AI Hiring Brief · 2026-07-20

32 frontier AI labs · 3 roles added in 24h · agihunt.info/en/jobs

Today at a glance

Apple goes managerial on agentic search: A new "Machine Learning Engineering Manager, Evaluation, Agentic Search Capabilities, Proactive" role appeared in Santa Clara — the first time Apple AIML has publicly posted a dedicated EM for agent-search evaluation, signaling that capability is moving from research into an organized engineering team with its own management layer.
Cursor plants a flag in Asia-Pacific: A "Forward Deployed Engineer" role opened simultaneously in Australia, Singapore, and Japan — FDE roles typically mean large-account implementation work, so three markets going live on the same day marks a clear shift from exploration to execution in the region.
ByteDance Doubao/Seed dwarfs everyone at week's end: At 7-day additions of 978 against a current open count of 965, the entire posting board appears to have been refreshed and expanded this week. The next-closest 7-day figure in the dataset is Scale AI at 195 — less than a fifth of ByteDance's volume.
Microsoft AI posts its sharpest single-day closure count: Nine roles closed in one day with no new additions, an unusual pattern for the largest global roster in this dataset by consistent weekly additions (+149 over seven days).
The quiet day is mostly a global-labs story: Only three companies recorded any daily addition. All Chinese labs in the dataset reported zero new postings on the day itself, even those that show large 7-day figures.

Company moves

Anthropic added one role today — "Software Engineer, Web Platform" (Engineering & Design - Product, San Francisco, CA) — alongside one closure, leaving the net count flat. At +33 over seven days and 411 open roles total, the steady pace of product-side and platform engineering additions continues; the web platform hire suggests the team is building out the user-facing infrastructure layer rather than pure research capacity.

Apple AIML added "Machine Learning Engineering Manager, Evaluation, Agentic Search Capabilities, Proactive" (Machine Learning and AI, Santa Clara) and closed two other roles. The job title is notable on its own terms: "Agentic Search Capabilities" and "Proactive" appear together in a single management posting, framing evaluation of agent-driven search as a discrete enough area to require its own EM. The 7-day figure of +23 against 352 open roles shows the broader team expanding steadily while this new role marks a qualitative step in how that expansion is being organized.

Cursor posted "Forward Deployed Engineer" (Solutions, Australia; Singapore; Japan) today, bringing its 7-day additions to 14 against 121 open roles. FDE positions are characteristically client-facing: they handle technical integration, onboarding, and deep deployment at enterprise accounts. Opening three Asia-Pacific markets simultaneously — rather than sequentially — suggests the commercial pipeline in those regions is already active and needs boots on the ground.

Microsoft AI recorded no new additions but closed 9 roles in a single day, its highest single-day closure count visible in this dataset's current window. Over the past seven days it has still added 149 roles and holds 496 open, so this does not reverse the week's expansion — but the asymmetry between daily closures and zero additions is unusual and worth watching in coming days.

Baidu AI (百度 AI) closed 25 roles in the past day with no additions, the largest single-day net reduction of any company in the dataset today. Its 7-day figure still shows +16, meaning recent additions and today's removals are running in opposite directions. At 399 open roles, the pool remains substantial, but the pace of closure is a signal worth tracking.

Meituan LongCat (美团·龙猫) closed 8 roles with no new additions, a small but consistent reduction from a base of 74 open positions.

Companies with no daily movement — including OpenAI, Google DeepMind, NVIDIA, Meta AI, SpaceXAI, Perplexity, Mistral AI, 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), and several others — showed unchanged daily counts; their 7-day figures are covered in the analysis below.

What it signals

The most arresting 7-day figure in the dataset belongs to ByteDance Doubao/Seed (字节跳动豆包/Seed): +978 additions over the week against a current open count of 965. The near-identity of those two numbers implies the posting board was substantially refreshed — a common practice on Chinese platforms to maintain visibility — but even accounting for that, the absolute volume is on a different order of magnitude from peers. OpenAI added 42 in the same window, Anthropic 33, NVIDIA 48. ByteDance's weekly figure exceeds the combined 7-day totals of those three by a factor of roughly eight. That gap reflects not only company size but a distinctly different hiring philosophy: a broad-front recruitment strategy that attempts to cover every technical direction simultaneously, as opposed to the more targeted, role-by-role approach visible at the major Western AI-native labs.

Several Chinese labs beyond ByteDance show 7-day additions that closely match or nearly equal their total current open counts: MiniMax (Hailuo) (MiniMax 海螺) +192 against 190 open, Tencent Hunyuan (腾讯混元) +61 against 60 open, 01.AI (零一万物) +53 against 50 open, SenseTime SenseNova (商汤日日新) +67 against 16 open, Alibaba Tongyi Qwen (阿里通义千问) +28 against 28 open. These ratios indicate posting-board refreshes rather than organic growth of open headcount, and should be read as a data characteristic of Chinese recruitment platforms rather than as a week-over-week expansion signal. The substantive question — which directions are being recruited for — requires looking at the role titles themselves, which the current dataset's Chinese-lab entries do not surface.

On the global infrastructure side, Scale AI at +195, Cohere at +142, Together AI at +63, and Cerebras at +107 over seven days all point to continued expansion in the lower and middle layers of the AI stack — data labeling, enterprise LLM deployment, compute infrastructure, and inference. These are not the headline model labs, but their collective hiring pace suggests the supporting ecosystem is still in an active buildout phase. Cerebras in particular reached 107 open roles against a 7-day addition of exactly 107, again suggesting a refresh, but the volume itself — for a silicon-focused compute company — is consistent with the ongoing hardware race around AI inference speed.

The most structurally meaningful single datapoint today may be Apple AIML's management-layer addition. The emergence of a dedicated EM for agent-search evaluation, combined with the broader pattern of Cursor deploying forward-deployed engineers into Asia-Pacific enterprise accounts, suggests that two inflection points are arriving roughly simultaneously: agent-capable products are complex enough to require organized engineering teams with specialized management (not just researchers), and the commercial deployment of those products has reached the point where local technical presence is a prerequisite rather than a nice-to-have. The contrast with the Chinese side — where comparable management-layer hiring around agent evaluation has not yet surfaced in this dataset's public signals — may reflect either a lag in public posting visibility or a genuinely different organizational model for taking agent capabilities to market.