DeepSeek is hiring 150 people at once!
What’s going on? They’re only about 300 to 500 people strong—this must be a big move.
Moreover, these 150 HC are not focused on AI research; they are entirely concentrated on two roles: Backend Development Engineer and Agent Elastic Computing Development Engineer, primarily targeting senior backend developers with 2 to 10 years of experience.
Cui Tianyi explained that as scale increases, complexity arises, and this complexity, in turn, demands further scaling.
Is this... the scaling law for team size?
DeepSeek elastic computing moves from paper to large-scale engineering
This time, the Agent Elastic Computing R&D Engineer role is divided into two directions: platform development and underlying systems.
This platform is a flexible computing platform, DSec, customized for Agents. DeepSeek Elastic Compute), recently released DeepSeek First disclosed in the V4 technical report.
According to the paper, DSec consists of three Rust components: the API gateway Apiserver, the Edge agent on each host, and the cluster monitor Watcher, which communicate with each other via a proprietary RPC protocol and run on DeepSeek Built on our proprietary 3FS distributed file system.
The job postings reveal plans for the next phase of engineering expansion.
To enhance reliability and efficiency, DSec requires modifications across the entire system stack: from the operating system to the virtual machine, through all levels of networking and storage, up to application-layer scheduling and control plane services.
This project involves many things that have never been done before, with no ready-made answers to copy.
The platform development and maintenance role is similar to a blend of product and development, requiring participation in understanding requirements as well as personally designing, developing, and maintaining the system.
Deep-level optimization and tackling more challenging problems—such as squeezing the performance of certain operating system components to SOTA levels or pushing them directly to hardware limits—is a plus.
The final statement, "Your work directly determines the diversity and efficiency of the Agent model training," is not empty words.
Because DeepSeek We will achieve SOTA and then open-source it, which will impact more than just the future DeepSeek the entire industry, not just agent infrastructure.
The agent spans the entire R&D lifecycle.
On the server-side development team, six areas cover the complete pipeline from model research to user-facing services:
Large model research platform, Agent framework components, R&D efficiency infrastructure, DeepSeek API, online services, data engineering.
DeepSeek The mission of this team is described as: staying close enough to the forefront of large model research to receive immediate feedback from the front lines. The reliability and usability of the platform directly determine the efficiency of large model research and the speed at which the upper limits of intelligence are raised.
The core mission of the large model research platform is to abstract the research process into platform capabilities, requiring "deep immersion in researchers' day-to-day workflows" to proactively identify issues rather than passively respond to requests. The ultimate goal is to "shorten the cycle from idea generation to validation and iteration."
The Agent framework component is responsible for building a unified Agent integration and execution framework.
The responsibilities in the R&D efficiency infrastructure domain cover a unified CI/CD system, observability and alerting systems, large-scale data transfer and governance across clusters, and operational automation, with a clear mandate to "explore the application of agents in automated diagnosis and RCA."
Two directions for external users directly serve DeepSeek the product.
DeepSeek The API direction must address the challenge of ultra-large-scale API services, delivering cutting-edge model research outcomes to global developers and massive users through stable and efficient APIs, while evolving in tandem with internal inference frameworks.
The online service direction requires architectural design and continuous iteration to support large model applications at a scale of tens of millions of daily active users.
Finally, the data engineering role supports all stages from model training to online services, with a tech stack including Spark, Flink, Kafka, ClickHouse, and Iceberg. Candidates must be able to independently complete end-to-end data modeling from event logging to business metrics.
Looking at the responsibilities across these areas together, they form a complete pipeline: the model executes tasks in a sandbox environment, the platform logs the operations and outcomes to generate training and evaluation data, and this data is then used to improve the model.
What kind of people is DeepSeek looking for?
All research-related roles emphasize the same thing: engineers must “immerse themselves in researchers’ frontline work environments,” proactively identify issues, and drive solutions from definition to implementation, rather than merely accepting and delivering requirements.
The job posting dedicates an entire paragraph to describing the ideal candidate. DeepSeek We expect the engineers we hire to already be deeply using agents in their daily development work.
One requirement is to proficiently use AI Agent tools for software development, enabling the writing of high-quality, reliable code in areas without direct experience, with AI assistance.
You are also required to keenly detect and promptly intervene when the Agent provides a problematic solution.
In this agent era, the importance of experience with specific languages or tools, or having worked on certain types of projects, is decreasing.
Development is shifting from "I know how to do it" to "I know what to ask, where things might go wrong, and how to prove it's correct."
Reference link:
[1]https://x.com/tianyi/status/2096975270578139273?s=20
[2]https://app.mokahr.com/social-recruitment/
This article is from the WeChat public account "Quantum Bit," authored by Meng Chen.