National standard for agent interconnection released: Why may the physical world still not be connected even with unified interfaces?
In late June, the State Administration for Market Regulation (Standardization Administration of the People's Republic of China) officially approved and released a series of seven national standardized guiding technical documents titled "Artificial Intelligence - Agent Interconnection".
Source: Press Conference on Agent Interconnection Standardization by the State Administration for Market Regulation
These seven parts sequentially cover the overall architecture, identity codes, identity management, agent description, discovery, interaction, and tool invocation. Like a precise zipper, they completely close the loop from "identity identification" to "collaborative interaction" of agents, filling the domestic standard gap in this field.
This document, led by the China Electronics Standardization Institute and participated in by more than 30 units including Huawei, Xiaomi, Lenovo, Ant Group, and major cloud providers, is officially defined as an "agile standardization arrangement during the industry cultivation period". Its intention is obvious: to be inclusive, reserve room for trial - and - error in innovation, and attempt to solve three deep - seated pain points in the industry:
Break through ecological barriers: The interfaces and protocols between agents of different manufacturers are not unified, forming agent islands, which seriously restricts large - scale collaborative applications.
Resolve the trust crisis: Agents lack a unified identity authentication and traceability mechanism. Cross - domain interaction faces security risks such as identity impersonation and data leakage.
Reduce innovation costs: The lack of general interaction and description specifications leads to repeated construction by enterprises and high costs for system integration and adaptation.
Meanwhile, the official also announced that it will speed up the development of detailed rules for agent auditing and transactions. Many media have interpreted this as "the end of information islands" and "the full - scale start of a major cycle of AI applications".
However, if we look back at the "interconnection history" of the technology industry, history has repeatedly given opposite lessons: The establishment of standards is often not the end of the game, but the start of a new round of covert battles.
Especially today when AI is accelerating towards the physical world, the core interests that giants are competing for lie neither in the underlying hardware entities nor in the cloud - based super - brains, but precisely in the invocation protocols that translate "intentions" into "actions".
Because hardware is just "muscle" that can be replaced at any time, and large - scale models are "brains in a vat" confined to the cloud. Only these invocation protocols are the "throat" connecting the virtual and the real.
Now, although the standards have been established, the islands in the interconnection history have always been caused by the mismatch of incentive mechanisms and control rights. If a consensus on commercial interests cannot be reached, simply unifying technical interfaces will not address the root cause.
This article will be divided into three parts to dissect the underlying logic behind this "breaking through the barriers of agents":
Part One: Expansion test. Is interconnection creating incremental value or "homogenizing" various enterprises?
Part Two: Covert battle for control rights. Why is the "right to translate intentions" the real strategic point in the era of agents and embodied intelligence?
Part Three: Machine autonomy situation. How should the industry cross the new islands and build a real AIoT agent?
The mirror of history: Mismatched interests and escaping control rights
The call for "interconnection" in the industrial Internet has been around for a decade, but it has always been difficult to make progress.
The McKinsey Global Institute (MGI) clearly calculated in its classic report "The Internet of Things: Beyond the Hype, Validating Real Value" that, on average, 40% of the total potential value that the Internet of Things can release, and nearly 60% in some scenarios, must rely on the interoperability between systems to be realized.
But in reality, the root cause lies in the device layer. The field networks have long been dominated by a few manufacturers, and incompatible industrial protocols are fragmented, such as Profinet dominated by Siemens and CC - Link dominated by Mitsubishi, and they are deeply coupled with equipment.
Why is it difficult to promote? Because the costs and risks of interconnection are borne by established giants, while the benefits flow to small and medium - sized enterprises and end - users. When the incentive mechanism runs counter to commercial interests, no one is willing to actively remove their moats.
The Matter standard for smart home on the consumer side provides a more confusing example.
On the surface, Matter has no shortage of appeal, with Apple, Google, and Amazon collectively endorsing it, and thousands of globally certified products. However, its actual implementation has awkwardly stopped at the *"lowest common denominator"*, that is, all parties only compromise on the most basic and non - differentiated functions to reach an agreement.
The embarrassment of implementation lies here: A large number of Matter - certified devices only stably open the most basic switch and dimming commands when crossing platforms. High - level functions that truly determine product premium and differentiated experience (such as AI - adaptive rhythm lighting, precise energy consumption monitoring, complex security linkage, etc.) are still firmly locked in the native apps of the giants. Even if a function is written into the Matter specification, it often takes a long time to be widely available on retail devices or fully opened in the apps of each ecosystem.
Although fragmentation seems to have been reduced, commercial competition has not stopped. Instead, new fragments have emerged at the "functional layer". Interconnection only stays at the "floor level", while the real control rights have escaped to the "private layer".
These two cases reveal a cruel truth: If interconnection touches the pricing power and customer stickiness and there is no new value compensation, incumbents will either resist or lock it at the "floor level".
Therefore, the core of testing the success or failure of standards lies in the "expansion test": Is it expanding the market or engaging in internal competition for a share?
TCP/IP has dominated the world because it has opened up an exponential new market through "bottom - layer decoupling". Before its birth, the network was a closed garden of giants such as IBM. After its birth, it transformed the network into a standard pipeline, giving rise to an ecosystem that emerged without barriers. When TCP/IP was adopted as a standard in 1983, there were only more than 500 hosts on the entire network. Around 1995, the number exceeded ten million, giving birth to a digital new world that was unimaginable before. It did not deprive hardware manufacturers of their profits but instead expanded the entire hardware market.
In contrast, the deep - level interconnection of Matter and industrial protocols is blocked because they essentially "pipe - ize" and "homogenize" enterprises.
