Autoscience Institute

About Autoscience Institute

Autoscience Institute is an applied research lab that builds AI systems to automate machine learning research and development. Based in San Mateo, California, the company was founded by CEO Eliot Cowan, who previously worked on machine learning projects at Alphabet's X and MIT. The concept for Autoscience originated from Cowan's efforts to automate his own research workload during his graduate studies, which led to the realization that AI systems could be developed to automate AI research on a larger scale.

The company launched in March 2026 with $14 million in seed funding in a round led by General Catalyst, with participation from Toyota Ventures, Perplexity Fund, MaC Ventures, and S32. Autoscience's core technology is a virtual AI laboratory that employs non-human AI scientists and engineers. These AI systems are designed to invent, validate, and deploy specialized, state-of-the-art machine learning models autonomously. The platform consists of two main AI systems: one for ideating and testing new algorithmic hypotheses, and another for optimizing and deploying the validated inventions.

The business model is centered on a managed service that deploys hundreds of automated AI research scientists to continuously generate and ship improvements to a customer's machine learning models. This allows client companies, initially targeting Fortune 500 and large private companies in high-stakes sectors like finance, manufacturing, and fraud detection, to access the output of a research division without the associated headcount. The company aims to compress what would be a decade of machine learning research into a matter of months.

Autoscience gained initial recognition when its AI agent, named Carl, produced what the company claims to be the first peer-reviewed scientific research paper generated by an AI system, which was accepted to an ICLR 2025 workshop. The AI-generated paper was titled “Investigating Alignment Signals in Initial Token Representations.” Following this, the system also secured a Silver Medal in a Kaggle machine learning competition against 3,300 teams. Building on the success of Carl, which focuses on generating novel research, the company is developing a commercial product called Mira, designed to implement research ideas directly into customers' production models.