AI Hiring Brief · 2026-07-06

19 frontier AI labs · 4520 roles added in 24h · agihunt.info/en/jobs

Today at a glance

• OpenAI's EMEA advertising machine: "Regional Client Partner, Ads Solutions" and "Customer Success Manager - Ads Solutions" landed across Dublin, Munich, and Paris simultaneously, alongside a new "Head of Scaled, Ads Solutions" in San Francisco and an "Executive Business Partner, B2B Ads Solutions" in New York — the clearest sign yet that OpenAI is building a dedicated ad-revenue business from scratch.
• Anthropic maps the entire financial services sector: Seven "Strategic Account Executive" titles covering investment banking & capital markets, insurance, retail & commercial banking, asset & wealth management, healthcare, life sciences, and a Canada FSI seat appeared in one window — the most granular vertical sales build-out visible in this snapshot.
• Meta commits to humanoid hardware: "System Safety Engineer, Humanoid Robotics," "Mechanical Design Engineer: Robotics Hardware Engineering," and "Mechanical Engineer: Electronics Packaging and Sensor Integration" arrived together, signaling that Meta's physical-AI program has moved from pure research into integration and safety engineering.
• Chinese labs build agent infrastructure, not just apps: DeepSeek opened its "Agent Harness 团队" (agent harness team) across researcher, engineer, and PM tracks; Moonshot AI added "Coding Agent 研发工程师｜Kimi Code" and "Agentic Growth Engineer"; StepFun listed "Agentic 基础设施开发工程师" (agentic infrastructure engineer) — a convergent bet on agent runtimes as a foundational layer rather than a product feature.
• Mistral plants its first Asian flag in Seoul: Seven-plus roles — "Applied Scientist / Research Engineer," "AI Deployment Strategist - Korea," fullstack and ML engineers — landed simultaneously, the most concentrated single-city market-entry signal in this dataset.

Company moves

OpenAI (722 open) leads in volume. Beyond the EMEA advertising cluster, a "Deal Lead, Special Situations (Semiconductors)" in Partnerships signals active chip-supply dealmaking, while a "Researcher, Alignment CoT Monitorability" and "Simulation Infrastructure Engineer" keep the core science track running in parallel with commercial expansion.

Anthropic (388 open) ties its FSI sales sweep directly to engineering depth: "Engineering Manager, Agent Runtime Platform," "Staff+ Software Engineer, Caching," "Staff + Senior Software Engineer, Inference Deployment," and "Research Engineer, Computer Use" all appeared alongside the commercial roles. In Europe, "Technical Specialist, Claude Code" (London) and "Partnerships Lead Southern Europe" (Paris) extend the geographic footprint.

Google DeepMind (89 open) posted two exceptional hires: "Senior Virologist" and "Principal Virologist, Responsible Development and Innovation," alongside "Threat Modeler Lead, CBRNE" (chemical, biological, radiological, nuclear, and explosive threats). These are rare in any AI lab and indicate a biosafety research program operating well beyond software safety. On the model side, "Research Scientist, Gemini Diffusion," "Research Scientist, Agent Post Training," and "Senior Security Engineer, Agentic Red Team" continue core research and extend adversarial testing to agent-specific attack surfaces.

Microsoft AI (492 open) covers the widest functional range in this snapshot. "Principal ML Architect" under Silicon Engineering, "Senior Researcher - AI Computer Architecture" (Cambridge, UK), and "Senior Business Intelligence Engineer - Agentic Development" are the standout additions — the last suggesting internal agentic tooling has become a tracked engineering discipline.

Meta AI (131 open) pairs its robotics hardware cluster with "Strategic Supply Chain Risk and Resilience Lead" in AI Infrastructure, flagging hardware-supply fragility as an operational concern. "AI Research Scientist, Generative AI for Materials Discovery" and "AI Research Scientist, Audio-Visual Understanding, FAIR" add breadth to the FAIR research footprint.

NVIDIA (565 open) shows the deepest silicon design commitment here: SRAM circuit design, ASIC hardware design, ASIC timing (multiple postings), DFT power methodology, and "Senior VLSI CAD and AI Automation Engineer" span the full chip development cycle. A "Senior Deep Learning Performance Architect - LPU" stands apart — suggesting NVIDIA is at minimum exploring accelerator architectures beyond the GPU for language workloads.

SpaceXAI (214 open) is almost entirely a construction story. Memphis and Southaven, MS dominate: "Program Manager (Data Center Construction)," "Manager, Mechanical Engineering," "Asset Manager, Energy Storage," "Water Treatment Operator." Research additions are thin — "Software Engineer, Ads Product" and "Member of Technical Staff - Imagine Safety" are the only notable non-infrastructure roles.

