AI 泡沫破灭之后——智能体分身网络落地探微
After the AI Bubble Bursts: A Close Look at Agent Avatar Networks in Practice
小满智能 · 2026-10-16 · v1.0
一、资本浪潮下的 AI 热点：数学会被替代吗？
1. AI Hype in the Capital Wave: Will Mathematics Be Replaced?
国外这些靠资本驱动的 AI 公司，不断在互联网上炸出各类头条新闻。最近热度最高的一个说法，就是数学会被 AI 替代。真正的 AI 落地，会在泡沫破灭之后，才脚踏实地融入社会生产、各类工作场景当中。
Capital-driven AI companies abroad keep producing all kinds of headlines on the internet. The hottest claim lately is that mathematics will be replaced by AI. Real AI adoption will only take root in production and in all kinds of work scenarios after the bubble bursts.
数学，确实是 AI 擅长的主场。AI 最不擅长的，其实是物理世界，面对真实宇宙它能力有限。它只能辅助、加速我们，帮我们构思解决方案。在物理领域的探索里，AI 能提供的帮助存在上限；物理实验、现实验证，天然就是 AI 的短板，即便 AI 可以提速，提升幅度也是有限的。
Mathematics is indeed the home ground where AI excels. What AI is least good at is the physical world; facing the real universe, its abilities are limited. It can only assist us and speed us up, helping us conceive solutions. In exploring the physical world, the help AI can offer has a ceiling. Physical experiments and real-world verification are naturally AI's weak spot, and even if AI can speed things up, the margin of improvement is limited.
数学领域则不太一样。数学本身是纯数理构建的理想建模世界，AI 诞生的根基就是数学，所以 AI 处理数学问题会相对擅长。但这不代表未来的数学家会失去价值。反过来看，我认为新一代数学家肩负着重大的历史使命，这是一场数学领域的变革。
The field of mathematics is different. Mathematics itself is an ideal modeling world built purely from mathematical structures, and mathematics is the very foundation from which AI was born, so AI is relatively good at handling mathematical problems. That does not mean future mathematicians will lose their value. Looked at from the other side, I believe the new generation of mathematicians carries a major historical mission. This is a transformation in the field of mathematics.
二、人机协同：搭建数学领域的智能体网络
2. Human-AI Collaboration: Building an Agent Network for Mathematics
数学，是最容易实现人机合一、搭建智能体网络用来探索与发展数学的标志性领域。对数学家而言，这个领域大有可为。传统数学家的工作模式将会被打破，必须借助智能体。人类本身信息输入输出的能力存在局限，需要把数学家组织起来，和 AI 协同开展探索，建立一套全新的数学探索范式。这和我之前讲的整体框架完全可以对应。
Mathematics is the signature field in which human-machine unity is easiest to achieve, and in which agent networks can be built to explore and develop mathematics. For mathematicians, this field offers a great deal of room. The traditional working mode of mathematicians will be broken; they must rely on agents. Human capacity for information input and output is limited, so mathematicians need to be organized to explore in coordination with AI, establishing a brand-new paradigm for mathematical exploration. This corresponds exactly to the overall framework I have described before.
AI 可以帮我们打开思路，生成各式各样的解决方案，但方案真伪与合理性，最终还是要依靠人类完成验证。数学家和 AI 工程师协同，开发适配人类思维输入输出带宽的 AI 智能体分身网络，把数学家的专业素养、基础能力封装成一个个智能体分身，组成智能体分身网络，驱动一个专门解决数学问题的大型 AI 集群。
AI can help us open up our thinking and generate all kinds of solutions, but whether a solution is true and reasonable must ultimately be verified by humans. Mathematicians and AI engineers work together to develop AI agent avatar networks matched to the input and output bandwidth of human thinking, packaging the professional competence and foundational skills of mathematicians into individual agent avatars. Together they form an agent avatar network that drives a large AI cluster dedicated to solving mathematical problems.
三、可复制的通用架构：主控智能体网络
3. A Replicable General Architecture: The Master Agent Network
这套模式，也可以复制到所有适合 AI 落地的行业。核心是搭建一套主控网络，也就是由众多智能体分身组成的网络，它蒸馏沉淀各行各业专家的知识，形成人类专家分身网络。人类必须参与主控网络的管控、优化与迭代。依靠这套主控网络，驱动 AI 集群开展探索。
This model can also be copied to every industry suited to AI adoption. The core is to build a master network, a network made up of many agent avatars that distills and accumulates the knowledge of experts from every industry, forming an avatar network of human experts. Humans must take part in the management, optimization and iteration of the master network. Relying on this master network, AI clusters are driven to explore.
