DeepSeek to hire 150 engineers for AI agent infrastructure, not model development

DeepSeek Expands Engineering Team for Infrastructure

DeepSeek announced plans to hire around 150 engineers, with a focus on server-side engineering and Agent Elastic Compute. This recruitment drive, announced by Cui Tianyi from DeepSeek's Harness team, is intended to support the company's expanding AI agent operations.

The new positions will cover areas such as operating systems, virtualization, networking, storage, scheduling, and control-plane services. These roles are distinct from AI research and model development, concentrating instead on the underlying systems that enable AI agent functionality.

Scaling Challenges for AI Agent Workloads

DeepSeek's existing backend systems require upgrades and maintenance to handle increasing workloads. A key component facing scaling challenges is DeepSeek Elastic Compute (DSec), the sandbox infrastructure designed for executing agent workloads during post-training and evaluation phases.

The company's goal is to manage the growth of concurrent AI agent sandboxes and associated backend processes, which include training and evaluation tasks.

DSec: Flexible Agent Environment Infrastructure

DSec supports four types of isolated environments for agent workloads through a single Python SDK. These include pre-warmed containers for simple function calls, Docker-compatible containers for persistent environments, Firecracker microVMs for stronger isolation, and QEMU virtual machines for tasks requiring a full guest operating system.

This infrastructure allows DeepSeek to handle a range of tasks, from basic tool calls to complex software engineering operations that need a complete OS. This approach addresses similar challenges faced by the broader industry as AI agents transition from development to production environments.