A customer-service agent built by one vendor cannot talk to a logistics agent built by another. A hospital bot and a city-governance bot speak different dialects. Multiply that by the thousands of AI agents (智能体) now entering Chinese workplaces, and you get a country full of brilliant but lonely machines — each useful on its own, none able to hand a task to its neighbour.
That isolation is no longer just an engineering nuisance. As agents move from answering questions to booking flights, screening medical scans and running factory shifts, the inability to interoperate has become a tax on the entire "artificial intelligence plus" (人工智能+) agenda.
Key takeaways
- China published GB/Z 185-2026, its first national standard system for agent interconnection (智能体互联), covering identity, description, discovery, interaction and tool use.
- The series spans seven parts released on 22 May 2026 and currently in force, built as guiding technical documents (国家标准化指导性技术文件) rather than mandatory rules.
- Every agent gets a unique "digital ID" (数字身份证) through an identity-code scheme, so cross-platform bots can authenticate before they collaborate.
- More than 70 research, industry and user organizations, led by the China Electronics Standardization Institute (中国电子技术标准化研究院), drafted the texts.
- Regional rollouts are already underway in Beijing, Shanghai and Shandong, with travel, government and telecom agents joining pilot zones.
- The framers explicitly aim to feed Chinese practice into ISO/IEC/ITU tracks, positioning the standard as a candidate global "common language."
The problem: a country of agent islands
For two years, Chinese companies have shipped agents at speed. But each vendor defined its own interfaces, identity scheme and message format. The result, in the words of standards officials, was "信息孤岛" — information silos where a banking agent and an e-commerce agent could not confirm who the other was, let alone divide a task. Enterprises paid the price in custom integrations and duplicated security work.
What the seven parts actually cover
The GB/Z 185-2026 series is structured as a closed loop. Part 1 sets the overall architecture (总体架构). Parts 2 and 3 handle the identity code (身份码) and identity management (身份管理) — the trust layer that issues each agent a digital ID. Part 4 is agent description (智能体描述), a capability profile answering "what can this agent do." Part 5 covers discovery (智能体发现), the matchmaking mechanism for "how do I find the right agent." Parts 6 and 7 define interaction (智能体交互) and tool invocation (智能体工具调用), the rules for cooperative work.
Together they let a logistics agent, a customer-service agent and a payment agent recognise, verify and delegate to one another without a bespoke bridge for every pairing. Standards officials describe the goal plainly: agents that are "连得上、信得过、管得住、可追溯" — connectable, trustworthy, controllable and traceable.
A digital ID, not just a label
The identity-code part is the linchpin. It adopts a layered OID (object identifier) scheme in which one code maps to exactly one agent, and two agents must complete mutual authentication before exchanging data. That turns identity from a nice-to-have into the first gate of any interaction — the foundation for audit, access control and accountability when something goes wrong. For a technology whose failures are easy to hide inside opaque models, a verifiable identity is a meaningful step toward governance by design.
From paper to pilot
Publication was only the start. In June and July 2026, promotion conferences in Beijing's Haidian district moved the text into practice; Haidian, home to many AI-native firms, volunteered as a validation ground. In late July, Shandong Mobile was named among the first batch of identity-code nodes (身份码节点), having already run a live pilot of 26 agents across government, office, medical and marine scenarios. In late August, Shanghai's Zhangjiang zone issued its first identity-code nodes to about 30 organisations including a speech-AI firm, Lenovo, ZTE and Volcano Engine. Travel platform Qunar signed on as the first online-travel adopter to trial the protocol.
Why this reaches past China
China's market regulator is running a broader "hundred national AI standards" push — 146 AI projects spanning computing power, large models (大模型), embodied intelligence and agents. Officials say they want Chinese proposals reflected in international bodies such as ISO, IEC and ITU. If the agent-interconnection protocol (AIP) becomes a reference elsewhere, it would give Chinese platform and chip makers a head start in writing the rules of multi-agent economies — much as earlier cellular and payment standards did.
Honest limitations
This article rests on official releases from the Standardization Administration and state-linked media; I have not inspected the full technical text of each part. The standards are guiding documents (国家标准化指导性技术文件), not binding law, so adoption depends on market incentives and procurement. Reported participant counts (70-plus organisations) come from organiser statements and were not independently audited. The global-influence claim is a stated intent, not a settled outcome. I have not assessed how the identity scheme interacts with existing personal-information and cross-border data rules.
What readers can do now
- If you build or buy agents, ask vendors whether their products conform to GB/Z 185-2026 and support the agent interconnection protocol (AIP) before committing to a custom integration.
- If you run a platform, pilot an identity-code node (身份码节点) so your agents can be discovered and trusted by external partners.
- Track the standards body's "pilot plan" (先锋计划) and open-source tooling; conformance today is cheap insurance against rework when adoption widens.