AI-Riesen setzen stark ein und richten ihre Kräfte ein. Tiefgehende Analyse von AI-Agenten: Warum ist dies die ultimative Form von KI?

Today, I read an interesting opinion that there are few new technologies that offer organizations more opportunities to accelerate their productivity and transform their business processes than Agentic AI. Its potential even seems to be greater than that of its relative, Generative AI (GenAI).

Moreover, I read a report from Huatai Securities stating that Generative AI is entering a new development phase where AI agents play the leading role.

The so - called Agentic AI essentially corresponds to the concept of Intelligent Agents that we are already familiar with. Recently, I participated in several events and tested numerous AI products based on Intelligent Agents. It is obvious that the concept of Intelligent Agents is becoming increasingly popular.

I have a clear impression that this might be the latest phase of development in the AI field, which has been in motion since the emergence of ChatGPT. Today, I would like to try to show you a more comprehensive global picture of the development of Intelligent Agents.

I. From the "Knowledgeable Brain" to the "All - around Worker": What Exactly are Intelligent Agents?

To understand why Intelligent Agents (Agents) hold so much promise, we first need to clarify the fundamental difference between them and the well - known Generative AI (GenAI).

If we say that GenAI, represented by ChatGPT, is a "brain" that has a lot of knowledge and can answer all questions, then AI agents are like "hands and feet" attached to this brain to turn it from a "conversationalist" into an "actor".

GenAI tools are good at generating content according to instructions due to their programming logic, but their ability to act ends here. In contrast, Intelligent Agents have higher capabilities:

They are assigned a goal and can then independently understand, plan, use tools, and interact with the environment to achieve this goal.

Let's take a simple example: Recently, I tested how an agent can generate an ultra - high - definition video film in a few minutes, even in ten minutes. Writing the script, creating the storyboards, selecting the music, and generating the images - tasks that would take a human team weeks - an agent can accomplish in one go.

Industry experts have proposed a clear development path for Intelligent Agents, which can be roughly divided into several phases: from L1 chat assistants that can only answer simple questions, to L2 workflow agents that need to be pre - programmed by humans, to L3 inference agents that can plan their own tasks like professionals. The currently hottest competition is in the area of L4 multi - agent systems, which can cooperatively deploy multiple agents with different capabilities to solve complex, interdisciplinary problems.

From this development path, it can be seen that the development of AI is evolving from the pursuit of "larger and stronger" single models to the creation of a cooperative "Intelligent Ecosystem".

This is the fundamental reason for the increasing popularity of the concept of Intelligent Agents - it marks the transformation of AI from a "tool" to a real "partner" and "digital workforce".

II. Global Tech Giants Draw Their Swords: The "Present" of the Agent Industry

The wave of Intelligent Agents is not just empty talk. Globally, tech giants have already invested massive resources in this future technology and shown their "trump cards" to bring this future concept into the present.

Microsoft: Integrate Agents into Every Corner of Productivity

Microsoft's strategy is "Copilot Everywhere". It endeavors to turn Copilot from an in - app assistant into a "Super - Agent" that works across the Windows operating system, the Office 365 package, the Teams collaboration platform, and the Azure cloud service.

In the future, Copilot will not only help you write emails or summarize documents, but it will understand complex instructions such as "Prepare a complete report for next week's sales meeting" and then independently retrieve data from Excel, create charts in PowerPoint, extract key points from Teams chats, and finally compile a complete presentation report for you.

Moreover, Microsoft has open - sourced the AutoGen framework to help developers develop powerful multi - agent applications. The goal is to create a huge, cooperative network of AI agents that deeply integrates the capabilities of agents into every phase of digital work.

Google: Define the Future Interaction with Multimodal and Universal AI

Google focuses on multimodality and universality. The impressive Project Astra, presented at the Google I/O Conference, is a good example of this.

The goal of Astra is to create a universal AI agent that can see, hear, speak, remember, and understand complex situations. In the demonstration, it could recognize the environment in real - time through the phone's camera, understand code, and even remember the storage locations of objects, showing its potential as an "all - around helper in daily life".