TCP/IP standardizes the "transmission channel", while Matter attempts to standardize the "control rights". Once fully interconnected, the experience difference between high - end brands and white - label devices will be smoothed out, and the ecological moats of some enterprises will instantly collapse. This kind of interconnection does not create incremental value but only transfers enterprise profits through price wars. Facing the requirement of "revolutionizing themselves", many enterprises, especially giants, naturally pay lip service.
So, returning to the interconnection of agents: Will the implementation of these seven national standards give rise to a prosperous "machine - autonomous economy" like TCP/IP, or will it fall into the defensive quagmire of large - model manufacturers fighting each other?
The key lies in whether this set of standards can, while breaking through barriers, point out the "new continent of exponential growth" for the entire industrial chain.
The "new disease" of embodied intelligence: The responsibility gate in the physical world and the data moat
So, where exactly is the "new continent" of agent interconnection?
Theoretically, it exists in the vast blue ocean of "machine autonomy": from the adaptive collaboration of cross - production - line equipment in factories, to the seamless service of multi - brand home appliances in households, and then to the autonomous handover of drones and unmanned vehicles in logistics networks. This continent promises extremely high system - level efficiency, which can not only create incremental value but also reshape the entire social production function.
However, this new continent currently remains more in the "imaginary" stage. Because when we use the "expansion test" to measure today's "intention execution layer", we will find that it not only inherits the "old disease" of interconnection but also catches the "new disease" of the era of embodied intelligence.
First, look at the "old disease": The zero - sum game of control rights and pricing power.
In the agent ecosystem, super - platforms with the most entry points (such as large - model manufacturers or terminal giants) have become the new "blockers". They will gladly accept the "interconnection floor" defined by the standards, but they will surely defend their private territories above the floor, such as the orchestration logic of multi - agents, long - term memory libraries, and complex action semantics.
Because once fully interconnected, hardware entity manufacturers may become "dumb actuators" that can be replaced at any time, facing the situation of being "pipe - ized", and the pricing power of the industry will irreversibly concentrate on the cloud - based invocation and orchestration layer.
What really sets agent interconnection apart from history is its "new disease": The "responsibility gate" in the physical world.
In the pure digital world, a cross - subject API call has almost zero cost. If it fails, you can just report an error and retry.
But in the physical world, each cross - manufacturer machine call is accompanied by real marginal costs, irreversible physical consequences, and clear exposure to responsibility.
Imagine: An agent of Manufacturer A calls a robotic arm of Manufacturer B to perform a high - precision action, resulting in equipment damage or production line shutdown. Who should be held accountable? Is it because Manufacturer A issued the wrong intention, or is it because Manufacturer B's execution accuracy is substandard? Before the mechanism for defining rights and responsibilities and commercial insurance is in place, rational participants will actively shrink the connection radius and limit the rights of cross - domain calls to avoid "physical - level" disaster risks.
Bug fixes can be made for digital - world bugs, but compensation is required for wrong actions in the physical world. This responsibility gate standing in front of embodied intelligence is a hard barrier that has never been encountered in the era of software interconnection.
The key to breaking the deadlock: Master the "invocation pricing right" and remove the barriers between responsibility and interests
It can be seen that the root cause of agent interconnection ultimately lies in the mismatched incentives and the lack of responsibility. Each solution needs to be precisely targeted.
To cure the "old disease" of mismatched incentives, the key is to establish a neutral pricing and clearing layer.
The key to breaking the deadlock is to make interconnection truly expand the market for adopters. Once physical actions can be standardized for invocation, each execution should be transformed into a measurable, priced, and clearable commercial event, just like the "API economy" in the digital world.
Only when the new revenue that hardware manufacturers obtain through open interfaces is sufficient to cover their losses from being "pipe - ized" can interconnection truly pass the "expansion test". The regulatory authorities revealed at the press conference that they will promote the "transaction and auditing" standards, which just confirms this point: The underlying commercial rules that determine the distribution of interests are the key to interconnection.
To cure the "new disease" of physical responsibility, the core is to build a binding mechanism between responsibility and compensation.
The physical risks borne by cross - domain calls need to be definable, transferable, and even covered by financial instruments. First, completely remove the "responsibility gate" from the system, and achieve confirming rights before invocation. Only then will the machines in the physical world dare to truly move towards deep connection.
The release of the seven national standards has paved a unified highway for agent interconnection, which is an extremely crucial first step. But the real problems in the intention execution layer have always been beyond the pure technical interfaces. This is like building a road, but the more difficult game has just begun: What kind of vehicles will run on the road, who will collect the tolls, who will be responsible for traffic accidents, and whether the control rights belong to the entire road or are monopolized by a few toll stations.
The lesson from history is that industries that regard the release of standards as the end often repeat the isolation in compromise. In the journey of embodied intelligence, the urgent task is to truly upgrade the unification of interfaces to the transfer of value and the attribution of responsibility.
Only by crossing this barrier can AIoT agents truly emerge from the imagination and move towards a prosperous "intelligent economy".
Conclusion
The evolution of the technology history seems to always cycle between "breaking old islands" and "building new high walls".
Putting large - scale models into the shells of robots only completes the awakening of individual "intelligence"; while enabling thousands of agents from different manufacturers to seamlessly collaborate, interact, and interconnect in the real physical world truly kicks off the prelude to the "machine society".
Jumping from the "information Internet" to the "agent Internet of Things", we are not only crossing the physical boundary between bits and atoms but also moving from simple "data exchange" into the deep - water area of "value transfer and shared responsibility".
Ultimately, interconnection has never been a purely technical proposition. It is an infinite game about interest distribution, risk pricing, and order reconstruction.
This article is from the WeChat official account "Internet of Things Think Tank" (ID: iot101). The author is Peng Zhao, and it is published by 36Kr with authorization.