AI Hiring Brief · 2026-08-02

32 frontier AI labs · 38 roles added in 24h · agihunt.info/en/jobs

Today at a glance

OpenAI moves into physical compute: Two Scaling additions — "Construction Manager, Industrial Compute" and "Industrial Compute" — are among the most operationally concrete roles the lab has posted, pointing toward owned or co-managed datacenter capacity. Paired with "Software Engineer, API Agents" and "Software Engineer, API Multimodal," the day blends bricks-and-mortar infrastructure with API-layer product work.
Microsoft AI posts 21 today, 222 this week: Silicon Engineering hires — "Principal Subsystem Engineer (Design or DV)" and "Principal Subsystem Design Manager" — alongside a "Quantum Engineer - Readout" signal vertical integration into custom hardware. A 7-day total of 222 is roughly four times OpenAI's 50, the next highest among global labs.
Scale AI builds a Frontier Agents engineering ladder: Four additions construct a complete function: "VP, Research" at the top, then "Staff Frontier Agents Engineer (Applied AI)," "Senior Frontier Agents Engineer (Applied AI)," and "Frontier Agents Engineer (Applied AI)" — a deliberate career tier, not ad-hoc gap-filling.
ByteDance Doubao/Seed closes 50 roles in a single day: The largest open roster in the dataset (912) absorbs its steepest single-day contraction with zero offsetting additions — the sharpest consolidation signal in the current window.
Perplexity and Cursor enter distribution mode: Perplexity adds iOS and Android "Member of Technical Staff (Computer Growth)" engineers; Cursor posts "Integrated Campaigns, AMER" in Marketing — both companies shifting from pure engineering builds toward commercial and mobile reach.

Company moves

OpenAI (754 open; 6 added, 6 closed today; 50 over 7 days): The two Scaling hires are the day's most operationally unusual additions — construction managers sit closer to real-estate and facilities work than to model research, suggesting OpenAI is managing physical build-outs at industrial scale. On the product side, "Software Engineer, API Agents" and "Software Engineer, Agent Productivity" extend the agentic surface at the API layer, and "Software Engineer, API Multimodal" signals continued investment in multimodal API capabilities. "Counsel, Litigation" in Legal rounds out a day that spans concrete and courtrooms.

Microsoft AI (547 open; 21 added, 30 closed today; 222 over 7 days): Silicon Engineering leads with "Principal Subsystem Engineer (Design or DV)" and "Principal Subsystem Design Manager," both distributed across Mountain View, Redmond, and Raleigh — a three-site silicon org that mirrors hyperscaler chip-design practices. "Quantum Engineer - Readout" is the day's most technically narrow hire. "Commercial Executive" in Commercial Sales and "AI Networking" in Software Engineering round out a day that touches custom silicon, enterprise sales, and datacenter networking simultaneously. Software engineer additions in Hyderabad and Bangalore continue an India engineering expansion visible across recent weeks.

Scale AI (215 open; 4 added, 0 closed today; 14 over 7 days): The four additions are structurally coherent — a "VP, Research" to anchor the function, then three rungs of "Frontier Agents Engineer (Applied AI)" (Staff, Senior, individual contributor) in Enterprise Engineering. Scale is building a dedicated agents engineering org with a defined career path, treating applied agentic work as a billable product surface rather than a research experiment.

Meta AI (139 open; 1 added, 5 closed today; 15 over 7 days): A single addition — "AI Research Scientist, Modeling & ToolCalling - MSL" in Menlo Park — against five closures. The pairing of modeling with tool-calling in one research role signals that Meta is investing in action reliability at the research layer, a prerequisite for autonomous agent deployments built on its open-source ecosystem.

NVIDIA (561 open; 3 added, 4 closed today; 37 over 7 days): "Senior AI Systems Engineer, Chip Design" continues NVIDIA's pattern of applying AI to its own design workflows. "Senior Software Engineer, AIOps and Observability" targets internal infrastructure reliability at fleet scale. "Senior Supply Chain Business Transformation Manager" reflects the pressure of managing hardware supply chains for AI datacenter buildouts globally — an operational challenge that has become as important as the engineering one.

Perplexity (87 open; 2 added today; 5 over 7 days): Both additions carry the "Member of Technical Staff" title in "Computer Growth" — one iOS, one Android — open across San Francisco and New York City. Framing what are effectively growth engineering roles under a technical staff title signals that Perplexity treats mobile expansion as a product-engineering problem, not a marketing motion.

Cursor (120 open; 1 added today; 4 over 7 days): "Integrated Campaigns, AMER" in Marketing is Cursor's first explicit demand-generation hire visible in this window. For a tool that grew almost entirely on developer word-of-mouth, a campaigns role points toward structured enterprise pipeline-building in the Americas.

ByteDance Doubao/Seed (字节跳动豆包/Seed; 912 open; 0 added, 50 closed today; 9 added over 7 days): Fifty closures in a single day with no additions is the starkest single-session contraction in the dataset. The 7-day picture — 9 added, 50 closed today — is consistent with active organizational rationalization rather than expansion.

Anthropic, Google DeepMind, Apple AIML, SpaceXAI, Mistral AI, Cohere, Cerebras, Lovable, Together AI, and all remaining Chinese labs posted no single-day changes.

What it signals

The most structurally significant pattern across today's data is the shift from software-only hiring to hardware-adjacent investment. OpenAI's "Construction Manager, Industrial Compute" is an unusual title for a model company; the role belongs to a world of contractors, permits, and physical plant, not foundation models. Its pairing with a general "Industrial Compute" Scaling role suggests OpenAI is building or co-managing owned compute facilities — a step toward infrastructure independence that hyperscalers established years ago and that frontier AI labs are now pursuing more deliberately. Microsoft AI's silicon engineering additions reinforce the same vector: design-verification engineers and subsystem design managers are chip-layer roles, not cloud-layer ones, pointing to custom silicon work that would give Microsoft more control over the AI compute stack it sells and runs.

The agents theme has progressed from individual job postings to deliberate org design. Scale AI's move to create a three-tiered "Frontier Agents Engineer" ladder in Enterprise Engineering is the clearest example: this is not a lab posting a handful of agents researchers but a company constructing a functional unit with defined seniority bands, implying that agentic enterprise products are far enough along to require their own workforce architecture. OpenAI's "Software Engineer, API Agents" and "Software Engineer, Agent Productivity" point in the same direction — the agentic API is now a first-class product with dedicated headcount rather than a shared-infrastructure feature.

The 7-day figures reveal a pronounced gap in pace between global and Chinese labs. Microsoft AI at 222 additions is a company-wide acceleration that spans silicon, software, security, data science, and sales simultaneously. Among Chinese labs tracked here, no company posted any new roles today, and ByteDance Doubao/Seed's 50 closures — from the dataset's largest roster — suggests that the period of aggressive Chinese frontier hiring is giving way to consolidation. Whether this reflects macroeconomic conditions, regulatory environment, or a natural maturation cycle after earlier build-outs is not determinable from hiring data alone, but the directional contrast with the pace of global labs is measurable and consistent across recent sessions.

Meta AI's single research addition — modeling paired with tool-calling — is a small but coherent signal. Reliable tool-calling is the capability bottleneck that separates demo-quality agents from production-grade ones; a dedicated research scientist hire in that space suggests Meta is treating this as a foundational research problem, not merely an engineering tuning task.