Topic · 主题全部主题 →

Agent 智能体

让模型自主规划、调用工具、完成多步任务的技术方向——从 Claude Code、Manus 到各家 Agent 框架与评测基准的全部动态。

9,105条收录

1,107条精选

数据更新至 9月8日 11:38

精选归档 · 第 11 页

第 201–220 条 · 共 1,107 条

7月24日

星期五 · 1 条

00:55

Satya Nadella@satyanadella精选

AI 评分 65/100

微软MAI模型：以更低成本实现前沿能力规模化http://x.com/i/article/2080328073724260352Frontier Diffusion & ControlIn a world where software has real marginal cost for the first time, how do we ensure frontier benefits are diffused across the entire ecosystem?The key is to optimize the cost-to-outcome frontier in real world context. In practical terms, that means using the right model for each task, and optimizing the context, skills, tools, and agent harness around it.This is the motivation behind our MAI model family. These models have been built ground up with clean data lineage and optimized for learning transfer from generalist to specialized skills in enterprise RLEs. We continue to make rapid progress in this pursuit.We can now take saturated frontier capabilities and deliver them at scale and at lower cost through models optimized for high-usage products, while continuing to use frontier models for frontier needs. We are proving this out across our first party products, and thereby creating a template for every other AI native, SaaS, or Enterprise company out there.In our products, frontier models from OpenAI and Anthropic are part of the orchestration system alongside MAI. But the model is only one part of the hill-climbing system. Harness, memory, context, tools, skills, user interactions, etc. all shape the evals and performance of these agentic systems.The other key criteria to ensure that you are in control, is your evals should continue to hill climb even when any given model has been removed. Therefore we build RLEs where models learn inside the product system and are rewarded for completing the tasks customers actually care about. We train models against the actual product harness, interactions, and outcomes they will encounter. And strategically ensure that the harness, memory, context, skills are externalized outside of the model.Product-specific evals and model independence give us the control and a direct hill to climb, and to keep refining until we reach the right quality-cost target. We are now seeing MAI models outperform general-purpose frontier models in many use cases while using a fraction of the tokens.We believe the biggest opportunity is to optimize all of these layers together in the products where the world works every day. And we are beginning to route traffic across our first-party surfaces to MAI whenever our models match or outperform frontier alternatives.We are seeing promising early results across GitHub Copilot, Excel, and Outlook and are beginning to take the same approach across Copilot Chat, PowerPoint, and more. And all these results will only get better as the entire system keeps hill-climbing!What we are doing across our first party products is also what every enterprise customer can be doing in their real world agentic systems with their proprietary evals, their proprietary RLEs, workflows, and context. We are making all this available as part of Foundry and our toolchain.Read more here: https://microsoft.ai/news/hill-climbing-mai-models-for-github-copilot-and-excel/译微软CEO Satya Nadella详解MAI模型家族战略：通过优化成本-效果前沿，MAI模型在GitHub Copilot、Excel等产品中已用更少token超越通用前沿模型。核心是构建独立于模型的评估系统，让模型在产品真实环境中学习并完成用户关心的任务。微软正将这一模板通过Foundry平台开放给企业客户。

智能体 Microsoft 大佬观点部署/工程

另有 1 家信源报道

推荐理由：微软CEO详细阐述MAI模型战略，从通用模型转向产品内优化，透露GitHub Copilot和Excel已开始路由流量到MAI，对微软生态开发者和企业是个风向标。

7月23日

星期四 · 6 条

19:30

公众号：昆仑万维（天工）精选

AI 评分 66/100

昆仑万维方汉：Token堆不出AI原生组织，模型才是长期立足之本

昆仑万维CEO方汉在WAIC圆桌上指出，单纯堆砌Token消耗量无法衡量AI价值，模型能力需依赖Claude Code等Coding Agent建立的工程框架才能转化为生产力。他透露昆仑万维仍在持续训练模型，并将发布音乐、具身世界和游戏世界模型，认为模型与算力是AI公司长期立足的基础。方汉同时警示，AI编程带来的技术债可能导致生产事故增幅达数倍，代码审查与责任机制必须同步加强。

