AI Jobs › Hiring brief archive

AI Hiring Brief · 2026-09-14

32 frontier AI labs · 4 roles added in 24h · agihunt.info/en/jobs

Today at a glance

  • A rare quiet day across the board: net additions were just 4 roles across all 32 labs tracked, against 27 closures — the softest daily print in this dataset in weeks, dragged down almost entirely by one company's cleanup.
  • ByteDance Doubao/Seed clears its board: 17 roles closed today with zero replacements posted, the largest single-day closure round of any lab.
  • Perplexity doubles down on agents: both of today's Perplexity additions sit in AI Research & Systems and reference agents directly — Engineering Manager (TLM, Agents) and Member of Technical Staff (AI Software Engineer, Agents), both in San Francisco.
  • Anthropic keeps building out Trust & Safety: its only new role today, Enforcement Operations Lead, Cloud Partners, sits under Safeguards (Trust & Safety) and is open across San Francisco, New York City and Washington, DC — enforcement capacity, not just policy, is being staffed.
  • China's labs went completely silent today: not one of the twelve Chinese labs tracked posted a new role; three (Baidu AI, Meituan LongCat, ByteDance Doubao/Seed) only closed listings.

Company moves

Anthropic added one role today, the Enforcement Operations Lead, Cloud Partners under Safeguards (Trust & Safety) — a multi-city enforcement hire that suggests Anthropic is scaling the operational side of trust and safety, not just research.

Perplexity added 2 roles and closed 2. Both additions land in AI Research & Systems and are explicitly agent-focused (an Engineering Manager and a Member of Technical Staff for Agents), pointing to a dedicated push on agentic products even as older listings roll off.

NVIDIA (英伟达) added a single role, Solutions Architect - Deep Neural Network Evaluation, open across 5 locations — a smaller but telling signal that evaluation and benchmarking infrastructure is getting dedicated headcount alongside NVIDIA's usual silicon and robotics hiring.

OpenAI closed 2 roles today with none added, a pause after a 7-day stretch that still added 57 roles — the second-highest weekly total in this dataset.

ByteDance Doubao/Seed (字节跳动豆包/Seed) closed 17 roles today with no new postings, its steepest single-day cut of the period, while its 7-day net was only 4 — a sharp deceleration relative to peers of similar size.

Baidu AI (百度 AI) closed 2 roles today, but its 7-day figure of 25 additions is the highest among Chinese labs tracked, indicating today's closures are a normal churn against an otherwise active week rather than a slowdown.

A handful of other labs registered smaller closure rounds with no replacements today: SpaceXAI (1 closed), Cohere (1 closed), and Meituan LongCat (美团·龙猫) (2 closed). Every other company in the dataset — including Google DeepMind (谷歌 DeepMind), Microsoft AI (微软 AI), Meta AI, Apple AIML (苹果 AIML), Mistral AI, Cursor, Lovable, Together AI, Scale AI, Cerebras, Pika Labs, Aleph Alpha, Genmo, Moonshot AI (月之暗面 Kimi), DeepSeek (深度求索), Zhipu AI (智谱 AI), StepFun (阶跃星辰), ModelBest (面壁智能), Tencent Hunyuan (腾讯混元), MiniMax (Hailuo) (MiniMax 海螺), 01.AI (Yi) (零一万物), SenseTime SenseNova (商汤日日新) and Alibaba Tongyi Qwen (阿里通义千问) — showed no change today.

What it signals

The 7-day numbers make clear that today's stillness is not the field slowing down broadly, just a lull layered on top of very uneven underlying paces. Microsoft AI added 156 roles over the past week — more than double OpenAI's 57 and well over triple Anthropic's 45, and dwarfing every other lab in the dataset. That gap has been consistent across recent editions and is now wide enough that Microsoft AI's weekly hiring volume looks structurally different in scale from the rest of the field, global or Chinese, rather than just temporarily ahead.

Among the labs that did move today, the shape of the roles is informative. Perplexity's two additions are both tagged to agents within AI Research & Systems, adding to a pattern across recent editions of agent-specific engineering and management hires rather than generalist research roles — consistent with a company treating agentic products as a distinct build track with its own management layer, not a side project bolted onto core search. Anthropic's addition continues a similar throughline on the safety side: an "Enforcement Operations" title, as opposed to a policy or research title, implies Anthropic is now standing up people who execute on Trust & Safety decisions across cloud partnerships, not just people who write the rules.

The closures tell a starker story about Chinese labs' current posture. ByteDance Doubao/Seed's 17 closures today, against a 7-day net of only 4, suggest the team is pruning its listings meaningfully faster than it's replacing them — a divergence worth watching if it continues, since ByteDance's open count (920) is the largest in the entire dataset and even modest proportional shifts there move a lot of absolute roles. By contrast, Baidu AI's 25 additions over 7 days show that not all large Chinese labs are pulling back at the same time; its 2 closures today read as ordinary turnover against a still-active hiring week. Smaller Chinese labs — StepFun with 10 added over 7 days and ModelBest with 7 — kept modest but positive momentum, while DeepSeek, Zhipu AI, Tencent Hunyuan and Alibaba Tongyi Qwen posted no net change at all over the past week, a sign that hiring intensity within China's frontier cluster is now quite bifurcated rather than moving as a bloc.

Taken together, the day's data reinforces a pattern from recent editions rather than introducing a new one: a handful of global labs (Microsoft AI, OpenAI, Anthropic, NVIDIA) continue to add at a steady clip with increasingly specific functional titles — agents, evaluation, safety enforcement — while most Chinese labs and smaller global players are either flat or, in ByteDance's case today, actively contracting their open listings.

← 2026-09-13 · All editions · 2026-09-15 →

All open roles at 32 frontier AI labs →