AI Hiring Brief · 2026-07-07

19 frontier AI labs · 69 roles added in 24h · agihunt.info/en/jobs

Today at a glance

ModelBest goes all-in on embodied AI leadership: All 4 new roles at the Chinese lab are lead-level positions spanning hardware, data, world models, and algorithms — a full-stack team-formation sweep that reads as a zero-to-one build, not incremental headcount.
Microsoft AI posts its heaviest single-day count: 31 new roles distributed across India, Australia, Canada, the Netherlands, Estonia, and multiple US locations, with security and CoreAI hires anchoring the mix — a global engineering replication play rather than a research cluster.
NVIDIA extends autonomous vehicles onto Chinese soil: "Carline Program Manager - Autonomous Vehicles" in Guangzhou is a rare vehicle-program management posting by an international lab in China, paired with agentic AI systems and obstacle foundation model roles to sketch a full AV stack.
Meta clusters on hardware-software co-design: Three roles — "AI Research Scientist, SysML," "Research Engineer, ML H-W/S-W Codesign," and "Software Engineer, Systems ML - Compilers / Backend" — arrived together, consistent with the MTIA custom silicon roadmap.
OpenAI formalizes its pricing architecture: Two simultaneous "Pricing Strategist" postings — one for API, one for GTM — alongside further EMEA commercial hires signal a shift from landing customers to systematically designing how value is captured.

Company moves

Microsoft AI added 31 roles today, its heaviest single-day count and bringing its 7-day total to 523. Engineering positions dominate, spread across Hyderabad, Bangalore, Sydney, Melbourne, Brisbane, Vancouver, Tallinn, and Schiphol. Two US postings stand out: "Principal Software Engineer - CoreAI" (Redmond / Mountain View) and "Principal Applied Threat Intelligence Manager" (Redmond / Reston). The geographic breadth points to deliberate replication of engineering capacity across delivery centers; the security and CoreAI pairings suggest AI is being woven into Microsoft's enterprise security product layer, not kept separate from it.

NVIDIA added 12 roles across three distinct threads. "Carline Program Manager - Autonomous Vehicles" in Guangzhou marks a vehicle-program management role on Chinese soil — uncommon for an international hardware company. "Senior AI and ML Engineer, Agentic AI Systems" targets inference-side agentic deployments. "Senior Architect - Molecular Dynamics" and "Senior Physics-Machine Learning Engineer - CAE" extend NVIDIA's push into scientific simulation. Thirteen roles closed on the same day, leaving net headcount roughly flat.

ModelBest (面壁智能) posted 4 new roles — every one a lead or expert-level position in embodied intelligence: "具身智能硬件负责人" (embodied AI hardware lead), "具身智能数据专家/负责人" (embodied AI data expert/lead), "世界模型专家/负责人" (world model expert/lead), and "具身智能算法专家/负责人" (embodied AI algorithm expert/lead), all in Beijing. Hiring four department heads simultaneously implies both a settled team architecture and capital sufficient to staff it at once.

Meta AI added 4 roles concentrated in systems-level AI: "AI Research Scientist, SysML," "Research Engineer, ML H-W/S-W Codesign," and "Software Engineer, Systems ML - Compilers / Backend" form a coherent cluster pointing toward compiler stack and custom silicon co-design work. The fourth, "AI Research Scientist, FAIR Chemistry" (New York / San Francisco), extends Meta's AI-for-science positioning independently.

OpenAI posted 5 roles: two "Pricing Strategist" positions (one under Finance for API pricing, one under Strategic Finance for GTM pricing), "Enterprise Field Marketer, EMEA" in London, "Customer Success Manager - Ads Solutions" in Dublin, and "Technical Program Manager, Strategic Initiatives" in San Francisco. Separate owners for developer-facing and enterprise-facing price design signals that commercial architecture is becoming structurally deliberate rather than improvised. Seven-day additions stand at 727, the highest absolute count among all 19 tracked companies.

