I looked through the OpenAI and Anthropic recruitment pages and discovered their plans they don't want to disclose.

Author: Jean-Stanislas Denain, Campbell Hutcheson

Compilation: Deep Tide TechFlow

Deep Tide Introduction: Researchers at Epoch AI have examined the public recruitment pages of OpenAI, Anthropic, xAI, and DeepMind, inferring the strategic directions of these companies from the distribution of job postings.

The conclusions are quite interesting: The proportion of sales positions at OpenAI and Anthropic has surged within a year, with the technical sales roles focused on helping clients “learn to use AI” growing the fastest;

OpenAI is developing a portable device with a camera that runs on self-developed chips, alongside 7 robotic positions;

Anthropic does not manufacture chips but is aggressively negotiating data center contracts. The recruitment pages are one of the few publicly available signals of strategy, and this analysis is rich in information.

The full text is as follows:

AI companies are tight-lipped about their strategies, but recruitment pages are public.

These positions hide clues: what products a company is developing, who they want to sell to, and which aspects they believe will become bottlenecks. A position for a “Camera ISP Software Engineer” suggests a device with a camera. Hiring for “Frontline Deployment Engineers” indicates that implementing AI in companies is quite challenging. A host of robot-related positions implies ambitions that go far beyond chatbots.

We analyzed the public job postings of leading foundational model labs, including OpenAI, Anthropic, xAI, and Google DeepMind. Here are our findings.

Key Findings:

The sales and sales-related positions at OpenAI and Anthropic have significantly increased over the past year. The proportion of GTM (go-to-market) positions at Anthropic rose from 17% to 31%, while OpenAI’s increased from 18% to 28%. The growth is concentrated in technical roles that help clients implement AI.

Job postings can provide insights into product roadmaps. For instance, both OpenAI and DeepMind are investing in hardware products (robots and consumer devices), while Anthropic is more focused on improving core products.

The recruitment pages also reveal different strategies for acquiring key resources (computing power and data). For example, OpenAI has 21 positions related to self-developed chips, while Anthropic has none.

A few notes: Job postings reflect who the company wants to hire, not the existing team. For instance, if a team has 20 open positions, it could mean a large team is expanding or that a new team doesn’t yet exist. A listing for a “Research Engineer” might be hiring one person, ten, or possibly no one at all.

GTM has become the largest recruitment category for OpenAI and Anthropic.

Over the past year, sales and related positions at OpenAI and Anthropic have both seen substantial growth: The proportion of GTM positions at Anthropic rose from 17% to 31%, while OpenAI’s increased from 18% to 28%. This is not surprising for companies experiencing rapid revenue growth and competing in an unsaturated market. Sales-related positions are currently the largest recruitment category for both companies. In contrast, research positions only account for 12% of open positions at Anthropic and 7% at OpenAI.

Chart: Changes in the proportion of various positions at OpenAI and Anthropic.

The fastest-growing subcategory is technical roles that help clients effectively utilize AI. Both companies are hiring for roles such as “AI Success Engineer,” “Partner AI Deployment Engineer,” “Solutions Architect,” and “Frontline Deployment Engineer,” with responsibilities focused on helping clients identify AI use cases and complete integration. Over the past year, the proportion of “Implementation Adoption” roles at Anthropic increased from 5% to 11%, while OpenAI’s rose from 11% to 17%. This indicates that clients are facing challenges in fully utilizing AI products, and bridging this gap is crucial in teaching clients “what AI can do.”

The geographical distribution of sales positions also reveals market focus. Over half of the sales positions at both companies are based in the United States (52% at Anthropic, 55% at OpenAI). Neither company has disclosed regional revenue distribution, but the concentration of hiring indicates that the U.S. remains the primary market.

Internationally, both companies are aggressively hiring in Europe and the Asia-Pacific region. Anthropic leans towards Europe (29% vs. OpenAI’s 21%), while OpenAI favors the Asia-Pacific (24% vs. Anthropic’s 19%). Growth in the Asia-Pacific is concentrated in Japan, South Korea, India, Singapore, and Australia. Notably absent are China, the Middle East, Latin America, and Africa. The focus on global sales by various labs indicates they do not believe they will be pushed out of the market by local competitors in Europe and the Asia-Pacific.

