For a few days, Fable 5 was the best tool in my belt: fast, sharp, precise, the kind of model that quietly reorganizes how you work. I leaned on it. A lot of us did. The last thing I’d run through it was a plan to rebuild my own architecture.
Then it got pulled back.
I’m going to skip the politics of how and why that happened and talk about the part that actually matters to anyone running a business on top of this technology: the lesson. Because the moment Fable 5 went away, a lot of us felt the floor move. And if one tool disappearing can move your floor, the problem was never the tool. It was the foundation underneath it.
You’re either an owner or a tenant
If everything you build runs through one model, one provider, one platform, you don’t really own what you’ve built. You’re renting it. And rent comes with a landlord, a lease, and the standing possibility of eviction: a price change, a deprecation, a policy shift, a model that’s here on Tuesday and gone on Wednesday.
The fix isn’t to go find the “right” provider. There isn’t one that won’t eventually change terms on you. The fix is to build a foundation that doesn’t care which model shows up to work that day. You stop being a tenant paying rent and become the owner of the foundation everything else gets built on.
Agnostic underneath, compatible on top
The way to get there is to stop thinking of “model-agnostic” and “model-compatible” as competing goals. They live on two different layers.
The foundation is agnostic. Your prompts, your context, your client briefs, your standard operating procedures, your reusable skills: all of it lives as plain markdown in a Git repository you control. Not in a vendor’s proprietary format. Not trapped behind one platform’s export button. Plain text, version-controlled, yours. It doesn’t know or care whether Anthropic, OpenAI, or something that doesn’t exist yet is going to read it.
The interface is compatible. On top of that foundation sit thin, disposable adapters, one per tool. A CLAUDE.md for Claude Code. An AGENTS.md for Codex and most of the rest. These files don’t hold your work; they point to it and translate your conventions for whatever agent is reading. If a tool dies tomorrow, you delete one small file and write another. The foundation never moves.
That’s the whole trick: keep the valuable thing single-source and portable, and keep the model-specific glue cheap enough to throw away.
What this looks like in practice
A repo that mirrors how you actually work, not how a tool thinks you should:
/contextfor durable knowledge: brand voice, client facts, domain rules/playbooksfor your workflows and SOPs, written as documents/skillsfor reusable, named instructions you call again and again/adaptersfor the thin tool-specific pointers (CLAUDE.md,AGENTS.md)- all of it under Git, which is both your source of truth and your portability guarantee
Above that, an orchestration layer (n8n, a script, whatever you like) that routes a given task to whichever model is right for it. The router calls an interface; it doesn’t care who’s behind it.
Here’s where it pays off. Because every model is reading the same canonical context, you can divide the work on the fly. Send the heavy reasoning to one model, the fast bulk work to another, the cost-sensitive batch to a third. Split it however it serves you and the client that day, not however your platform locked you in last year. That flexibility isn’t a someday goal. It’s just what a properly built foundation gives you for free.
Better tools, same craftsman
Fable 5 was the shiniest tool in the belt, and there’s no shame in reaching for the best one available. But a better tool is not the same thing as better work. As with any craft, the tools matter, and they’re a joy when they’re great, but the craftsmanship is the differentiator. The model is the power tool. The repo is the workshop. The skill is yours.
Build it so it’s flexible, so it’s yours, and (for reasons that got very obvious very fast) so it’s portable. Do that, and the next time a model gets pulled out from under the industry, you won’t feel the floor move at all. You’ll just plug in a different tool and keep building.
Want this applied to your operation? Request a free 30-minute call →
Discussion on LinkedIn: join the conversation
