工具观察

Grok Build

上周推了新版本,支持TUI (终端里鼠标点击)里调用Grok Imagine 做图做视频。然后比较有意思的两点:1、Plan mode,可以手动指定修改部分。2、大任务自动拆分:碰到那种前后端测试一起做的活儿,Grok Build 会自动开几个子代理并行处理。观察了一下,它确实会把前端、后端、测试拆开同时进行,比串行快不少。3、无头模式:把 Grok Build 塞进自动化脚本或者 CI/CD:

grok -p "分析这个代码库的安全问题" --output-format streaming-json

Musk说Gork Build接下来会利用上收购cursor获得之前程序员的编程习惯,迭代会很迅速。

Grok还是做起了开发工具这条路,反而让我有点担心接下来行业的发展:如果只有coding开发能收上来足够的费用来撑起来接下来模型发展,感觉天花板小了…

Anthropic

这次的Encyclical反响还蛮好,有人把这个和1888年的Revurm Novarum和方济各的Fratelli tutti来比。

微观一点的预计会有一批产品名会从这篇里面出

https://www.nytimes.com/2026/05/25/world/europe/pope-leo-encyclical.html?smid=nytcore-android-share

OpenAI

GPT-5.6信息泄露:OpenAI is internally testing GPT-5.6 (codenamed iris-alpha). Expected as soon as June, this flagship update reportedly features a massive 1.5 million token context window, specialized "Pro" agentic workflows, and a major leap in commercial-grade UI/front-end code generation. [1, 2, 3]

Gemini 3.5 Pro 预计6月出,Claude Opus 4.8预计也是6月,然后Claude Mythos Preview预计要开放个中小企业申请。

上周Hacker News上讨论度高的部分Vibe coding产品

nah

nah is a permissions guard built in pure Python with zero required dependencies that works out of the box. The main classifier maps tools deterministically into an intent taxonomy in milliseconds. An optional LLM resolves qualified ambiguous asks.