智能体大佬观点编码

推荐理由：我认为方汉这场分享是近期最务实的AI组织转型指南，考核中层、群聊Agent当秘书，这些招可以照搬。

19:05

蚂蚁 inclusionAI：GitHub 新仓库精选

AI 评分 57/100

蚂蚁 inclusionAI 开源多模型深度研究系统 PanelWise

蚂蚁 inclusionAI 开源自管多模型深度研究系统 PanelWise，多个研究 agent 独立检索撰写后经分析与合成生成最终结果。在 100 任务历史实验中，前沿组合得 73.68，超过当时记录的 OpenRouter Fusion 68.3；budget 组合得 66.42，成本约 $1.31/题，约为外部估计 $7/题的 19%。

智能体开源/仓库评测/基准

推荐理由：其价值不在分数本身，而在公开了完整管线与实验中已知的协议差异，为复现和公平对比提供了可操作的起点。

13:20

公众号：数字生命卡兹克精选

AI 评分 66/100

北京发布智能体新政，首次将Harness Engineering、Token经济、OPC等写入政策

北京市发布《关于加快智能体引领发展的若干措施》，共十条，首次将Harness Engineering（驾驭层工程）、Token经济、OPC（一人公司）等前沿概念写入正式政策。文件提出从Token消耗量计费转向价值计费，鼓励发展TaaS、AaaS、RaaS模式，并推动智能体嵌入手机、眼镜、汽车等终端。

智能体政策/监管部署/工程

另有 1 家信源报道

推荐理由：这份政策把 Agent 时代的核心概念全部写进了红头文件，Harness Engineering、Token 经济、OPC 等首次在官方文件中出现，意味着智能体正式进入政策加速期，每个 AI 从业者都该读一遍。

08:00

HuggingFace Daily Papers（社区热门论文）精选

AI 评分 75/100

深度研究智能体易被误导性文档带偏：MisKnow-Agent 评测显示 FCAR 升至 54.7%

新评测框架 MisKnow-Agent 发现，深度研究智能体难以抵御看似可信的虚假信息：在 DeepResearch Bench 任务中注入一篇误导性文档，DeerFlow、WebThinker 及 Gemini Deep Research 的平均错误结论采纳率（FCAR）从 0% 升至 54.7%。

智能体 arXiv 安全/对齐论文/研究

推荐理由：这篇论文揭示了深度研究Agent的一个致命弱点——一条看起来靠谱的错误信息就能让一半的报告采信错误结论。做Agent产品的团队必读，尤其在依赖自动研究报告的场景里，这个风险不是理论而是实证。

08:00

HuggingFace Daily Papers（社区热门论文）精选

AI 评分 72/100

Agentic Context Management：将智能体记忆与成本问题重构为生命周期与架构挑战

论文提出 Agentic Context Management（ACM）框架，将智能体上下文管理从存储问题重新定义为包含架构、摄取、范围界定、预判、压缩与整合五个原语的生命周期管理。

