1. OpenAI’s Core Chip Engineer Clive Chan Defects to Anthropic, Just Before Mass Production
Clive Chan, the second hardware employee at OpenAI, announced his departure on X and revealed he joined Anthropic this week. Chan joined OpenAI in January 2024 and was one of the earliest technical leaders of the company’s in-house chip project, spearheading the collaboration with Broadcom on a 10GW custom AI accelerator (TSMC 3nm process). Last October, he revealed the chip would enter mass production in 9 months — that timeline is essentially now. Chan cited “talent, values, ambition” as his reasons for joining Anthropic. Reuters had previously reported that Anthropic has expressed interest in building its own chips but remains in early stages; Chan’s arrival is likely to accelerate those plans.
AI Pulse View: The talent war is intensifying. OpenAI’s in-house chip took 30 months from design to mass production, and its core engineer is leaving for a direct competitor right at the finish line — signaling that AI chip self-development is becoming the new battleground among top players. If Anthropic can rapidly build a chip team with this hire, it will take a critical step toward compute independence.
Source: QbitAI
2. Ant Group Launches Overseas AI Payment Solution for Global Agentic Commerce
Ant International officially launched the Agentic Mobile Protocol (AMP), targeting global e-wallets, super apps, and digital banking services. AMP addresses the core challenges of payment convenience, security, and cross-market interoperability for AI agent commerce. The protocol integrates five capabilities: Agent Identity, Authorization, Payment, Settlement, and Trust — along with a KYA (Know Your Agent) certification system and AgentSafePay fund protection mechanism. Ant International connects 2 billion consumer accounts and 150 million merchants across 200+ countries and regions. Industry forecasts project the global agentic commerce market to grow from $5.7 billion in 2025 to approximately $28 billion by 2030.
AI Pulse View: The explosion of agentic commerce requires not just capable models but trustworthy payment infrastructure. AMP extends Ant’s mature “dare to pay, dare to compensate” guarantee to agent transactions, filling a critical gap in global agentic commerce standards for the mobile payments ecosystem — providing AI companies going global with a full-stack solution from transactions to trust.
Source: QbitAI
3. Amap Releases ABot-Earth0.5: The World’s First 3D-Native City World Model
Amap (Alibaba Group) released ABot-Earth0.5, the world’s first fully 3D data-trained, production-ready 3D-native urban world model. Unlike traditional 3D city modeling that follows a “collect first, fit later” approach, ABot-Earth0.5 trains directly on 3D data, building native understanding of three-dimensional space and generating 3DGS-format urban scenes end-to-end. Users can generate a 3D city from a single satellite image or text prompt on a consumer-grade single GPU, with efficiency approximately 1000x faster than traditional methods. The model pioneers a compression-generation framework directly targeting 3DGS point clouds and introduces a sliding-window inference mechanism for kilometer-scale continuous urban generation. Currently open for beta testing.
AI Pulse View: 3D-native world models are core infrastructure for embodied AI and spatial computing. By bypassing the 2D distillation path and training directly on 3D data, ABot-Earth0.5 dramatically lowers the barrier to 3D city scene generation, offering a cost-effective solution for autonomous driving simulation, digital twins, and metaverse applications.
Source: QbitAI
4. 2026 Forbes Self-Made Women Billionaires: Anthropic President Surges to #2
Forbes released its 2026 list of America’s self-made women billionaires, with Anthropic President and co-founder Daniela Amodei jumping to #2 with a net worth of $15.5 billion — a nearly 13x increase from $1.2 billion a year ago. This surge follows Anthropic’s recent mega-fundraising round of $65 billion at a $965 billion post-money valuation. This year’s list features 43 self-made women billionaires, a record high, with AI significantly boosting the wealth of many honorees.
AI Pulse View: Anthropic’s near-trillion-dollar valuation and its co-founders’ exponential wealth growth reflect the capital frenzy in the AI industry. Amodei, who left OpenAI with her brother Dario and five others in 2020 to found Anthropic, has now become one of the most commercially valuable figures in the AI safety赛道.
