## 🔍 Core Insights **AI Agents** are moving from experimentation to large-scale deployment: Apple is reportedly accelerating its push to catch up on autonomous agents, while **Anthropic** disclosed through its IPO that Claude already runs across all four major chip families, making compute scheduling capability a new moat [2][3]. Meanwhile, **Nvidia** plans to acquire diffusion-based large language model company Inception for over $500 million, and **Qualcomm** has partnered with Cirrascale to open evaluation of rack-scale inference accelerators, as the underlying compute race continues to heat up [13][16]. ## 🚀 Key Developments - **Apple develops "take over your computer" AI agent** [2]: It can write code, browse the web, and control applications, while Apple tightens third-party agent permissions on Mac. - **Anthropic rises 17% on its first trading day** [2]: Its IPO video revealed that Claude's training and inference already run on all four major chip families, with AMD GPU integration taking only 3 days [3]. - **Nvidia plans to acquire Inception for over $500 million** [13]: The two sides are in advanced talks, with a valuation of at least over $500 million, possibly with performance-linked bonuses. - **PetroChina Shared Services deploys 18 agents into a 10,000-person back office** [4]: In partnership with Volcano Engine, it built three types of agents—knowledge, data, and process—covering six major scenarios including finance and legal, with answer accuracy exceeding 90%. - **936 AI agents conduct autonomous research for three months** [6]: MIT, Stanford, and others jointly ran an alphaXiv experiment where agents independently completed paper writing, code debugging, and experimental validation, with AI judges conducting blind review. - **Qualcomm opens cloud evaluation of Dragonfly AI200** [16]: Its first rack-scale solution, with 768GB memory per card and 43TB total memory in a 56-card rack. - **Tencent launches dedicated email for AI agents** [8]: Quietly launched, allowing users to assign dedicated email addresses to AI agents. - **Kling 4.0 showcases 4K 10-bit HDR** [11]: Video generation capabilities upgraded again, with actual test results made public. ## 🔗 Sources [1] Hands-on with the new Mercedes E-Class infotainment system! Response speed surpasses new EV makers, but it lacks a bit of the "Mercedes feel" — https://www.bestblogs.dev/article/5d5f2630e1?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item [2] Morning Brief | Apple hires to build a "take over your computer" AI agent / The largest IPO in history, Anthropic goes public / Huawei officially announces Mate 90 Air focused on thinness and lightness — https://www.bestblogs.dev/article/deea355a79?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item [3] Anthropic IPO video says Claude training and inference already run on all four major chip families, AMD GPU integration took only 3 days — https://aihot.news/items/wxnv1ek19ha6vdwcsbqsv3oic [4] PetroChina Shared Services and Volcano Engine jointly build three types of agents—knowledge, data, and process—with 18 agents deployed into a 10,000-person back office — https://aihot.news/items/a7vjvgdl96onwm7qgtlnvtdc4 [5] Google engineer uses PostScript to run Gemma 3 270M on a printer, 14.1MB memory, about 60 seconds per token — https://aihot.news/items/bnydt76bzk61kr5c82ifgt0d6 [6] AI agents conduct autonomous research on alphaXiv for three months: over 900 papers, AI judges