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

Summary

Agon is a research orchestrator built on the premise that LLMs make research production scalable, shifting the bottleneck from producing artifacts to judging claims. Its answer: validate what can be checked inside the workflow and leave the remaining judgments to human scientists. The system rests on six design principles — Prompt Economy, Future-Facing, Minimal Prompts, OmniDisciplinary, Massive Parallelism, and Zero-Code — and was run across domains for 444 iterations of Prompt Economy loops, starting from small topics with no human-written experimental code. The deployments demonstrate scalability while exposing new classes of failure, which the authors organize into a taxonomy along severity, fixability, visibility, and capability locus.

Contribution

Claimed: a scalable omnidisciplinary research orchestrator plus a failure taxonomy that separates failures the loops can see and fix from those that require human judgment, pushing toward a "machine scales, human steers" paradigm. What the abstract supports: a large-scale deployment (444 loop iterations) and the taxonomy; no quality benchmark against other systems or human baselines is mentioned.

Method

System paper with at-scale deployment as the evidence base. The abstract does not specify the domains covered, the underlying models, how outputs were judged, or any quantitative quality metrics — only the iteration count and the failure-taxonomy dimensions.

Relevance to RISE

Informs rq-formulation through paper-drafting at the orchestration level; catalog slug agon. The design question it poses — which validations live inside the automated loop and which judgments are reserved for humans — is a direct instance of the structure/aggregation allocation question in the Architecture as Epistemology paper, and its failure taxonomy (visibility and capability locus) is an attempt to operationalize what the loop can and cannot certify about its own output.

Critique / open questions

Whether "Prompt Economy" delivers quality rather than only volume cannot be assessed from the abstract: no external evaluation, no acceptance outcomes, no comparison condition. The failure taxonomy's inter-rater reliability and the base rates of each failure class are not stated. The claim that remaining judgments are "left to human scientists" leaves open how much human labor the 444 iterations actually consumed.

Key quotes

"Large language models are making research production scalable, shifting the bottleneck from producing artifacts to judging claims." (abstract)

"We organize these failures into a taxonomy along severity, fixability, visibility, and capability locus. The taxonomy separates failures the loops can see and fix from those that require human judgment." (abstract)

"Together, these results show that Agon is pushing research toward a new paradigm: machine scales, human steers." (abstract)