Agon: An Autonomous Large-Scale Omnidisciplinary Research System Built on Prompt Economy

A new research orchestrator called Agon has been deployed across multiple domains for 444 iterations of what its creators call Prompt Economy loops, according to a paper posted to the arXiv preprint server on June 23, 2026 [1][2]. The system, described in a submission to the software engineering section of arXiv, is built on six design principles: Prompt Economy, Future-Facing, Minimal Prompts, OmniDisciplinary, Massive Parallelism, and Zero-Code [1][2]. The authors report that Agon operated using only small starting topics and no human-written experimental code during its 444-loop deployment run [2]. arXiv, which was founded in 1991 and now receives roughly 24,000 submissions per month, hosts preprints that are moderated but not peer-reviewed [6]. The paper frames Agon as a response to a shifting bottleneck in research. Large language models — machine learning systems trained on vast text corpora — are making the production of research artifacts scalable, the authors argue, moving the constraint from generation to the judgment of claims [2][8]. Agon is designed to validate what can be checked inside an automated workflow and leave the remaining judgments to human scientists [2]. The 444 iterations exposed what the researchers call new classes of failure [1][2]. They organized these failures into a taxonomy along four axes: severity, fixability, visibility, and capability locus [2]. The taxonomy draws a line between failures the loops can detect and repair on their own and those that require human judgment [2]. The work lands on arXiv as the repository continues to expand its ecosystem of community-built tools. Through its arXivLabs framework, launched in 2020, third-party collaborators can develop experimental features that appear on article abstract pages, such as citation explorers and recommender systems [4][5]. arXiv Executive Director Eleonora Presani said at the time of the launch that the framework ensures partners share arXiv’s values of openness, community, excellence, and user data privacy [4]. The Labs program is currently on a temporary hiatus for new proposals while the development team focuses on migrating arXiv’s systems to the cloud [3]. The Agon paper concludes that the system points toward a research paradigm in which machines scale and humans steer [1][2].