Government sales are also a key focus area. OpenAI and Anthropic each have 10 government sales positions, covering federal civilian, defense, and state and local governments. OpenAI has one position specifically targeting national security, while Anthropic has two. xAI has two positions for sales to international governments, located in London and Dubai, as well as one aimed at the U.S. government. These roles suggest that government will be an important revenue source for foundational model labs.

Unlike Anthropic, OpenAI, and xAI, DeepMind’s recruitment page shows almost no sales activity, as the distribution of Gemini is managed by Google’s existing sales organization.

Recruitment pages reveal new product directions for OpenAI and DeepMind.

Job postings also indicate what each company is developing. Anthropic has 5 product and engineering positions focused on improving Claude Code, while OpenAI has 10 similar positions aimed at enhancing Codex. Each company has one engineering position related to financial services. OpenAI also has 3 positions focused on new features for ChatGPT Health and OpenAI for Healthcare.

However, when it comes to existing products, the perspective from recruitment pages isn’t perfect. It’s hard to determine whether a position is expanding existing functionalities or building entirely new products, as platform or infrastructure roles often span multiple product lines. Therefore, recruitment information is most valuable when revealing “new bets.”

First, OpenAI is developing a consumer hardware device. This project has 15 open positions. These roles reveal several details: a “Camera ISP Software Engineer” position describes building an imaging system for a battery-powered portable device; a “Research Engineer” role focuses on directly running Transformer models on the device; and an “Operating System Engineer” position mentions self-developed chips. Taken together, this seems to indicate a portable device with a camera that runs on self-developed AI chips and can operate AI models at the edge. Additionally, two hardware and operations positions based in Singapore suggest preparations for manufacturing. DeepMind is also betting on hardware, with two XR glasses development positions, one of which implies that voice commands will be a core interaction method.

Beyond hardware, OpenAI has 2 positions for social products in incubation stages, and 1 position for an “Employment Platform” aimed at helping users train skills, obtain certifications, and match with employers. Anthropic has 1 research product manager position specifically exploring entirely new product categories, along with another for consumer new products.

Recruitment pages also reveal the strategies for acquiring computing power and data.

The recruitment efforts of various labs showcase their different approaches to core inputs (computing power and data). The most notable divide is between building internal computing infrastructure vs. outsourcing procurement.

OpenAI has 21 positions related to self-developed chips (primarily engineering roles), accounting for 3% of all current positions. Anthropic has not pursued the self-developed chip route and has chosen another path: multiple positions focus on collaborating with external partners to design and build data centers, including a “Data Center Mechanical Engineer” responsible for guiding external companies on cooling and mechanical system design, and a “Data Center Design Execution Lead” who connects Anthropic’s technical needs with third-party delivery partners. Anthropic also has 3 legal positions specifically responsible for negotiating data center or co-location hosting contracts.

Another prominent direction in recruitment is the reinforcement learning training environment. Anthropic has multiple positions focusing on building environments for training new capabilities, including a team for “Environment Expansion” responsible for constructing RL environments and managing vendor relationships, as well as a “Universes” team building hyper-realistic long-duration agent training scenarios. OpenAI is also hiring researchers for a “Synthetic RL” team to develop RL training methods based on self-play, simulators, and synthetic feedback.

Unlike OpenAI, Anthropic, and DeepMind (which do not have dedicated human annotation positions), xAI’s recruitment shows a different data strategy. It has 27 human data annotation positions, indicating that xAI prefers to keep data annotation work in-house. It’s also interesting that xAI is willing to publicly recruit for these positions. Other labs similarly rely on large-scale human annotation but typically opt for outsourcing or do not publicly recruit.

Conclusion

Job postings are an imperfect signal source, but they are one of the few publicly accessible windows into the evolution of leading AI labs. The current picture is that these companies are heavily investing in sales and product implementation, expanding into new product categories, and competing for key resources such as computing power and data. As various labs continue to expand and their strategies gradually diverge, their recruitment pages will remain one of the best observation windows.