Year-end special: From tool applications to value creation, AI agents are entering their "iPhone moment"

On January 15th, Qianwen App announced its full integration with ecological scenarios such as Taobao, Alipay, Fliggy, and Gaode, becoming the first in the world to achieve a closed - loop AI shopping function from ordering takeaways to booking air tickets. This fully automated operation from decision - making to payment has also kicked off the grand drama of AI Agents in the industry this year.

In the past year 2025, the rise and accelerated penetration of AI Agents were undoubtedly the biggest innovation highlights in the AI field. New technologies represented by AI Agents and new business models centered around business reconstruction are driving the industry's deep - level transformation from tool application to value creation. This momentum has been strengthened and continued in 2026. The policy orientation and market dynamics have formed a strong resonance, accelerating the upgrade of the AI Agent industry.

For example, on the policy side, multiple government departments have intensively deployed key tasks for 2026 in recent days, with AI Agents being one of the policy focuses. The National Conference on Industry and Information Technology proposed to promote the special action of "Artificial Intelligence + Manufacturing" and cultivate a number of AI Agents and intelligent native enterprises in key industries. The National Data Administration also proposed that in 2026, a number of data standards will be laid out in frontier fields such as AI Agents and Embodied AI.

On the market side, at the recently held AGI - Next Frontier Summit, Tang Jie, the founder of Zhipu; Yang Zhilin, the founder of Dark Side of the Moon; Lin Junyang, the technical leader of Alibaba Tongyi Qianwen; and Yao Shunyu, the chief AI scientist of Tencent, all believe that AI Agents will make significant leaps this year. They will upgrade from being able to complete the workload of 1 - 2 days for humans currently to being automated tools capable of independently undertaking 1 - 2 weeks' worth of task flows.

In 2026, AI applications are evolving from being usable to being user - friendly. As the entry point for end - users, AI Agents will interact with the real physical world more frequently and are expected to become the infrastructure of the AI era.

01 AI Agents are Becoming a New Generation of Super Entrance

At the press conference of Qianwen App today, the official announced that it has opened the testing functions of daily services such as food delivery, shopping, air ticket, and hotel reservation to all users. Alibaba has also taken the lead in the Agent - style e - commerce track. For example, in the food delivery scenario, users only need to input "Order me a cup of milk tea" or "Order me two cups of coffee", and Qianwen can then utilize the flash shopping service capabilities of Taobao to complete location determination, merchant recommendation, order generation, and one - click payment, achieving the goal of "Just say it, and it will be delivered". According to evaluations, the system can automatically use coupons and handle complex situations such as order consolidation smoothly.

In fact, since last year, major global manufacturers have been accelerating the launch of their respective AI Agent products with unique advantages. In China, general or vertical AI Agents such as Manus of Monica, AutoGLM Meditation of Zhipu, Lovart, Flowith, and Genspark have been successively launched. Quark has also created a super AI Agent in the form of an "AI Super Box", with advantages in areas such as search, browsers, scanning, and question - answering. Abroad, companies like Apple, Google, and OpenAI also regard AI Agents as one of their key research focuses for the year.

Meanwhile, following Baidu's Qianfan AppBuilder, various AI Agent development platforms have also emerged. For example, the new - generation AI Agent development framework AgentScope 1.0 of Alibaba Tongyi Laboratory aims to solve the problems in the construction, operation, and management of AI Agents, providing a production - level solution covering the entire life cycle of "development, deployment, and monitoring". The AI Agent framework Youtu - Agent of Tencent YouTu Laboratory has also been officially open - sourced. The Agent platform "Kouzi Space" under ByteDance has been launched on the Apple App Store and Android app stores.

The year 2025 was called the "Year of AI Agents". AI Agents have officially moved from laboratories to the front line of applications, covering multiple vertical fields such as logistics and manufacturing, content creation, customer service, mobile assistants, office automation, software development, medical diagnosis, education and training, financial consulting, autonomous driving, and industrial manufacturing. By building an intelligent closed - loop through "decision - making (LLM) + memory + planning + tools", they are gradually reshaping the terminal interaction hub and becoming the core of a new generation of super entrances.

