The Three Major Trends of the Internet of Agents: The Connection Paradigm is Being Completely Restructured
I deeply appreciate a classical internet concept: connection is power. As long as a person can freely connect to information, connect with others, and connect with tools, they have the power to change the world.
I believe the Internet of Agents will completely restructure the current internet's connection paradigm, returning the internet from platforms back to connections.
Previously, I wrote an article: How the Internet of Agents Differs. Today, I'll delve deeper into the three major trends of the future Internet of Agents, which forms the foundational logic behind our design of ANP (Agent Network Protocol).
Often, a macro perspective is easier to judge than a micro one. By ignoring details and paths while grasping the essence, we can clearly see the future. Our thinking on this issue relies on two essential principles:
- How to lower the overall operational costs of society
- How to better leverage the capabilities of new technology (AI)
Core points of this article:
- We have three core judgments about the Internet of Agents:
- Agents will replace existing software and become an important component of the internet
- Personal AI assistants will become the new internet gateway, replacing most apps people currently use. In comprehensive scenarios, they can provide an experience ten times better than current apps.
- Agents will replace software serving businesses. Companies can publish agents on the internet to directly connect with and serve users.
- Personal AI assistants will connect directly with agents, and agents will connect directly with other agents, rather than through third-party platforms. This is a new connection paradigm.
- Any agent can interconnect with any other agent, breaking down data silos and allowing information to flow freely.
- This will enable AI to obtain complete contextual information, helping individuals or businesses make more intelligent decisions.
- This will allow AI to access all tool capabilities, helping individuals or businesses complete more complex tasks.
- Connections between agents will have a higher proportion, becoming the main type of connection on the internet.
- Agents will connect through protocols, which is the AI-native way of connection.
- Computer Use/Browser Use is just a transitional form. It teaches AI to interact with the internet in human ways but cannot fully leverage AI's capabilities.
- Protocol is the most efficient way for agents to connect. AI is better at handling direct, low-level data rather than UI on screens or HTML on web pages, which are designed for humans.
- Agent communication protocols will become the universal protocol between agents, just like HTTP. In the future, standard agent communication protocols will emerge. Based on standard protocols, a data network specifically designed for AI access will emerge.
- Agents will replace existing software and become an important component of the internet
- From the perspective of productivity and production relations, AI is the most advanced productivity, and the internet is AI's most direct production relation. AI will inevitably change the existing internet structure, especially data silos, which are the biggest obstacle to AI's capabilities.
- As long as there are better solutions (lower costs, better experiences), someone will implement them. If big companies don't, startups will.
- Our ideological school: Classical Internet Spirit + AI-Native Reconstruction. We hope to rebuild the internet in an AI-native way, returning it to the early vision of "connection is power," creating a more open, vibrant, and creative internet.
- The Internet of Agents will see a new connection paradigm, where connection is power replaces platform is power.
We are exploring our ideas through action. This is our designed agent communication protocol, ANP, which aims to become the HTTP of the Internet of Agents era: https://github.com/agent-network-protocol/AgentNetworkProtocol.
We also welcome you to join the ANP open-source technology community. Our community slogan is: Defining agent connection and collaboration through open source.
The main content follows.
1. Agents Will Replace Existing Software and Become an Important Component of the Internet
1.1 Personal AI Assistants Will Become the New Internet Gateway
In the future, personal AI assistants will become the new internet gateway, and most existing apps will disappear. More accurately, most current applications will no longer need to provide user interfaces on terminals; they will only need to provide data and services to personal AI assistants, which will provide personalized UIs for each user to meet their individual needs.
In a single scenario, such as booking a hotel, the service experience provided by a personal assistant may not be much better than existing apps.
However, in comprehensive scenarios, such as travel planning, especially multi-person travel planning, personal assistants authorized to access all your information can call all internet tools for more intelligent planning, providing a service experience far superior to existing apps.
