Problem

Autoscience targets the human bottleneck in AI development: machine-learning teams may have substantial compute and data, but too little researcher capacity to generate, test, and iterate on new ideas at scale. The company frames this as an economic productivity problem in which the limiting resource is the time and expertise required to run experiments, discover better model designs, and translate findings into working systems.

The clearest use case is accelerating enterprise ML research and development. Instead of waiting for human researchers to investigate many possible improvements, an autonomous system can run experiments against a company’s own code, infrastructure, and evaluations, helping teams discover models that outperform their existing baseline more quickly.

Product / Service

Autoscience is building AI systems for autonomous AI research, with products aimed at both industry and academia. Its enterprise offering, Mira, operates within a customer’s technology stack, reads the codebase, runs experiments, and ships verified improvements as pull requests. The company says Mira can begin delivering verified improvements within a week and is offered through a pilot conducted alongside Autoscience’s research team.

Its academic-oriented system, Carl, is designed to hypothesize, run ablations, write full-length research papers, and submit them for peer review. Together, these products promise to convert portions of the ML research workflow from manual investigation into an automated, experiment-driven process, increasing the number of ideas a team can test without linearly increasing its human research staff.

Market

Autoscience competes in the emerging category of AI research and development automation, particularly autonomous agents for machine-learning experimentation, model improvement, and scientific discovery. The supplied company materials do not identify named direct competitors, so the most defensible comparison is with the broader set of AI-agent, ML-platform, and automated model-development products that help researchers write code, run evaluations, and optimize models.

The company appears to be early but has meaningful technical and financing traction. It reports that Carl became the first AI system with a peer-reviewed paper, at an ICLR 2025 Workshop, and Autoscience announced a $14 million seed round in March 2026 backed by General Catalyst, Perplexity Fund, Toyota Ventures, S32, and MaC Ventures. Its website also advertises enterprise pilots and an active research and engineering hiring program; revenue and customer counts are not disclosed in the available materials, so its commercial status cannot be established as either revenue-generating or pre-revenue.