The Rise of the Agentic Web: Future of AI Collaboration

1

Introduction

With the further development of large models in multimodal understanding, logical reasoning, and other areas, the Agent is expected to see large-scale application in 2025, replacing humans in solving more and more daily tasks. AI experts generally believe that 2025 will be the year of the agent explosion.

By the end of 2024, Gartner has also listed agentic AI as one of the top ten technology trends for 2025, predicting that by 2028, at least 15% of daily work decisions will be autonomously made by agentic AI, compared to 0% in 2024.

At CES 2025, Jensen Huang also emphasized that there will be many commercial applications for Agents in 2025.

This article starts from Agents and focuses on the development of Agentic Web.

2

What is an AI Agent

According to a blog post by OpenAI‘s former VP of Safety Research, Lilian Weng, Agent = Large Language Model (LLM) + Planning + Memory + Tools + Action. In an LLM-driven Agent system, the LLM acts as the brain of the Agent and consists of several key components such as Planning, Memory, Tools, and Action.

3

Cooperation of Multiple AI Agents

With individual Agents having significant autonomy, researchers gradually discovered that problems can be solved through collaboration among multiple Agents.

In 2023, Stanford University’s Human-Computer Interaction research team created an AI town consisting of 25 Agents to simulate social interactions. This AI town features infrastructure like universities, apartments, cafes, and bookstores, where Agents reside and conduct daily activities, even though they are unaware of living in a simulated environment. These Agents can work, chat, organize social activities, make new friends, and even fall in love, each with unique personalities and backstories. This social simulation demonstrates the adaptability and behavioral diversity of Agents in complex human social environments.

The research team at Tsinghua University has attempted some early explorations with AI employees, creating a virtual software company named ChatDev, operated by various Agents with different roles, including CEO, CPO, CTO, programmers, code reviewers, testers, and designers. These Agents form a multi-agent organizational structure, tasked with the mission of “revolutionizing the digital world through programming.”

Fudan University’s tech team developed a one-stop AI employee development and edge service platform named AI2Apps, replicating Devin’s development model for AI employees in any position. The team independently developed a browser-based Agent operating system, providing a full suite of developer tools, including terminals, file management, Agent debugging, code editors, and browser sandbox environments, enabling developers to build and combine AI employee teams under the leadership of the