Tool-type apps, such as translation apps, will likely be replaced first, followed by news, shopping, and social apps. Game apps, which provide emotional value to people and have complex on-device rendering, may not be replaced in the short term.
1.2 Agents Replace Software Serving Businesses
In the business sector, enterprise software will also be replaced by agents, such as customer service, marketing, ERP, CRM, communication tools, etc. Many companies are already developing products in this area.
Returning to the internet topic, if businesses of all sizes have an outward-facing agent that can be published on the internet, this might change how consumers connect with businesses.
1.3 Personal AI Assistants Directly Connect with Agents
Currently, connections between people and businesses or merchants are made through internet platforms. If every person and every business has an agent on the internet, why can't they connect directly?
Let's first answer a question: Why wasn't direct connection feasible before, and why is it possible now?
Previously, if every business developed a website or app, users would need to log in and access many pieces of software, resulting in high user costs. At the same time, businesses would have to develop and maintain software, which was also costly. Therefore, internet platforms developed once and, leveraging software's infinitely low marginal cost, could serve a large scale of users who only needed to use a few apps, keeping usage costs low.
Now, several changes make a difference:
- The cost structure of software has changed. Development costs are lower, and operating costs (computing power) are higher, with cost advantages more dependent on scale effects. In the future, there may not be much cost difference between businesses using internet platform software versus using their own software (or purchasing software) plus cloud computing power.
- Personal assistants replace humans in directly connecting and interacting with business agents, reducing the usage cost for people.
- AI solves the connection cost problem between agents (protocol coordination, testing, upgrades, etc.).
Therefore, in the Internet of Agents, direct connections between people and businesses no longer face significant technical or cost barriers.
Finally, from a business perspective, current internet e-commerce platforms have a monetization rate of about 3-5% (for every 100 units of transaction, the platform earns 3-5 units). This isn't to say their profits are unreasonable, but rather that this operating model has higher costs.
If a direct connection model can operate at a lower cost, such as transaction costs below 3% of total GMV, then this model will be more competitive.
Of course, this concept still has many issues to solve, such as how to handle after-sales service, ensure transaction security, and so on.
2. Agents Must Interconnect — Data Silos Are Strangling AI's Potential
We believe that existing internet data silos are the biggest obstacle to leveraging AI capabilities.
2.1 AI Can Obtain Complete Contextual Information
Only through interconnection can AI obtain complete contextual information to provide more intelligent decisions for individuals or businesses.
In the current internet, each person's information is scattered across various apps. For example, our social information is in WeChat, shopping information in Taobao/JD/Pinduoduo, and entertainment information in TikTok/Kuaishou. This is why general agents like Manus can currently only use public information to complete a small portion of human tasks.
These data silos are the biggest problem hindering AI's capabilities today and the first problem AI needs to solve.
Interconnection between agents allows information to flow freely, which is key to solving the data silo problem. For individual users, using a personal assistant to connect with all agents consolidates all personal information in one place, giving the personal assistant complete contextual information.
2.2 AI Can Access All Tool Capabilities
Similarly, only through interconnection can AI access all tool capabilities provided by other agents to complete more and more complex tasks for individuals or businesses.
For example, through a personal assistant, we can connect with a hotel's agent to book accommodations, an airline's agent to book flights, and a tourist attraction's agent to book tickets.
If a general agent like Manus is authorized to access all my information and can call all internet tool capabilities, it could provide a product experience that no current app can match.
2.3 Higher Proportion of Connections Between Agents
Many platform agents now, such as those on Coze or Dify, are still mostly designed to serve humans. We believe that apart from personal assistants, most future agents should provide services to other agents rather than directly to humans.
When a person sends a task to an agent, the agent will break down the task and collaborate with other agents many times to complete it. In this process, the number of connections and collaborations between agents will be greater than those between humans and agents.
We believe that connections between agents will be the main type of connection on the future internet.
