Researchers built a dashboard so writers don't lose track of what their AI co-writer just did to the plot.
Tencent researchers published NarrativeSteward, an open-source authoring environment where AI agents draft and revise interactive narratives while organizing the results into linked outlines, worldbuilding notes, and story graphs instead of one long wall of generated text. The system keeps change records and runs execution checks so authors can verify what an agent actually altered before accepting it. In a 12-participant study, writers said it was easier to request revisions and understand what had changed compared to using a general-purpose AI agent. The code is posted on GitHub under Tencent's account.
The real problem with agentic writing tools was never generation, it's governance of generation. Once an agent can independently draft and revise large chunks of a narrative, the bottleneck shifts to whether a human can audit and redirect that work without re-reading everything from scratch. NarrativeSteward's bet is that structural visualization and change tracking, not better prompts, are what make delegating to an AI collaborator actually usable.
Twelve participants is a small sample, and feeling like you understood the changes is not the same as the resulting stories being any good.