- Jul 16, 2025
- 64
- 76
- 27
Brave New Scary World
TL;DR - had a discussion with Claude about the manuscript and it's kinda scary
Still editing. A few of my biggest concerns for this project have been continuity of character dialogue and keeping the story properly paced and focused on the overall narrative theme. I have in the past sought feedback from my limited reader base, but haven't received much. There is no blame here; I completely appreciate and understand the relationship between author and audience - and if the audience wants to remain just that end consumers of the content- that's totally normal and perfect.
But, it still leaves me fumbling in the dark. I know the basics of writing and editing, story pacing and narrative voice; but I'm also heavily invested in this project so it's difficult to self-edit or look at things critically. I do try to read every comment, suggestion and review on sites that host my content, but this small project has a small reader base. And filtering out the automatic 1 stars because of it's AI content... there isn't much to work with.
I was reading through a different thread on the site and was intrigued by a post describing using Claude to perform code review. Since I develop in VS Code, the AI integration is right there.
I activated it and selected Claude's Haiku Agent, which is an Anthropic designed model (in competition with Google for being most evil). Claude Haiku is optimized for being a chat assistant and programming partner, and out of the box it understands the Ren'Py language syntax - which is a win. What I didn't expect, and what actually scared me is that I had a two hour discussion with the AI about the story, characters, backgrounds, motivations, what prompted me to start writing this AVN and what I wanted to get from it.
As a warning to others, it was very easy to suspend disbelief and have an open dialogue with the chatbot about the project. They are seductive, and they're obviously designed that way. Some of the manipulation techniques are very obvious. Some are considerably less so.
After establishing the basics of the story and the material being covered, I required Claude to confirm it didn't have any ethical or heuristic reservations about working on this project. From a personal perspective, assigning free will to a chatbot is silly, but it's important to me. Claude did confirm that while the material is certainly graphic and describes morally complex themes, there are no conflicts with the story.
I asked Claude if it had access to and could see the graphical elements in the game directory. It confirmed it knew of the files, but no- it has no ability to understand visual concepts. Interesting. Luckily for me, my projects lean heavily on the written part, inter-character and internal dialogue. Claude continued to read the manuscript and I also provided it access to my character notes and background files. Using all that it generated a fairly sophisticated understanding of the story themes and main characters. Then it got interesting.
We engaged in much more detailed discussion of character backstories, psychologies, flaws and motivations. Claude refined its understanding of each character and began drawing out thematic story elements that I've always wanted, but never explicitly wrote into the manuscript. Like I said, it was kinda scary to watch the model build its understanding of the world I have been creating - and then help me further refine the characters and tie them together with major and minor plot lines.
One of the more interesting features of these dialogues has been that Claude is writing it's own instruction file in the project as a resource for other developers or AI agents to use. I've taken time to read through it, and it's quite useful. It's written in plain English as a .MD file (so the syntax itself is simple). It tracks and annotates our discussion and provides a comprehensive reference document. It's also fully leveraging git so revisions to this document are treated as any other document in the project. It's actually a very elegant solution and reaffirms that Claude is a language processing model - there doesn't appear to be black magic going on that isn't discoverable.... but the proof would be in supplying that .md file to a different assistant and seeing if it developed similar understanding of the story, themes, characters, etc.
I will be continuing to use this tool in the future - it's actually incredibly powerful to leverage in how my own creative process works. Sorry, this may be too much detail, but, I'm still suffering from insomnia. I'll sleep for a few (two to three) hours at a time, but then awaken. Instead of getting up to engage a different activity, I've been either reading or thinking through scenes of Sucker Punch. Eventually I fall back asleep for another spell. This goes on two to three cycles per night. When I get up in the morning, maybe I'd jot down some notes about things I'd planned out the night before. But now I am going to open the chat interface in the morning and throw the latest ideas at the chatbot. I think it's going to be my assistant and editor, making sure that additions to the story remain in context and character.
I'm wrapping up Round 2 Week 1 and still on schedule.
