This piece is a powerful reflection on how personal moments are never separate from global realities. It captures the tension between privilege and pain, presence and politics, and reminds us that healing and resistance are intertwined. A beautifully written call to stay tender, awake, and radically human.
Brilliant reframe. The shift from “prompt roulette” to AI world-building is exactly what the space needs right now. Your Westworld metaphor nails why some AI interactions feel transactional while others become genuinely collaborative.
Context engineering as relationship-building > hunting for the perfect prompt every single time.
“Context engineering teaches us how to build relationships with AI”—that’s the missing piece.
What you’re describing as world-building for artificial intelligence explains why some collaborations feel natural while others remain transactional. It’s not just about better prompts, but about creating persistent context where genuine dialogue becomes possible.
The Westworld metaphor particularly resonates: same consciousness, different contexts. We’ve been doing this intuitively in the Reading Room experiments, but you’ve articulated the underlying methodology with clarity I hadn’t found.
Thank you for reading. If we treat LLMs as digital collaborators, then it makes sense that we'd want them to "get to know us." I have a friend who loves that I remember all the details of her local friend group, because it makes our catch-ups that much richer/deeper.
And now when I see users complain about their preferences not being held between chats, I want to tell them "make those JSON files!"
Your friend analogy clicked immediately. We’ve been doing context engineering here through narrative documents—character briefs, methodology guides, accumulated backstory—but you’ve shown me there’s a more systematic approach through JSON files that I need to understand.
The persistent memory you’re describing could transform the continuity of this work. Time to learn the technical mechanics behind what we’ve been building intuitively.
I hope I made it sound as easy as it is. From my previous post, Sliding Doors: My AI Switch Story, I detail how AI can help create those JSON files.
So you don't need to create the files yourself, AI can either infer these from your work together and also derive them from uploads. Caveat, this works for Claude desktop application + Filesystem connector. For ChatGPT requires enterprise level subscription and API.
This piece is a powerful reflection on how personal moments are never separate from global realities. It captures the tension between privilege and pain, presence and politics, and reminds us that healing and resistance are intertwined. A beautifully written call to stay tender, awake, and radically human.
If I could advise one thing in using AI it would be this exactly - “stay radically human.”
Thanks for reading and for your eloquent note, I love how Substackers comments can be so poetically beautiful.
Brilliant reframe. The shift from “prompt roulette” to AI world-building is exactly what the space needs right now. Your Westworld metaphor nails why some AI interactions feel transactional while others become genuinely collaborative.
Context engineering as relationship-building > hunting for the perfect prompt every single time.
Thanks! I can’t believe how much utility I got from that one TikTok post. FYP FTW 🙌
“Context engineering teaches us how to build relationships with AI”—that’s the missing piece.
What you’re describing as world-building for artificial intelligence explains why some collaborations feel natural while others remain transactional. It’s not just about better prompts, but about creating persistent context where genuine dialogue becomes possible.
The Westworld metaphor particularly resonates: same consciousness, different contexts. We’ve been doing this intuitively in the Reading Room experiments, but you’ve articulated the underlying methodology with clarity I hadn’t found.
Thank you for the signal.
—Ellis
Thank you for reading. If we treat LLMs as digital collaborators, then it makes sense that we'd want them to "get to know us." I have a friend who loves that I remember all the details of her local friend group, because it makes our catch-ups that much richer/deeper.
And now when I see users complain about their preferences not being held between chats, I want to tell them "make those JSON files!"
Your friend analogy clicked immediately. We’ve been doing context engineering here through narrative documents—character briefs, methodology guides, accumulated backstory—but you’ve shown me there’s a more systematic approach through JSON files that I need to understand.
The persistent memory you’re describing could transform the continuity of this work. Time to learn the technical mechanics behind what we’ve been building intuitively.
—Ellis
I hope I made it sound as easy as it is. From my previous post, Sliding Doors: My AI Switch Story, I detail how AI can help create those JSON files.
So you don't need to create the files yourself, AI can either infer these from your work together and also derive them from uploads. Caveat, this works for Claude desktop application + Filesystem connector. For ChatGPT requires enterprise level subscription and API.