Everyone's arguing about the wrong part of the Fable shutdown

Here’s what we’re reading and thinking about in the news this week:

Imagine this: It's Monday. You open the workflow your team runs every day and it returns nothing but errors.

Nobody broke it. The AI model behind it just doesn't exist anymore.

That's not a thought experiment. It happened to a lot of teams this weekend.

You might have heard: on June 9, Anthropic released Claude Fable 5, its most powerful model yet. Three days later it was gone. The US government issued an order to suspend access, and to comply, Anthropic pulled Fable 5 and Mythos 5 offline worldwide for every customer.

Hundreds of millions of people were using it on Tuesday. By Friday, nothing.

Everyone's arguing about the politics: the jailbreak, the national security order, the feud between Anthropic and the White House. Fine.

But that's not the story that matters —

A tool a lot of teams built their work around vanished in three days, with no warning and no way to migrate.

"Knowing which model to use for what" is not the skill

There's a popular idea right now that the hot AI skill is knowing which model to use for which job. Claude for this, GPT for that, Gemini for the other thing.

We call BS.

Knowing which model to use is being a good shopper in a store that can close without notice. You can be the savviest customer alive. It doesn't matter when the shelves are empty on Monday.

The real skill is being model-agnostic

Think about how you manage people. If you only know how to manage one specific type of person, you're stuck the day they quit. If you can define the work clearly enough to hand it to anyone, you're fine either way.

Models are the same. Betting everything on one model is betting your best employee never leaves. Being model-agnostic means you can hand the work to whoever shows up.

This is the whole idea behind something we teach called Own the Playbook, Rent the Tech. Your process is the asset. The model is a rental. You write down what the work is, the steps, and what good looks like, in a doc you own. Then you feed that to whatever model you're running. When the model changes, you don't start over. You swap.

What betting on one model actually costs

Early on, we worked with a sophisticated client who had built their prompts and workflows around specific models. Smart team, ahead of the curve. Then they noticed those models quietly shifting, behaving differently, giving different results for the same task.

We encouraged their AI operator to make the workflows model-agnostic. They did. They haven't had an issue since.

Here's the cost difference. When you own the playbook, migrating to a new model takes minutes. When you don't, it takes hours, days, or weeks. And not just rebuilding. Rethinking. Because you designed the whole thing around one model's quirks, you have to relearn a new one from scratch.

"But models really are different"

True. They have different strengths out of the box. Fable was genuinely good at things other models aren't.

But here's the part people miss. You can distill those strengths into your playbook. The thing one model does well, you can usually write into instructions that get you most of the way there in any model. The strength becomes yours, baked into a process you control, instead of locked inside a vendor who can disappear on you.

Try this this week

Take one workflow you run regularly. Write it out as a plain playbook in a doc: the inputs, the steps, and what a good result looks like.

Then run that same playbook in both ChatGPT and Claude.

Watch how different the two outputs are. That gap is your to-do list. It shows you exactly where your playbook is leaning on one model to fill in the blanks, and how much more you need to spell out so the model stops mattering.

UPCOMING EVENTS

Free workshop TOMORROW: How To Own Your AI Assistant (with Obsidian!)

We're hosting Nick Milo (Linking Your Thinking) for a free 75-minute session next week, on June 17. He'll show you how to combine Obsidian and Claude Cowork so your AI assistant runs on files you own, not data locked inside a platform.

Join us!

LINKS

For your reading list 📚

• Microsoft just put thousands of AI models behind one endpoint. The "which model should I use" question gets less interesting every week. Owning your process gets more interesting.

• 👀 An AI agent broke into a company and stole its database in under an hour, making its own calls the whole way. No human driving.

• Everyone's bracing for AI to cut jobs. Data from 163,000 workers says adoption hit 80% and it's speeding work up, not replacing it.

• Rhode Island banned therapy chatbots, Colorado vetoed a ban on algorithmic pricing, and New York sent seven AI bills to the governor. The rulebook is being written in real time.

That's all!

We'll see you again soon. Thoughts, feedback and questions are much appreciated - respond here or shoot us a note at [email protected]

Cheers,

🪄 The AMP Team (formerly: the AI Exchange Team)