Why Chinese AI Lacks Its Own Terminology and How to Change That

Chinese Large‑Model Landscape

Chinese companies have released large language models such as DeepSeek, Zhipu (智谱), Qwen (千问) and MinMax. Their capabilities are comparable to foreign models, with a gap of roughly six months to a year.

Emerging Application‑Layer Terminology

Vibe Coding – coined by Andrej Karpathy in February 2025 on X. Describes a programming mindset guided by intuition or “feeling”.

MCP – “Model Connection Protocol”, an open standard announced by Anthropic in November 2024 that defines a unified interface for AI platforms to access external resources.

Agent Skills – an Anthropic open standard released December 2025. Provides a lightweight, modular description for composing tool‑use capabilities of AI agents.

Specification‑Driven Development (SDD) – traces to NASA engineering practices of the 1960s. In September 2025 GitHub released the open‑source Spec Kit which defines a Markdown‑based SPEC.md format, treating specifications as the “truth source” for AI agents.

Harness Engineering – introduced by Mitchell Hashimoto and later endorsed by OpenAI, Anthropic and thought leaders such as Martin Fowler. Refers to systematic engineering of AI “harnesses” that expose model capabilities as reusable services.

Fast‑Follower Dynamics in Chinese AI Tooling

Chinese developers often replicate newly released foreign AI tools within weeks:

After the release of Cursor , multiple domestic AI IDEs appeared.

Following Claude Code , Chinese products rushed to add command‑line interfaces.

When OpenClaw entered the market, several firms began building local variants to capture market share.

This rapid implementation emphasizes speed over original protocol design, leading to reliance on externally defined standards.

Implications of Relying on External Standards

When protocols and abstractions are defined abroad, Chinese engineering effort often reduces to “solving problems set by others”. The cost manifests as intensive overtime, 996 work schedules, and a focus on short‑term API‑call metrics rather than long‑term architectural innovation.

Path Toward Original Standards

To shift from follower to leader, the community should:

Allocate dedicated time for engineers to design and document new abstractions rather than only responding to weekly release pressures.

Encourage open‑source publication of protocols (e.g., a GitHub repository defining a new connection standard) so that they can be adopted globally.

Prioritize abstract‑layer design—such as defining a SPEC.md schema or a modular Agent Skills descriptor—over merely optimizing API throughput.

Given the depth of talent, the volume of open‑source contributions, and the near‑parity of model capabilities, Chinese teams have the technical capacity to originate world‑leading application‑layer standards.