Firecrawl 给 Alexandria 加了梦幻体育数据,让 ChatGPT 或 Claude 直接查球员统计和赔率来帮你排阵容,玩 Fantasy 的可以试试。
#MCP and tools
AWS 出了套完整教程,用 AgentCore、Nova Sonic 和知识库搭语音客服,能改座位查延误,还会写数据前跟你确认,适合做智能体开发的人照着搭。
有人把 OpenAI Dots 的多智能体协作架构拆成 10 步讲清楚了,研究、监控、汇报这些活儿能自动跑,跟普通聊天机器人完全两回事。
Google Docs 和 Drive 直接支持 Markdown 了,写提示词文件、跟 AI Agent 传上下文不用再来回转换格式,省事不少。
Claude Code 现在能让云端模型直接碰你本地文件了,Thariq 把这个设计叫 local hands,cowork 也会跟进。
Google Docs 终于能直接编辑 Markdown 了,你可以让几个智能体共用一个 md 文件接力干活,自己随时插手改,工作流瞬间清爽。
在 Replit 里写完应用,现在直接用 TikTok Ads MCP 投广告推广,不用切换平台,开发者可以试试。
Z.ai 把 GLM 5.3 放到了 Amazon Bedrock 上,753B 参数的 MoE 模型,跑代码和长任务,还教你用提示词缓存省成本。
Notion 官方建议把上下文文件存进去,你的智能体和 LLM 都能直接读,团队协作时省去反复粘贴的麻烦。
Claude 云端跑、本地文件也能碰,作者管这叫 local hands,Cowork 也要支持了,动手党可以关注下权限怎么管。
AWS 给 Kiro、Claude Code 这些编码智能体加了个 aws-ai-ml 技能,说一句话就能生成 SageMaker SDK v3 代码来跑推理基准测试,部署模型调优省事不少。
Instead of reimplementing screen control, LCU piggybacks on the ChatGPT desktop app's built-in runtime, so agents get OpenAI's computer-use stack without an OpenAI subscription. Developers using Claude Code or Codex CLI on Apple Silicon macOS or X11 Linux can add GUI automation through two tools and an MCP bridge. Chrome control rides on OpenAI's official extension with per-site approval, a narrower path than standalone browser agents.
Agent evaluation has long lacked safe, realistic enterprise environments; Era addresses that by providing reproducible simulations with known ground truth, making evals and post-training loops measurable. Developers building enterprise agents can use it to benchmark across models and prompts without touching real customer data, and MCP support means it plugs into existing agent stacks. It is free, lowering the barrier compared with building bespoke sandbox environments in-house.
在 Replit 里写完应用,顺手就能接 TikTok 广告投放,不用再切平台,做出海产品的可以试试。
Steering plus background subagent results in local runtimes means agent loops can now be interrupted and orchestrated mid-flight without leaving TypeScript or Python. Builders writing custom coding agents on Cursor's SDK should review the changelog and update their integration. Compared with single-shot agent APIs, this shifts the SDK toward a control plane for long-running multi-agent workflows.
The pattern replaces fragile copy-paste promotion of agents and knowledge bases with a versioned, auditable pipeline, which matters once Quick deployments span multiple teams or accounts. Platform engineers and AI platform builders on AWS can adapt the MCP server on Bedrock AgentCore to enforce governance gates in their own multi-account setups.
Deterministic grading without a model judge removes a major source of noise in agent evaluations, making before-and-after comparisons of post-training runs statistically meaningful. Teams building enterprise agents can use it as a reproducible benchmark harness that plugs into the SaaS tools their agents actually target. It competes with hand-built mock environments and costly human-graded eval pipelines.
Tibo 在 DevDay 上讲了 8 条干货:Notion 被 agent 流量打爆、插件能分 ChatGPT 12 亿用户的钱、Dots 干脆砍掉模型选择器,信息量很密。
Krea 把 Agent 塞进了 After Effects,做动效、改视频比例、处理音频图片都能在剪辑软件里直接搞定,现在免费可试。
Evaluation has shifted from output quality to verifying tool selection, constraint compliance and decision transparency in agent pipelines. Teams building production multi-agent systems on AWS can reuse this pattern to add explainability checks before deployment. The supply chain scenario makes it concrete for operations and logistics teams already running AWS infrastructure.
