Hello! I'm your financial analysis assistant. This article from "Tsinghua Management Review" explores a very core and urgent question: As AI agents become smarter and more capable, will familiar platforms like Taobao, WeChat, and Douyin be replaced by AI, or will they evolve together with it?
The article's main point is clear: We shouldn't think in a black-and-white binary framework. It's not about AI completely disrupting the platforms, nor about the platforms and AI always working in harmony. The future landscape will be one of "vertical stratification and dynamic competition."
To help you understand this better, I've broken down the long article into five key points and explained them in simple language.
---
1. What is an "agent," and how is it different from previous robots?
In the past, we talked about "intelligent customer service" or "automated programs," which were just advanced tools that followed rules. If you asked about return policies, they would look up the information in a database and provide you with an answer.
Today's agents, especially those based on large language models (LLMs), are more like "thinking employees":
- They can perceive: They understand your vague requests (for example, "Find me a sports jacket for autumn that costs under 500.")
- They can make decisions: They decide where to search, how to compare options, and which one to choose.
- They can execute: They can automatically open browsers, compare prices, place orders, and make payments—all without your direct intervention.
The article categorizes agents into three types:
- Cognitive agents (strategists): Good at thinking, analyzing, and suggesting solutions. Examples include medical diagnosis assistants and investment advisors.
- Interactive agents (butlers): Good at communication and coordination. Examples include autonomous driving systems and advanced customer service.
- Executive agents (workers): Good at performing tasks and automating processes. Examples include warehouse robots and form-filling tools.
The key point: In reality, a good agent often combines multiple roles, being able to think, communicate, and act.
---
2. Why is there such a heated debate? The "disruption camp" vs the "coexistence camp"
There are two main views in the industry about the relationship between AI agents and traditional platforms (like Taobao or Microsoft Office), similar to the debate over whether phones will replace computers:
🔴 Disruption camp (AI is the killer):
- Logic: Used to, you had to search on a platform like Taobao to buy something; now you just ask an AI, and it buys it for you.
- Consequences: The platform loses its traffic sources and data control. Data from user interactions with AI is either stored locally or with the AI company, leaving the platform without valuable information, which weakens its algorithms and potentially disintegrates it.
- Example: Character.AI allows you to chat directly with AI characters without going through social media platforms.
🟢 Coexistence camp (AI is a helper):
- Logic: Developing a powerful AI is too expensive for most individuals or small companies. Large platforms have the money, data, and users, so AI can be integrated into them to make them more useful.
- Consequences: The platform provides the infrastructure (computing power, data), and AI offers intelligent services, benefiting both parties.
- Example: Microsoft's Copilot, integrated into Teams and Outlook, doubles the efficiency of writing emails and holding meetings, making Microsoft more profitable and making users more dependent on its services.
The core conflict: The disruption camp fears that platforms will be "de-intermediated" (losing their profit margin), while the coexistence camp sees platforms as essential infrastructure (like water, electricity, and gas) that AI cannot do without.
---
3. What's the truth? Platforms are no longer just "markets," but "multi-layered ecosystems"
The article argues that the debate stems from a narrow understanding of platforms.
- Old thinking (horizontal, single-layer): A platform is simply a marketplace where buyers and sellers meet, and the platform makes money by monopolizing traffic. In this view, AI seems like a new vendor at the market entrance, taking away customers.
- New thinking (vertical, multi-layer): A platform is a multi-layered ecosystem:
- Foundation: Computing power, chips, cloud storage (e.g., Alibaba Cloud, NVIDIA).
- Middle layer: Large models, general algorithms (e.g., Tongyi Qianwen, GPT).
- Upper layer: Specific applications, agents, user interfaces (e.g., DingTalk, Taobao app, autonomous driving software).
In this new perspective: AI agents don't aim to destroy the ecosystem; instead, they could become a key component of it.
- Example: Huawei's Ascend ecosystem: The foundation is its chips, the middle layer is the Pangu large model, and the upper layer includes applications like mining and healthcare.
- Value is distributed across all layers.
⚠️ Beware of false harmony: Coexistence is not static. Once underlying technologies (like chips or models) advance, they can impact upper-layer applications (e.g., OpenAI providing APIs that allow developers to bypass certain platforms); upper-layer applications (e.g., Tesla) may integrate lower-layer technologies for better experiences (e.g., by developing their own chips).
Conclusion: It's a dynamic process where control over key layers determines who has the power.
---
4. How can different types of companies find their place in this multi-layered ecosystem?
As the ecosystem changes, so do the competitive strategies for companies in different positions. The article offers advice for four types of players:
🏗️ Infrastructure providers (like NVIDIA, Huawei):
- Challenge: Technology evolves rapidly; how to retain existing users?
- Core competitiveness: Software-hardware integration + smooth upgrades.
- Example: NVIDIA's CUDA: Developers are accustomed to its software environment, and switching to a new chip is costly. The new chip must be compatible with existing software to ensure a smooth upgrade experience.
- Key: Create a high switching cost so the ecosystem relies on your products.
🧠 General large model developers (like DeepSeek, OpenAI):
- Challenge: Trying to be good at everything leads to limited impact; how to focus?
- Core competitiveness: Modular design + strategic focus.
- Technology: Design models as modular building blocks with general capabilities and specific plugins for different scenarios.
- Strategy: Focus on high-value areas (e.g., coding, finance) and provide tools for partners in less critical areas (e.g., education, entertainment).
- Key: Balance versatility and specialization.
🛠️ Vertical application and agent developers (like ShuiZhu Intelligence, industry SaaS):
- Challenge: How to ensure your applications remain relevant as models change?
- Core competitiveness: Unique data + rapid optimization.
- Example: Design AI that understands industry-specific preferences and standards.
- Build a feedback loop: The more users use your app, the more accurate your data becomes, and the better the model performs, making it harder for competitors to replicate.
- Key: Convert expert knowledge into usable data for the models.
🏢 Multi-layer platform ecosystem operators (giants like Alibaba, Tencent, Amazon):
- Challenge: Managing multiple layers and capturing value.
- Core competitiveness: Promote cross-layer interactions + control key assets.
- Example: Alibaba Cloud supports Tongyi models, which makes DingTalk more useful, and DingTalk users generate data that feeds back into Alibaba Cloud.
- Key: Identify and control critical assets (e.g., data, computing power, protocols). These are often the most scarce and difficult to replace, and they give you a dominant position in the ecosystem.
---
5. Practical advice for leaders:
The article concludes with three management tips:
1. Reframe your thinking: Stop focusing solely on traffic. Build a multi-layered ecosystem based on computing power, data, models, and applications.
- Example: Traditional giants like Alibaba have advantages but also face challenges. They need to be open to external innovation and avoid becoming complacent.
2. Find your niche: Most companies can't cover all layers. Focus on your core strengths.
- Example: If you make chips, focus on software-hardware integration; if you develop models, focus on modular design; if you make applications, focus on unique data.
- Tip: Avoid blind diversification. Forging ahead in unrelated areas often leads to failure.
3. Track evolving trends: Identify the most valuable assets. What will be scarce in the future—high-quality data in a specific industry or standards that connect systems?
- Example: Anticipate and occupy valuable positions before others do, as this can give you a competitive advantage.
In summary: The future business world won't be about AI vs platforms but about a competition of multi-layered ecosystems. Those who understand their position and control the key layers will succeed. For individuals, this means smarter, more personalized tools; for companies, it means that "data assets" and "ecological niches" are more important than just traffic.