Core Automation Recruits Researchers From Anthropic, DeepMind

Core Automation, a startup founded by former OpenAI researcher Jerry Tworek, has recruited researchers from Anthropic and Google DeepMind and announced it is "building the world's most automated AI lab." Tworek lists himself as CEO and cofounder on his X profile. One early hire, Rohan Anil, who previously worked at Google DeepMind and Anthropic, said Tworek "nerdsniped" him into leaving Anthropic a few weeks ago to join the new venture.

The company message centers on automating research workflows and systems that "optimize and automate work, starting with research itself." The article does not disclose technical stack, funding, or product milestones. For practitioners, an "automated AI lab" typically implies investment in reproducible experiment orchestration, automated evaluation, and integrated model lifecycle tooling. Likely building blocks include:

• •experiment orchestration and metadata tracking to accelerate iteration and eliminate manual bookkeeping
• •automated evaluation pipelines and benchmark suites to reduce researcher overhead and speed comparative analysis
• •model and data pipeline automation, including hyperparameter search, continuous training triggers, and programmatic dataset curation

Talent movement from top labs into focused startups is a persistent signal that researchers value tight feedback loops and engineering leverage. Core Automation frames itself differently from consumer-facing startups: the priority is researcher productivity and systems-level automation rather than direct productization or model scale alone. That approach can yield disproportionate returns if it materially shortens experiment cycles or automates routine research tasks, allowing smaller teams to compete with scale advantages enjoyed by large labs.

Monitor early engineering hires, open-source contributions, technical blog posts, and any experiment-management or evaluation tooling the company releases. Those artifacts will reveal whether Core Automation is delivering novel automation primitives or repackaging existing MLOps patterns.