WeChat Open Platform released the ‘Guidelines for Developers to Access the WeChat AI Ecosystem,’ guiding mini-program developers to integrate their mini-program capabilities into WeChat AI. This marks a crucial step for WeChat AI to evolve from a conversational tool to a platform. After integration, mini-programs can be recommended and invoked by WeChat AI; those not integrated cannot be invoked. The platform offers two modes: Auto mode, where the platform automatically analyzes the mini-program source code and allows AI to operate directly; and Development mode, where developers voluntarily open features, which after platform evaluation and review, can be invoked by AI. Currently, WeChat AI is still in beta testing, and users cannot experience it yet. Meituan has officially announced its integration with WeChat Agent. According to @MaxForAI, Meituan was invited to beta test as early as early 2025 and had developed a product called ‘Kangaroo Bao Bao,’ but it could not launch due to insufficient model capabilities at the time. Many bloggers highly praised this move, considering it the opening of China’s largest GEO traffic entry point to developers. However, @cellinlab raised a question: Will mini-program entities similarly refuse to be invoked by AI, just as WeChat refuses other Agents to invoke WeChat?<\/p>\n
Sources:<\/p>\n
Notion directly called out Anthropic’s Opus 4.7 and 4.8 models on its official status page for performance degradation, causing a significant increase in failure rates for users using these models in Notion AI. Notion has removed all Anthropic models from the model selector and redirected requests. @op7418 further shared screenshots comparing the status pages of OpenAI, Anthropic, and Google, pointing out that Anthropic couldn’t even maintain 99% availability. This incident reflects that Anthropic’s stability issues in enterprise application scenarios are becoming a real pain point; no matter how powerful the model, insufficient reliability can lead to direct abandonment by downstream products.<\/p>\n
Sources:<\/p>\n
According to the Financial Times, OpenAI is preparing the biggest overhaul of ChatGPT since its launch in 2022, with internal executives bluntly stating ‘Chat is dead.’ The overhaul direction is to integrate the coding tool Codex, multi-step AI Agent, image generation, and third-party applications (known to include Canva and Booking) into a unified interface, creating a cross-platform personal AI assistant. The overhaul is driven by clear commercial motives: ChatGPT currently has 900 million weekly active users and monthly revenue of $2 billion, but most users use the free version, and the company has yet to become profitable, with enterprise customers contributing about 40% of revenue. OpenAI completed a $122 billion funding round in March this year, with a valuation of $852 billion, and is preparing for an IPO. Meanwhile, Anthropic has submitted a confidential S-1 filing to the SEC, with a valuation of $965 billion, surpassing OpenAI by valuation; Google Gemini also reached 900 million monthly active users. @dotey’s long-form analysis points out that this overhaul is both a product upgrade and a new story to tell investors.<\/p>\n
Sources:<\/p>\n
@dotey shared a complete Claude Design development workflow: first use Claude Design to generate a design draft in HTML + CSS + React + data.js format, then hand it over to Claude Opus 4.8 to implement an MVP, iterating through small versions to gradually improve. He believes React is more suitable for Agent-driven development than pure HTML because React naturally forms a structured tree structure, making it easier for Coding Agents to understand and modify. Due to the poor experience of Claude Desktop, @dotey reverse-engineered the core logic of Claude Design and combined it with Cursor’s web markup feature to create a locally runnable Skill. He also emphasized the importance of Agent self-verification capability for long-running tasks: long runs without clear error feedback may just waste tokens.<\/p>\n
Sources:<\/p>\n
Kimi officially launched Kimi Work, an AI Agent product running on the local desktop, supporting up to 300 sub-Agents working in parallel. Key features include: web automation operations (search, scroll, click, input) via browser extension, built-in calls to Yahoo Finance and World Bank global market data, and a memory system to record user preferences and historical decisions. Supports macOS (Apple Silicon) and Windows. Meanwhile, @MaxForAI noted that the Kimi Developers account gained 40,000 followers in three days, commenting ‘How you treat developers is how developers treat you,’ implying that Kimi’s investment in developer relations is paying off.<\/p>\n
