The era of AI agents is accelerating, and publicly traded companies are actively deploying strategies.

“Future AI agents may have even more intelligence than humans, and humans will live in a world with hundreds of millions, even billions, of AI agents,” predicted Meta CEO Zuckerberg in an interview program this July. A report from Bank of America also pointed out that AI evolution is on the eve of the third wave—Agentic AI (artificial intelligence with agency) is about to arrive.

In the second half of this year, AI Agents have been extremely popular, becoming a hot area that tech giants such as Microsoft, Apple, Google, OpenAI, and others are actively developing. In the domestic market, large-model vendors, internet companies, and listed firms are all actively deploying AI Agent strategies.

As more companies launch AI Agent products and ecosystems, industry insiders expect 2025 to be the breakout year for AI Agents. Galaxy Securities predicts that China’s AI Agent market size will surge to 852 billion yuan by 2028, with a compound annual growth rate (CAGR) of 72.7%.

Vendors initiate the AI agent competition

AIAgent, short for artificial intelligence agent, also known as an “AI intelligent agent,” is a system driven by large language models as its core. It possesses autonomous understanding, perception, planning, memory, and tool-using capabilities, enabling it to automatically perform complex tasks. Unlike traditional AI, AI Agents can independently think and call tools to gradually achieve specified goals.

For example, if a user wants to go out for dinner, a large language model can suggest dining locations and provide relevant restaurant information. An AI Agent, however, can not only recommend restaurants but also search for cuisines and venues within the user’s budget, execute reservations, add the trip to the calendar, and send reminders.

Since the second half of this year, AI Agents have been highly sought after, with tech giants like Microsoft, Apple, Google, OpenAI, and Anthropic announcing significant progress.

In June, Apple showcased its latest AI achievement, Apple Intelligence, at its developer conference. In November, Microsoft announced over ten commercial AI Agents at the “Microsoft Ignite 2024” conference. Google followed closely, also announcing a full push to promote commercial AI Agents, releasing a series of incentive activities and products, and specifically launching one of the few global markets for commercial AI Agents. OpenAI plans to release a new AI Agent codenamed “Operator” in January 2025.

In the domestic market, vendors such as Zhipu are also focusing on AI Agents, and a competition around “AI Agents” is quietly underway.

On October 25, Zhipu released AutoGLM, an intelligent agent capable of understanding voice commands to interpret user intent, simulate human behavior, and automatically complete tasks such as ordering takeout, booking flights, and reserving hotels. On November 29, Zhipu launched an upgraded version supporting autonomous execution of long-step tasks exceeding 50 steps, and it can flexibly switch between different apps during task execution.

Zhipu believes that AI large models are shifting from Chat to Act. In the future, a unified AI intelligent agent will operate various hardware devices to improve human-computer interaction efficiency. With increasing computing power, models and terminal-cloud collaborative architectures suited for AI-native devices will gradually emerge. Smart devices such as smartphones, PCs, cars, glasses, and home appliances are rapidly entering the market.

Besides Zhipu, companies like ByteDance, Baidu, Tencent, Alibaba, and others are also deploying AI Agents, including ByteDance’s Coze Agent platform, Baidu’s Qianfan AgentBuilder development tool, and Kunlun Wanwei’s Tian Gong skyAgents.

Multiple listed companies deploying AI Agents

As domestic vendors compete in AI Agents, many A-share listed companies are also actively deploying related products.

For example, iFlytek has launched AI assistants for education, healthcare, judicial, and government sectors. Jincai Hulian’s “Xinzhiyue Tax & Finance Large Model” adopts an open architecture, combining its tax and finance products with an open platform to build a Tax & Finance AI Agent.

ZKJ Jin Cai’s AI Agent development platform offers functions such as agent creation, multi-model invocation, and workflow definition. It can automatically route and select the most suitable large model based on industry scenarios to create agents. The company states that ZKJ Jin Cai’s AI Agent supports multi-intent understanding, can automatically execute multiple tasks based on user instructions or context, and, after learning user preferences, provides personalized services with highly human-like natural language interactions.

