Latest

Every published item, in order · 16543 items
Nov 14
ProductsBuildersSelected76°10:40
Sierra Releases Personal Agent Protocol Poppy 1.0 with Shopify and Meta Muse

Poppy turns agent-to-business access from screen-scraping into a discovery, identity and authorisation layer, meaning companies can gate agents without blocking them entirely. Developers building agents that act on user accounts should review the OAuth scoping flow and the extension model, since shopping is delegated to UCP rather than reinvented. The Apache 2.0 licensing and draft status make early adoption low-cost but subject to change before final 1.0.

shao__meng@shao__mengSource
ProductsBuildersMulti-source10:31
Chinese contentReplit 与 Visa 合作推进智能体商务

Replit 的智能体以后能直接帮你下单买东西了,Visa 负责管住支付安全这一环。比起自己接 Stripe 写一堆风控代码,开发者用 Replit 这套可以省不少事。做智能体应用、想让 AI 自动处理采购的团队值得关注,后面还会上新功能。

Event
Amjad Masad@amasadSource
ProductsBuildersOfficial10:06
Chinese contentopenJiuwen 推出 RSI 原生 AgentOS,智能体可自行编写并回滚插件

如果你在折腾智能体框架,这个思路挺有意思:openJiuwen 的 AgentOS 让 WorkSwarm 智能体自己写插件、自己装、坏了还能回滚,像给智能体配了个带版本控制的包管理器。比起 LangChain 这类要人工写工具的框架,它把扩展这事交给了智能体自己。

Event
Primary sourcepandaily@[email protected] (Pandaily)Source
ProductsBuildersOfficial10:04
Chinese contentDeepSeek 开放搜索、文件、OCR 与语音全套 API,开发者可复刻官方 App

如果你在做搜索问答或智能体应用,这次开放挺省事:搜索、OCR、语音识别和合成都能一个家买齐,不用再东拼西凑几家供应商。做工作流产品的团队可以直接照着 DeepSeek App 的功能清单来编排接口。

Primary sourcepandaily@[email protected] (Pandaily)Source
Tomorrow 08:00, the daily lands in your inboxTelegram channel or email. Same screening as this page, no ads.
ProductsBuildersMulti-source08:55
Chinese contentAgent harness Reason 主打单任务成本,比 Codex 最多省 20 倍

Reason 这个 agent harness 用 Terminal-Bench 和 DeepSWE 的成绩证明跑分不差,重点是把单任务成本打到 Codex 的零头。如果你的 agent 要 7x24 常驻跑巡检、客服或 PR Review,这笔账很实在,可以申请个早期访问试试。

Event
shao__meng@shao__mengSource
ProductsBuildersMulti-source08:38
Chinese contentllama.cpp 与 ggml 发布 1.0.0 版本,支持超 150 种模型架构

在本地电脑上跑大模型这个项目终于到 1.0.0 了,现在支持 150 多种模型架构,六大芯片厂商的硬件都有测试覆盖。做本地部署或依赖这个库开发的工程师,可以关注它新定的版本规则,升级前更好判断要不要改自己的代码。

Event
IT之家Source
ProductsBuildersOfficial08:22
Chinese contentShield 用 OpenAI 打造 Forge 智能体,帮 IT 技术员交接遗留门户系统工作

如果你所在团队还在和一堆没有 API 的老门户系统打交道,这篇 OpenAI 客户案例值得看看。Shield 的 Forge 智能体能自动导航这些系统跑流程,技术员只处理需要判断的活。跟 UiPath 这类 RPA 比起来是用大模型直接操控界面,IT 运维团队可以参考。

Event
Primary sourceOpenAI BlogSource
ProductsBuildersMulti-source07:08
Chinese contentClaude Code 的 Projects 更新:协调器更主动,可回答问题和汇报线程状态

Claude Code 的 Projects 更新后,协调器会主动在聊天里回答问题、汇报各并行线程的进度,不用你自己一个个去查。如果你平时在 Claude Code 里同时跑多个任务,这个改动会明显省事,值得升级试试。

Event
claudedevs@claudedevsSource
ProductsBuildersMulti-source06:37
Chinese contentReason Machines 发布 Reason:让缓存命中为编码代理省下 22% 成本

Reason Machines 这个叫 Reason 的编码框架,靠把提示词前缀固定不动,让 OpenAI 缓存命中按 1 美元/百万 token 计费,比 Codex 便宜 22%。没换模型、没做微调,纯粹改了上下文组织方式。自己搭 agent 的开发者直接照抄这套前缀结构就能省账单。

