East Asian Technology Intelligence
Japan & China tech news — translated, contextualized, and delivered for Western readers.
Free. Unsubscribe anytime.
This story ran in Issue #100, alongside three other stories. Anthropic recently shared that it uses 30,000 internal AI agents. These agents now handle over a quarter of the company’s own research and development work. This change shows that AI is no longer just a tool for engineers. Instead, AI is becoming an engineering team that works alongside human developers.
The company scaled its core R&D autonomy from 1% to 26% in just six months. Anthropic calls this stage Agentic Level 4 autonomy. The system does not just write code. It can now break down tasks, plan projects, and run processes in parallel. This growth represents a structural change in how companies build advanced AI.
Chinese tech outlets like QbitAI focused heavily on the design of Claude Code. They analyzed its multi-agent system and the Coordinator tool that manages Git branches and shared memory. This coverage shows a deep interest in practical engineering and deployment. Chinese media prioritized these concrete systems over flashy statistics or abstract capabilities. For these observers, the real story is how Anthropic built the system.
This internal work reflects the Japanese philosophy of monozukuri, which means the relentless pursuit of perfection in making things. Anthropic is applying this manufacturing mindset directly to software. The company has built an assembly line where AI agents work to create better AI models. Western media often ignores this systematic process because it prefers to cover major breakthroughs in model performance.
Yet, we cannot assume that this internal process will lead to better products for customers. The danger is that AI agents might optimize only for metrics that other AIs can easily measure. This focus could create models that lack the intuitive understanding or common sense that human users value. Anthropic must manage this feedback loop with great care.
To see if this method works, we should watch Anthropic’s next major model release. We must check its performance on qualitative tests, not just quantitative benchmarks. We should also see if other developers adopt the Coordinator system. Finally, we must watch for similar data from other major AI labs like Baidu and Alibaba to see if this setup becomes the new industry standard.
This story appeared in AsiaAI.FYI Issue #100. Get this in your inbox each week — subscribe free.