Apple AIML (353 open) surfaces two robotics titles ("Senior Robotics Software Engineer," "Senior Robotics Simulation Software Engineer - Special Projects") plus two "Senior Swift Software Engineer: Agentic OS Experiences" seats and a "Senior Engineering Manager, Applied AI & Agentic Experiences," pointing to on-device agentic features approaching productization.

Mistral AI (177 open) pairs its Seoul launch with a Paris security hiring surge: "Deputy Director, Safety & Security HQ," "Applied AI Engineer, CyberSecurity," "Pentester, Offensive Forward Deployment Engineer," and "CyberSecurity Engineer, Incident Response Lead" — four security roles in one window, consistent with preparation for enterprise procurement scrutiny.

Cursor (112 open) is building a federal sales organization at speed: "Regional Vice President, Federal Sales," "Regional Director, SLED," "Regional Director, Federal Systems Integrators," and "Strategic Account Executive, Federal Sales — DoW" appeared alongside healthcare VP roles — an unusually broad government-market entry for a coding AI company.

Perplexity (78 open) is deepening both its API platform and agent capabilities ("Engineering Manager (TLM, Agents)," "Member of Technical Staff (AI Software Engineer, Agents)"), with a "Customer Success & Enterprise Support Lead, APAC" in Tokyo marking Asia-Pacific commercial entry. Lovable (73 open) is hiring security engineers ("Penetration Tester," "Application Security Engineer," "Staff / Principal Software Engineer, Security") alongside its first North American sales leadership, a pattern typical of a product transitioning to enterprise sales.

Among the Chinese labs, Zhipu AI (智谱 AI, 237 open) is the largest, with business development roles for Americas, Middle East, and Singapore, and a "法务总监-资本市场与投融资方向" (legal director, capital markets & investment financing) hinting at corporate finance activity. Baidu AI (百度 AI, 412 open, 2 closures) is the most commercially mature, with cloud-sales coverage across finance, automotive, and healthcare verticals. Meituan LongCat (美团·龙猫, 72 open) runs a nearly pure research agenda — embodied intelligence, agentic coding, agentic RL, and interpretability researchers dominate. ModelBest (面壁智能, 112 open) leans toward enterprise delivery: multiple "政企大模型解决方案工程师" (government & enterprise LLM solution engineers) and a "具身智能真机部署与评测工程师" (embodied intelligence physical deployment and evaluation engineer) reflect a B2B delivery model. StepFun (阶跃星辰, 148 open) splits between model training engineers and hardware-adjacent roles including "WiFi开发工程师" and "云台开发工程师" (gimbal engineer), consistent with device-AI integration work.

What it signals

Because added_1d equals added_7d across every company in this dataset, this is the inaugural snapshot for this tracking series — a cross-sectional portrait of where these organizations are staffed today rather than a measure of single-day acceleration. The 4,515 visible openings span 20 companies at very different scales and maturity levels, which makes the shared patterns more meaningful than the individual numbers.

The clearest structural signal is that commercial infrastructure is now expanding as fast as, or faster than, technical headcount at the largest global labs. OpenAI's EMEA advertising apparatus and Anthropic's FSI sales matrix are not exploratory; they are organized, geographically distributed, and role-specific enough to indicate real revenue targets behind them. The agent-runtime engineering sitting beneath these commercial pushes — Anthropic's "Engineering Manager, Agent Runtime Platform," caching and inference-deployment engineers — suggests the technical scaffolding for agentic products requires significant investment before it can reliably underpin enterprise SLAs at scale.

NVIDIA's chip-design depth is a counter-signal to any narrative that software is eating silicon. Hiring across the full chip development cycle, with a title pointing toward LPU-class accelerators, indicates that the next round of AI compute differentiation will come from silicon architecture, not just software optimization. Meta's parallel moves in systems-ML infrastructure and humanoid hardware reinforce the point: the competitive frontier is increasingly physical — compute substrate, robot bodies, custom accelerators — not purely algorithmic.

The Chinese labs present a more homogeneous picture than their different scales might suggest. The convergence on agent infrastructure — harness frameworks, eval systems, agentic coding pipelines — across DeepSeek, Moonshot AI, and StepFun reflects a shared architectural conclusion that winning in the agent era requires controlling the runtime, not just the model. Whether executed through a focused team of 36 visible openings (DeepSeek) or a broad 237-role campaign (Zhipu AI), the direction is the same, and the unanimity suggests the bet is driven by technical conviction rather than competitive mimicry.