Source: 36Kr
5. New Paradigm for Agent Self-Evolution: Sun’s Team Proposes OpenSkill, Achieving SOTA on Multiple Benchmarks
Researchers at Lehigh University led by Assistant Professor Lichao Sun proposed the OpenSkill framework, enabling AI agents to acquire executable, transferable skills without relying on target task supervision signals. OpenSkill operates in three stages: open-world knowledge acquisition, non-leaking skill evolution, and zero-shot target evaluation. Results show OpenSkill achieves state-of-the-art automated performance across multiple benchmarks, and learned skills can be directly transferred to weaker models. Resources are publicly available on GitHub.
AI Pulse View: The core bottleneck in agent self-evolution has been the dependency on high-quality feedback signals. OpenSkill bypasses this limitation through an “unsupervised + virtual testing” paradigm, allowing agents to continuously accumulate experience and iterate in real deployment scenarios. This approach has practical significance for lowering the barrier to agent deployment.
Source: 36Kr
6. VentureBeat: Agentic AI Solved Coding — and Exposed Every Other Problem in Software Engineering
VentureBeat analysis points out that as agentic AI becomes central to engineering workflows and drives an explosion in code generation, enterprise leaders face a key question: If we’re shipping code faster than ever, why aren’t our products improving at the same rate? The article argues that writing code was never the rate limiter — defining the right requirements, integrating with complex systems, and maintaining software under real-world conditions have always been the hard parts. When agents flood organizations with massive amounts of code, the hard parts only get harder.
AI Pulse View: This analysis reveals an overlooked truth: the improvement in AI coding capability is just step one. Requirements definition, system architecture, testing validation, and operations management are the deep waters of software engineering. Enterprises need to rethink their engineering governance frameworks rather than simply assuming “more code = better products.”
Source: VentureBeat
7. VentureBeat: When Claude Changed, Everything Changed — Managing AI’s “Blast Radius” in Production
VentureBeat shares a real-world production war story: a team’s data reporting system built on Claude Sonnet 3.5 experienced widespread failures after upgrading to Sonnet 4.5 — the model began folding request parameters into the description field, causing API calls to lose filter conditions, and for the first time started asking clarifying questions instead of returning structured output. The team was forced to roll back to 4.0, but the rollback cost was enormous since they had already built new integrations against 4.5 in the interim. The article introduces the concept of “infinite blast radius”: the downstream effects of an LLM model upgrade cannot be enumerated in advance because both the input space (natural language) and failure modes (anything the model might do differently) are unbounded.
AI Pulse View: A classic lesson in LLM engineering practice. Models are not deterministic components, and version upgrades cannot be treated like traditional library upgrades. Enterprises need to establish “model change management” mechanisms — including strict regression testing, canary releases, and rapid rollback capabilities — treating LLMs as unpredictable external dependencies rather than reliable infrastructure.
Source: VentureBeat
8. AI Entrepreneurs Assemble! “2026 New Generation AI (Shenzhen) Innovation Competition” Officially Launches
The 2026 New Generation AI (Shenzhen) Innovation and Entrepreneurship Competition has officially launched, opening registration for AI entrepreneurs. The competition focuses on AI technological innovation and industrial applications, aiming to discover and cultivate promising AI startups and promote AI technology adoption in Shenzhen and the Greater Bay Area.
AI Pulse View: Shenzhen, known as a hardware and manufacturing hub, is accelerating its transition into the AI industry. Such competitions provide AI entrepreneurs with channels to connect with capital, technology, and markets — and serve as a window into the trends of China’s AI startup ecosystem.
Source: QbitAI
Other Updates
- Qualcomm × GAC Aion × WeRide: GAC Aion N60 autonomous driving competition takes runner-up; WeRide WRD 3.0 end-to-end solution debuts at Qualcomm summit QbitAI
- Analog Computing Breakthrough: Chinese team proposes new architecture — “let matrices go analog, let logic go digital” — redefining computing paradigms QbitAI