36氪_让一部分人先看到未来

In Zukunft könnten Experten-Agents möglicherweise ein Niveau erreichen, auf dem es "Nur der Sieger gewinnt" gibt.

One-sentence Introduction

Based on its self-developed Multi-Agent architecture, it is an intelligent agent service provider that deploys solutions into the production processes of enterprises.

Team Introduction

Yang Jinsong (CEO): Former product/commercialization director at DAMO Academy, former head of ByteDance's Feishu AI, and former head of Amazon AWS's aPaaS platform. He led the launch of Alibaba Lingjie and Tongyi - Alicemind, and managed product revenues exceeding 2 billion yuan.

Dr. Wang (Chief Scientist): A graduate from Columbia University. He was a research scientist at Alibaba DAMO Academy and Google Research, with 12,000 citations on Google Scholar.

Other core team members come from Alibaba DAMO Academy, Tencent, ByteDance, Amazon AWS, Google, etc.

Financing Progress

Recently, it completed an angel - round financing of tens of millions of yuan. The lead investor was Linge Venture Capital, followed by Jimen Asset Management, and the old shareholder, Sinovation Ventures, continued to participate. This round of financing will be mainly used for product R & D and market expansion.

Product and Commercialization Status

In the domestic market, AutoAgents uses the "Lingda" platform (Agent Builder) to meet the market demand for enterprise - level agents. It solves problems such as data security, permission management, and system integration that enterprises are concerned about during the application of large models and provides mature technical solutions.

Currently, AutoAgents' products have served leading customers in industries such as electricity, finance, the general Internet, and manufacturing. It builds a hybrid human - AI workflow for enterprises, greatly improving their operational efficiency and innovation vitality. The company obtained tens of millions of commercial contracts in 2024. Currently, it is the Agent application product with the highest market share in the electricity industry, serving the State Grid and more than a dozen of its subsidiaries.

The company's products have been introduced by 5 cloud providers, offering more than 100 open - source and domain - specific models, and are being promoted on a large scale through more than 20 industry partners.

In the overseas market, AutoAgents has also launched a standardized product, Agents Pro, a social media operation tool for SMBs, adopting a free - trial and community - spread model.

Different from Agent platforms such as Coze and Dify, which are more consumer - oriented, AutoAgents' differentiation lies in its focus on the enterprise - level market. AutoAgents does not simply provide tools. Instead, through standardized Agent products and industry solutions, such as providing more granular permission control, data dashboards, and database docking capabilities, as well as hybrid - cloud/All - in - One deployment solutions, it helps enterprises deliver service results and achieve "pay - for - results".

To be enterprise - level ready, AutoAgents integrates components such as enterprise - level RAG, AI Coding, Text2Agent, and visual workflows. In terms of deployment, AutoAgents supports hybrid - cloud/All - in - One deployment and can adapt to domestic computing power.

At the product's core, Lingda introduces a unique multi - agent collaboration mechanism, which can solve the context window limitation in the multi - step reasoning process. It supports the generation of Agent applications from a single sentence, greatly improving development and deployment efficiency.

During task execution, Lingda can break down complex tasks and assign them to different professional agents, with a coordinating agent overseeing the overall work.

Currently, Lingda can also help humans with "real work". AutoAgents supports Anthropic's MCP protocol, enabling agents to more efficiently discover and call external tools.

Lingda can also simulate human operations (similar to OpenAI Operator). With the built - in Docker sandbox system, agents can independently browse the web, retrieve data, and call common software to complete specified tasks.

After the DeepSeek craze this year, the market generally believes that this is a crucial moment for the "Year of Agents". AutoAgents has also developed the "Yuanzhi" assistant for developers and individuals, which will be launched in 2025. It is an autonomous agent product that can run in real - world business scenarios, capable of self - planning and completing research and analysis tasks in professional fields.

Source: AutoAgents

With Agent fine - tuning technology, AutoAgents has been able to enhance agents' tool - calling capabilities, optimize collaboration efficiency, and improve code generation quality. AutoAgents has accumulated more than 20 patents, software copyrights, and other intellectual property achievements in this field and has published papers at multiple international top - tier conferences.

After this round of financing, AutoAgents will continue to achieve rapid commercialization and will launch products for the consumer market. It also plans to expand into overseas markets.

Founder's Thoughts

Yang Jinsong

• The core difference between enterprise - level agents and personal agents is that the former need to meet the strict requirements of enterprises in terms of data security isolation, hierarchical permission systems, and in - depth system integration. General agents focus on ease of use and universality, while enterprise - level agents require in - depth customization to adapt to complex business scenarios.

• Currently, the implementation of Agent technology still faces many challenges. Even though inference models such as DeepSeek R1 have powerful capabilities, in actual enterprise applications, a large amount of engineering transformation is still required, in - depth adaptation with existing toolchains, and integration of small models in the field to effectively control hallucinations and ensure the reliability and safety of output results.

• The final outcome of Agents may present a "winner - takes - all" situation. In specific vertical fields, agents that can effectively accumulate industry know - how, precipitate best practices, and conduct in - depth training with high - quality data will build deeper competitive barriers.

• The future development direction of Agents lies in the in - depth collaboration between AI experts and industry experts to re - plan enterprise workflows, automate complex work processes, and enable humans to focus on high - density decision - making and responsibility - taking. AutoAgents will develop into a "1 + N" business model in the future, that is, produce various Agent products and solutions through a technical platform, implement pay - per - service - volume, and obtain continuous revenue from the global service value chain, which can break through the revenue ceiling of traditional software sales.

• Enterprise software is shifting from "tool - based payment" to "result - based payment". The core value of Agents lies in being service - result - oriented. Through standardized Agent products and industry solutions, it directly creates business value rather than simply providing tools. The fundamental purpose of enterprises choosing Agents is to solve actual business problems, rather than simply paying for models.

• Inference models represented by DeepSeek R1 have significant value in expanding Agents' ability to solve open - ended problems, especially suitable for C - end applications such as code writing and novel creation. However, in enterprise - level applications, it is still necessary to carefully evaluate their applicability in scenarios, security alignment capabilities, and hallucination control levels to avoid having overly high expectations for the model's capabilities.

What "Intelligent Emergence" Wants to Say:

Agents have become an undeniable topic in 2025, but the industry is still in its early stage. AutoAgents' core competitiveness lies in its ability to achieve usability in enterprise - level production environments. It has been successfully implemented in large - enterprise scenarios with low fault - tolerance rates, such as electricity and finance.

AutoAgents has a clear understanding of the capabilities and boundaries of current large models. In different industries, it has enterprise - level Agent products, and in the overseas market, it also has products for emerging business scenarios such as community marketing. The company not only entered the market early but is also rapidly seeking commercial implementation. Both its product matrix and market strategies are relatively clear.