Chinese Tech Daily — 2026-06-08#

Top Story#

Apple’s WWDC26 marked a massive pivot in its artificial intelligence strategy with the unveiling of “Apple Intelligence” and a completely rebuilt Siri AI. By combining on-device models with Google’s Gemini technology for server-side processing, Apple aims to make the iPhone a central dispatcher for users’ digital lives. However, due to domestic regulatory requirements, Chinese users will face delays as these flagship AI features will not be available in the initial regional rollout.

Engineering & Dev#

Agent workloads push infrastructure to the limit: GitLab announced layoffs affecting 14% of its staff (about 350 employees) to reinvest in its infrastructure, specifically citing that AI agent workloads are straining their existing Git architecture. The company has initiated a next-generation refactoring of Git to support a 100x scale increase, preparing for a future where autonomous agents consume resources at machine scale.

Xiaohongshu deploys “Seal” enterprise AI agent: Social media giant Xiaohongshu successfully rolled out its internal AI assistant, “Seal,” to all employees using the OpenClaw framework. To handle the engineering challenges of managing tens of thousands of instances, the team implemented a lossless context pruning mechanism inspired by JVM garbage collection, successfully reducing token inference costs by 50%.

Shopify boosts GraphQL execution speed by 15x: Shopify engineering revealed “GraphQL Cardinal,” a redesigned execution engine that replaces traditional depth-first traversal with a breadth-first execution model. By coordinating resolver execution per query layer, the new approach significantly reduces redundant computations and garbage collection overhead in heavily nested e-commerce queries.

Tencent revamps internal AI token quotas: A leaked internal memo reveals that Tencent is shifting from per-capita AI token distribution to an output-based dynamic allocation system. The company will prioritize token quotas for employees who can demonstrate clear productivity gains, signaling a more rational, efficiency-driven approach to internal AI infrastructure costs.

LLMs are turning into Large Coding Models: An analysis of Arena AI benchmark data suggests that tech giants are disproportionately investing resources into improving the coding capabilities of foundation models rather than their general language skills. This trend is largely driven by the clear return on investment that AI coding assistants provide in enterprise SaaS environments.

Products & Digital#

Kimi Work launches as a local agent swarm: Kimi has introduced “Kimi Work,” a desktop-based general-purpose AI assistant that can coordinate up to 300 sub-agents to handle complex workflows. The tool features deep local file integration, automated browser control via WebBridge, and built-in access to global financial data to autonomously generate reports and presentations.

Universal Installer tackles Android sideloading: As Android makes sideloading increasingly difficult, a new Shizuku-based tool called Universal Installer is gaining traction. It elegantly handles complex package formats like APKS and OBB data, bypasses strict system prompts, and even integrates the VirusTotal API to check file hashes before installation.

Huawei’s Maextro V800 interior leaks: Leaked details of Huawei’s upcoming Maextro V800 luxury MPV showcase an extreme focus on high-end executive comfort. The vehicle, which weighs over 3.1 tons, features factory-standard crystal wall lamps and a built-in machine capable of serving freshly ground coffee and hot tea to second-row passengers.

Xiaomi 17T series officially released: Xiaomi launched the 17T and 17T Pro models in China, positioning them as “all-around imaging flagships”. The devices boast massive 7000mAh batteries, MediaTek Dimensity processors, and Leica optical triple-camera setups.

News & Commentary#

China imposes security reviews on overseas investments: Beijing has issued new State Council regulations requiring Chinese companies to undergo national security reviews before expanding their operations and investments abroad. Legal and trade experts interpret this as a strategic move to construct an economic fortress, ensuring that capital, top talent, and critical intellectual property remain within China’s borders amidst decoupling pressures from the West.

Reflections on the Silicon Valley Chinese AI community: Following a major AI meet-up in Silicon Valley, a Chinese tech executive observed that the biggest hurdle for Chinese AI startups in the US isn’t product quality, but a lack of established relationships and networking savvy. In the AI era, where machines pre-screen backgrounds before business meetings even happen, building a verifiable public footprint and earning trust have become critical prerequisites for closing deals.

Notion briefly disables Anthropic models over “degraded performance”: An official status update from Notion announcing a temporary ban on Anthropic’s Claude Opus 4.7 and 4.8 models caused a stir in the AI community, as users feared the models had silently lost their reasoning capabilities. Twelve hours later, Notion executives clarified that the issue was merely a brief infrastructure outage and not a fundamental downgrade in the AI’s intelligence.

Also Noted#

• Trailing slash bypasses AWS API Gateway auth: Security researchers discovered that simply appending a slash to an AWS HTTP API route can completely bypass Lambda authorizer authentication.
• Ant International launches Agentic Mobile Protocol (AMP): The new framework aims to standardize secure mobile payments and identity verification for autonomous AI agents making purchases globally.
• ClickHouse challenges Elasticsearch in log analytics: ClickHouse has integrated inverted indexes for full-text search, allowing it to perform log analysis up to 6 times faster than Elasticsearch at massive scales.
• F5 introduces Token-Based Load Balancing (TBLB): F5 is shifting its load balancing strategy for AI inferencing to route traffic dynamically based on token cost, model capability, and KV cache status.