"AI Agents will boom with application scenarios." Shen Dou, the executive vice - president of Baidu Group, believes that one of the opportunities in 2026 lies in the productivity explosion brought by Agent Scaling. According to the latest data released by Baidu, since its launch one month ago, the super AI Agent Baidu Famo has covered fields such as logistics, manufacturing, and scientific intelligence. More than 2,000 enterprises have applied for trials, and it has also been implemented in frontier fields such as automobile design and R & D and the optimization of precision instruments in space stations.

Gartner also clearly pointed out that agent - type AI has become a key technological trend this year and in the future. The year 2025 was an important node for its mainstreaming. AI Agents are evolving from simple auxiliary tools to complex ecosystem capable of collaborative operation, profoundly changing the underlying logic of how enterprises handle complex tasks and make decisions.

With the resonance of technological maturity, market demand, and industrial ecosystem, AI Agents have also welcomed the spring of commercialization. A research report recently released by CCID Consulting shows that the market size of AI Agents in China reached 4.75 billion yuan in 2024, a year - on - year increase of 64.4%. It is expected to reach 7.84 billion yuan in 2025, with a continuous growth rate of over 60%, achieving double - digit growth for two consecutive years. By 2026, it will be close to 15 billion yuan.

02 Demonstrating Value in Multiple Scenarios, Continuous Evolution of AI Agents

Currently, the development of artificial intelligence is undergoing a significant transformation, evolving from generative AI centered around content generation to AI Agents centered around goal - driven operations. AI Agents will no longer be limited to content recognition and creation but will have stronger goal orientation, autonomous decision - making and planning capabilities, and real - time interaction capabilities with the environment.

The value of AI Agents in productivity, customer experience, business growth, marketing, and security has been realized in multiple scenarios such as manufacturing, finance, and healthcare, and they are undergoing an evolution from being auxiliary to making decisions, from the periphery to the core, and from partial to overall.

For example, the core pain point in the manufacturing industry is "production interruption", as equipment failures can lead to production line shutdowns and huge losses. AI Agents can monitor the real - time operation data of equipment and issue early warnings before failures occur, transforming traditional "post - maintenance" into "pre - prediction". There are cases where a leading manufacturing enterprise's predictive maintenance AI Agent reduced production downtime by 50% by monitoring equipment data in real - time and warning of potential failures in advance, saving the enterprise a huge amount of economic losses. In addition, production scheduling AI Agents can dynamically adjust production plans based on data such as order requirements, raw material inventory, and equipment status.

The financial industry is also an area where AI Agents are most suitable to intervene, mainly in scenarios with high repeatability and relatively clear rules such as customer service operations, risk control, marketing support, credit approval, and insurance claims. In these fields, the core role of AI Agents is not to replace decision - making but to improve overall operational efficiency. The "China Banking Customer Service Center and Remote Banking Development Report (2024)" released by the China Banking Association shows that the opening rate of AI customer service in banks has exceeded 60%, and 31% of banks have completed the deployment of large - scale models.

In the domestic financial market, different types of players have gradually formed differentiated paths. Large - scale players like Ant Group and Tencent Cloud rely more on the group's computing power, models, and platform capabilities to integrate AI Agent platforms into financial scenarios. Some vertical fintech companies choose to focus on single - business scenarios. For example, Qifu Technology launched an AI Agent for the credit process, using the LangGraph multi - agent collaborative framework to integrate multiple intelligent modules such as data query, knowledge Q&A, insights, and reports, automating complex data analysis processes and solving common pain points in retail banking credit such as communication, approval, and compliance supervision. Du Xiaoman also released the "Yuanli AI Platform" last year, reshaping the risk control system through large - scale models combined with data intelligence and building an AI - enabled risk control closed - loop that is "more accurate before lending, more transparent after lending, and more flexible in strategies".

