Latest

Every published item, in order · 4309 items
Nov 12
How-toBuildersMulti-source16:32
Chinese contentGrokBot 团队单月合入 11000 个 PR,回退率约 0.5%

GrokBot 团队一个月往生产合了 11000 个 PR,回退才 0.5%。她的秘诀是把每次纠正智能体的错误都变成确定性检查,而不是往 agents.md 堆字。正在用 Claude Code 或 Cursor 跑智能体工作流的人可以直接抄这个做法。

Event
AI Will@FinanceYF5Source
How-toBuildersMulti-source16:02
Chinese content用好 Coding Agent 的关键:把协作当管理来做,附 12 条实践

这篇把和 Claude Code、Cursor 这类编程 Agent 的日常协作拆成了管理流程:选任务、写要求、跟进度、验收上线。里面 12 条实践都偏可操作,不是空谈。每天让 Agent 写代码的人照着做,能少走不少弯路。

Event
AI Will@FinanceYF5Source
How-toBuildersMulti-source15:43
Chinese content工程师新定位:用好 Coding Agent 关键是学会管理而非写 Prompt

这条帖子把用 Coding Agent 的日常拆成了一套管理动作:选任务、写要求、验收、决定上线,还给了12条具体实践。如果你每天和 Cursor、Claude Code 打交道但总觉得产出不稳,可以对照着检查自己缺了哪一步。

Event
AI Will@FinanceYF5Source
Chinese content用 YouArt Desktop 自动生成开源打字游戏「键盘小岛」的演示视频

老程序员用 YouArt Desktop 给自己的开源打字游戏「键盘小岛」做演示视频:把 GitHub 链接丢给它,Agent 自己打开游戏、操作录屏、连 PR 历史都自动提取,连中文字体缺字问题都顺手修了。自己做过项目的开发者可以试试这种创作 Agent 加 Coding Agent 的分工。

shao__meng@shao__mengSource
How-toBuildersMulti-source15:35
Chinese content20个开源Skill把Claude Opus 5.5配成创意工作室

有人把20个开源Skill打包成Claude Opus 5.5的创意工作室,从视频到游戏开发都有现成配置,GitHub星数超77万。做内容或独立开发的可以直接抄清单,比自己调提示词省事,想扩展Skill用法的也值得看看别人怎么拆角色。

Event
AI Will@FinanceYF5Source
Chinese content单条提示词生成 83 秒手绘风机甲动画短片,代码已开源

有人把完整提示词和代码都贴出来了,一条 prompt 就能在浏览器里跑出一部 83 秒、39 个镜头的手绘风机甲打海怪短片,还自带配乐、离线可播。想研究 AI 视频生成怎么写长分镜提示词的,这份拆解值得直接抄作业。

AI Will@FinanceYF5Source
How-toBuildersMulti-source15:33
Chinese content用 Fable 5.5 以纯代码生成83秒动画短片,8组Agent并行3小时内完成

有人拿 Fable 5.5 不用视频模型,纯靠8组Agent写 Three.js 代码渲染出一部83秒短片,39个镜头1992帧,最后就一个HTML文件。比 Sora 按帧生成省钱得多,想精确控制画面的开发者可以照这个流程试试。

Event
AI Will@FinanceYF5Source
Tomorrow 08:00, the daily lands in your inboxTelegram channel or email. Same screening as this page, no ads.
Chinese content开发者逆向 Anker 录音豆蓝牙协议并自建录音管理软件

这位开发者自己逆向了 Anker 录音豆的蓝牙协议,还给它做了 MCP,录音能直接在 Claude Code、Codex 这些工具里被调用,不用翻设备。自己接千问 ASR 本地转录,绕开了原厂绑定。喜欢折腾硬件和想复用录音数据的人可以等他开源。

berryxia@berryxiaSource
How-toBuildersMulti-source12:39
Chinese content用 Claude Fable 5.2 把 2D 户型图转成可漫游 3D 样板间,单份报价 5000 至 10000 美元

这篇把一个具体的赚钱路子拆得很细:Fable 5.2 能把户型图变成能走动的 3D 房子,一套卖 5000 到 10000 美元。从找客户、免费出样品到 Blender 加 three.js 建模的流程和提示词全给了,做设计外包或想接单的朋友可以直接照着跑。

Event
AI Will@FinanceYF5Source
Chinese content从22年底算起:MCP、memory、agents 都会被大模型本身淘汰

这篇推文只用了几句话就给了一个判断:等模型能 RSI,MCP、memory、agents 这些外挂都会被模型自己吃掉,agent 只是模型的一个实例。作者建议把精力放在定义目标、约束和评估上,而不是搭框架。适合正在纠结要不要深耕 agent 工程的开发者对照思考。

Yangyi@YangyixxxxSource
How-toBuildersMulti-sourceSelected12:20
Stanford CS329Z publishes agent engineering course materials

The signal here is that agent harness internals, normally proprietary engineering inside Claude Code and Codex, are now teachable with a US$5 budget and MIT-licensed test suites. Self-learners get a concrete path: clone the repo, make the failing tests pass, and only pay for the model-scored portion. Practitioners and educators across ASEAN and China can reuse the material freely with attribution, making it a direct alternative to paid agent frameworks and bootcamp courses.

