Source: BohuFN

From general-purpose large models to industry-specific large models, the new frontier of AI is now focused on AI agents, as AI begins its journey from the pedestal to the masses.

Recently, several tech giants have unveiled AI applications centered around "agents." Ant Group launched three new AI products and introduced an independent AI-native app called "Zhibaoxiao," while also initiating a plan to build an AI agent ecosystem on Alipay and releasing the "Treasure Chest" AI agent development platform.

Tencent's Yuanbao brand's AI agent section officially went live, inviting 11 partner companies such as Tongcheng, Weimob, and CR Sanjiu to create high-quality AI agent applications. OpenAI is also rumored to be planning to launch a new AI codenamed "Strawberry" this fall, which is significant for further developing AI agents.

The new moves by tech giants seem to be sending a clear signal: in the process of AI commercialization, "agents" serving the general public are becoming a widely favored application direction.

As AI agents gain traction, competition around "applications" will become increasingly fierce. How can different companies position themselves competitively? What direction should we strive towards to build a "super ecosystem" of AI agents?

01 Tech Giants Embrace AI Agents

With the popularity of ChatGPT, the concept of "AI agents" has garnered widespread attention in the industry. These agents, akin to the intelligent assistants seen in science fiction, can execute tasks based on human instructions, conduct autonomous understanding through search, analysis, and research, and continuously optimize feedback content.

However, AI development cannot be achieved overnight. Over the past year, major tech companies have raced to develop general-purpose large models. To date, there are over 100 large models in China with over 1 billion parameters.

After the "hundred-model battle," tech companies faced the same dilemma: finding suitable application scenarios for general-purpose large models is not easy. As homogenized applications increase, they can even become somewhat redundant.

As a result, tech companies have accelerated their exploration of AI agents. Last year, Bill Gates discussed his vision for AI, mentioning that agents will change the way people interact with computers. In the future, people will no longer need to use different apps for different tasks; they can simply tell their devices what they want to do using natural language.

Robin Li, Chairman of Baidu, has also frequently mentioned "AI agents" in public, stating that they represent the future trend of generative AI. Not only can they engage in dialogue, but they also possess reflective and planning capabilities.

From general-purpose to industry-specific large models, tech companies aim to empower industries and scenarios with large model capabilities. However, accurately understanding the needs of enterprises and users is not easy, and inadvertently, it can become "innovation for innovation's sake."

The emergence of AI agents, however, signifies that tech companies are starting to take the initiative. Instead of focusing on uncovering demands, they are using AI to enhance user needs, allowing users to articulate their demands and then fulfilling them.

Currently, AI agents have been implemented in various industries such as education, finance, and healthcare. Mark Zuckerberg once predicted that there could eventually be billions of AI agents, potentially outnumbering humans.

While there are many players in China's AI agent field, the competition among tech companies is clear: large enterprises compete for ecosystems, while smaller enterprises vie for applications.

On one hand, major companies like Baidu, Alibaba, Tencent, and ByteDance have launched one-stop AI agent development platforms, such as Alibaba's "Cheese Cake AI," ByteDance's "Kouzi," and Baidu's Wenxin AI Agent Platform.

Due to their extensive user bases and scenario matrices, internet giants can leverage their strengths in content ecosystems to provide channels for AI agent development and monetization, further enriching the AI agent ecosystem.

Recently, Baidu released its first AI agent alliance solution, claiming to have established a closed-loop AI agent ecosystem that integrates agent distribution and monetization. According to Robin Li, Baidu's Wenxin AI Agent Platform has attracted 100,000 enterprises and 600,000 developers. Currently, 18% of search results are generated by AI, and the daily distribution volume of AI agents has surpassed 10 million.

On the other hand, large model startups like Zhipu AI and Mianbi AI focus on AI agent applications in vertical scenarios.

For example, facing the consumer market, Moon's Dark Side has launched Kimi, which excels in multilingual dialogue and long-text processing. For the business market, Mianbi AI aims at intelligent terminals like smartphones and cars with MiniCPM. Deep Engine Technology focuses on building an AI agent development platform tailored to the financial industry.

From a practical perspective, most consumer-facing AI agents are positioned as intelligent assistants, offering conversational services and personalized features. Business-facing AI agents, on the other hand, need to provide more professional and customized services, including data annotation and model fine-tuning, to meet specific enterprise needs in operations, management, and production.

Currently, some large model startups are attempting to create AI agent stores or platforms. Kimi launched the Kimi+ AI Agent Store in May this year, featuring over 20 official AI agents. DingTalk released an AI assistant marketplace where users can create AI agents based on their ideas and needs, with over 700 AI assistants available within just one month.

However, compared to internet giants, AI agent stores or platforms run by large model startups tend to focus more on application scenarios they are familiar with. They collect user data feedback through AI agents while attracting new traffic, essentially helping large models find commercial monetization pathways.

Therefore, if AI agents become the websites of the AI era, with numerous agents forming a new AI ecosystem, this "super ecosystem" is likely to flourish within internet giants. Leveraging years of accumulation and layout , internet giants possess not only substantial traffic but also rich ecosystem matrices and extensive scenario deployments. They can provide strong data support and application scenarios for AI agents, accelerating their market promotion and user acceptance, ultimately fostering a smarter, more interconnected, and efficient super ecosystem.

02 Ant Group Takes the Lead in the "Ecosystem Battle"

As a result, internet giants are focusing on building AI agent ecosystems. For instance, Ant Group recently launched AI life assistants such as "Zhibaoxiao," an AI-native app focused on services, and AI financial and health assistants, along with the AI agent development platform "Treasure Chest" on Alipay.

