Figure 1
This figure illustrates two contrasting workflows for transforming research outcomes into scholarly communication. On the left, the traditional publication pathway is depicted, where experimental results, datasets, simulations, and analyses are interpreted by authors and manually distilled into a linear narrative manuscript. This manuscript becomes the primary and final container of knowledge, optimized for sequential human reading but largely disconnected from the underlying data and reasoning once published. After publication, the document remains static, with updates or corrections requiring new submissions or separate publications. On the right, the Agentic Publication workflow presents a fundamentally different approach. Research outcomes populate a structured and continuously evolving knowledge base rather than being immediately compressed into prose. This knowledge base preserves data, metadata, methods, assumptions, and interpretations as addressable units. A custom large language model (L L M), configured for the scientific domain and constrained by validation, provenance, and governance mechanisms, operates on this structured content to generate an Agentic Publication. The Agentic Publication functions as an active knowledge object rather than a single document. It supports conversational querying, programmatic API access, and interoperability with other agentic publications, enabling both human users and machine agents to interact with the research. Importantly, the figure shows that traditional manuscripts are no longer the authoritative source but become derived artifacts. From the same agentic core, the system can automatically generate a print-ready manuscript that conforms to journal standards while preserving traceability to the underlying knowledge. The figure emphasizes a shift from document-centric publishing to knowledge-centric scientific communication.Simplified overview of how research outcomes yield in a traditional manuscript and an Agentic Publication (AP). The knowledge-based including structured and unstructured content (text) is coupled with a suitable AP LLM considering the AP properties described in this manuscript (red arrow), resulting in an AP. APs can be used for basic “chat with your paper” but also provide APIs for computational access (inc. to the data). APs can interact with other APs, but also auto-generate print-ready versions of traditional manuscripts. These auto-generated manuscripts are outputs of the AP for human consumption and accessibility, but the AP itself remains the primary citable entity