AI Hiring Brief · 2026-09-19

32 frontier AI labs · 71 roles added in 24h · agihunt.info/en/jobs

Today at a glance

Autonomous vehicles at NVIDIA: NVIDIA posted three roles aimed squarely at self-driving software — a Senior Integration Engineer, End-to-End Model for Autonomous Vehicles, a Senior Context Fusion AI Engineer, and a Senior Perception Software Engineer — while closing 11 of its own openings on the same day.
DeepMind spans safety to generative media in one day: Google DeepMind's four new roles cover four distinct fronts — Safety Oversight, Responsible Frontier AI Research, Robotics RL, and Generative Media — a rare same-day spread across the lab's safety and product research lines.
Together AI's single-day spike is almost entirely internships: 12 of Together AI's 15 additions over the past 7 days landed today, and the visible postings are all Summer/Winter 2027 research and engineering internships covering Frontier Agents, GPU Programming, Model Shaping, and Inference — a university-recruiting push, not headcount growth.
Moonshot AI trims faster than it hires: 月之暗面 Kimi (Moonshot AI) added 8 roles but closed 13, and the new postings skew toward investor relations, compute planning, and brand/PR rather than research — a commercial-facing reshuffle rather than a research build-out.
ByteDance's Doubao/Seed doubles down on agents and robotics: new postings include an agent-focused large-model architect for TikTok Search, an Agent-direction algorithm engineer for Volcano Engine (火山引擎), and a robotics software product manager under Seed.

Company moves

OpenAI added 9 roles (64 over 7 days) against 12 closed. New postings pair data-center-adjacent finance and workplace roles (Head of Finance – Data Centers, Global Transportation Programs Senior Manager) with commercial and legal hires (Applied AI Engineer, Quants in London; Deputy General Counsel, Commercial) — infrastructure and enterprise go-to-market scaling in parallel.

Anthropic added 9 roles (39 over 7 days, 10 closed), almost entirely sales and policy: Enterprise Account Executives for banking (Zürich) and pharmaceuticals/life sciences (Munich), a National Security Partnerships Lead, and a Technical Program Manager, Public Sector — continued push into regulated-industry and government accounts.

Google DeepMind (谷歌 DeepMind) added 4 roles (25 over 7 days, 4 closed) spanning safety oversight, responsible AI research, robotics RL, and generative media — a small but deliberately diverse slate.

Meta AI added 2 roles (15 over 7 days, 7 closed), both in AI infrastructure: an Engineering Manager for ML Compilers (Triton/Kernel DSLs) and a systems engineer for AI training-data infrastructure.

NVIDIA (英伟达) added 3 roles (30 over 7 days, 11 closed), all three tied to autonomous-vehicle perception and end-to-end modeling.

Apple AIML (苹果 AIML) added 2 roles (22 over 7 days, 2 closed): an Applied AI Science manager for GenAI model autograding/evaluation, and a senior SDET for Siri AI quality — both evaluation-focused, consistent with Siri's ongoing overhaul.

SpaceXAI added 2 roles (18 over 7 days, 3 closed), continuing its data-center buildout (Iron Worker, Memphis-area) alongside an events-operations hire.

Perplexity added 2 roles (6 over 7 days, none closed): a Search Core/Indexing engineer and a research fellowship.

Together AI added 12 roles today (15 over 7 days, none closed) — see above; nearly all are 2027 internships.

Scale AI, Cursor, Lovable, Cohere, and 01.AI (零一万物) each posted one or two roles today with no closures — a Payroll Analyst and a Public Sector Technical Program Lead at Scale AI, a Singapore-based AI Deployment Manager at Cursor, a Stockholm product engineer at Lovable, two IT Support Specialists (Toronto, Korea) at Cohere, and an AI Test Engineer at 01.AI — routine ops and support hiring rather than a directional shift.

Cerebras added 2 roles (8 over 7 days, 3 closed): a Network Systems Architect and a Senior Staff AI Accelerator Performance Architect — silicon hiring continues.

Moonshot AI (月之暗面 Kimi) — see above; net negative with commercial/ops-heavy postings including Technical IR｜投资者关系（技术方向）(technical investor relations) and Compute Planning｜算力规划专家 (compute-planning specialist).

StepFun (阶跃星辰) added 2 roles (19 over 7 days, 2 closed): a large-model data operations specialist (大模型数据专家运营) and a senior technical-brand operations role (资深技术品牌运营).

ModelBest (面壁智能) added 1 role (10 over 7 days): a multimodal LLM algorithm engineer for OCR/text understanding (多模态大模型算法工程师，OCR/文字理解方向), posted across four cities.

Meituan LongCat (美团·龙猫) closed 3 roles today with no additions.

Baidu AI (百度 AI) added 3 roles (18 over 7 days, 3 closed): a Java engineer for its after-sales tooling platform, a Search API AI product manager, and an algorithm engineer for large-model annotation/evaluation across all modalities.

ByteDance Doubao/Seed (字节跳动豆包/Seed) added 3 roles (24 over 7 days, 7 closed) — see above; agent and robotics postings dominate today's slate.

SenseTime SenseNova (商汤日日新) closed 2 roles today with no additions — the week's other pure-contraction signal alongside Meituan LongCat.

Unchanged today: Microsoft AI (0 added, 0 closed today, despite 169 added over the trailing 7 days), plus Mistral AI, Aleph Alpha, Genmo, Pika Labs, DeepSeek (深度求索), Zhipu AI (智谱 AI), Tencent Hunyuan (腾讯混元), MiniMax (Hailuo) (MiniMax 海螺), and Alibaba Tongyi Qwen (阿里通义千问) posted no visible changes.

What it signals

The 7-day totals show a wide gap in pace between the two groups tracked here: the global labs added 438 roles over the past week versus 92 for the Chinese labs — roughly 4.8x. Microsoft AI alone accounts for 169 of the global figure, nearly double China's entire weekly total, though its own additions went silent today (0 added, 0 closed). Even stripping Microsoft out, global labs still added around 269 roles to China's 92, so the gap isn't a single-company artifact.

Composition differs as much as volume. Global additions this week lean into infrastructure and enterprise expansion happening together: OpenAI is staffing data-center finance and legal functions alongside applied-AI sales; Anthropic is opening regulated-industry accounts in banking and pharma across Europe; NVIDIA and Meta are hiring into perception software and ML-compiler infrastructure respectively. On the China side, the largest movers — ByteDance's Doubao/Seed, Baidu, StepFun — are posting product, operations, and brand roles more than frontier-research ones, and Moonshot AI is net-negative for the day (8 added, 13 closed), with its new roles concentrated in investor relations and PR rather than model work. That points to a maturing, productization-first phase for several Chinese labs even as their overall pace trails the global cohort.

One theme does show up on both sides simultaneously: agentic and embodied AI. DeepMind posted a Research Scientist, Robotics RL role the same day ByteDance's Seed unit posted a robotics software product manager and two Agent-direction roles (one for TikTok Search, one for Volcano Engine) — evidence that robotics and agent tooling are being staffed concurrently across both ecosystems, not sequentially.

A second cross-cutting thread is evaluation and safety tooling. Apple posted a GenAI model autograding/evaluation manager and a Siri AI-quality SDET on the same day DeepMind posted Safety Oversight and Responsible Frontier AI Research roles — a signal that as consumer-facing deployments (Siri, Gemini) scale, quality and safety evaluation functions are being built out as standing infrastructure rather than one-off research efforts.