Cursor officially launched Design Mode, allowing users to update UI directly by pointing, drawing, or voice. @dotey subsequently released a community Skill called “baoyu-design” that runs Claude Design locally, best paired with Opus 4.8: describe the screen to generate exquisite HTML, click any element in the preview to modify it. He also wrote a dedicated tool to parse HAR files, decrypting the Protocol Buffers binary communication between Claude Design and the server for easy viewing of request prompts. Additionally, @dotey compared the design output of GPT-5.5 and Opus 4.8 using the same prompt to generate a Reader Mac App interface, concluding that “aesthetically, Claude Opus 4.8 > GPT 5; for design-related tasks, Opus is still the way to go.”<\/p>\n
Sources:<\/p>\n
@vista8 received an invitation to test Raycast’s new AI tool Glaze. In practice, with just a one-sentence requirement and one debug session, he built a music radio app based on the Spotify Web SDK in only 10 minutes. Glaze bundles development, logging, and publishing together with complete context, making development and publishing highly efficient. Its built-in Store supports one-click installation of existing apps, offering a near-native experience. However, he also noted that Glaze’s programming model is “somewhat unfriendly and a bit dumb,” even scolding developers. Another blogger tested Glaze and developed a Pandora-like music radio in 10 minutes, stating that “for developing Mac mini-tools, Glaze is enough from now on.”<\/p>\n
Sources:<\/p>\n
Several bloggers discussed the capability boundaries of different AI models. @vista8 shared a subjective front-end aesthetic ranking from a live discussion with Tom: Claude Opus 4.8 > Kimi 2.6 > GPT 5.5 > DeepSeek V4 Pro > GLM 5.1 > DeepSeek V4 Flash. @dotey believes ChatGPT’s DeepResearch is the best, while Gemini excels in search capabilities; he usually uses both and compares results. One blogger tested the Claude + Qwen 3.7-max combination, spending no more than 10 yuan, and personally felt the effect was better than GPT-5.5 and slightly below Opus 4.8. Overall, Opus 4.8 leads in design aesthetics, but GPT-5.5 is stronger in search and engineering tasks; different models should be chosen for different scenarios.<\/p>\n
Sources:<\/p>\n
A 70,000-character resignation letter allegedly from Alibaba DingTalk circulated on social media, sparking discussions about workplace culture. @MaxForAI wrote a long post criticizing the “identity vetting” phenomenon in the discussion\u2014many people, instead of discussing the content of the letter, questioned the author’s age, gender, experience, and platform for speaking. @imwsl90 shared observations of different types of bosses based on personal career experience, mentioning the overtime culture of bosses from Alibaba and Huawei backgrounds. @fankaishuoai summarized the management style of DingTalk’s former head, Wuzhao, as “unconditional obedience upward, unconditional exploitation downward.” These discussions represent personal opinions on social media; the specific content and authenticity of the original resignation letter have not been officially confirmed.<\/p>\n
Sources:<\/p>\n
Grok Build released two versions, v0.2.31 and v0.2.32, on June 7. @XFreeze posted the changelog for v0.2.31, and @elonmusk confirmed that “Grep timeout issue fixed in latest Grok Build.” Subsequently, @aksheyd reported a crash\/panic issue during web_fetch, which was fixed in the latest stable version 0.2.32. Grok Build also added .envrc loading functionality, which directly passes the user’s .envrc file into the Agent’s shell environment variables.<\/p>\n
Sources:<\/p>\n
@xiaohu cited a Financial Times report that OpenAI is preparing the largest overhaul of ChatGPT since its launch. The company plans to transform ChatGPT into a “super app” combining coding tools and AI agents, adding products that executives believe will generate more revenue. Currently, this information comes from a single media report; OpenAI has not officially confirmed specific overhaul plans.<\/p>\n
Sources:<\/p>\n