Behind this success is the strong performance of Google's Gemini model, especially its inherent ability for multimodal perception and "Tool Use", which enables it to call different APIs to perform real - world tasks.

For enterprise users, Google offers the Vertex AI Agent Builder to help them quickly develop agents for specific business scenarios.

OpenAI: The Key Milestone on the Road to AGI

As a pioneer of the current AI wave, OpenAI sees Intelligent Agents as the key path to Artificial General Intelligence (AGI). The GPTs launched by OpenAI can be regarded as the first attempt to develop agents, which allows users to create individual versions of ChatGPT for specific tasks.

But OpenAI's ambitions go far beyond that. It is actively working on the development of next - generation agents that can act autonomously in the computer desktop environment, use the browser, and apply various software to perform complex tasks. These agents will be able to interact with the digital world like humans, from booking flight tickets to managing complex projects, and will actually extend human capabilities.

NVIDIA: Provide the "Armory" for the Era of Intelligent Agents

In this competition, NVIDIA plays the indispensable role of an "arms dealer". It not only provides powerful GPUs for global AI companies but also a comprehensive platform for the development and operation of agents.

Tools like NIM (NVIDIA Inference Microservices) enable developers to easily package models into callable services, which is the foundation for the development of agents.

Recently, NVIDIA even released the "GR00T" project for humanoid robots, showing its ambition to transfer the capabilities of agents from the digital world to the physical world.

Of course, in this global competition, Chinese tech companies cannot be ignored. Companies like Baidu and 360 have already launched multi - agent platforms for the public that can perform complex tasks, showing the global synchronization of development in this area.

III. "Digital Employees" Become a Reality: How Will Intelligent Agents Change Industries?

After all these high - profile technologies, one naturally wonders how these "AI agents" will change our work and lives. Simply put, all industries will get a group of tireless and high - performance "digital employees".

We all get annoyed by the robotic support that only says "How can I help you?" Future agent - based support will be different. They will have more autonomy and can retrieve your data, understand your questions, and actually handle your affairs like real humans.

Within companies, these "digital employees" will play an even stronger role. A warehouse management agent can keep an eye on the inventory around the clock and, as soon as it notices that an item is running low, independently re - plan the delivery route and delivery time.

For programmers, many tedious and repetitive programming tasks can be delegated to AI agents. They can write new functions, review code, and even find bugs in real - time. Even in cool areas like "Digital Twin" (creating an identical model of a real device in the computer), agents can analyze different data, simulate the operation of the device, proactively inform you of potential malfunctions, and even jointly assist in repair planning.

Of course, in addition to the advantages, there are also risks. The most direct challenge is network security. Imagine if hackers also deploy "Intelligent Hacker Agents", they can launch fast and fierce automated attacks. This forces us to build our own "Security Agents" team. Future network defense and attack will probably be a battle between two groups of AI agents.

Does it sound like the future is already here but still a bit far away? In fact, although the path is full of hope, there are still some hurdles to overcome.

The biggest problem is that the agents developed by different companies still cannot communicate well with each other. There is a lack of unified standards and interfaces, which makes it difficult to cooperate across platforms and between companies. Once this problem is solved, the capabilities of agents will be "omnipotent".

IV. The Road is Long, but the Goal is Beautiful: Challenges and Future Prospects

We are currently in a very critical starting phase. Although the videos of all - powerful AI assistants look like magic, there will still be a lot of work to do before they are really widespread.

What should we do? The experts' advice is very practical:

Start cautiously, but start now. Each of us and every company should actively inform and explore what these AI agents can do for us. In particular, we should find the applications that can actually bring returns. You can start with small pilot projects and give your AI agent a "key" to let it run in the digital world and gain experience.

To the original question: Are Intelligent Agents the latest phase of AI development? The answer is yes. They mark the evolution of AI from a passive "content generator" to an active "task executor". This is a fundamental leap.

Now is the best time to explore Intelligent Agents. We should learn from existing success stories, start small, start building and testing, and give the agents the "digital practical key".

This article is from the WeChat account "New Mango x AI", written by Green and published with the permission of 36Kr.