AI Agents: Current Status, Technological Progress and Development Trends

Click on the “Turing Artificial Intelligence“, select the “star” public account

The artificial intelligence insights you want will be delivered to you first

Copyright Statement

Reprinted from Wang Jiwei, copyright belongs to the original author, used only for academic sharing

Although many are discussing intelligent agents now, it is still necessary to understand AI Agents before talking about this industry.

1. Definition and framework of AI Agents

Under this definition, AI Agents are mainly divided into perception, planning, and action. After perception, planning begins, decisions are made, and then actions are taken. After the action is completed, it enters a loop of observing the environment, continuing perception, further planning, optimizing, and finally taking better actions. This is the simplest expression of AI Agents.

AI Agents have exploded at this event node for three main reasons:

4. Industry Data

Next, let’s look at some industry data. Here, we have collected and organized relevant data from five industries.

5. AI Agent Panorama

With more AI Agent product solutions being launched, more companies and teams involved in AI Agents are gradually emerging, and the industry landscape is becoming clearer.

International Panorama

Domestic Panorama

In the domestic market situation, Wang Jiwei’s channel reviewed two industry reports. The left image is the “China AI Agent Industry Research Report” released by Ginkgo Light Year in April, where they created the 1.0 version of the China AI Agent ecosystem map based on the market situation at that time. Of course, this image also reflects the ecological structure of the domestic intelligent agent industry last year.

The right image is the second quarter report released by InfoQ. It can be seen that there are significantly more intelligent agent products. From April to June, some entrepreneurial products emerged, and some large companies also launched related products. Of course, this was the market situation six months ago, and now there are even more products.

6. Product Status

The current forms of AI Agent products and services include the following types. Common AI Agent products include chat assistants, coding assistants, AI search, etc.

The current AI Agent products have the following common characteristics:

Intelligent Agent Construction Platform

Overall Assessment: Still in the Early Stage of AI Agents

Currently still in the early stage of AI Agents. More intelligent agents are more like chatbots, capable of performing relatively complex tasks, but there is still a long way to go to the ultimate goal of autonomous agents.

Even so, the application trend is already unstoppable. AI Agents are widely used in multiple industries such as customer service, programming, and content creation, especially showing significant results in China’s e-commerce and education sectors.

Technological advancements enable AI Agents to work autonomously, demonstrating human reasoning and creative thinking. Concerns about safety and ethical issues are being raised, and breakthroughs in multimodal interactive interfaces are being achieved. The application of AI Agents in research and other fields is expanding, and it is expected to promote enterprise stratification and application focus in the next 5-10 years, gradually revealing commercial value.

7. Issues and Limitations of AI Agents

Although AI Agents have gradually realized commercial use in many fields, they still face some issues and limitations due to current technological, ecological, and user acceptance factors.

Here, Wang Jiwei’s channel summarizes nine points of shortcomings in AI products, including limitations in interaction capabilities, stability issues with random outputs, and exception handling problems. For details, please refer to the left table in the image.

AI Agent application deployment also faces several challenges, and here we directly quote the survey and summary from langbase’s “state-of-ai-agents” report, as shown in the right chart above.

8. AI Agent – AI Agentic Workflow – AI Agentic AI

AI Agents are intelligent entities capable of perceiving the environment, autonomously understanding, making decisions, and executing actions. Agentic Workflow refers to a workflow that accomplishes tasks through predefined multi-step large language model (LLM) calls. Agentic AI represents the peak of AI’s capabilities and behaviors, including the ability to act independently, learn, and adapt.

The development of AI Agents towards Agentic Workflow and the rise of Agentic AI are driving efficiency improvements and digital transformation in the industry. These technologies change the operational models of enterprises, enhance customer experiences, and bring revolutionary changes to decision support and automation services.

At the same time, they also extend the application value chain, transforming industry formats. Despite facing technological challenges, they bring unprecedented development opportunities to the industry.

Second Part: Technological Progress

1. AI Agent Technology Stack

The technology of AI Agents has developed to the point where the technology ecosystem has basically formed, and various technologies used to build AI Agents are constantly improving.

The two images below show the market panorama compiled by investment firm Aura Ventures in July and August last year. They included the technology section, listing relevant companies or products in this difficult-to-understand map.

You can clearly see that under each technology and solution, some representative companies are listed. For a specific interpretation of this image, please refer to section 15.2.4 of the book on industrial patterns.

The right image is the latest technology stack statistics released by Letta in November this year. It also presents the AI Agent building process from a technological perspective, indicating which technologies are provided by which technology suppliers. In terms of technology alone, many technology companies have gained market recognition in just over half a year.

In terms of technology stack, you can focus on understanding the “AI Agents Stack” image. The left image can be used as a reference because it was created earlier, but it allows for an overall understanding of the AI Agent market structure.

2. AI Agent Technology Ecosystem Map

At the beginning of the article, we first introduced the AI Agent technology architecture proposed by Weng Liliang. Visualizing this architecture with technology and enterprises allows us to see the AI Agent ecosystem map drawn from the perspective of technology suppliers by Activant Capital, which is the left image. The technology-oriented architecture diagram helps us better understand intelligent agents.

The right image is a visualization of related technology vendors and also serves as a quadrant diagram of technology maturity and market growth beliefs, where each technology and its representative vendors have an appropriate position in this quadrant. Through this image, we can also see the market development potential of these technologies and products.

3. AI Agent Technological Progress

Here, we will briefly introduce the technological progress of AI Agents.

