What do frontier AI companies' job postings reveal about their plans?
The last twelve months have made something clear about the frontier AI labs: they know how to build models. What they are still figuring out is how to make you use them.
Epoch AI analyzed open job postings at OpenAI, Anthropic, xAI, and Google DeepMind and the results tell a consistent story about where these companies are actually investing. At Anthropic, go-to-market roles now represent 31% of all open positions, up from 17% a year ago. At OpenAI, that share grew from 18% to 28%. Research — the thing these companies were founded to do — makes up just 12% of open roles at Anthropic and 7% at OpenAI. The data supports reading this as a shift in what kind of company each lab is becoming. Epoch AI frames the trend as evidence that the labs are evolving into enterprise software companies that also happen to build foundation models.
The sharpest growth within GTM is a specific subtype: roles dedicated to helping customers actually adopt AI. Adoption-focused roles at Anthropic doubled from 5% to 11% of open positions; at OpenAI they grew from 11% to 17%. These are not classic sales roles. They are AI Success Engineers, Partner Deployment Engineers, Solutions Architects, and Forward Deployed Engineers — people whose job is to walk a buying company's engineers through integration, show them where the leverage is, and debug whatever is blocking deployment. Epoch AI identifies these as a distinct hiring category, and the labs' own job descriptions confirm the technical nature of the work.
The implication cuts both ways. It means the adoption gap is real: enterprises are buying frontier AI but are not deploying it at scale without dedicated help inside the buying organization. That is a solvable problem — and the hiring signals the labs intend to solve it. But it also raises a question worth sitting with. Dario Amodei, Anthropic's CEO, said in May 2025 that AI could wipe out half of all entry-level white-collar jobs. The technology that might eliminate those roles is, right now, creating demand for the people who need to teach enterprises how to use it. The displacement is not here yet. The consulting layer to make the displacement possible — that is growing fast.
OpenAI has 21 open roles tied to its custom silicon effort, the most concrete sign yet that the lab is building its own hardware rather than relying entirely on commercial chip suppliers. Anthropic has taken the opposite approach: zero internal chip roles, but multiple openings for engineers to oversee datacenter design and construction with external partners, including a Data Center Design Execution Lead who bridges Anthropic's technical requirements and third-party delivery.
The hardware bets diverge further. OpenAI has 15 open roles describing a battery-powered portable device with a camera, running on custom silicon, with AI models executing on the edge — a description that tracks closely to what a portable AI device would need to be. Two Singapore-based hardware and operations hires suggest manufacturing conversations are underway. OpenAI also has seven robotics roles focused on training robots in simulation at scale, with the postings referencing soft components and production scaling. DeepMind has nine roles implying a humanoid robot with dexterous hands and two roles for XR glasses. Anthropic, by contrast, has no open roles that suggest new hardware. Its product and engineering hiring is concentrated in Claude Code improvements and a general research product manager role aimed at entirely new product categories.
xAI's hiring tells a different story about inputs. It has 27 open roles for human data work — labeling, quality control, data operations — which Epoch AI flags as suggesting a deliberate choice to keep that function in-house rather than outsourcing it. Anthropic and OpenAI do not advertise comparable roles at scale, implying they use external vendors or automated pipelines for this work. The contrast is not trivial: the choice to label data in-house versus outsourcing it shapes what kind of training signal your model gets and who is accountable for its quality.
On government sales, both OpenAI and Anthropic have 10 open roles targeting federal civilian, defense, and state and local buyers. Anthropic has two roles specifically targeting national security; OpenAI has one. xAI has two international government roles based in London and Dubai and one targeting the US government. This is a market all four labs expect to compete in seriously, even as the regulatory and contractual ground under frontier AI vendors remains contested.
Geographically, more than half of each company's sales roles sit in the US — 52% at Anthropic, 55% at OpenAI — suggesting the American enterprise market is still the primary prize. Both are hiring aggressively across Europe and Asia-Pacific, with Anthropic tilting toward Europe and OpenAI toward Asia-Pacific, concentrated in Japan, South Korea, India, Singapore, and Australia. Notably absent: China, the Middle East, Latin America, and Africa. The labs are not competing in markets where they expect national AI champions to block them.
Epoch AI's analysis comes with the standard caveats for job posting data: open roles do not tell you about current headcount, and a single posting might yield one hire or ten or none. A cluster of roles can signal a genuine bet or a headhunter's wish list. But across four companies, the signal converges: the labs are not primarily competing on research headcount anymore. They are competing on distribution, adoption, and the ability to turn model capability into workflow change inside large organizations. That is a different business than the one they started in.
Source: Epoch AI (https://epochai.substack.com/p/what-do-frontier-ai-companies-job)