Sources:<\/p>\n
ByteDance was reported to have issued a press release declaring that Gu Quanquan (@QuanquanGu) is not the head of the AI4S pharmaceutical direction. Subsequently, Gu Quanquan was forced to post a statement on Xiaohongshu to prove her identity, claiming she is indeed the pharmaceutical head. @MaxForAI commented that this confirms previous rumors of internal conflict in the AI4S direction were not unfounded. This incident reflects that organizational structure and personnel battles in the emerging AI for Science direction are intensifying at big companies.<\/p>\n
Sources:<\/p>\n
The National Security Ministry published an article under the pseudonym ‘Anping’ issuing a security warning about ‘AI relay station’ type tools. The article stated that the Central Cyberspace Affairs Commission is deploying a nationwide special action called ‘Qinglang \u00b7 Rectifying AI Application Chaos,’ advising users to choose legitimate platforms, desensitize sensitive data, and promptly address abnormal charges or data leaks. @MaxForAI reposted this article and hinted that it is related to the current phenomenon of many users accessing overseas AI services through relay stations. This signals regulatory attention to AI relay\/proxy services.<\/p>\n
Sources:<\/p>\n
Multiple bloggers showcased practical results of AI video tools. @Chengzilhy shared a complete case of producing a China high-speed rail FPV flying video for a client: referencing a red line path map to write detailed flight prompts, flying from valleys, viaducts, panda bamboo forests, and into the interior of the high-speed train carriage. The client’s leadership was satisfied with the effect. @xiaohu open-sourced a video translation tool he had used for half a year, supporting a one-command pipeline of download, transcription, translation, polishing, and burning, automatically adding any subtitles to any foreign language video. These cases show that AI video tools have moved from ‘showing off skills’ to ‘delivering business results.’<\/p>\n
Sources:<\/p>\n
@xiaohu cited data indicating that as of May 2026, over 80% of the code merged into Anthropic’s codebase was written by Claude, while before the release of Claude Code in February 2025, the ratio was only single digits. This data outlines the recursive evolution path of an AI company ‘using its own AI to build itself,’ and also explains from the side why Anthropic has been able to quickly catch up with OpenAI in valuation.<\/p>\n
Sources:<\/p>\n
@Gorden_Sun released a new AI PPT skill polished for two weeks: first use GPT to generate a luxury PPT in image format, then convert it to a fully editable PPTX file, including three independently usable sub-skills. @AlchainHust announced that he has rewritten Huashu Design from scratch as version 2.0, which has gained 16k+ stars in two months of open source. The new version has significantly improved visual style diversity and material accuracy. AI-generated PPTs are moving from ‘usable’ to ‘good to use,’ with the tool chain’s maturity rapidly improving.<\/p>\n
Sources:<\/p>\n
Statistics: Number of timeline scans=360, Number of bloggers hit=42, Total tweets hit=192, Weighted tweet score=163.35, Original tweets=111, RT tweets=27, Fetch attempts=2, Boundary coverage status=tail_confidently_crossed_target_boundary<\/p>\n","protected":false},"excerpt":{"rendered":"
WeChat AI takes a key step toward platformization, OpenAI integrates tools for profitability as a personal assistant, and Anthropic’s model stability issues prompt Notion to disable them, raising concerns.<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[5],"tags":[19],"class_list":["post-1337","post","type-post","status-publish","format-standard","hentry","category-brief","tag-x--ai-"],"_links":{"self":[{"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/posts\/1337","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/comments?post=1337"}],"version-history":[{"count":0,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/posts\/1337\/revisions"}],"wp:attachment":[{"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/media?parent=1337"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/categories?post=1337"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/en.blog.liu-qi.cn\/index.php\/wp-json\/wp\/v2\/tags?post=1337"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}