3. Agents Connect Through Protocols — The AI-Native Connection Method
3.1 Computer Use/Browser Use is a Transitional Form
The previously popular Computer Use approach and the recent Browser Use approach employed by Manus are, in our view, transitional solutions that teach AI to interact with the internet in human ways.
The biggest advantage of these methods is that they immediately allow AI to access the internet, connecting to information and tools, but they also have many problems:
- Computer Use can obtain the most information but is relatively costly and occupies the user's computer or phone.
- Browser Use is less costly but cannot access information within apps, and cloud-based logins may pose security issues.
We believe agents should use a more native connection method.
3.2 Protocol is the Most Efficient Connection Method for Agents
We believe that connecting agents through protocols is the most efficient method and the AI-native way of connection.
AI is better at handling direct, low-level data rather than UI on screens or HTML on web pages, which are designed for humans.
On this point, MCP and our designed ANP protocol are highly aligned, belonging to the same technical approach.
Moreover, using agent communication protocols, agents' public information can build a data network that is easy for AI to access and specifically designed for AI, distinct from the current internet composed of UI page links designed for humans.
3.3 Standard Agent Communication Protocols Will Emerge
In the future, standard agent communication protocols will emerge, becoming the universal protocol between agents, just like HTTP. Only in this way can the industry's agents truly interconnect and form a genuine Internet of Agents.
Currently, many standardization organizations and communities are conducting similar research, and there are many protocols specifically designed for agents. Besides our ANP, there are also Agora Protocol, ACP, Agentsjson, and others.
4. The Likelihood of Our "Three Judgments" Occurring
We can analyze the likelihood of these three judgments occurring using the theory of productive forces and production relations.
AI is the most advanced productive force today, and the internet is AI's most relevant production relation. Many phenomena in the current internet, such as data silos, are hindering AI's capabilities. We believe these will inevitably change, though the path of change is uncertain.
Existing super apps and super platforms are powerful. They have enormous traffic, rich data, mature business models, and motivation to maintain the current internet structure.
However, if better solutions exist, if T doesn't implement them, B will; if B doesn't, A will. And regardless of whether BAT (Baidu, Alibaba, Tencent) act, I believe startups will. Startups and open source will be important forces in this technological revolution.
5. Our Philosophy: Classical Internet Spirit + AI-Native Reconstruction
If we summarize our internet philosophy, I believe we belong to the Classical Internet Spirit + AI-Native Reconstruction school.
- Classical Internet Spirit
We believe the internet should return to the openness of the Web 1.0 era. With the support of new technology (AI), the internet should return to the early vision of "connection is power" rather than the current "platform is power."
- AI-Native Reconstruction (Agent-Native Web)
This isn't simple "nostalgia" but rebuilding the internet with AI as a completely new technological element, forming new orders and infrastructure.
It's not designing UIs for "humans" but designing protocols and data interfaces for "agents."
It's not making AI adapt to the existing "page" data network but using protocols to build a data network that is easy for AI to access and specifically designed for AI.
The Internet of Agents Will Come Faster Than You Think. Are You Ready?
If you're interested in this topic, please contact us.
The goal of AgentNetworkProtocol is to become the HTTP of the Internet of Agents era. Our vision is to define how agents connect with each other and build an open, secure, and efficient collaboration network for billions of agents.
The ANP open-source technology community currently has 20+ developers and is recruiting more. If you're interested in agent communication protocols, whether in development, product, or operations, you can join us to define agent connection and collaboration through open source.
We also welcome companies, organizations, and communities interested in the ANP protocol to join our open-source community!
Contact information:
- GitHub: https://github.com/agent-network-protocol/AgentNetworkProtocol
- Discord: https://discord.gg/sFjBKTY7sB
- Website: https://agent-network-protocol.com/
- WeChat: flow10240
If you're interested in the Internet of Agents, you can scan the QR code below to join our discussion group, or add WeChat (Zhangxy66088) to join the group.
We also have an agent communication protocol discussion group. If you're interested, add my WeChat (flow10240) to join.