DeepSeek Releases New Update During National Day Holiday, Post-90s Lead Responds: Aiming for Innovations That Global AI Giants Will Follow

Deep News
Yesterday

After a new version released during the Spring Festival took the world by storm, DeepSeek has been "working overtime" again, rolling out another update during the National Day holiday. On October 3, the agent framework of China's leading large model developer DeepSeek, DeepSeek Harness, released a new version that strengthens plugin capabilities and cross-platform support, and adds an experimental Claude Code Mods compatibility layer. This means that some extension features developed by U.S. AI giant Anthropic's Claude Code could, in theory, run within the DeepSeek Harness framework after adaptation.

The update quickly drew attention in technical communities, and on October 4, DeepSeek Harness lead Cui Tianyi published a post in response. He said the main purpose of building this compatibility layer was to verify whether the extension capabilities that Claude Code Mods provide to plugin authors are roughly a subset of what the DSH "everything is a plugin" architecture offers. At present, it is still not possible to make all Claude Code Mods run seamlessly in DeepSeek Harness. If this is deemed a goal worth pursuing, it may be technically achievable in the future.

The core philosophy of DeepSeek Harness is that "everything is a plugin." Cui Tianyi said this concept was established from the very beginning of the project, before the first line of code was written, just as the concept of "open source" was established from the founding of DeepSeek before the release of its first model. He said open source is their original aspiration, not something they were forced into. Making their Agent Harness product open and open source with the most extreme scalability possible is also their original aspiration, not something done under pressure or by imitating others.

Cui Tianyi mentioned that a considerable part of his daily work is devoted to recruiting and interviewing for DeepSeek. DeepSeek's goal is to achieve AGI, and the goal of the DeepSeek Harness team is to advance AGI from the perspective of Harness. In his view, only doing innovative things truly counts as "advancing AGI." If they copied Claude Code or copied Codex, and made a product that was basically no different from some other Agent Harness in terms of both technical philosophy and user experience, that would definitely not be a matter of advancing AGI.

"What advances AGI must be innovative work. This Claude Code Mods compatibility experiment is an attempt by us to test the extensibility of the DSH plugin system through a concrete implementation. In the DeepSeek Harness team, we hope to create some world-class innovations, innovations that influence the industry's development and influence how everyone views and defines Agent Harness, this emerging thing, and we hope to create innovations that AI giants around the world will recognize and follow," Cui Tianyi said.

Cui Tianyi noted that whether viewed from the perspective of the team or the product, DeepSeek Harness is still very young and has many imperfections. There is still a long road ahead, and AGI is far from reaching its endgame. Public information shows that DeepSeek's Agent Harness team was formed in March this year. Its lead, Cui Tianyi, is a post-90s professional who studied computer science at Zhejiang University, holds six ACM Asia Regional Contest gold medals, worked for nine years at top quantitative firm Jane Street, and joined DeepSeek in March this year. A Harness refers to a software support framework built around a large language model, designed to help AI Agents manage and execute more complex tasks. Simply put, it can be understood as the entire operating system needed for a large model to truly become an Agent. On August 13, DeepSeek Harness v0.1 (DSH) developer preview was officially opened for testing to developers worldwide, with source code simultaneously released under the MIT license. After the project launch, it quickly sparked heated discussion among developers on social platforms such as X, and within less than 24 hours it amassed more than 70,000 stars on the developer community GitHub.

Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.

Most Discussed

  1. 1
     
     
     
     
  2. 2
     
     
     
     
  3. 3
     
     
     
     
  4. 4
     
     
     
     
  5. 5
     
     
     
     
  6. 6
     
     
     
     
  7. 7
     
     
     
     
  8. 8
     
     
     
     
  9. 9
     
     
     
     
  10. 10