Apple AIML added 5 roles with notable geographic spread: "Machine Learning Engineer - Health AIML" (Cupertino), "Staff ML Engineer - Ads ML Infrastructure" (New York), "AI Application Engineer" (Shanghai), "AIML - Machine Learning Researcher, MLR" (Cupertino), and "Computer Vision Researcher - Video Restoration" (Herzliya, Israel). The Shanghai and Herzliya postings indicate sustained investment in both the China market and Apple's Israeli R&D node.

Anthropic added 2 roles: "Industry Principal, Life Sciences" (Sales, San Francisco / New York) continues the vertical enterprise sales build-out; "Software Engineer, Research Infrastructure" is a research-support engineering hire. Seven-day total reaches 390.

Google DeepMind added 1 role — "Gameplay Programmer, Games, Inception, DeepMind" in London, assigned to the Inception games research team — while closing 9, for a net of −8. That is the sharpest single-day net contraction among all tracked companies today.

SpaceXAI added 1 role: "Member of Technical Staff - RL Inference" (Palo Alto), targeting reinforcement learning at inference time and directly linked to iterative model quality improvement.

Mistral AI added 2 back-office roles: "Accounts Payable Manager" (Paris) and "Legal Counsel, Patent Attorney" (Palo Alto). No research or engineering movement.

StepFun (阶跃星辰) added 1 role: "资金专员" (treasury/fund operations specialist) in Shanghai's Xuhui district — a finance back-office post carrying no technical signal.

Baidu AI (百度 AI) added 1 role — "渠道经理" (channel sales manager, J101386) in Beijing — while closing 27, the largest single-day closure count among all tracked companies, for a net of −26.

Meituan LongCat (美团·龙猫) closed 4 roles with no additions; Cursor closed 3 with no additions. The remaining companies — Perplexity, Lovable, Moonshot AI (月之暗面 Kimi), DeepSeek (深度求索), and Zhipu AI (智谱 AI) — reported no movement today.

What it signals

The most telling pattern today is how differently global and Chinese companies are approaching the embodied AI problem. ModelBest, a software-native Chinese lab, is attempting a full-stack physical intelligence build by hiring four department heads at once across hardware, data, world models, and algorithms. That seniority distribution is a signal: you hire lead-level talent when you are forming the function, not scaling it. NVIDIA, by contrast, is approaching the same domain from its incumbent compute position — the Guangzhou automotive program manager, the obstacle foundation models perception engineer, and the agentic AI systems engineer layer delivery and go-to-market capability on top of an existing platform. Both vectors are converging on the same physical-world deployment problem from opposite starting points.

Meta's three-role systems-ML cluster is worth watching cumulatively rather than in isolation. "SysML," "H-W/S-W Codesign," and "Compilers / Backend" are not adjacent hires by coincidence — they represent the three disciplines required to make a custom silicon investment pay off at the software layer: system-level ML optimization, co-design of hardware behavior with training workloads, and a compiler stack that can target non-standard accelerator architectures. If this pattern continues over weeks, it points toward a Meta inference infrastructure that is progressively less dependent on third-party silicon, with real competitive implications for NVIDIA's data-center revenue.

The 7-day figures reveal a structural divergence in hiring intent between Microsoft and the frontier labs. Microsoft's 523 additions are overwhelmingly distributed engineering roles across delivery centers; OpenAI's 727 and Anthropic's 390 are weighted more toward commercial, vertical sales, and research infrastructure. Microsoft is scaling a product already in customers' hands. OpenAI and Anthropic are building out the commercial architecture — pricing, vertical sales, customer success — that will eventually harvest value from the technical lead they established earlier. OpenAI's dual Pricing Strategist postings on the same day are the clearest single-day expression of this: one role to design developer-facing API pricing, another to design enterprise GTM pricing, implies that revenue capture is now complex enough to require functional specialization rather than a generalist approach.

Baidu AI's net of −26 today, driven by 27 closures against a single channel-sales addition, continues a pattern of contraction rather than expansion. Google DeepMind's −8 net, with its sole addition being a games research programmer, suggests scope tightening rather than ambition reduction — consistent with consolidating around fewer high-priority threads. Both are meaningful counterweights to the expansion narrative dominating the top of the leaderboard.