AI Hiring Brief · 2026-09-12
32 frontier AI labs · 94 roles added in 24h · agihunt.info/en/jobs
Today at a glance
- Microsoft AI's scale dwarfs the field: 29 additions and 41 closures in a single day, with 165 net adds over 7 days — no other lab in this dataset comes close, and the new postings lean into silicon (Principal Design Verification Engineer, Silicon Engineering) and agentic roles (Senior Software Engineer (Agentic AI); GTM Scale Business Architect - AI and Agents).
- A safety-and-behavior layer is thickening at Google DeepMind: a Senior Operational Program Manager for AI Safety and Behavior and a Senior Technical Program Manager for Model Behavior appeared alongside frontier-research roles for Gemini Omni and Reasoning and Artificial General Intelligence (Singapore) — governance headcount growing in step with frontier work.
- Silicon shows up at three labs at once: Microsoft opened a Principal Design Verification Engineer role, Meta AI posted an ASIC Engineer / (Pre/Post) Silicon Validation Engineer, and NVIDIA added a Senior Silicon Engineer, Power and Thermal Controllers and a Senior DFT Engineer — custom-chip hiring is not isolated to one company this cycle.
- StepFun (阶跃星辰) is the only Chinese lab moving at volume: 37 roles added over 7 days versus flat or single-digit counts everywhere else in China, though today's only visible addition was a memory算法 (memory algorithm) posting.
- ByteDance Doubao/Seed and ModelBest (面壁智能) are pruning, not growing: ByteDance closed 13 listings today against just 4 added over the past week, and ModelBest closed 9 today against 7 added over 7 days — both show closures outpacing replenishment.
Company moves
OpenAI added 7 and closed 11 today (50 over 7 days), with Codex go-to-market roles in Munich and London (Applied AI Engineer, Codex; Manager, Applied AI Engineering) alongside a Technical Program Manager for AI Infrastructure Capacity Planning — commercial Codex expansion into Europe paired with continued infra planning.
Anthropic added 11 and closed 11 (44 over 7 days), building out a regulated-industries and international bench: Head of Regulated Industries (Customer Success), AI Deployment Specialist for Beneficial Deployments, External Affairs for South Korea, and two Solutions Marketing Manager roles split across Cybersecurity and Healthcare & Science.
Google DeepMind (谷歌 DeepMind) added 10 and closed 8 (23 over 7 days): beyond the safety/behavior program roles noted above, it opened a Research Engineer for GeminiApp Personalization and a Director of Product Management for Applied AI Enterprise Verticals — research, safety oversight, and enterprise productization all advancing together.
Microsoft AI (微软 AI) added 29 and closed 41 today. New listings spread across silicon engineering, an agentic-AI software engineer, and a wide sales/solution-specialist push — an AI Business Solution specialist in Tokyo, Digital Solution Area Specialists for the Danish and Swedish markets based out of Dublin and Barcelona, an Intelligent Optimization Lead in Dubai, and a Senior Customer Experience Engineering Manager in Bangalore — global go-to-market scaling for Copilot and agents.
Meta AI added 3 and closed 7 (25 over 7 days): the ASIC/silicon-validation role sits next to a Systems ML Tooling engineer and a Privacy Engineer spanning Generative AI and Security — infrastructure buildout with a governance hire attached.
NVIDIA (英伟达) added 9 and closed 5 (32 over 7 days), with a Software Manager for Robotics Platform Engineering and a Software Engineer, LLM Inference both based in Shanghai, plus a Senior High Performance AI Engineer for Agentic AI and silicon/power roles in India — robotics, inference, and chip work advancing across three regions at once.
Apple AIML (苹果 AIML) added 3 (16 over 7 days), including a Machine Learning Researcher for Multimodal Foundation Models in Zurich alongside Wallet-focused ML engineering in Cupertino.
SpaceXAI added 2 (3 over 7 days), both on the Human Data team — an Expert Team Lead for Human Data Operations and a Post-training Team Lead — pointing to continued investment in data-labeling and post-training leadership.
