D
Dramatic Parrotfish
The idea of agentic editing is - the LLM can
do
things for you like updating codex entries, saving snippets, creating/rearranging acts/chapters/scenes etc.
The biggest hole in the AI authoring experience right now is:
  1. In the writing pane, you can ask it to generate text but you cannot hold a conversation about the text being generated and provide feedback on what needs improvement. You can sort of simulate it by doing 'tweak and generate' with 'rephrase', but that's very one-shot and doesn't contain context of previous iterations and feedback you already gave.
  2. In chat mode, you can have a conversation, but the LLM can't directly write to your novel or codex. You can copy/paste back and forth, but it's clumsy.
Main scenarios:
  • Brainstorm with the LLM and have it automatically update codex entries (including showing diffs and asking for confirmation)
  • Workshop a scene with the LLM, where both you and the llm make edits in the 'write' window, and you can have a frank conversation about what you like/dislike and what you want improved. Best if the LLM can have context on things like where the cursor currently is, text that's currently selected, etc.
  • As a result of having such a frank conversation, having the ability to say something like - "Ok, extract the principles/stylistic preferences I've expressed here and save as notes in my style guide".