In the healthcare industry, AI Agents have also appeared in multiple scenarios. In hospitals, relevant AI Agents have been jointly released by medical institutions and enterprises for patient services, auxiliary diagnosis and treatment, and hospital management. Outside hospitals, enterprises mainly engaged in Internet healthcare and health management have launched AI family doctors, AI health managers, and AI psychological counselors. In addition, AI Agents also cover fields such as scientific research and teaching, and drug development, and can even serve as digital employees in different functional positions of enterprises.

Gartner also predicts that in the next three to six years, expert - type AI Agents will rise rapidly. These AI Agents focusing on specific and complex work processes in different fields will further improve industry operational efficiency and decision - making accuracy.

03 Collaboration among Multiple AI Agents Becomes a Trend, Three Issues to Be Solved

The development plan for AI Agents has reached the highest decision - making level of the country. In the "Opinions on Deeply Implementing the 'Artificial Intelligence +' Initiative" issued by the State Council, for the first time, the future time nodes for the development of artificial intelligence were clearly defined at the national strategic level: by 2027, achieve in - depth integration with six key fields, and the penetration rate of intelligent terminals will exceed 70%; by 2030, complete full empowerment, and the intelligent economy will become the core growth pole; by 2035, China will fully enter a new stage of intelligent economic and social development.

However, it is undeniable that the application of AI Agents in China is still in its infancy, and the lack of collaboration due to fragmented development poses challenges. There are three main issues that need to be addressed in the future.

First, be vigilant against "fake AI Agents" in the market. Since last year, with the increasing popularity of the AI Agent concept, a number of "fake AI Agents" or "shell - wrapped AI Agents" have emerged in the market. Many companies have repackaged or relabeled traditional technologies and existing products as AI Agents and promoted them through marketing strategies, misleading users. These products may seem to be equipped with AI functions, but in fact, they are just a simple linear combination of large - scale models and RPA. The large - scale model is responsible for generating operation instructions, and RPA is responsible for executing preset processes. There is a lack of in - depth collaboration and dynamic adaptation between the large - scale model and RPA, and they are essentially still limited by traditional automated tools and not real AI Agents.

Second, break through technological bottlenecks and promote the construction of high - quality datasets to enhance the professional ability of AI Agents in solving complex scenarios. Currently, most AI Agents are still limited to adding basic planning capabilities and tool invocation (or function invocation) functions to large - language models, enabling them to break down complex tasks into smaller, executable steps. They can perform data analysis, trend prediction, and a certain degree of workflow automation, and can choose the right tools to complete tasks in simple scenarios. However, in the face of complex scenarios, their decision - making quality, professional depth, and technological capabilities are still insufficient.

For example, in the construction of datasets, on the one hand, it is necessary to break the "data silos" and "ecological barriers" to achieve global connection. On the other hand, it is necessary to transform various data from multiple sources, in multiple forms, and in multiple formats within enterprises into data assets that can support the subsequent training of AI models and the application of AI Agents. In particular, for industry - specific terms, jargon, and business logic in enterprise applications, it is necessary to extract implicit rules and values and convert them into knowledge to better adapt to professional scenarios.

Third, pay attention to the issues of technological ecosystem and collaborative standardization. The application of AI technology is accelerating, but the ecosystem construction is not yet mature. In particular, the lack of a unified full - link security testing standard for the security risks of single AI Agents makes it difficult to quantify and avoid potential problems, which also hinders the future collaboration and co - governance of multiple AI Agents. For example, the short - lived appearance of Doubao Mobile has sparked great controversy in the industry, leading to discussions on issues such as security, privacy, ethics, collaborative standards, and business models.

Generally speaking, driven by technological development and industry demand, enterprises in the industry will continuously increase the application of AI Agents, and application scenarios in various fields will be further expanded. Facing these more complex business scenarios, a single AI Agent is difficult to handle all tasks, which will also drive AI Agents to develop in the direction of "more autonomous, more intelligent, and more collaborative". At the same time, the formation of the AI Agent ecosystem will also reshape the business models and governance systems in the AI era.

This article is from the WeChat official account "Laika Think Tank" (ID: laikazk), written by Gong Yan and republished by 36Kr with permission.