Track Hyper | Giant Race: The New Entry Competition of Intelligent Frameworks

Author: Zhou Yuan / Wall Street News

On September 2, Tencent officially open-sourced the intelligent agent framework Youtu-Agent.

Information from Tencent indicates that on the WebWalkerQA benchmark, the accuracy based on DeepSeek-V3.1 reached 71.47%, setting a new record for open-source models.

Alibaba, Tencent, and Microsoft are all working on open-sourcing intelligent agent frameworks, with Microsoft having done something similar two years ago. What exactly are their goals?

Wall Street News notes that different companies emphasize different keywords in their narratives: Alibaba emphasizes "developer-friendly," Tencent emphasizes "application landing," while Microsoft directly embeds the intelligent agent into the Office and Copilot ecosystem.

These seemingly scattered actions actually point to the same unproven future—intelligent agents may become a new digital entry point in the AI era.

The problem is that the value of this entry point has not yet been scaled and validated in real business operations.

Thus, the giants choose to first test the waters in the open-source community, handing the stage to developers and early users, while also positioning themselves: if the future indeed unfolds as currently anticipated, they will not lose the first-mover advantage.

Therefore, this competition for intelligent agent frameworks is destined to be more than just a simple technical contest.

Conservative Tencent and Aggressive Alibaba

The giants suddenly appear so "selfless" in opening up intelligent agent frameworks, seemingly generous in their stance. However, within the industry, open-source is often interpreted as a low-cost market entry strategy.

The biggest issue with intelligent agents currently is "whether there are real scenarios that can run them": from file management, paper retrieval to data analysis, existing frameworks provide many functions, but these functions are scattered across the daily work fragments of most enterprises and have not yet proven to significantly reduce costs or increase revenue.

In other words, the imaginative potential of intelligent agents is enormous, but there is a lack of large-scale empirical evidence. No enterprise is willing to make significant investments in unproven technology.

If the giants rashly push the framework into commercial use, they would bear extremely high uncertainty and opportunity costs.

Thus, open-source has become the most prudent approach: allowing developers to experiment, the community to incubate possible scenarios, and whoever's framework is used by more people will have the initiative in future standard competitions.

This is not merely "openness," but more like an insurance policy: it showcases technological presence, shifts risk, and can also delineate spheres of influence in an intangible way.

When the market matures and scenarios are validated, they can tighten ecological barriers again; such cases are not rare in history.

Tencent's newly launched Youtu-Agent is described as an open-source framework capable of supporting tasks such as local file management, data analysis, and paper research.

The official statement claims that based on the WebWalkerQA benchmark, this framework has set a new testing record—but there is no need to cheer for it, as this is just a routine action to showcase technological strength.

In fact, from the perspective of product positioning, Tencent's intention in launching this framework is not to "define a new entry point," but rather to conduct a cautious exploration Tencent's long-standing strategic preference has been to "deeply cultivate within its existing ecosystem," such as social networking, gaming, and enterprise services. In contrast to Alibaba's proactive exploration of AI organizational forms, Tencent's attitude towards intelligent agent frameworks appears more cautious.

For instance, this time, the features provided by Tencent are very practical—documents, data, and papers are all scenarios that researchers and developers would use. This choice avoids "over-promising," but also shows that Tencent places more importance on "first landing on real applications, then discussing ecosystem scale."

This resembles a form of risk hedging. Because if intelligent agents cannot be embedded on a large scale into daily workflows, it will be difficult to become a true platform entry point.

Therefore, rather than loudly proclaiming the future, it is better to find a thread of real demand in localized applications.

Compared to Tencent's caution, Alibaba's AgentScope 1.0 appears much more aggressive: emphasizing "development, deployment, and monitoring" for full lifecycle management, attempting to build a one-stop multi-agent development platform.

Such design goals reflect Alibaba's consistent strategy—entering from the platform level, creating an infrastructure, and then expanding a complete ecosystem around it.

In Alibaba's narrative, intelligent agents are not just tools; they may also become catalysts for new organizational forms (such as AI organizations). Alibaba hopes that through the framework, developers can build complex applications without worrying about operational and monitoring issues, leaving the complexity to the platform to solve.

In contrast to Tencent's cautious probing within its own ecosystem, Alibaba's choice represents a larger-scale bet.

In the past, Alibaba gained experience in promoting platform-based products through multidimensional explorations in cloud computing, collaborative office, and retail technology.

AgentScope 1.0 is thus packaged as a "universal framework," aiming to cover as many potential scenarios as possible. However, this broad strategy also means that Alibaba must wait for more scenarios to naturally emerge; otherwise, the platform's value is difficult to realize.

Microsoft: On a Different Path?

Compared to domestic giants, Microsoft's approach appears bolder and more confident, or perhaps more pragmatic and direct. This is mainly because Microsoft launched a similar framework two years ago.

Therefore, Microsoft now chooses to directly embed the capabilities of intelligent agents into the Office suite and Copilot.

The logic behind this approach is clear: users are already using Word, Excel, and Outlook daily, so embedding intelligent agents as enhanced features creates natural scenarios.

Microsoft is not in a hurry to compete for framework standards in the developer community; instead, it leverages its existing user base to turn intelligent agents into ready-made productivity tools.

For the vast number of Microsoft users—especially enterprise users—this path is actually more acceptable: enterprises do not need to learn an additional new framework, nor do they need to worry about deployment and monitoring; they can gradually experience the value of intelligent agents simply by continuing to use familiar software.