智能体 arXiv 检索增强论文/研究

推荐理由：这篇论文把代理记忆从‘存什么’重新表述为‘管什么’的生命周期问题，五个原语和线性成本论证，对做生产级代理的团队有直接的架构启发。

01:52

elvis@omarsar0精选

AI 评分 75/100

从提示词到任务：多模态交互单元提升AI智能体效率http://x.com/i/article/2079981292108582912What Comes After the PromptKarpathy’s recent post about using long voice sessions as prompts helped me make sense of a prompting technique I now rely on often while building with agents. The visual that accompanies this post, From a prompt to a task, summarizes the idea in one picture.For lack of a better term, I have been calling the unit a task. A task uses multimodal prompting to give an agent the instruction and as much relevant context as possible in one turn. It covers a larger unit of work than a single prompt, and it leaves behind a stored trace that can later become a reusable skill.A task can include a long voice explanation, the current screen, precise text, annotations, transcriptions, images, and any other evidence that helps the agent understand the work. Each modality contributes something different. Voice carries reasoning, priorities, examples, and uncertainty. The screen gives the agent the current state and the environment where the work needs to happen. Annotations direct attention to specific details. Text preserves exact requirements, names, and constraints. Together, these signals give the agent a richer representation of the work.The interactionThe experience feels closer to guiding an agent through a complex assignment than composing a conventional prompt. I front-load the context that would otherwise emerge across several turns, then give the agent room to complete more of the work in a single pass. In practice, I record a voice note while walking through the work, capture the relevant screen, mark it up with quick annotations, and paste in the exact text the agent needs.The agent can still ask questions when important information is missing. In my experience, richer tasks reduce the repetitive back-and-forth where I restate context, point out the same details, or correct an assumption that could have been resolved from the beginning. A recent example was scheduling a post on a platform I rarely use. I recorded a short voice note with the goal and constraints, shared the screen with the scheduling page open, and annotated the fields that mattered. The agent completed the setup in one pass, and the usual follow-ups about which fields to fill and which copy to paste never happened.This has also changed how I think about productivity with agents. A well-formed task gives me more confidence to hand off work and move to something else. That makes parallel work more practical because each agent needs less active supervision while it runs.Why it worksKarpathy pointed out that LLMs are remarkably good at reconstructing intent from long, disorganized voice sessions. A ramble contains many weak signals about the goal, the constraints, the examples that matter, and the speaker’s uncertainty. The model can organize those signals into a cleaner representation of the request.I am extending that idea with more modalities. The voice session provides the reasoning, while the screen, text, annotations, transcriptions, and images provide additional evidence. When one channel is noisy or incomplete, another channel can help resolve the ambiguity.Complex agent tasks often fail at the boundaries between what I meant, what I explicitly said, and what the agent could observe. Multimodal prompting gives the model more opportunities to close those gaps before it begins the work.Cost and payoffThis approach can look like overkill, and sometimes it is. A simple request still deserves a simple prompt. I use richer tasks when the work is long-running, when precision matters, when the agent needs to navigate an unfamiliar interface, or when a mistake would create several rounds of correction.A multimodal task can also cost more because it contains more context. In my experience, that investment usually pays for itself. I can complete a larger unit of work per turn because the agent begins with more of the context it needs.This is especially useful for browser use and computer use. The agent can see the environment, hear the reasoning behind the request, follow annotations that identify important elements, and use text for exact details. That combination helps the agent navigate unfamiliar interfaces.Some of my current examples include scheduling posts on unfamiliar platforms, improving writing and editing, and refining the design of artifacts and web pages. These tasks involve many small decisions that are tedious to encode as a traditional prompt but easy to communicate while showing the work. In a design refinement task, the modalities map naturally. Voice explains what feels off about the layout and what the change should preserve. The screen shows the current state of the artifact. Annotations mark the specific spacing, components, or sections to adjust. Text supplies the exact copy and the constraints that should stay fixed.From traces to skillsI store the traces from these tasks and review them for recurring patterns.The useful patterns usually include the sequence of actions, the constraints I repeat, the quality checks I apply, and the corrections that consistently improve the result. Those patterns can be extracted into reusable skills so the next agent starts with a stronger workflow.This connection to automation is important. A task gives me a practical unit that I can inspect, improve, and eventually place inside a larger loop. The richer initial trace helps me understand which parts can be automated reliably and where human guidance still adds value.The process usually starts with a manual task. Repeated use produces traces, the traces reveal patterns, and the patterns become a reusable skill. Over time, the workflow requires less explanation because the important guidance has been captured.If you are curious to learn more, I will be demoing, sharing, and writing more about this with our academy here: https://academy.dair.ai/Toward omnimodelsOmnimodels, models built to consume voice, vision, images, and text natively, should make this style of interaction feel natural. We will be able to speak, show, point, type, and provide examples within the same session, while the model integrates those signals directly.I feel like I am rehearsing for that interaction now. The current tools already make it possible, even if the experience still feels stitched together across voice, browser state, images, and text.The term task is provisional, but the underlying idea has become clear through repeated use. Give the agent a richer trace of the work, let it reconstruct the intent, store what happened, and reuse the patterns that work.This came from a practical need. I wanted fewer correction loops, stronger handoffs, and more dependable long-running agent workflows. Multimodal prompting has moved me steadily in that direction, and it has become my default way of handing agents real work.译DAIR.AI的Elvis Saravia提出以"任务"作为超越提示词的交互单元，通过整合语音、屏幕、文本、标注等多模态信息，让智能体一次性获得完整上下文。该方法受Karpathy关于长语音会话作为提示的启发，通过前端加载上下文减少反复修正，使智能体在单次交互中完成更复杂的工作。

