Andrej Karpathy has released AutoResearch, an open-source system designed to automate AI experimentation and refine machine learning models [1, 2].
The tool addresses a primary bottleneck in AI development by reducing the amount of manual labor required to train and optimize systems. By automating the trial-and-error process of experimentation, the framework allows developers to iterate on model performance more rapidly than traditional manual methods [2].
Karpathy first introduced the concept during a keynote in San Francisco on June 17, 2025, where he said, "Software Is Changing (Again)" [1]. The project later moved to a public GitHub repository in early March 2026 [1].
Industry players are already attempting to integrate the tool into their workflows. Shopify reported a 53% speed increase [1] through the use of the system. However, reports indicate that this specific claim remains unmerged and has been flagged within the project's development community [1].
David Ondrej said AutoResearch is "an open source system designed to refine AI systems through automated experimentation" [2]. The system focuses on creating a loop where the AI can test hypotheses and adjust its own parameters without constant human intervention.
This shift toward automated research loops represents a move away from static training sets toward dynamic, self-optimizing environments. As the framework spreads, it may change how developers approach the constrained optimization of large language models and other complex AI architectures [1, 2].
“"Software Is Changing (Again)"”
The release of AutoResearch signals a transition in AI development from manual tuning to algorithmic optimization. If automated experimentation becomes the standard, the speed of AI iteration will likely increase, though the friction seen in Shopify's unverified speed claims suggests that quantifying these gains across different hardware and software environments remains a challenge.