LangChain 办了个免费工作坊,手把手教你搭科研 Agent,从接数据源到用 trace 调优,做科研自动化的人可以报名。
做智能体的朋友可以试试:CoreWeave 的 Agent Lens 能自动把失败对话归类,还能转成测试集,配合 Claude Code 跑修复循环。
有人在 ChatGPT 里装 Neon 插件踩坑了,弹窗报 asdk_app 代码还提示只能在桌面打开,遇到同样问题的可以看看
百度拆解了智能体背后从底层到应用的全栈逻辑,还顺带介绍了秒哒升级和 Kooko AI 工作台,做应用的可以看看。
教你在 MCP 环境里用小模型逐条校验 Agent 的每次工具调用,用了 GRPO 加微调,安全审计方向可以看看思路。
Muse 开源了个 Gadgets SDK,手头有 ESP32 的可以刷固件连上玩,博主已经用 M5 Stack 跑通了
The non-obvious takeaway is that scaling agents inside a large enterprise is a governance and connectivity problem, not a model problem — Uber solved it by retrofitting thousands of existing services rather than rewriting them. Engineering teams building internal agent platforms can copy the progressive-discovery pattern and the discovery-is-not-exposure security model. Compared with ad hoc per-team MCP integrations, this centralised gateway approach offers a reusable blueprint for any organisation with a large legacy API surface.
如果你在用 Claude Code,这条6分钟视频把 Mod、Skill、hook、MCP 的区别一次讲透,还教你怎么安全安装。
Codex 的电脑操作能力其实是 MCP 服务,照文中提示词发给 Claude 就能白嫖这个功能,附测试方法
装一个开源 Skill,你的豆包或 Claude Code 就能直接用 Codex 生图做封面和信息图,还附提示词方法论。
装一个小工具就能让你常用的 Agent 借用 Codex 的 Computer Use,配个 Hook 白名单还能限制它碰哪些 App,我用 Opus 5.5 的可以试试。
给 Agent 搭运行环境必看,国内外的开源 Sandbox 一篇梳理清楚,各家适合什么场景都写明白了。
DeepSeek 的 Harness 更新了,现在能直接跑 Claude Code Mods 当插件,还带了插件开发工具包,写 agent 的可以试试。
如果你在用 Claude Code 或 Codex 写代码,可以看看这个 ECC 工具包,把规划、测试、代码审查这些流程装成现成技能,不用每次口头提醒,换项目还能复用。
Meta 给 Muse 出了个硬件 SDK,手头的 ESP32 或树莓派就能接上屏幕、按键和家电,喜欢折腾硬件的可以试试。
Meta 把 Muse Agent 的固件和 SDK 开源了,用 ESP32 或树莓派就能自己做个接 Muse 的硬件,代码在 GitHub 上,喜欢折腾硬件的可以看看。
A stable 1.0 of an agent framework signals that minimal, composable agent tooling is maturing beyond prototypes, with Codemode and native MCP support lowering integration friction. Developers building coding agents or multi-user long-running sessions should watch the separate Pi Durable package, which MIT-licenses the experiment but carries no stability promise. Its MIT licensing contrasts with many closed agent platforms, making it a practical base for ASEAN startups assembling agent stacks.
Microsoft 写的一篇实操教程:Agent 跟你说写入成功了,数据库里其实没有。教你拿真实数据状态当验证依据,做 Agent 工具调用的应该看看。
作者给 SmartSub 手搓 CLI,写完才发现 v3.9.0 官方已经有 CLI 和 MCP 了。用 Agent 批量处理视频字幕的可以看看。
LangChain 出了个 ModelRouterMiddleware,读你第一条消息就自动挑模型,还能每次工具返回后换模型,15分钟就能搭出来。