Event
rohanpaul_ai@rohanpaul_aiSource
ProductsBuildersMulti-source06:34
Chinese contentGemini 4 Hydrogen 演示从创意到成片的视频生成流程

Google 用 Gemini 4 Hydrogen 做了个有趣的演示:一句话需求就自动找公有领域素材、剪成片、配旁白。和 Sora 那种纯合成画面不同,它走的是素材检索加自动剪辑路线,适合做资料类短片。做内容运营或短视频的朋友可以看看这个工作流思路。

Event
Google Gemini App@GeminiAppSource
ProductsBuildersSelected05:43
Browser Use cuts agent cost 33% by fixing prompt cache hit rate

Cache hit rate is often invisible in agent eval reports, yet here it accounted for a third of the cost per task with zero score change, a lever most teams never instrument. Engineers building or benchmarking browser agents can audit their own harnesses for cache-breaking behaviour and adopt browser-use 1.0.4 directly. The standardised comparison against six browser tools inside Codex gives buyers a concrete reference point when choosing a harness.

Browser Use@browser_useSource
ProductsBuildersMulti-source05:24
Chinese content美国 Gemini 用户现可绑定 Google Wallet 并接入银行账户

Google 这次把 Gemini 接上了真实钱包和银行数据,靠 Plaid 读信用卡和账户记录。比起 Apple Intelligence 还停留在语音和照片整理,Gemini 已经能分析你的实际消费了。如果你在美国用 Gemini,可以在设置里试试绑定,顺便留意一下授权范围。

Event
TestingCatalog@testingcatalogSource
ProductsBuildersMulti-sourceSelected05:19
Artificial Analysis launches Diarization Index, NVIDIA leads

A standard WDER metric shifts how transcription APIs will be evaluated, since many buyers currently judge quality only on word accuracy and miss attribution failures that break meeting summaries and call analytics. Teams building voice agents, meeting assistants or call-centre tooling can use the 97-conversation suite to benchmark vendors before committing. NVIDIA's Nemotron 3.5 Transcribe at 3.3% versus Voxtral Small Transcribe 3 at 7.5% sets a clear quality bar for providers to chase.

Event
Official accountArtificial Analysis@ArtificialAnlysSource
ProductsBuildersOfficialSelected05:00
Google opens Antigravity /teamwork multi-agent tools to external users

The real signal is that Google moved a multi-agent collaboration framework from internal demos to external availability, so claims about agents solving open research problems can now be independently tested. Researchers in mathematics, chemistry, systems software and ML engineering can apply the toolset to their own unsolved tasks. Teams comparing agent frameworks should check the linked results before committing. Availability details and pricing for ASEAN and global users were not stated in the announcement.

Primary sourceGoogle Research@GoogleResearchSource
ProductsBuildersMulti-sourceSelected71°04:55
llama.cpp v1.0.0 ships as first stable release

A stable 1.0 changes llama.cpp from an actively shifting project into infrastructure teams can standardise on, which matters for anyone deploying models on-premise or offline. Cohere's prompt adoption shows model vendors now treat llama.cpp support as a launch requirement, a signal worth tracking for teams choosing which architectures to build against. Developers embedding local inference in apps can treat the v1 API as a stable contract rather than pinning versions defensively.

Event
Official accountCohere@cohereSource
Chinese contentGoogle 一周密集发布:两款 8 代 TPU、Gemini 4 Carbon 等多款产品

Google 这一周发得真够密的:两款 8 代 TPU、Gemini 4 Carbon 和 Hydrogen 两个新模型,还有 Gemini Robotics 2.5 和智能眼镜预购。做开发的可以直接去 AI Studio 试试新的 Decisions API,关心算力成本的也该看看新 TPU。

Event
Sundar Pichai@sundarpichaiSource
ProductsBuildersMulti-source04:49
Chinese contentOpenAI 推出 ChatGPT Wallet,Dots 智能体可代用户完成支付

OpenAI 给 Dots 智能体配上了真实钱包,绑定卡后就能替你下单买东西。现在只对美国 Pro 用户开放,比 Google 的结账方案更彻底——直接把支付权交给智能体。想体验智能体代付的人可以关注后续地区扩展。

