DeepSeek's 150 Hires All Go to Engineering: DSec Agent Compute Stack Rework From OS to App Layer

量子位 · wechat · 2026-09-08

DeepSeek opened 150 roles (on a team of only 300-500), all concentrated in server-side development and Agent elastic compute engineering, targeting senior engineers with 2-10 years of experience. Lead Cui Tianyi framed it as a Scaling Law of headcount: scale breeds complexity, which demands more scale.

Elastic compute: from paper to large-scale engineering

Six server-side tracks span the full R&D chain

LLM research platform, Agent framework components, dev-efficiency infra (CI/CD, observability, exploring Agents for auto-diagnosis and RCA), DeepSeek API (ultra-large-scale serving), online services (tens of millions of DAU), and data engineering (Spark/Flink/Kafka/ClickHouse/Iceberg). Together they form a loop: models execute tasks in sandboxes → the platform records traces → data improves models.

The ideal candidate is expected to already develop with AI Agents daily—writing quality code in unfamiliar domains with AI assistance while catching Agent mistakes—marking a shift from "knowing how" to "knowing what to question and how to verify."

Related event: DeepSeek opens around 150 hiring slots for senior backend and agent infra engineers(3 posts)→

Original post →

More from coding & agent

coding & agent channel →