Problem

rekursiv.ai targets the human bottleneck in machine-learning research. Its thesis is that AI progress is increasingly constrained not by compute or data, but by how quickly researchers can generate, test, interpret, and refine ideas. Human research teams are expensive and scarce; experiments can take days, most fail, and the rate of progress is limited by how many hypotheses experts can investigate in parallel.

The clearest use case is automated ML experimentation and optimization. In one proof of concept, four AI Scientists ran 984 experiments over 30 days on Sudoku-Extreme, producing an algorithm that achieved 97% accuracy versus an earlier 85% state of the art while using 167 times less training compute. The company says that setup cost roughly $500 for a month, illustrating the potential to turn research that normally requires scarce expert labor and substantial compute into a much cheaper, continuously running process.

Product / Service

The company is building a cockpit for autonomous research: a user defines a problem, and the system dispatches a fleet of AI Scientists to form hypotheses, design evaluations, run experiments, review one another's work, interpret results, and trace claims back to evidence. Its agents are organized into complementary roles such as Scientist and Analyst, can coordinate experiments through a shared lifecycle, and can run unattended while still allowing a human researcher to redirect the work or contribute ideas. The site currently invites prospective users to request access, suggesting a controlled platform or research-engagement delivery model rather than a broadly self-serve product.

The benefit is an iterative scientific loop rather than a conventional coding agent that merely executes instructions. By running many experiments in parallel and learning from failures, rekursiv.ai aims to discover methods that human researchers might not try, reduce the cost of experimentation, and create a compounding system in which each discovery improves subsequent research. The longer-term product vision is a platform where humans and millions of AI Scientists collaborate on difficult research problems.

Market

rekursiv.ai sits in the emerging autonomous AI research, AI Scientist, and machine-learning discovery category. Its closest publicly documented analogue is Sakana AI's The AI Scientist, which automates idea generation, code writing, experiment execution, result analysis, paper production, and review. The distinction in rekursiv.ai's positioning is its emphasis on fleets of collaborating, self-improving researchers and on scaling the discovery loop, rather than only generating individual research papers.

The company appears to be very early stage. Y Combinator lists it as founded in 2026, part of its Summer 2026 batch, active, and based in San Francisco with a two-person founding team. Public traction is primarily technical: the company reports novel algorithm discoveries, the Sudoku-Extreme result, and an earlier example in which four AI Scientists and $500 produced a machine-learning paper. The public materials reviewed do not disclose paying customers, revenue, or a commercial-scale deployment, so rekursiv.ai is best characterized as pre-commercial or at least pre-revenue on currently available evidence, with access still requested directly through the company.