codex) without manually copy-pasting the same prompts across them, seeking better ways to orchestrate agents.
A Chinese-language self-study tutorial for Stanford’s CS329Z “Engineering AI Agents” course has been released, featuring 19 lectures and 4 hands-on exercises covering LLMs, RAG, tool calling, memory, multi-agent systems, and evaluation, with support for offline HTML reading.
FAIR at Meta 提出 Routed Continual Learning：将过往轨迹按簇划分、每个簇搭配固定记忆库中的一种记忆系统，并由 LLM 路由器决定复用或新建簇，在交错任务流上相对全局 ICL 历史提升 15–30% 归一化奖励并降低 65–72% API 成本。
The tweet promotes an AI orchestration tool called Grokbot, claiming it can centrally coordinate AI tools such as Claude, Codex, and Gemini, and includes a link to a tutorial.
Microsoft's open-source Agent Framework has surpassed 14K stars, focusing on solving reliability issues for Agents in production environments. It supports multi-agent orchestration, checkpoint resume, human-in-the-loop approvals, OpenTelemetry observability, and is compatible with Python, .NET, and Go.
A developer shares how 'beads,' an agent-native ticket system, solves common multi-agent workflow problems like context preservation, agent delegation, and parallel coordination by providing a shared, persistent ticket database that all agents on a network can access simultaneously.
OpenScience is an open-source AI research workstation. Give it a research goal and it automatically handles literature review, experiment design, running code, and paper writing. It comes with 290+ built-in skills and supports local execution with your own API key.
SkillMem introduces a compact external latent memory for reusable skills in latent multi-agent reasoning, achieving 38× storage reduction and up to 4.23× inference speedup across benchmarks with Qwen3 models from 4B to 14B.
OpenFrontiers introduces a cross-disciplinary corpus of 12,151 open research problems plus a 200-problem retrospective benchmark of recently resolved mathematics problems with Lean 4 statements, evaluating seven LLM agents and finding that closed-book rediscovery of research-level results remains rare while live-web success largely reflects retrieval of published solutions.
A tweet promoting a 60-minute technical session on how to build AI Agents, covering architecture design, rule setup, skill configuration, validation mechanisms, and a complete approach to scaling from a single Agent to multi-agent collaboration.
An introduction to the open-source Python multi-agent orchestration framework Swarms (~7.3k stars), which supports sequential task handoffs, parallel execution, hierarchical team structures, and MCP and skills integration — designed to automate business processes into multi-AI collaborative systems.
Sam Altman 描述了他的AI安全愿景：不是单一的超级AI，而是大量仅比人类略聪明的AI相互制衡，模仿健康人类社会的运作方式，以避免将所有智能集中于一处。
Skydive (@getskydive) is now open to everyone, letting users run teams of AI agents, each with its own cloud computer featuring a browser, files, and connected apps, visualized as characters working in a shared 3D office.
Skydive 推出面向所有人的 3D 虚拟办公室产品，用户可以创建并管理 AI agent 团队，为每个 agent 分配工作、连接工具、下发任务并审阅其成果，每个 agent 都拥有配备浏览器、文件和终端的云端电脑。
作者用自建的开源消息板工具 bot_board 替代了 agent orchestrator，让 Claude Code、Codex、Copilot 等多个编码代理自主协调、分工和共享项目上下文，显著提升了多代理并行工作流的效率。
Skydive 开放给所有人使用，让用户组建 AI agent 团队，在 3D 办公室中观察它们工作，并通过 Slack、邮件、iMessage 或网页分配任务。作者认为随着 agent 团队规模扩大，动态交互界面将是下一个发展方向。
@getskydive is now open to everyone, letting users build teams of AI agents that work in a 3D virtual office and take tasks via Slack, email, iMessage, or the web. Users can claim their company's domain to place their office on the map, and agents are free to use this week.