Summary of Key Points
DeepSeek recently announced the recruitment for Harness-related positions, with plans to double the size of all departments and continue to intensify its focus on the AI Agent domain. The company has just completed a Series A financing round of 51 billion yuan, with investors including Tencent, JD.com, NetEase, and CATL, among others. Its development strategy is shifting from renting computing power to building its own data centers, and it is creating competitive barriers through industry collaboration, distinguishing itself from large model companies like Zhipu and Alibaba.
Breakdown and Interpretation
1. What exactly is Harness, which has become so popular in the AI community?
Harness can be considered the “housekeeping system” for AI models—it determines which tools an AI model can use (such as data retrieval, code writing), which resources it can access (like databases and the internet), how different AI assistants (Agents) can work together, and when to stop working. Without Harness, even the most powerful AI models are merely capable but do not know how to utilize their capabilities.
For example, if you ask an AI to write a market report,Harness would instruct the model to retrieve the latest data, analyze competitors, and organize the information into a document, while also automatically checking for errors. DeepSeek’s recruitment of Harness professionals is aimed at turning its models into practical Agent products (core formula: Model + Harness = Agent). However, there are very few people who understandHarness, which is why DeepSeek is urgently seeking candidates.
2. Why is DeepSeek making such a strong focus on Agents and Harness?
Agents represent the “battlefield” of the current AI industry—basically, they are AI assistants that can complete tasks independently (such as writing code or managing work processes). Both Zhipu and Alibaba Cloud are competing in this area: Zhipu has gained popularity with its AI Coding technology, which once had a market value of over one trillion yuan; Alibaba Cloud also emphasizes AI Coding and Agents as key areas of focus.
DeepSeek began planning for its Agent strategy in the second half of 2025, and this recruitment is another step forward. Harness is crucial for turning models into practical applications: without it, models cannot convert their capabilities into useful actions. For instance, while DeepSeek’s models can write code, they needHarness to guide them on how to identify issues, make modifications, and conduct tests.
3. What are the hidden synergies in its 51-billion-yuan financing round?
DeepSeek’s investors were carefully selected, each with a specific role:
- Tencent, JD.com, NetEase: Help DeepSeek apply Agents in e-commerce, social media, and office scenarios (for example, JD.com uses Agents for automated customer service, while Tencent uses them to optimize WeChat functions).
- CATL: Solve the issue of high energy consumption associated with AI computing power. CATL specializes in energy storage (such as photovoltaic storage), which can help DeepSeek reduce costs and ensure stable power supply.
- Huawei Chips: Collaborate with DeepSeek’s models (similar to how Google’s Gemini model works with its own TPU chips) to make the models run faster and more efficiently.
These collaborations will help DeepSeek establish unique barriers in the market.
4. Is it cost-effective to shift from renting to building its own computing power?
Most AI companies currently rent computing resources (such as servers from Alibaba Cloud or Tencent Cloud) because it is cheaper and more flexible. However, DeepSeek may decide to build its own data centers:
- For leading companies, building their own computing power is more cost-effective in the long run:
- Higher usage over time leads to lower costs (for example, buying 1000 servers and using them for 5 years is cheaper than renting them for the same period).
- Customization options are available (adjusting server configurations to optimize model performance).
- Avoiding external risks (such as sudden price increases or shortages in rental resources).
Although the initial investment is significant, it can lead to better resource control and cost savings in the long term.
5. Differentiation among large model companies: What makes DeepSeek stand out?
The paths taken by major AI companies are becoming increasingly distinct:
- Zhipu: Rising rapidly thanks to AI Coding, with a high market value.
- Alibaba: Developing everything from chips to computing power and from models to applications in-house.
- DeepSeek: Following a strategy of “industry collaboration + Agents + Harness + self-built computing power”—a more modest approach, but its ecosystem is gradually taking shape. For example, its partnerships with CATL for energy management and Huawei for chip technology could give it an advantage in the field of General Artificial Intelligence (AGI).
In summary, DeepSeek’s actions indicate its ambition to go beyond being just a model company; it aims to build a complete ecosystem from computing power to practical applications, securing a strong position in the AI competition.
(Note: The above analysis is for reference only and does not constitute investment advice.)