Don’t Marry the Model: How to Make Your AI Work Portable

by Lisa Peyton and her team of AI super heroes

On the morning of June 12, two of Anthropic’s frontier models were available like any other day. By that evening they were gone. Not down for maintenance, not rate-limited. Gone, for a huge chunk of users, pulled by a US government export control directive that landed with no warning and no transition window. All over a jailbreak claim that Anthropic looked at, disputed, and called a likely misunderstanding it’s working to reverse. Gravitee wrote up the whole mess, and it’s worth ten minutes of your time.

So let me get this straight. A tool people were doing real work on vanished in an afternoon, over something the company that built it says probably isn’t even a real problem.

If that doesn’t make you wonder where your work actually lives, nothing will.

Now the good news, and it’s the part the panic headlines skip right over. You’re not as exposed as that story makes you feel. You’re already more resilient than you think. You just have to stop leaving the keys to your work inside one model’s front door.

Your Work Was Never In the Model

Let’s clear something up, because it changes how you react to news like this.

You’re not dependent on Claude. Or ChatGPT. Or Gemini. You’re dependent on the CONTEXT you’ve built with whatever model you’ve been living in. The client background you fed it. The brand voice you spent three sessions calibrating. The decisions you made and the dead ends you ruled out. The campaign you were three prompts away from nailing.

That context is the asset. The model is just the room it happened to be sitting in this week. Treat the model as the asset and every suspension feels like a heart attack. Treat your context as the asset and a suspension is a minor annoyance. You grab your stuff and walk to the next room.

The Engineers Already Solved This. So Can You.

When those two models went dark, the advice flying around for engineering teams was architectural. Put a proxy in front of your models. Build a gateway so that when one provider gets pulled, you reroute to another one without rewriting your whole app. Smart, and exactly right, if you’re running production systems with agents calling models all day long.

You’re probably not doing that. You’re a marketing consultant in a chat window with a model that knows your client cold. No engineering team, no gateway. What you’ve got is a brain, a browser, and a deadline.

So you need the consultant’s version of the same idea. Not a proxy. A go-bag. One file you can grab and carry into any model the second you need to, so you’re back up and running in about a minute instead of rebuilding three weeks of context from memory.

The Lock-In Is the Product

Quick gut check on the hype machine while we’re here.

A lot of the AI noise is built to make you feel like your chosen tool is irreplaceable and like switching would cost you everything. That panic isn’t an accident. The more locked in you feel, the better it is for everyone selling you the lock. The dependency IS the product.

The move that actually protects you is the unglamorous one nobody’s running a course on. Own your context. Keep it portable. Become the kind of practitioner who can lose access to any single tool on a Tuesday and just shrug.

The Hack That Daylene Handed Me

I’d love to tell you I came up with this. I didn’t.

One of my standout students, Daylene Long, handed me this hack, and I’ve run it at the end of every working session since. It’s almost embarrassingly simple, which is exactly why it works.

Before you close the tab, you have your AI write a handoff file. One plain markdown document that captures everything a fresh model needs to pick up right where you left off. Markdown matters here because it’s plain text. It opens anywhere, in any tool, and any model reads it instantly. No lock-in, which is the whole point.

Then when your model gets suspended, deprecated, geo-blocked, or just starts behaving strangely after some update, you don’t start over. You open a new tool, paste in the file, and you’re running.

What Goes In the Go-Bag

A handoff file is not a chat log. A chat log is mostly noise, and half of it is you typing “no, not like that.” The handoff is the signal, written so the next model gets up to speed fast. Here’s what it captures:

• Session context: what you were working on and where things stand right now
• Working preferences: how you like things done, so the next model doesn’t relitigate your style
• Decisions with the reasoning: what you chose and why, so nothing gets re-argued from scratch
• Verbatim language locked: the exact wording you’ve already approved and won’t let drift
• Open questions: what’s still undecided and what’s riding on it
• Artifacts created: what exists, what state it’s in, and where to find it
• Next steps: the specific moves you want to make next, so you start in motion instead of cold

Preserve the why behind every call, not just the call. A file that says “we went with option B” is useless. A file that says “we went with option B because the client hates jargon and option A was full of it” gets the next model defending your decision instead of reopening it.

Do It At the End, Every Time

This is the whole trick, and it’s a habit, not a heroic act.

You don’t build the go-bag when disaster hits. You build it as your close-out ritual, every single session, while the context is still fresh in the model’s memory. I’ve turned mine into a skill I run on command, so it takes one line and about twenty seconds.

Picture future-you. It’s 5pm on a Friday, a directive just dropped, your model is dark, and your client deliverable is due Monday. A total sh*t storm. Except present-you spent twenty seconds packing the go-bag, and just saved that person’s whole weekend. Be the friend who packs the go-bag.