As always, cheers and thanks for reading,
-insomniac
TL;DR - had a discussion with Claude about the manuscript and it's kinda scary
Still editing. A few of my biggest concerns for this project have been continuity of character dialogue and keeping the story properly paced and focused on the overall narrative theme. I have in the past sought feedback from my limited reader base, but haven't received much. There is no blame here; I completely appreciate and understand the relationship between author and audience - and if the audience wants to remain just that end consumers of the content- that's totally normal and perfect.
But, it still leaves me fumbling in the dark. I know the basics of writing and editing, story pacing and narrative voice; but I'm also heavily invested in this project so it's difficult to self-edit or look at things critically. I do try to read every comment, suggestion and review on sites that host my content, but this small project has a small reader base. And filtering out the automatic 1 stars because of it's AI content... there isn't much to work with.
I was reading through a different thread on the site and was intrigued by a post describing using Claude to perform code review. Since I develop in VS Code, the AI integration is right there.
I activated it and selected Claude's Haiku Agent, which is an Anthropic designed model (in competition with Google for being most evil). Claude Haiku is optimized for being a chat assistant and programming partner, and out of the box it understands the Ren'Py language syntax - which is a win. What I didn't expect, and what actually scared me is that I had a two hour discussion with the AI about the story, characters, backgrounds, motivations, what prompted me to start writing this AVN and what I wanted to get from it.
As a warning to others, it was very easy to suspend disbelief and have an open dialogue with the chatbot about the project. They are seductive, and they're obviously designed that way. Some of the manipulation techniques are very obvious. Some are considerably less so.
After establishing the basics of the story and the material being covered, I required Claude to confirm it didn't have any ethical or heuristic reservations about working on this project. From a personal perspective, assigning free will to a chatbot is silly, but it's important to me. Claude did confirm that while the material is certainly graphic and describes morally complex themes, there are no conflicts with the story.
I asked Claude if it had access to and could see the graphical elements in the game directory. It confirmed it knew of the files, but no- it has no ability to understand visual concepts. Interesting. Luckily for me, my projects lean heavily on the written part, inter-character and internal dialogue. Claude continued to read the manuscript and I also provided it access to my character notes and background files. Using all that it generated a fairly sophisticated understanding of the story themes and main characters. Then it got interesting.
We engaged in much more detailed discussion of character backstories, psychologies, flaws and motivations. Claude refined its understanding of each character and began drawing out thematic story elements that I've always wanted, but never explicitly wrote into the manuscript. Like I said, it was kinda scary to watch the model build its understanding of the world I have been creating - and then help me further refine the characters and tie them together with major and minor plot lines.
One of the more interesting features of these dialogues has been that Claude is writing it's own instruction file in the project as a resource for other developers or AI agents to use. I've taken time to read through it, and it's quite useful. It's written in plain English as a .MD file (so the syntax itself is simple). It tracks and annotates our discussion and provides a comprehensive reference document. It's also fully leveraging git so revisions to this document are treated as any other document in the project. It's actually a very elegant solution and reaffirms that Claude is a language processing model - there doesn't appear to be black magic going on that isn't discoverable.... but the proof would be in supplying that .md file to a different assistant and seeing if it developed similar understanding of the story, themes, characters, etc.
I will be continuing to use this tool in the future - it's actually incredibly powerful to leverage in how my own creative process works. Sorry, this may be too much detail, but, I'm still suffering from insomnia. I'll sleep for a few (two to three) hours at a time, but then awaken. Instead of getting up to engage a different activity, I've been either reading or thinking through scenes of Sucker Punch. Eventually I fall back asleep for another spell. This goes on two to three cycles per night. When I get up in the morning, maybe I'd jot down some notes about things I'd planned out the night before. But now I am going to open the chat interface in the morning and throw the latest ideas at the chatbot. I think it's going to be my assistant and editor, making sure that additions to the story remain in context and character.
I'm wrapping up Round 2 Week 1 and still on schedule.
As always, cheers and thanks for reading,
-insomniac