In the mobile internet era, user demands are met through various apps and mini-programs, with tech companies relying on app development for profit. However, the trend towards "lighter" apps and mini-programs indicates that users do not necessarily need numerous apps and may even grow tired of being "enslaved" by them.

With the continuous development of the AI industry, more natural human-computer interaction methods will emerge, and AI agents will serve as the bridge. Apps will be decomposed into smaller services, with AI agents assuming the role of intelligent coupling. In this context, AI agents are poised to become new traffic entry points, making their early deployment crucial for the commercial ecosystems of major enterprises.

Taking Ant Group's AI applications as an example, they are built upon the Alipay ecosystem but can further empower Ant Group to construct a more comprehensive AI agent ecosystem. The combination of the two is a "1+1>2" scenario.

Firstly, AI applications centered on Alipay possess stronger scenario distribution capabilities. Alipay serves billions of users across various scenarios such as transportation, government affairs, healthcare, and finance. Data shows that 600 million users use Alipay for medical consultations, and 500 million for daily travel. These will become landing scenarios and a vast user traffic base for Ant Group's AI agents.

Taking "Zhibaoxiao" as an example, it connects with Alipay's digital lifestyle ecosystem, enabling users to quickly book tickets, order food, hail taxis, and search for nearby entertainment options by conversing with "Zhibaoxiao," thereby bringing AI into people's daily lives.

Secondly, Ant Group provides sufficient computing power to support its AI agent ecosystem. Recently, Ant Group announced the latest progress in its self-developed Bailing large model, which can directly understand and train multi-modal data including audio, video, images, and text. Through multi-modal models, it realizes ACT technology, enabling AI agents to possess certain planning and execution capabilities.

Within the industry, native multi-modality is considered a necessary path towards AGI, enabling AI to better comprehend the complex information of the human world and align with human interaction habits during application.

Thirdly, Ant Group leverages its efficient connectivity to link AI agents with real-world business services. Ant Group's digital financial business can be considered the "capillaries" of various industries, boasting mature business mechanisms and closed loops. AI agents can leverage these channels to enter the physical business world, fulfilling personalized and deeply customized scenario and ecosystem needs.

Of course, Ant Group is not the only internet giant with scenarios, computing power, and connectivity capabilities. Each company has its unique advantages.

For instance, Baidu excels in search and is closer to traffic entry points. Robin Li once stated that search is the primary entry point for AI agent distribution. In recent years, ByteDance has also actively deployed in local lifestyle and finance, attracting substantial user traffic with its content advantages.

Therefore, although the new battle for AI agents begins at traffic entry points, the ultimate test lies in enterprises' ecosystem operation capabilities. Companies need sufficient scenario traction at the front end, smooth scenario coordination within, innovative models externally, and continuous user experience optimization to become excellent "traffic entry points" in the era of AI agents.

03 Easy Deployment, Challenging Breakout Hits

While tech companies strive to become the "App Store" of the AI agent era, "super apps" remain crucial, as they bring richer traffic value to the AI agent ecosystems of major enterprises.

However, while deploying AI agents is relatively easy, creating breakout hits is challenging. Getting AI agents to be truly "used" and achieve commercial monetization poses a significant challenge for enterprises developing AI agents.

From the B2B perspective, numerous AI agent applications have been deployed in vertical fields such as finance and healthcare in recent years. In finance, AI agents can manage wealth and assess risks. In healthcare, they assist in intelligent diagnosis and image interpretation.

However, there are still many challenges in exploring commercial models. Firstly, the low tolerance for errors in healthcare and finance makes the accuracy of large models a significant obstacle to adopting generative AI in the finance industry.

Secondly, the business logic of vertical AI agents differs from that of large models. While the former focuses on solving problems in vertical industries, the latter creates demand and attracts traffic. These different business logics often limit the development of vertical AI agents, making it difficult for them to move from niche markets to general markets and resulting in significant uncertainty in profitability prospects.

On the B2C front, AI agents' functionalities are still relatively limited. Although some AI agents demonstrate practicality in specific scenarios, they primarily rely on language models and risk becoming mere "chatbots," failing to showcase overwhelming advantages over traditional apps or mini-programs.

Current AI agents must continuously enhance their "thinking and understanding" capabilities, achieving substantial breakthroughs in content depth and interaction quality to meet the diverse and demanding needs of scenario integration.

Therefore, while creating AI agents currently has a low barrier to entry – with users able to create dozens in a single day – they cannot yet replace traditional app markets.

Perhaps, "super apps" are still in the making, but good soil is essential for their growth. Currently, major enterprises are actively building AI agent ecosystems, partly to explore their commercial closed loops and bridge technology empowerment, scenario application, and traffic distribution. Only by finding commercialization possibilities can AI agents ensure a promising future, which is a crucial prerequisite for further fostering the AI agent ecosystem.

As the AI agent ecosystem enriches, AI agents will possess better collaboration capabilities. After all, the basic abilities of individual AI agents, such as understanding, generation, logic, and memory, have inherent limitations. By deploying AI agents in scenario collaboration, they can potentially evolve further.

From this perspective, while Ant Group's AI applications may not be the "top-tier" in the market, their pioneering deployment in scenarios signifies greater evolutionary potential.

Currently, China's AI agent market is rapidly evolving, with the participation of diverse enterprises injecting vitality into the market. However, when large model capabilities will advance to endow AI agents with true imagination and creativity remains to be seen – practice, as they say, is the sole criterion for testing truth.

*The cover image and accompanying images belong to their respective copyright holders. If the copyright owners believe that their works should not be made available for public browsing or free use, please contact us promptly, and we will make immediate corrections.