Core Automation hires researchers from Anthropic and Google DeepMind

Core Automation has begun hiring away senior researchers from Anthropic and Google DeepMind, according to Business Insider, in a move that signals how quickly today’s AI labs can be assembled by recruiting rather than inventing. The startup is led by a former OpenAI researcher and markets itself as “the world’s most automated AI lab,” a slogan that doubles as a promise to investors and a warning to competitors.

The immediate story is a familiar one in technology: a new lab arrives with a compelling narrative, a small founding team, and a list of departures from better-known rivals. But the deeper point is what that talent flow implies about where competitive advantage sits. If researchers can move in a pack from one frontier lab to another, the differentiator is less the published paper and more the surrounding machinery: proprietary training data, long-term compute contracts, and the ability to ship products that lock in customers. Those assets are harder to poach than people.

The branding around “automation” also reads as a positioning play inside the AI economy. Labs that automate their own research workflows—evaluation, data curation, synthetic data generation, code-writing, experiment orchestration—can plausibly run more iterations per dollar and per researcher. That matters in a sector where the constraint is no longer just ideas, but the cost of turning ideas into trained systems. A lab that can compress the cycle from hypothesis to model update can spend the same compute budget more effectively, even if it never matches the absolute scale of the incumbents.

For the established firms, this kind of spinout is both a leak and a hedge. When top staff leave, they take tacit knowledge about internal tooling, model behavior, and what failed as well as what worked. Yet the same ecosystem also helps incumbents: it pushes up the market price of AI talent and normalises compensation packages that only the largest balance sheets can routinely afford. The result is a market where “competition” often means reshuffling a small pool of specialists while the capital-intensive parts—chips, power, and distribution—concentrate further.

Core Automation’s pitch suggests it is trying to compete on operational tempo rather than raw size. Whether that works will depend on what the company can secure that is not portable: access to compute, a defensible data strategy, and a product surface where its models can earn money before the next hiring wave changes the map again.

For now, the most concrete fact is also the simplest: a new AI lab has started by taking people from the labs that already dominate the field.