New Indochina’s subsidiary Shanghai Shierqu is dedicated to researching and developing General Purpose AI Agent technology, focusing on multimodal AI technologies and products.

In October, XinkaiPu launched “XiaoMei Classmate,” a campus life AI product based on the PanGu large model and Tongyi Qianwen large model.

On December 9, when asked whether it has already deployed AI Agents, ZKJ Chuangda responded that the company has been advancing product and technology development in this area. For example, its Rubik Avatar integrates multiple AI technologies, including machine learning, natural language processing, and computer vision, based on edge-side intelligence and naked-eye 3D technology, utilizing Qualcomm’s high-performance chips and the Kanzi 3D rendering engine for real-time rendering.

Naxing Shares recently replied to an investor inquiry, stating that its subsidiary Unique Network leverages cutting-edge large-model AI technology and has deep collaborations with leading AI companies like Microsoft, providing efficient, intelligent AIGC applications and solutions for industry clients. Its “Xiaolu” is an AI assistant based on AIGC, supporting intelligent chat and knowledge-base Q&A.

Coolte Intelligent responded to an investor question on December 17, stating that its Coolte AI Agents 2.0 aims to become a universal digital intelligence tool and assistant, aiding traditional enterprises in digital transformation. The 2.0 version is currently under development.

2025 may see an AI Agent commercial explosion

Industry consensus suggests that 2025 will be the year of AI Agent breakout.

A report by Credit Suisse and Shanxi Securities indicates that AI Agents are expected to be deployed in batches in 2025, driving increased demand for computing power and expanding investments in AI infrastructure.

Debon Securities also predicts that by 2025, AI Agents could experience a surge, with the terminal market achieving a qualitative leap. As the Agent network matures, rising market penetration will create a strong business model loop for the entire AI ecosystem.

“Currently, tech giants like Apple, Google, and OpenAI regard Agents as a key focus for 2025. That year may mark the first major breakout for Agents, further accelerating AI application deployment,” said Zong Jianshu, an analyst at Yangtze Securities.

Wanxing Technology’s head of government and enterprise business, Tang Fangxin, sees AI Agent development as transitioning from “experiments” to “applications,” though it has not yet matured. However, with ongoing improvements in computing power and advances in machine learning and natural language processing, AI Agents’ abilities to understand user intent, deliver personalized services, and perform complex tasks are continuously improving. Currently, AI Agents are already used in customer service, smart home, and personal assistant fields, and are expanding into more industries and scenarios.

Tang Fangxin believes AI Agents will continue evolving, gradually moving from “usable” to “well-used.” They are already deployed in some commercial scenarios—for example, Wanxing’s interactive digital human business can significantly address needs in exhibition hall explanations, smart presentations, speeches, and corporate greeting scenarios, using AI digital humans to introduce services and revolutionize traditional interaction methods, enhancing audience experience.

However, Tang Fangxin emphasizes that deploying AI Agents involves collecting and processing large amounts of user data. Therefore, security and privacy are critical, potentially raising concerns about data misuse and leaks. Additionally, AI Agents depend on local computing power and large models, which require high technical maturity, integration, and operational capabilities. Currently, the focus should be on expanding from single scenarios to related ones, consolidating mature pathways and solutions, while strengthening security and privacy regulation. An open mindset is essential to explore and embrace this technology.

“From ‘training’ to ‘inference,’ AI intelligent agents will be the essential pathway to the era of general artificial intelligence,” states a Galaxy Securities report. As large models rapidly iterate and upgrade, AI intelligent agents based on large-model technology are entering a stage of rapid development. Their rise is reshaping the AI industry chain and creating new investment opportunities. By 2028, China’s AI agent market is expected to surge to 852 billion yuan, with a CAGR of 72.7%. The AI intelligent agent industry chain is a diversified, highly collaborative ecosystem with broad future market potential.