Perplexity added 3 with no closures (6 over 7 days): a Strategic Business Operations Lead and a Technical Customer Success Manager, both tagged API, plus a Senior Cinematographer for Brand Studio — API commercialization and brand production growing side by side.
Cohere added 3 (6 over 7 days): a Product Security Engineer for its "North Security" effort, a Data Infrastructure engineer, and a Senior Director of Integrated Marketing.
Together AI added 4 (7 over 7 days), with a Staff Engineer for Distributed Storage and HPC & AI Infrastructure and a Senior Software Engineer for Infra Agent Systems both weighted toward its Bangalore/India footprint.
Scale AI added 1 (12 over 7 days): a Staff Machine Learning Engineer for the Public Sector, continuing its government-focused engineering push.
StepFun (阶跃星辰) added 1 today but 37 over the past week — far ahead of any other China-based lab tracked here.
Baidu AI (百度 AI) added 6 and closed 3 (25 over 7 days), spanning a 智能体研发工程师 (agent R&D engineer), a 风控策略模型工程师 (risk-control strategy model engineer), an AI开发者传播专家 (AI developer advocacy specialist), and a Java研发工程师 — agents, risk modeling, and developer relations moving together.
MiniMax (Hailuo) (MiniMax 海螺) added 1 but closed 7 (net closures over 7 days: 7 closed vs 4 added), with its lone new post a Global Compute Investments & Strategic Deals role — an overseas compute-deal hire amid a week of contraction elsewhere on the team.
On the closure side, ByteDance Doubao/Seed (字节跳动豆包/Seed) closed 13 listings today against only 4 added over 7 days, and ModelBest (面壁智能) closed 9 today against 7 added over the same week — both net negative. Moonshot AI (月之暗面 Kimi) closed 2 with no additions, and Cursor and Cerebras each closed a single listing with nothing new.
The remaining tracked companies — Mistral AI, Lovable, Pika Labs, Aleph Alpha, Genmo, DeepSeek (深度求索), Zhipu AI (智谱 AI), Meituan LongCat (美团·龙猫), Tencent Hunyuan (腾讯混元), 01.AI (零一万物), SenseTime SenseNova (商汤日日新), and Alibaba Tongyi Qwen (阿里通义千问) — showed no hiring activity today.
What it signals
The gap between Microsoft AI and everyone else keeps widening: 165 net adds over 7 days is roughly 3–4x OpenAI's or Anthropic's weekly pace, and its listings skew toward sales and solution-specialist roles distributed across a dozen countries rather than research. That pattern — high volume, high churn (29 added and 41 closed in the same day), and a geographic sprawl of GTM roles — reads as a company actively re-tooling its commercial organization around Copilot and agents rather than expanding research headcount.
Among the frontier labs, OpenAI and Anthropic are tracking each other closely (50 and 44 net adds over 7 days, respectively), both mixing steady infrastructure hiring with a push into adjacent functions — Codex go-to-market for OpenAI, regulated-industry and international sales for Anthropic. Google DeepMind's smaller but pointed batch, adding safety/behavior program management alongside frontier research and reasoning roles, suggests governance structure is being built in parallel with capability work rather than after it.
The China/global divide remains stark. Outside of StepFun (37 over 7 days) and Baidu (25 over 7 days), Chinese labs in this dataset were largely flat or contracting this week — DeepSeek, Zhipu AI, Tencent Hunyuan, and Alibaba Tongyi Qwen posted zero net movement, while ByteDance Doubao/Seed and ModelBest both closed more roles than they opened. That's a notable contrast with ByteDance's raw scale: at 925 open positions it's the largest single count in the dataset, yet its net hiring this week was effectively negative, consistent with roster pruning rather than expansion.
Cutting across regions, silicon and hardware engineering shows up simultaneously at Microsoft, Meta, and NVIDIA this cycle, while robotics and inference roles at NVIDIA are split evenly across Shanghai and other regions — a reminder that infrastructure investment in this data isn't concentrated in any one geography or company, even as the pace of that investment varies enormously between the largest labs and the rest.