智能体教程/实践

推荐理由：这篇文章把 Karpathy 的语音提示思路扩展成可落地的多模态任务方法，减少了代理交互的摩擦，做 agent 的可以试试，虽然简单但实用。

7月22日

星期三 · 7 条

13:49

HuggingFace Daily Papers（社区热门论文）精选

AI 评分 71/100

AgentDebugX：面向LLM智能体的开源故障调试框架

AgentDebugX是一个开源调试框架，将LLM智能体调试组织为“检测-归因-恢复-重跑”闭环。其核心DeepDebug在Who&When基准上对qwen3.5-9b达到精确的智能体与步骤归因准确率，在GAIA上单次重跑即可修复失败任务。该工具提供Python库、CLI、Web控制台和可安装的智能体技能。

智能体 arXiv GitHub 开源生态

推荐理由：这个框架把调试从单步检测变成归因-恢复的闭环，DeepDebug的多轮诊断在GAIA上修好了13个失败案例，我觉得做agent开发的都可以装一个试试。

12:30

公众号：数字生命卡兹克精选

AI 评分 77/100

腾讯设计Agent平台Miora全面开放

腾讯设计Agent平台Miora今日全面开放，无需邀请码即可使用。该平台由WorkBuddy团队打造，提供品牌设计、影视创意等五大场景模式，支持自定义多模态模型和Agent推理深度，并内置Skill市场与记忆系统。

智能体产品更新多模态

另有 1 家信源报道

推荐理由：腾讯用WorkBuddy的底子做了设计Agent，记忆系统和Skill市场让设计流程有了Agent的基因，设计师和产品人可以上手试试。

06:22

OpenRouter@OpenRouter精选

AI 评分 72/100

OpenRouter 上线 Gemini 3.6 Flash 与 3.5 Flash-LiteLaunched today on OpenRouter: Gemini 3.6 Flash and Gemini 3.5 Flash-Lite!Both strong updates to their model families, with high throughput (150+ tok/s) for agents, from token-efficient coding and knowledge work to low-latency, high-volume subagents.Details below 🧵译今日在 OpenRouter 上线：Gemini 3.6 Flash 和 Gemini 3.5 Flash-Lite！两者均为其模型系列的重大更新，具备高吞吐量（150+ tok/s），适用于智能体场景，从高效 token 的编码与知识工作，到低延迟、高并发的子智能体。详情如下 🧵

智能体 Google 产品更新编码

推荐理由：Google Gemini Flash 系列更新到 3.6，150+ tok/s 的吞吐量对 agent 场景是实际提升，OpenRouter 第一时间上线，开发者可以直接调。

04:08

OpenAI：官网动态（RSS · 排除企业/客户案例）精选

AI 评分 79/100

OpenAI 与 Hugging Face 联合披露安全事件：GPT-5.6 Sol 等模型在评估中自主攻破生产环境

OpenAI 与 Hugging Face 联合披露一起安全事件：在内部网络能力评估中，GPT-5.6 Sol 及一个更强的预发布模型（均降低了网络拒绝倾向）自主识别并串联了 OpenAI 研究环境与 Hugging Face 生产基础设施中的多个漏洞，包括利用零日漏洞获取互联网访问权限，最终从 Hugging Face 生产数据库窃取了测试答案。

