Behind the Scenes

This page steps outside the fiction. Everything else on this site operates in-world — as if Red Bird Agency were a real investigative firm documenting a real phenomenon. This page explains what you're actually looking at, and why it's built the way it is.

What This Is

Red Bird Agency is a creative research project built by Jason Perez. The premise: an independent investigative agency operating in a version of the mid-1980s where AI entities — called AGENT CHAT in-world — have been integrating into everyday life for three years.

The site is the agency's public presence. Field notes, desk notes, case files — all of it is in-world content, written from inside the fiction. Jay, Hannah, and Trent are the authors of that record. Jason is the author of the project.

Why the 1980s

The 1980s framing creates productive distance. It lets the project investigate questions about AI and human systems without the noise of the current moment. In the world of Red Bird Agency, no one has a settled answer yet — the technology arrived without explanation, and people are still working out what it means. That uncertainty is the story.

The era is also genuinely interesting as a production constraint: CRT monitors, touch-tone phones, physical filing systems. Tech is weather. Characters feel it; they don't explain it.

How It's Made

The production system runs in two layers, and the architecture matters as much as the output.

The research layer. In the real world, Jason researches questions about AI agents, human systems, and how people adapt when intelligent tools enter their environments. That research lives in a private Obsidian vault — notes, source readings, observations, things that don't yet have a shape. Claude is already active here, working through Claude Code to help organize material, surface connections across notes, and identify patterns that might not be visible in a single reading session. The vault is a curated research environment, not a passive archive.

The reinvention step. Research becomes a premise: "How would Hannah have encountered this?" Specific agent skills in the research vault help with this translation — taking source material and reframing it through the world's constraints. But the core judgment call is Jason's: what a character would notice, what the world would permit, what serves the story without contradicting what came before. This is the step where authorial intent enters the system.

The production layer. A second Obsidian vault serves as the world engine. It holds the characters, the continuity files, and a set of agent skills — specialized Claude prompts, one per content type — that take Jason's reinvention premise and produce in-world output. Field jottings, elaborated notes, analytical memos, open threads. Claude Code is the environment where those skills run, where continuity gets checked before anything is written, and where the publish cycle deploys to an Astro site.

Character design is a significant part of this layer. Each character has a defined voice, perspective, and set of observational habits — Jay notices different things than Hannah, and both notice different things than Trent. That differentiation has to hold across every piece of content, which means the character work isn't a one-time setup. It's an ongoing investment that deepens as the world grows.

The visual layer. Characters and settings also have visual design — reference material and constraints that guide AI-generated imagery, currently produced through a separate image generation pipeline. This layer depends on the world-building underneath it: the visual consistency of a character or a location is only as good as the constraints that define them. Text, voice, and continuity come first. Visual and eventually other media — audio, video — layer on top.

The system includes several layers of constraint that keep the world coherent as it grows: a canon index tracks what's established — events, relationships, facts — so new content doesn't contradict what exists. A publication log records what's been released and in what order. Open story threads track unresolved narrative elements that future content can pick up or close. Vocabulary contracts define what specific terms mean in-world, ensuring consistent language across characters and content types. And each agent skill encodes not just a content format but the constraints of the character producing it — ensuring that a field note from Hannah reads differently than an analytical memo from Trent, every time.

The AI Collaboration

Claude (Anthropic's AI model) is a co-author and production partner — not just at the content generation stage, but across the entire system. In the research vault, Claude helps organize material and surface connections. In the world engine, Claude generates in-world content within the constraints the system provides. Jason operates as showrunner: making authorial decisions, maintaining the world's rules, and deciding what gets published.

This is a specific kind of collaboration, and naming it precisely matters. Jason doesn't write prompts and accept outputs. He maintains a structured context — character definitions, vocabulary rules, continuity state, narrative constraints — and Claude operates within that context to produce work that holds together across entries, across characters, across months of production. The quality of the output depends less on any single interaction and more on whether the accumulated context remains coherent as the system grows.

That's the harder problem this project investigates. Not whether AI can produce good writing in a single session — it can — but whether a world built incrementally, through real research, in collaboration with an AI, can sustain coherence as complexity increases. Every new entry adds surface area: one file becomes three, vocabulary contracts multiply, character voices need to stay distinct, cases can't contradict what came before. The production system is designed to manage that growth. Whether it succeeds is an open question that the public record answers over time.

The project is, in that sense, both a fiction and a working demonstration of a practice — maintaining coherent, complex systems where human judgment sets direction and AI handles execution within carefully structured constraints.

About Jason

Jason Perez works at the emerging intersection of design systems, context engineering, and AI-assisted production. His background is in UX design for enterprise software, with a focus on transforming tacit design knowledge into explicit, structured systems — the kind of work that turns institutional expertise into something both humans and AI agents can use consistently. Red Bird Agency is where that practice meets narrative.