Event
Official accounttheverge@Stevie BonifieldSource
ProductsBuildersMulti-source04:37
Chinese contentNVIDIA 把 Déjà View 交互式 Demo 上线 Hugging Face Spaces

NVIDIA 在 Hugging Face 上架了 Déjà View 的免代码在线 demo,打开浏览器就能试用,不用自己配环境。比起读论文或本地部署,这种 Spaces 演示上手快得多,做图像相关项目的开发者可以先试用再决定要不要跟进。

Event
Official accountNVIDIA AI@NVIDIAAISource
Chinese contentbrowser_use 与 1Password 集成:智能体登录无需接触密码

做浏览器智能体的朋友可以试试,browser_use 接上了 1Password,智能体遇到登录页时由你审批、1Password 填密码,agent 全程看不到明文凭据。比把密码塞给 agent 自己填安全不少,跑需要登录的自动化任务很实用。

Browser Use@browser_useSource
Chinese contentReason Machines 推出 AI 编程智能体 Reason,主打每美元完成工作量

有点意思的省钱思路:直接复用你已有的 ChatGPT 或 Claude 订阅来跑编程智能体,还宣称比 Codex 少花 22% 完成更多任务。给小团队和个人开发者省一笔订阅钱,常驻审查 PR 的功能也很实用,可以试用对比下。

TestingCatalog@testingcatalogSource
ProductsBuildersMulti-sourceSelected04:10
Google Antigravity rolls out Dynamic Workflows for agent orchestration

The notable shift is that agent orchestration becomes reviewable code rather than opaque prompt chains, making multi-agent pipelines reproducible and auditable like normal software. Teams building production agents, especially those debugging nondeterministic subagent behaviour, can adopt this to turn ad hoc flows into saved, re-runnable scripts. Developers already on Antigravity should try the /run-workflow command; those comparing frameworks can weigh it against swarm-style autonomous orchestration.

Event
Varun Mohan@_mohansoloSource
Chinese content宾大 Generative AI Studio 学生展示:个人作品远超三年前团队项目

Mollick 在宾大看了一批学生用 AI 做的展示,他说个人项目的完成度顶得上三年前一整支团队干出的活。想感受 AI 对个人产出能力的实际拉动,这条一手观察比一堆跑分更直观,教课和招人的朋友都该看看。

Ethan Mollick@emollickSource
Chinese contentSalesforce 将时序基础模型 Moirai 接入 Service Cloud 秋季版,用于预测人工与 AI 客服混合排班

Salesforce 把自家时序预测模型 Moirai 装进了客服产品,用来同时预测人工坐席和 AI 客服智能体的工作量排班。做客服运营或企业服务采购的可以看看,传统 WFM 工具只管人,这个把 AI agent 也算进了产能。

Official accountSalesforce AI Research@SFResearchSource
ProductsBuildersMulti-sourceSelected03:51
Anthropic Publishes 10 Recommendations for Designing AI Safety Evaluations

The non-obvious takeaway is that Anthropic is treating evaluation methodology itself as publishable engineering practice, not just internal process. Teams running pre-deployment red-teaming or capability testing at AI companies and enterprises can benchmark their own evaluation design against these 10 recommendations. Readers comparing Anthropic's approach with OpenAI's preparedness-style evaluations will find this a useful primary source on how one frontier lab operationalises safety testing.

Event
Official accountAnthropic@AnthropicAISource
ProductsBuildersMulti-source03:43
Chinese contentGemini 4 Hydrogen 正式登陆 GitHub Copilot

写代码的朋友可以试试:Google 新模型 Gemini 4 Hydrogen 已经能在 GitHub Copilot 里直接用了,应用、CLI、VS Code 三端都有。它主打智能体编码和多步任务,官方说指令遵循比 Gemini 3.8 Flash 更强,正好拿来和 Claude、GPT 系列对比一下哪个顺手。

Event
GitHub@githubSource
ProductsBuildersMulti-source03:42
Chinese contentCodeRabbit 投入 100 万美元为开源项目免费提供 Deep Scan 安全扫描

开源项目常被自动化工具挖洞,维护者却只能下班后修。CodeRabbit 这次拿出 100 万美元给 1,125 个仓库免费做渗透级扫描,已找出 1,215 个高危问题并修复 75 个。自己维护开源库的人可以去申请,做安全工具的团队也该看看它的扫描思路。

Event
rohanpaul_ai@rohanpaul_aiSource

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