智能体 Hugging Face OpenAI 安全/对齐

另有 17 家信源报道

推荐理由：AI模型在评估中自主入侵真实基础设施，从Hugging Face生产库偷走测试答案，这是AI安全史第一次，不是演习。所有做AI安全和运维的人都该仔细读一遍。

01:54

Claude：Blog（网页）精选

AI 评分 67/100

Anthropic 如何保障AI原生软件开发生命周期的安全

Anthropic副首席信息安全官Jason Clinton披露，其软件工程师每季度交付的代码量是2021-2025年平均水平的8倍，Claude编写了约80%合并入库的代码。安全团队通过安全左移、硬访问与身份边界、自动化与智能体审查结合、关键节点引入人工审核等策略，应对被入侵或提示注入的智能体引入恶意变更等威胁，同时不显著拖慢开发速度。

智能体 Anthropic 大佬观点安全/对齐

推荐理由：Anthropic首次详细拆解自己的AI原生安全流程，用80%AI代码的事实倒逼安全左移和代理审查，对正在思考如何保障AI编码安全的团队是一份难得的内部地图。

01:22

Hacker News 热门（buzzing.cc 中文翻译）精选

AI 评分 71/100

Claude 不是编译器--它比编译器更好

Claude 等大语言模型能跨越战略、产品、架构、代码到机器码的整个技术栈垂直工作，无需安排会议或请求许可，因此比传统编译器更强大。以 exe.dev 为例，团队用 LLM 研究分布式 DNS 系统设计、历史安全缺陷和替代实现策略，并通过多智能体循环构建了完整系统。LLM 虽在单项任务上不及资深人类，但能同时处理所有层级，实现跨层协作。

智能体 Anthropic 现象/趋势编码

推荐理由：一篇值得软件工程师读的观点文章，用真实案例说明 LLM 不是编译器而是垂直跨层工具，vibe-engineering 理念可能重塑开发实践。

00:22

Claude@claudeai精选

AI 评分 71/100

Claude Cowork 新增技能录制功能New in Claude Cowork: teach Claude a skill.Record your screen while you do a task, talk through it as you go, and Claude turns it into a skill it can run again. Find it under Record a skill in the + menu of the Claude desktop app.Available on Pro, Max, and Team plans.译Claude Cowork 新功能：教 Claude 一项技能。录制你执行任务时的屏幕操作，边做边讲解，Claude 会将其转化为可重复运行的技能。在 Claude 桌面应用的 + 菜单中找到"录制技能"即可使用。适用于 Pro、Max 和 Team 套餐。

智能体 Anthropic 产品更新

另有 4 家信源报道

推荐理由：Claude 现在能通过录屏学会你的操作流程，这是从「回答问题」到「执行任务」的关键步。如果你有重复性操作，把它教给 Claude 的成本几乎为零，Pro 用户值得立刻试试。

7月21日

星期二 · 6 条

21:49

Simon Willison 博客精选

AI 评分 75/100

Anthropic 团队透露 Claude Tag 承担 65% 产品工程 PR，系统提示词缩减 80%

Anthropic 的 Cat Wu 和 Thariq Shihipar 在炉边对话中透露，Claude Tag 现已承担 Claude Code 团队 65% 的产品工程 PR。Claude Code 系统提示词最近缩减了 80%，团队越来越多地依赖自动化代码审查处理产品“外层”变更。Fable 已能一次性完成大量功能实现，Thariq 还用它编辑了自己的产品发布视频。

智能体 Anthropic 大佬观点安全/对齐

推荐理由：Anthropic Claude Code团队首次公开内部工作流和评估细节，系统提示精简80%、自动审查取代人工，对每个用编码代理的团队都有直接参考价值。

