Custom GPT Migration for Creative Workflows: Why I Built Aurelda Studio

When OpenAI announced that custom GPTs were moving toward retirement, I assumed I had a migration problem. I had built The Aurelda GPT to help me work with a large creative universe: three novels, character and visual bibles, lore, editorial standards, reference images, audio, video, and publishing work.

Then the better question appeared: why recreate the old thing at all? If I was going to migrate it, I could redesign how AI worked with Aurelda. That turned a technical transition into a case study in AI creative consulting: what happens when an AI system is built around a creator’s real body of work instead of a stack of clever prompts?

The Custom GPT Solved the First Problem

The original Aurelda GPT gave ChatGPT context. I no longer had to explain from scratch who Mo’an was, how Lumina worked, or why a generic fantasy image was not automatically an Aurelda image.

But Aurelda kept expanding. Final manuscripts sat beside developing manuscripts, character bibles, visual references, architecture guides, sound canon, editorial rules, and production notes. The problem was no longer whether the AI knew Aurelda; it was whether it knew which part of Aurelda mattered for the task in front of us.

Why Aurelda Studio Is Different

OpenAI’s plugin model allows reusable skills and reference material to live together in one system. Skills tell ChatGPT and Codex how to approach repeatable workflows, which made it possible to organize Aurelda Studio around responsibilities instead of forcing every task through one set of instructions.

Canon Stewardship handles story truth, continuity, and the boundary between established and developing material. Visual Canon handles likeness, environments, artifacts, Lumina, and approved imagery. Editorial, Publishing & Localization handles essays, SEO, metadata, and public-facing language, while Audio & Video Production handles trailers, narration, storyboards, and production planning.

They all serve the same world, but they do not need to think the same way. Better AI is not always about giving a model more information; often it is about giving it clearer rules for when that information should matter.

The Real Work Was Not Prompt Engineering

Building Aurelda Studio clarified what I mean by AI creative consulting. The valuable part was not writing a magical master prompt. It was deciding what the system should trust, what it should question, what it should never invent, and when it should stop and ask me.

For Aurelda, that meant creating a source hierarchy. The final trilogy manuscripts hold the highest narrative authority, approved bibles govern their subjects, and draft material can inform developing work without silently becoming final canon. If two sources genuinely conflict, the AI is supposed to surface the conflict rather than improvise an answer.

That is a fantasy-world problem, but it is also a business problem. Brands, nonprofits, founders, authors, and creative teams all accumulate knowledge in different places. The harder AI question is often not “How do we generate more?” but “How do we help the system understand what is authoritative, what is contextual, and what still requires human judgment?”

The Best Test Was an Essay I Could Not Get Right

Aurelda Studio proved its value while I was working on an essay about cultural appropriation, the white-savior trope, and the way Aurelda might be judged from its imagery before someone had read the books. My earlier drafts were defensible, but they felt like a legal brief. I was explaining why the work should not be misunderstood instead of saying what I actually wanted to say.

The conversation changed when the AI stopped trying to improve the draft and started asking better questions. What did I really see when I looked at Mo’an? What had living in Mexico changed in me? Why did I keep building Aurelda when almost nobody seemed to be watching?

Those questions led somewhere a rewrite prompt had not. The finished essay became less about proving Aurelda innocent and more about relationship, belonging, grief, and the danger of assuming that the first thing we recognize is the whole story. One of its central ideas was that Jason, the autobiographical outsider in Aurelda, needs Aurelda more than Aurelda needs rescuing by him.

Why This Matters for Aurelda

Aurelda is no longer only a trilogy. It now lives across writing, visual canon, audio, video, websites, essays, publishing systems, and experiments that have not been invented yet.

Aurelda Studio gives that work a structure that can grow. Different skills can handle different jobs, relevant source material can be consulted when needed, and final canon can remain distinct from work still in development. The goal is not to automate Aurelda; it is to make the technology better at supporting the way Aurelda is actually created.

Human-Led AI Is Still the Point

OpenAI is scheduled to retire custom GPTs on December 11, 2026, with migration toward plugins underway. For me, that deadline forced a larger question: was I preserving an old AI tool, or building a better creative system?

I chose the second. Aurelda Studio does not decide what the story becomes, what belongs in canon, or what I should say. It helps me hold a complicated body of work together, notice contradictions, retrieve the right guidance, and ask better questions before I make those decisions.

That is also the kind of AI creative consulting I want to do for other people. The most useful system is not the one that replaces the human at the center; it is the one designed carefully enough to understand what that human is trying to protect, build, and carry forward.