This is a way of "bringing ecology through applications." In contrast, Alibaba and Tencent prefer to first define developers (including individuals and enterprises) through frameworks, and then expect the natural growth of application scenarios Microsoft's choice effectively skips this uncertain phase. However, the risks it faces are actually greater than those of its domestic counterparts. Since Microsoft only provides a supplementary AI tool, companies can choose to use it or decide not to, so sunk costs do not need to be overly considered.

Wall Street Insights notes that behind this is Microsoft's resource endowment: the usage habits of hundreds of millions of users of productivity software, as well as Azure cloud as the underlying support.

Microsoft's framework advantage lies not in attracting developers through open source, but in the "strong binding" of its ecosystem—users have almost no room to escape.

However, this does not mean that Microsoft has no interest in open-source agent frameworks. As early as September 2023, Microsoft launched a similar agent framework—AutoGen: used for building and managing multi-agent systems.

The core idea of AutoGen is to allow multiple agents (which can be large language models, tool invocation modules, human participants, etc.) to collaboratively complete complex tasks through dialogue interaction.

These agents can have different roles and capabilities, such as some focusing on logical reasoning, some excelling at invoking external tools (like code execution, database queries), and some responsible for coordinating task processes.

Why do these giants choose to create agent frameworks?

On the surface, each company's reasons for launching frameworks are different: showcasing technical strength, serving developers, promoting scene implementation. But the deeper reason actually points to one question—where is the entry point?

If the future application world consists of countless agents, then whoever's framework becomes the de facto standard will have the opportunity to define interaction rules and allocate traffic entry points.

This is no different from the competitive logic of operating systems, browsers, and mobile app stores in the past.

Because of this, even though agents have not yet been widely embedded in enterprise practices, companies must lay out their strategies in advance.

Open source is the easiest approach to be accepted; frameworks are the most natural competitive carriers. These technological layouts may not immediately bring revenue, but they can seize the initiative in the formation of future order.

This also explains why, even knowing that the scenarios have not yet been validated, the giants still unanimously push the frameworks to the forefront.

Open source is not the end, but a prelude to a war for entry points.

The Real Value to B Has Yet to Be Realized

Although it seems that everyone is working hard, the excitement around agent frameworks does not equate to market maturity.

So far, no company has been able to present large-scale enterprise-level cases proving that agents can significantly reduce costs or improve efficiency. In fact, most applications are still in the demonstration or pilot stage.

This means that the current competition is more of a "discourse power game" rather than a commercial realization. The news effect brought by the release of agent frameworks may even be more valuable than real applications.

The giants are well aware that no one dares to assert that agents will definitely become the next entry point like mobile applications. But in a situation where risks and opportunities coexist, absence equals giving up.

Therefore, the current wave of agent frameworks is more like a strategic defense—first occupying a position (commonly known as "positioning"), and then seeing if the market will grow demand Therefore, in reality, this is both a precautionary measure and a necessity.

It is worth noting that in the current AI public opinion environment, "open source" and "breaking records" are often seen as breakthroughs.

However, from the actual situation, open source seems more like a posture: it merely conveys a signal of "we are present," but does not automatically imply industrial value.

For developers, the significance of Youtu-Agent is that it provides an experimental field to quickly test applications; however, Tencent is likely to focus more on which scenarios truly possess conversion value.

Whether Tencent's framework can continue to evolve depends on two factors: first, whether Tencent is willing to maintain long-term investment in the community, and second, whether it can address shortcomings in areas such as cloud computing, security, and compliance.

Without these two conditions, Youtu-Agent may become a "silent project" after a wave of popularity.

Alibaba bets on platformization, Tencent explores specific scenarios, while Microsoft enters through applications (in fact, Alibaba also has similar approaches, such as DingTalk, which is similar to Microsoft's Copilot, and even more comprehensive and deeply embedded); the three paths each have their own logic, yet converge on the same track: the agent framework may be the new entry point for the future.

To briefly explain the similarities and differences between DingTalk and Copilot: both Microsoft and Alibaba are actually pursuing "dual-line advancement": on one hand, there is the framework layer (Microsoft AutoGen, Alibaba's AgentScope), providing developers with the ability to build complex agents; on the other hand, there is the application/scenario layer (Microsoft's Copilot, Alibaba's DingTalk AI organization), directly embedding AI into existing workflows, allowing enterprise users and ordinary employees to perceive it immediately.

The difference between the two is that Microsoft leans more towards "gaining reputation from tool entry points," with AutoGen being an exploration at the developer level, and Copilot being the implementation in productivity scenarios, relying on Office's user base to naturally push AI in front of users.

Alibaba, on the other hand, is "walking on two legs," with AgentScope attempting to establish platform standards, while the AI organization experiments on DingTalk are essentially doing similar "embedded applications" as Microsoft, but the goal is internal management and collaboration within enterprises.

In other words, the strategies of Microsoft and Alibaba are not entirely different, but rather two interpretations of the same logic: they both aim to seize potential standards at the framework level while cultivating real demand at the application level.

The only difference is that Microsoft relies on a global office suite, while Alibaba depends on DingTalk's penetration rate in domestic enterprises.

Returning to the initial question, the value of the giants' covert battle for the agent framework entry has yet to be validated on a large scale. The giants choose open source to disperse risks and to seize potential dominance.

The true significance of open-sourcing the agent framework may not lie in the current technical functions, but in who can bind more developers and applications to their ecosystem before future standards are formed.

From this perspective, this competition is far from settled. The current wave of open source is merely the prelude to a new round of battles