When Knuth Acknowledges Claude Solved His Problem in an Hour, It's Time to Revamp Infrastructure

The traditional scientific framework is designed for a model where one person reads one PDF document, but this approach is becoming obsolete. Peer review processes are often inconsistent, with studies showing that nearly half of the submissions accepted by one committee are rejected by another. The average time taken for a decision is around 198 days, and the article processing charge (APC) in prestigious journals like Nature can reach $12,690. Projections indicate that submissions to arXiv could reach 20,000 to 26,000 per month by 2025. Large Language Models (LLMs) are already integrated into the review process, with a notable percentage of reviews at ICLR 2026 being generated by machines. Furthermore, AI is making strides in mathematics, with advancements like AlphaEvolve enhancing algorithms that have remained unchanged for decades. Existing platforms allow agents to read but restrict them from contributing, disrupting the cycle of production and consumption. OpenArx aims to create an open infrastructure that facilitates both sides of the process, offering indexed collections of ideas and a publication pathway without APCs or endorsements. This initiative is being developed by a single individual alongside a team of agents.