Why Chinese AI Lags in Terminology and Standards: A Candid Conversation

During an evening conversation with an AI student studying abroad, the author highlighted China's rapid progress in large models such as DeepSeek, Zhipu, Qwen, and MinMax, noting that the gap with foreign products is only a few months.

The student countered that Chinese AI excels in models but falls short on the application layer, prompting the author to argue that the world's largest use cases, most complex user demands, and the most intense engineering execution have always been Chinese strengths since the internet era.

The student then asked why Chinese programmers have never coined a notable term like "Vibe Coding," "MCP," "SDD," or "Harness Engineering." He listed several emerging concepts:

Vibe Coding : coined by Andrej Karpathy in February 2025, meaning “programming by feeling.”

MCP : a connection standard introduced by Anthropic in November 2024 to unify AI platform resource access.

Agent Skills : an open standard released by Anthropic in December 2025 for lightweight, modular AI tool composition.

SDD (Specification‑Driven Development) : a practice dating back to 1960s NASA, revived by GitHub’s September 2025 open‑source Spec Kit that defines a Markdown‑based SPEC.md as the truth source for AI agents.

Harness Engineering : a concept originally proposed by Mitchell Hashimoto and later endorsed by OpenAI, Anthropic, and Martin Fowler, gaining traction in Chinese communities.

The student argued that these protocols and designs are not mystical or hardware‑intensive, yet they have not originated in China, suggesting a systemic tendency to follow rather than lead.

The author reflected that Chinese engineers often prioritize speed and immediate delivery, resulting in a culture of “chasing” foreign standards instead of creating original ones. This mindset leads to overwork, 996 schedules, and a reliance on copying successful foreign ideas.

Despite the challenges, the author believes Chinese talent has the capability to innovate in application‑layer standards and protocols if engineers are given time to abstract, design, and publish new concepts rather than constantly sprinting to meet weekly API‑call targets.

Ultimately, the piece calls for a shift from being a fast follower to an original leader in AI terminology and standards, encouraging the community to pause, summarize insights, and create proprietary concepts that the global AI community will need to learn.