AI Briefing, March 25 — Issue #145
Editorial standards and source policy: Editorial standards, Team. Content links to primary sources; see Methodology.
## 🔍 Key Insights
The domestic AI music model **Kunlun Tech’s Mureka V8** has topped a globally authoritative leaderboard—securing #1 in *both* vocal and instrumental generation. **DeepSeek** has launched a large-scale hiring drive focused squarely on Agent development, signaling a clear strategic pivot. Meanwhile, **Google’s TurboQuant** and **Alibaba Cloud’s JVS Claw** have delivered critical advances—in low-level inference optimization and Agent tooling, respectively [3][1][13][14].
## 🚀 Highlights
- **Kunlun Tech’s Mureka V8 tops Artificial Analysis’ Music Model Leaderboard** [3]: Outperforms Suno and Udio in *both* vocal and instrumental generation—making it the first Chinese AI music model to claim global #1.
- **DeepSeek opens 17 Agent-focused roles—with strong preference for “Vibe Coding” talent** [1]: A definitive signal that its engineering focus has fully shifted toward Agent productization and real-world deployment.
- **Tencent’s WorkBuddy now supports WeChat-controlled “cyber assembly lines” in live testing** [4]: A full-spectrum AI agent workspace is live—dramatically lowering the barrier to automation and content creation.
- **Fliggy launches its first end-to-end travel Skill, “flyai”, open-sourced under the MCP protocol** [5]: Developers can plug in and use flight, hotel, visa, and other travel services instantly.
- **Figma officially opens its Canvas to AI Agents—enabling bidirectional sync with MCP and Skills** [10]: Design and code collaboration has entered a native multi-agent era.
- **Google releases TurboQuant: a lossless KV cache compression algorithm for LLMs** [13]: Delivers faster inference—key infrastructure for deploying large models on-device or at high concurrency.
- **Alibaba Cloud’s JVS Claw goes fully public—with new voice input and Skill toggle features** [14]: Its entry-tier subscription starts at just ¥39, significantly lowering the bar for enterprise-grade AI Agents.
- **Littlebird secures $11M in funding to build a context-aware assistant via “screen reading”** [22]: Tackles two core pain points head-on: clunky interaction and context forgetting.
## 🔗 Sources
[1] DeepSeek Is Hiring Aggressively for Agent Roles—17 Open Positions, “Vibe Coding” Talent Strongly Preferred — https://www.bestblogs.dev/article/8c787d05
[3] Chinese AI Music Just Quietly Claimed Global #1 — https://www.bestblogs.dev/article/92e76163
[4] Is WorkBuddy Going Wild? Tencent AI Experts Test It Live: Controlling a “Cyber Assembly Line” via WeChat — https://www.bestblogs.dev/article/d29f9504
[5] The First End-to-End Travel Skill Is Here! Fliggy’s “flyai” Launches on ClawHub & GitHub — https://www.bestblogs.dev/article/d674e826
[10] Figma Officially Opens Canvas to AI Agents—Supports MCP Protocol — https://www.bestblogs.dev/status/2036660040422101193
[13] Google Releases TurboQuant: An LLM KV Cache Compression Algorithm — https://www.bestblogs.dev/status/2036651972917616693
[14] Alibaba Cloud JVS Claw Goes Fully Public: Adds Voice Input & Skill Toggle, Lowers AI Agent Adoption Barrier — https://www.bestblogs.dev/article/c10710d7
[22] The Ultimate Form of AI Assistants? Why Investors Are Backing This “Screen-Reading” Startup — https://www.bestblogs.dev/article/6a8b2f1e