Event
宝玉@doteySource
How-toBuildersMulti-source74°12:15
Chinese contentFable 5.2 一句话提示词 41 分钟生成单文件 FPS 游戏

这个案例最有价值的是那套自检提示词:用 OfflineAudioContext 让 AI 自己听音频、用 agent-browser 截图看报错,等于给盲盒式生成加了质检。第 ④ 步缓存命中率 90.9% 才把成本从 $60.57 压到 $15.96,做 agentic 生成的人一定要算这笔账。建议从 $5 以内的音效台步骤开始跑。

Event
berryxia@berryxiaSource
How-toBuildersOfficial12:11
Chinese content一篇Claude Code实操笔记:用skill系统从vibe coding走向正规工程

写代码的人看这个:作者把Claude Code的skill系统玩明白了,连安装、创建自定义skill的命令都给出来了,还顺手写了测试、部署相关的模板。附赠的省钱配置也挺实在,每天3美元跑agent,比很多人烧几十美元省多了。

Primary source歸藏(guizang.ai)@op7418Source
How-toBuildersSelected12:07
Together AI: Match Inference Bottlenecks to Your Traffic Shape

The practical shift here is treating provider benchmarks as workload-specific rather than universal — a provider that wins on decode speed can lose badly on prefill-heavy or agent workloads. Engineering teams evaluating inference vendors or deciding between self-hosting and APIs should benchmark with traces from their own traffic before committing. Concrete comparison: run the same prompt set against at least two providers and measure the metric matching your traffic shape, not aggregate throughput.

Official accountTogether AI@togethercomputeSource
How-toBuildersMulti-source71°12:03
Chinese content斯坦福新开 CS329Z 智能体工程课程,第一讲公开:智能体系统详解

斯坦福这学期新开的智能体课免费放出来了,第一讲是 Diyi Yang 主讲,配中英字幕。她用 ReAct 论文、OpenClaw 删邮件这些例子把智能体的循环结构和记忆设计讲得很透,想做智能体的开发者可以直接当入门课看。

Event
宝玉@doteySource
How-toBuildersOfficial11:39
Chinese content在 Anthropic 的封锁之下,持有 Claude 账号成了一种套利空间

这条帖子聊的是个很实际的话题:Claude 被限制访问后,有个账号能干什么、能不能把订阅费赚回来。作者没给完整教程,但评论区吵得挺热闹,做副业或者靠 AI 接单的人可以看看思路。

Event
Primary source歸藏(guizang.ai)@op7418Source
How-toBuildersOfficialSelected10:30
Guizang skill guides Claude through iterative AI game development

The takeaway is that multi-round iteration plus a domain checklist beats one-shot prompting for complex projects like games. Developers using Claude or similar agents can fork the guizang skill to inherit combat design, game-feel and build-size optimisation practices without redoing the research. Anyone packaging their own project experience into skills will find the requirement-interview, tool-research, self-verify loop a useful template.

Primary source歸藏(guizang.ai)@op7418Source
Chinese contentMicrosoft Research 播客解析红队、蓝队与紫队网络安全实践

Microsoft Research 的人聊 AI 时代怎么做安全测试,把红队、蓝队、紫队各自干什么讲得很清楚。做安全或想给自家系统做攻防演练的工程师可以听听,比单纯讲攻防对抗的视角更完整。

Event
Official accountMicrosoft Research@MSFTResearchSource
How-toBuildersSelected04:00
PyTorch Maintainers Show Single-GPU Testing for Distributed Training

The non-obvious point is that most distributed-training failures can be caught on a single GPU with stubbed communication and static SPMD checks, saving cluster hours. Engineers preparing large PyTorch runs, or teams with limited GPU access, can apply the fake process group pattern directly. The same keynote also introduces torch_remat and FlexShard, worth tracking for memory-efficiency workloads.

Official accountPyTorch@PyTorchSource
How-toBuildersSelected02:40
Perplexity Finds CTF Tasks Poorly Measure Agent Security Skills

The non-obvious point is that even security-specialised models can underperform on real red-teaming work, so vendor claims of security fine-tuning deserve scrutiny. Teams evaluating agents for security-sensitive deployments in finance, infrastructure or SaaS should design internal task suites instead of relying on CTF leaderboard scores. The finding applies regardless of region, since evaluation methodology, not model origin, is the bottleneck.

Official accountPerplexity@perplexity_aiSource
Chinese contentLlamaIndex 让 AI 智能体用标准账号对 LlamaParse 做安全测试

LlamaIndex 团队让智能体用普通账号花四天攻自己的 LlamaParse,把解析、抽取、索引到 MCP server 全链路试了一遍。最扎心的发现是抽取会读到人眼看不到的文本,隐藏注入风险直接摆上台面。做 RAG 和文档智能体的开发者都该看看这套测法。

LlamaIndex@llama_indexSource
How-toBuildersMulti-source01:41
Chinese content实测:新注册Apple账号订阅Claude触发风控,最终被永久封号

想给Claude订阅省事的朋友看过来,作者把新Apple账号订阅踩坑的全过程记录下来了:72小时风控、客服battle、最后永久封号。新号直接订阅基本会被卡,老号和其他区可能更稳,踩坑前可以参考。

Event
Gorden Sun@Gorden_SunSource
Chinese contentDeepLearning.AI 推出 Tabular 基础模型课程,用 AI 智能体做表格预测

DeepLearning.AI 和 Prior Labs 出了门免费课,教你把表格基础模型直接当智能体的预测工具,不用再为每个数据集单独训模型。做数据分析、搭数据科学 Agent 的朋友可以看看,省掉不少训练和调参时间。

DeepLearning.AI@DeepLearningAISource
How-toBuildersSelected01:21
SGLang Megakernels Lift Muse Spark Open Decoding to 1,109 Tokens/s

Kernel-launch overhead, not compute, is often the real ceiling in multi-GPU inference, and this write-up shows concrete numbers and code proving it. Engineers building or tuning serving stacks on Blackwell-class hardware get CUDA/PTX examples and measured timelines they can adapt. Teams benchmarking against vLLM or TensorRT-LLM deployments will find the launch-count reduction a useful reference point.

Official accountLMSYS Org (SGLang)@lmsysorgSource

Only the latest 2000 items are listed; for older ones see the daily archive