AI Agent technology based on large language models is rapidly developing and iterating. By the second half of 2024, large models will be evolving towards multimodal capabilities. In October, OpenAI’s o1 model opened the post-training era, featuring applications of reasoning, vision, and contextual protocols, greatly promoting the application of AI Agents in more scenarios and fields.

You can see the left part of the above image, which is the prospect of large language model development. The image is divided into six parts, and I have made brief notes from left to right. In just over two years, large models have undergone multiple iterations, continuously enhancing and increasing the functionality and capabilities of AI Agents.

The right side shows the current seven mainstream RAG (Retrieval-Augmented Generation) technology architecture diagrams. RAG technology, which effectively addresses the long-term memory issues of AI Agents, has developed various technology architectures in just a few years.

The AI Agent technology framework is continuously innovating. Here, I will list some of the AI Agent technology frameworks launched by representative tech companies Microsoft and Google. This chart lists eight technology architectures and solutions from Microsoft, which is only a part of it; in fact, they have launched many related technologies.

Google is also starting to focus on AI Agents now. In the technology community, AI Agent technology frameworks are flourishing, and it is expected that by 2025, there will be a surge in multi-intelligent agents, GUI (UI) intelligent agents, and edge intelligent agents.

Here, I will also list some open-source and closed-source AI Agent projects. The two tables below list the open-source and closed-source projects that have launched AI Agent technology frameworks, products, and solutions. The left is open-source projects, and the right is closed-source projects.

These projects come from the intelligent agent sandbox technology supplier e2b’s GitHub repository awesome-ai-agents, which accumulates intelligent agent projects through collection and submission. Currently, there are 110 open-source projects and 105 closed-source projects.

Of course, this does not mean that these are the only AI Agent projects currently available; there are many related projects that are not listed here. This is just a brief introduction to these projects; for specific introductions to each project, you can refer to GitHub.

Third Part: Development Trends

Finally, let’s look at some development trends for AI Agents in 2025. Here, I list five trends as follows:

• Significant increase in AI Agent adoption
• Multimodal Agents enhance user experience
• Multi-Agent systems become popular
• AI Agent clusters and AI Agent networks
• Vertical AI Agents are poised to emerge

Regarding development trends, Wang Jiwei’s channel will not elaborate further; you can understand it in conjunction with the PPT content.

1. Significant increase in AI Agent adoption

The adoption rate of AI Agents will significantly increase in the coming year, with organizations across various industries planning to use them for cross-departmental tasks such as email generation, coding, and data analysis, etc. According to a Capgemini report, 82% of organizations plan to integrate AI Agents by 2026.

Deloitte predicts that by 2025, 25% of enterprises using GenAI will deploy AI Agents, increasing to 50% by 2027. Gartner predicts that by 2028, at least 15% of daily work decisions will be made autonomously by Agent AI, and 33% of enterprise software applications will include Agent AI.

In the coming year, specialized AI Agents will also emerge in fields such as finance, retail, and healthcare.

2. Multimodal Agents enhance user experience

The rise of multimodal AI Agents marks a significant advancement in AI capabilities, enabling them to process multiple input data types such as text, images, audio, and video, bringing extensive applications to various industries.

For example, in healthcare, these Agents provide more comprehensive and accurate diagnostic suggestions by analyzing medical imaging, patient records, and symptoms. In retail, they create more intuitive shopping assistants by combining visual recognition and natural language processing.

In creative industries like advertising and design, multimodal AI Agents can generate content that combines text and images, understanding the nuances of visual and verbal communication, which is highly valuable for creating targeted marketing materials and personalized content.

As the influence of AI Agents in specific industries grows, multimodal AI is at the forefront of transformation, with its ability to process and synthesize various types of information closely resembling human cognitive processes, becoming an essential tool in complex decision-making scenarios.

3. Multi-Agent systems become popular

Multi-Agent systems are becoming popular due to enterprises’ demand for complex solutions, becoming the center of development. AI Agents will collaborate to solve problems, perform multi-layer decision-making tasks, share information, coordinate actions, and manage complex cross-departmental workflows. For example, in logistics, they can optimize supply chains, manage inventory, and predict demand fluctuations.

By 2025, more organizations will deploy multi-Agent systems to optimize business processes, with Agent orchestration platforms like OpenAI Swarm and Microsoft’s Magentic AI leading this trend, helping enterprises collaboratively deploy and manage multiple Agents.

4. AI Agent clusters and AI Agent networks

In 2025, AI Agents will enter a higher level of collaboration and cooperative working stage. Salesforce AI research leader Silvio Savarese predicts that in the coming year, AI Agents will work in groups like ants to solve daily tasks and business challenges, redefining productivity and problem-solving capabilities on an unprecedented scale.

AI Agents will seamlessly integrate into life, with individuals owning personal Agents and organizations deploying specialized Agents. These Agents can be obtained through platforms like Agentforce, customized for specific tasks, and collaboratively achieve common goals. In the future, AI applications will focus on creating and customizing Agents that execute collaborative strategic tasks and decisions, whether in personal or business environments.

5. Vertical AI Agents are poised to emerge

Vertical AI Agents focus on specific industries or fields, using AI technology to automate tasks, improve efficiency, and partially replace human labor. Industry insiders predict that their future market size could be ten times that of the SaaS market and is expected to replace SaaS in multiple areas.

Starting from 2025, vertical Agents will rapidly occupy the market with more mature technologies and growing user demand.

Their main advantages are high specialization and customization, performing better than general intelligent agents in specific fields, with high efficiency, rapid response, and high stability and reliability optimized through extensive training. Their application fields are wide, covering healthcare, finance, customer support, market research, and analysis.