A team of 37 researchers from Stanford, CMU, Michigan and other top schools argue that our centuries-old paper format is breaking down in the AI era. They point to two hidden costs: the “narrative tax,” where failed experiments and dead ends vanish behind a tidy success story, and the “engineering tax,” where papers skim over critical details—tricks, pitfalls and code—that an AI would need to reproduce results. Their answer is ARA, or “AI-ready research packages.” Think of each paper as a living bundle: full datasets, executable pipelines, complete code, decision logs and documentation of every wrong turn. Instead of feeding AI agents a polished summary, you give them the full journey, warts and all. That means no more head-scratching over missing parameters or buried Slack debates. They flip the question from “How can AI help humans write papers?” to “What should research look like when AI reads and runs it?” In their view, impact will be measured by how easily an AI can trace every step, rerun experiments and build on the work—rather than by narrative flow or page count.