融合人类专家经验的主控智能体网络，能够快速核验 AI 集群产出的结果，持续校正探索方向，引导 AI 朝着我们想要攻克的问题去研究。主控分身网络本身具备一定自主性，部分场景下可以自主发起决策、开展探索；探索结果回流到网络内部做核验。如果网络无法完成验证，就交由人类专家组成的人工网络复核，为主控网络提供验证支撑。
A master agent network that incorporates the experience of human experts can quickly verify the results produced by AI clusters, continuously correct the direction of exploration, and guide AI toward the problems we want to tackle. The master avatar network itself has a degree of autonomy; in some scenarios it can initiate decisions and exploration on its own, and the results flow back into the network for verification. If the network cannot complete the verification, the matter goes to a human network made up of experts for review, providing verification support for the master network.
这里最核心的需求，就是专业知识校验能力。放到数学领域，就是提出问题、拆分决策分支、完成结果验证的能力。未来人类要解决的，就是规模化验证、规模化决策的难题。这也是本轮 AI 泡沫退去之后，百亿级智能体网络集群组网落地的核心目标。
The most central requirement here is the ability to validate professional knowledge. In the field of mathematics, that means the ability to pose problems, divide decision branches, and complete result verification. What humanity needs to solve in the future is verification at scale and decision-making at scale. This is also the core goal for landing networked clusters of ten-billion-scale agent networks after this round of the AI bubble recedes.
目前我构想的是这套基础模式，同时我们也可以借助 AI 智能体，去探索更高效的组网与协作方式。这正是组织能力、验证能力、决策能力的一个范例。
What I have conceived so far is this basic model, and at the same time we can use AI agents to explore more efficient ways of networking and collaborating. This is a good example of organizing capability, verification capability and decision-making capability.
四、长期主义：我们的科研底色
4. Long-Termism: The Foundation of Our Research
最后，我想用《卜算子·咏梅》来收尾。新一代科研从业者、新一代数学家，其实已经在用这套模式开展工作了。
Finally, I would like to close with the poem Song of Divination: Ode to the Plum Blossom. The new generation of researchers and the new generation of mathematicians are in fact already working with this model.
我们中国人，向来擅长长期主义与持久战。西方这套模式更看重短期收益，容易急功近利。这不是说他们的探索没有价值，但他们很难坚持长期研究。而我们，习惯沉下心默默耕耘。
We Chinese have always been good at long-termism and protracted effort. That model in the West values short-term returns more and easily becomes eager for quick success. This is not to say their exploration has no value, but they find it hard to persist in long-term research. We, by contrast, are used to settling down and working quietly.
毛泽东主席这首词写得极好，体现出不为名利所动的大无畏精神：
Chairman Mao Zedong wrote this poem superbly, expressing a fearless spirit unmoved by fame and gain:
已是悬崖百丈冰，犹有花枝俏。
俏也不争春，只把春来报。
Even at a hundred-fathom cliff of ice, a branch still blooms beautifully.
Beautiful as it is, it does not vie for spring; it only announces spring's arrival.
我用这几句词，形容国内这些扎根实践、利用 AI 智能体探索数学以及各基础学科全新研究范式的一线工作者。他们就是这样，默默无闻、勤恳深耕。
I use these lines to describe the frontline workers in China who are rooted in practice, using AI agents to explore a brand-new research paradigm for mathematics and the other basic disciplines. That is how they are: unknown, diligent and deep in their work.
俏也不争春，只把春来报。
待到山花烂漫时，她在丛中笑。
It does not vie for spring; it only announces spring's arrival.
When the mountain flowers are in full bloom, she smiles among them.
这是多么崇高的境界。等到 AI 真正落地、智能体时代百花齐放的时候，很多人或许不会记得这些深耕底层、默默为世界做出贡献的一线工作者。
What a lofty realm this is. By the time AI truly lands and the age of agents flourishes in full variety, many people may not remember these frontline workers who have plowed deep into the foundations and quietly contributed to the world.