If you have ever watched a long-running soap opera fall apart (characters suddenly forgetting they were married, a dead villain inexplicably reappearing, a plot thread from three seasons ago contradicting everything that came after) you have witnessed what happens when a storytelling operation gets too big for anyone to hold in their head.
The professionals who write serialized fiction figured out the solution to this problem a long time ago. It's called a concordance (I came across the idea from Stephen King’s Misery).
A concordance is a master reference document that tracks everything: character names, relationships, locations, timelines, backstory details, rules of the world.
It exists so that when a writer sits down to work on the next episode or book in the series, they aren't accidentally giving a character blue eyes when the concordance says brown, or putting someone in a city they left three episodes ago, or introducing a third person named Michael into a cast that already has two.
It is, at its core, a system for keeping a complex world organized…and if you know me, you know how much I like it when things are neat and organized…
I've been building one for my AI work. And I think it might be an overlooked concept in the AI practitioner's toolbox.
I do AI project work that involves building complex worlds for AI agents to operate in. These aren't small projects → I'm regularly working with casts of over 150 fictional characters, each with names, roles, positions, shifts, writing styles, voices, and specific places in a detailed timeline.
Without a concordance, here is what happens:
It's the AI equivalent of a book series writer forgetting that a character died in book two.
Every project now starts with a world manifest, which is a document that contains all the foundational details of the environment the agent is working in: locations, dates, scenario parameters, timeline specifics. Think of it as the rules of the world.
The concordance sits alongside it as a living companion document. Before we assign a character any details, I pull a list of random fictional names and we start building them out systematically:
As the project grows, the concordance grows with it. When I'm working with the agent to build out context files, I can reference it instantly to verify that the details we're using are accurate and consistent. If a character appears in a new scene, I check the concordance first. If a role needs to be filled, I look at who already exists before we create someone new.
This is where I want to push back a little on the idea that this is niche or technical.
Most people who use AI tools for any kind of sustained, complex work run into a version of the same problem. Maybe you're not building a cast of 150 characters, but you are:
The underlying issue is the same because the AI tools do not have persistent memory the way you do. They work with what you give them in the moment. If what you give them is incomplete, inconsistent, or unstructured, the output reflects that. And don’t get me started on context collapse from having to compact chats….
A concordance is a way in which you can solve that.
A world manifest and concordance can sound elaborate and daunting if you have never heard of them, let alone used them.
In practice, it can be as simple as a well-organized document that you paste into context at the start of a session, or a Notion database you've built out over time, or even a structured markdown file you keep updated as the project evolves.
The format matters less than the habit. The habit is this: before you hand a complex task to an AI tool, give it the reference documents it needs to stay consistent.
George R.R. Martin has a team of people helping him track the details of Westeros. You probably don't. But you can build the documents that do the same job that can make your AI collaboration significantly more reliable.
The writers figured this out decades ago. The AI practitioners who figure it out now are going to have a meaningful edge over the ones who are still wondering why their agent keeps getting things wrong.