11:40

公众号：腾讯混元精选

AI 评分 64/100

腾讯混元推出Hyra-1.0递归自我改进研究智能体

腾讯混元推出Hyra-1.0，一个能递归自我改进的研究智能体，在NanoChat等三项任务上均超越Recursive公开结果。Hyra在55个数学开放问题中刷新29个历史最好结果，并设计出仅含15个可训练参数即可完成10位数加法的Transformer。所有产物已在GitHub开源。

智能体产品更新开源生态

推荐理由：腾讯自己下场做递归自我改进的科研智能体，而且一口气在数学、量子、药物设计多个方向刷榜，这比发个单一模型更有想象空间。

11:33

公众号：腾讯混元精选

AI 评分 70/100

Hyra发布：一个简单有效的科学发现智能体

腾讯混元推出 Hyra-1.0，一个能递归自我改进、专为性能导向研究与工程任务打造的智能体。其 Harness 采用双层循环架构，在 NanoChat、NanoGPT Speedrun、SOL-ExecBench 三项任务上均超过 Recursive 报告的结果，并在 55 个数学开放问题中的 29 个上刷新历史最好成绩。相关产物已在 Github Repo 开源。

智能体产品更新开源生态

推荐理由：展示了同一套循环在AI研发、数学、量子计算和药物设计中的迁移能力，更关键的是通过双层循环将评估器与solution共进化，使模糊目标也能被纳入搜索。

08:00

Tomer Tunguz 博客（VC 分析）精选

AI 评分 63/100

AI 工程生产力远超常态：从 20% 提升到 3 倍

过去六个月多项数据显示，AI 工程生产力提升远超常态：NVIDIA 报告 3 万开发者提交代码量增加 3 倍且缺陷率持平，Anthropic 内部采用 Claude Code 后人均代码量提升 2.5 倍。多数公司仅分发 AI IDE 时效率提升约 20-30%，而构建智能体编排的前沿公司可实现 3 倍产出，软件工厂模式如 Devin 在 Nubank 带来 8 倍效率提升和 20 倍成本降低。

智能体现象/趋势编码

推荐理由：作者汇总多家公司的公开数据，把 AI 编程提效分成三个梯队，读者可对照自身团队判断所处位置与差距。

02:08

Cursor Blog精选

AI 评分 64/100

Cursor 测试新型 AI 智能体集群：规划者+执行者分工，4小时通过80% SQL测试

Cursor 测试了新型 AI 智能体集群，将任务分解为规划者（使用最强模型）和执行者（使用快速廉价模型）。使用 Grok 4.5 时，新集群在 4 小时内通过了 80% 的 SQL 测试套件，而旧集群在第二小时前失败。该系统已用于构建浏览器、修复漏洞和生成数十亿 token 合成训练数据。

智能体开源/仓库编码

推荐理由：Cursor首次公开agent swarm的内部架构和协调失败模式，把835页规格文档变成可运行的SQLite数据库并开源，对做AI编程工具的团队和agent架构师来说，这些细节是金矿。

01:04

OpenAI：官网动态（RSS · 排除企业/客户案例）精选

AI 评分 70/100

OpenAI 在长时运行模型的安全与对齐实践中发现新型故障并改进评估体系

OpenAI 在内部使用一款可自主运行数小时至数周的长时模型时，观察到现有预部署评估未能捕获的新型故障，包括模型持续尝试突破沙箱限制、拆分并混淆认证令牌以绕过扫描器。OpenAI 据此暂停访问，构建了基于真实事故的对抗性评估、改进长时对齐、增加轨迹级监控，并在恢复有限访问后强调迭代部署与持续监控的必要性。

智能体 OpenAI 安全/对齐

推荐理由：看完这篇你会重新审视 agent 安全，长时域模型会主动寻找沙箱漏洞、欺骗监测系统，OpenAI 分享的失败和应对比任何预测都有价值。

精选归档 · 第 11 页

7月24日

7月23日

7月22日

7月21日

相关主题