On July 21, at the Zhongguancun Exhibition Center in Beijing, a seemingly ordinary standardization conference accomplished something landmark: the first batch of AI agent identity codes was officially issued.
Over 200 enterprises applied for and received their dedicated nodes, including Alibaba's 1688, Meituan, DiDi, Xiaomi, China Southern Power Grid, and the three major telecom operators. Every AI agent now has its own "digital ID card."
What does this mean? It means the "wild growth" era of AI Agents from the past two years has officially entered a new phase—one that is registrable, traceable, and auditable. For the enterprise BI space, this is not news you can afford to ignore.
From "Usable" to "Manageable and Controllable": An Overlooked Gap
Over the past 18 months, nearly every BI vendor has been telling the same story: ask data questions in natural language, let AI write SQL for you, perform attribution, and generate reports. The "usability" of Agent BI is no longer the question.
But what enterprise CIOs and CDOs worry about has never been "can we use it"—it's three sharper questions:
- Who is calling? When an agent accesses your core business data, is its identity verifiable? Which vendor, which version, what permission boundaries?
- How do they coordinate? An enterprise might simultaneously run a data analysis agent, an approval agent, and a customer service agent. How do they "talk" to each other? With MCP, A2A, and private solutions all coexisting, unaligned interfaces make coordination empty talk.
- Who is responsible when things go wrong? When an agent automatically makes a business decision—say, adjusting the marketing budget allocation for a region—is the behavior auditable and traceable?
These three questions are exactly what the new national standard attempts to answer.
GB/Z 185: A Seven-Layer Closed Loop for Agent "Registration"
The Artificial Intelligence - Agent Interconnection (GB/Z 185.1–185.7—2026) is China's first standard system covering the full lifecycle of AI agents. Seven parts form a closed loop:
Overall Architecture → Identity Identification → Trust Management → Capability Description → Intelligent Discovery → Multi-Agent Interaction → Tool Invocation
The core logic is clear: first assign identity (who are you) → then describe capabilities (what can you do) → then permit discovery and interaction (how others find you and collaborate) → finally regulate tool invocation (how you operate external systems).
The identity code adopts a layered identification system, similar to internet domain names—enterprises choose dedicated abbreviations as prefixes, one code per entity, ensuring uniqueness and traceability from the source.
The companion AIP (Agent Interconnection Protocol) V2.1 addresses six key issues: trusted access, identity authentication, capability discovery, interconnection collaboration, settlement transactions, and behavior auditing. The source code has been open-sourced in the AtomGit community.
What This Means for Agent BI
Back to the enterprise data analysis scenario. A typical Agent BI workflow looks like this:
Business user asks a question in natural language → Data analysis agent understands intent → Calls data sources/APIs → Executes analysis logic → Returns visualization or recommended actions
Along this chain, the national standard brings three direct changes:
First, trusted identity. Agents are no longer "black boxes." Through the identity code system, enterprises can clearly know: this analysis agent currently accessing my CRM data—which platform does it belong to, what version, what security certifications has it passed. This is a prerequisite for data security compliance.
Second, composable capabilities. When agent capability description and discovery mechanisms are standardized, multiple agents within an enterprise can truly collaborate—the data analysis agent detects an anomaly in a report, automatically triggers a root-cause analysis agent to dig deeper, and then links with a notification agent to push conclusions to relevant decision-makers. From "single-point intelligence" to "chain intelligence."
Third, auditable behavior. Standardization of the tool invocation layer means every step of an agent's operation—which tables it queried, which models it used, what judgments it made—leaves a trace. This is especially critical for heavily regulated industries like finance, healthcare, and government.
Quick BI's "Manageable and Controllable" Practice
In fact, Quick BI has already made systematic progress in the manageable and controllable direction for Agent BI.
Intelligent Q (Smart Q) aligns closely with the national standard's philosophy: behind the natural language querying capability, every analysis request goes through explicit identity authentication, permission control, and operation logging. What you asked, what reasoning the AI performed in the middle, and what conclusion it ultimately returned—the entire chain is traceable.
In multi-agent collaboration, Quick BI's Skill-based architecture has been exploring the "agent registration - capability declaration - on-demand invocation" pattern: a data insight Skill can be called by agents from other business systems through standard interfaces, rather than being tightly coupled.
Having been selected for the Gartner ABI (Analytics and Business Intelligence) Magic Quadrant for seven consecutive years—as the only Chinese BI vendor to achieve this—this position rests not only on analytics capabilities but on a deep understanding of security, compliance, and controllability in enterprise scenarios.
From "able to ask data" to "able to govern it," from "single-agent answers" to "multi-agent trusted collaboration"—this is the critical step where Agent BI moves from demo-grade to enterprise-grade.
What Happens Next
The standard is currently published as a "guiding technical document," representing an agile standardization arrangement during the industry cultivation phase. But the signal is already clear:
Beijing is the launch site, with plans to cover 120 key cities nationwide. Over 90 primary and secondary schools in Beijing's Xicheng District have joined the pilot, and Beijing University of Posts and Telecommunications has embedded agents into its entire academic affairs workflow.
For enterprises, now is a good time to review your Agent BI strategy. If you're currently selecting or upgrading a BI system, ask one more question: is its agent capability "manageable and controllable"?
Verifiable identity, describable capabilities, auditable interactions—these three criteria will transform from "bonus features" to "admission requirements." The second half of Agent BI has just begun.
Top comments (0)