You spent six months building a custom GPT for your sales team. Onboarding scripts. Tone guidelines. Objection handlers. Vertical-specific call-prep prompts. The team uses it every day. Adoption is high. New hires get up to speed in a week instead of a month. It’s the most successful AI deployment the company has done.
Six months in, you get the email. The platform is changing pricing. The next tier triples your monthly bill or you lose three of the features your team relies on most. The team is happy with the tool. You spent six months making it work. The switching cost looks gross.
You sit down to look at the export options.
There aren’t really any.
You can copy and paste the system prompt. You can export a flat file of recent conversations. You can’t move the project memory the platform built about your accounts over six months. You can’t migrate the team-specific learnings the model has absorbed from custom instructions and feedback. You can’t bring the connections the tool has made between your CRM and your knowledge base and your customer history because each one is platform-specific glue.
What looks like the tool we built is half the tool.
The other half is sitting in the platform’s vault. You can’t touch it. You can’t take it with you. You can re-pay them. That’s the offer.
Here’s the thing. Your competitor down the street is in the same trap, with a different platform, paying a different lease. The lease has always been there. The platforms have always been competing on stickiness. The conversation has been about pricing or capability. The conversation should have been about portability.
Building on rented land
The phrase isn’t new. SEO operators have been talking about building on rented land for fifteen years. Build your business on Facebook’s algorithm, the algorithm changes, your traffic dies. Build on Google’s local pack, Google updates the local pack, your map listings disappear. Build on Amazon’s marketplace, Amazon changes the seller agreement, your margins shrink overnight.
Operators learned the lesson, partially. They diversified. They built email lists. They built their own websites. They paid for SEO instead of renting reach from a platform.
The lesson got applied to the demand side. It didn’t get applied to the supply side.
The supply side is where you do the work. Where the team writes prompts. Where the agents run. Where the project memory lives. Where the team’s accumulated learnings about how to talk to your customers about your products get stored, refined, and called back. That side has been quietly platform-locked for the last twenty-four months while everyone was paying attention to the demand-side lock-in.
Your team’s six months of careful prompt engineering is the new rented land. Same dynamic. Different surface.
What the platforms are actually selling
A pause. The platforms aren’t doing anything sneaky. They’re competing.
The features that make AI tools work better are the same features that make them harder to leave. Persistent memory. Project workspaces. Custom instructions that survive across sessions. Integrations with your other tools. Fine-tuning. Per-team or per-organization context that compounds over time.
Each one is a real productivity gain. Each one is also a stickiness mechanism. The platform that gives you the better experience holds the more valuable lock-in. From the platform’s perspective, that’s good business. From yours, it’s a lease that’s accruing whether or not the lease shows up on a line item.
I mean. Right? The math is symmetrical. The platform makes your team more productive by learning your team. The learning has value. The learning lives on their server. You bought their product; they got your data; their data became the thing your team relies on; your team relies on it more than they rely on your own repository of how-we-do-things. That’s how the lock-in compounds.
Six months in, the platform isn’t a tool you use. It’s an asset class you own a partial claim on. Your partial claim is what you can copy and paste. Theirs is everything else.
The Professional Recipe has been telling you this
The Professional Recipe has seven ingredients. Every AI station needs all seven. Training. Context. Guardrails. Examples. Output Over Process. Measurement. Feedback Loop.
Ingredient #2 is Context. What’s true about this customer, this job, this moment, this business. Last week’s piece (the compound knowledge base argument) said Context just got cheap to build. Point AI at the raw material you’ve been generating for a decade, let it organize and maintain a wiki, the kitchen never makes the same Context mistake twice.
That argument is half the picture.
The other half. Where does the Context live. The compound knowledge base sitting inside your custom GPT, your Claude project, your specific platform’s project memory, is doing the work. It’s also doing the work inside someone else’s house. If the house changes the rules, the Context goes back to being expensive overnight.
Context is now both cheap to build and expensive to lose. Those two facts are the whole story. Operators who only see the first half are building monuments in a city that keeps changing the property lines.
The Recipe doesn’t say anything about where Context lives. The ingredient just says Context. The ingredient is doing its job whether it’s stored in your folder on a local drive, in a Claude project, in a custom GPT, or in a flat text file that you own and can copy. The platforms blur this on purpose. “Just put it in our project workspace, it’ll be smarter.” That’s true. The trade is the part that doesn’t get said out loud.
The two postures
Operators land in one of two postures on this. I’ll name both.
The default posture. “The platform is working. Why would I worry about something that hasn’t happened yet?” This is the posture most teams have. They’ve built something that works. The team is happy. The output is good. Worrying about portability feels like worrying about hypotheticals. The cost is invisible. The benefit is visible. So they keep going.
The earned posture. “The platform is working AND I’m aware I’m building inside someone else’s house. I’m choosing the lease on purpose, with a copy of what matters in my own filing cabinet.” This is rarer. The operators who land here usually got there through a previous version of the same lesson. They had email lists that got reset by a platform change. They had social followings that got crushed by an algorithm update. They learned. When AI showed up, they extended the lesson into the supply side automatically.
The first posture isn’t wrong. It’s just unpriced. Every operator in the first posture is paying lease they haven’t decided to pay. Some of them will get to the end of their AI deployment lifecycle and never feel the bill. Most won’t.
The piece I’m writing isn’t get out of the first posture this week. The piece is price the lease. Decide on purpose. The lease isn’t free. Pretending it’s free isn’t a strategy. It’s a posture.
What porting actually looks like
A practical note. “Own your context” in 2026 doesn’t mean refusing to use platform features. It means keeping a parallel record, in a format you control, of the things that would be expensive to rebuild.
The system prompts you’ve refined. The custom instructions you’ve layered. The vocabulary your team has standardized. The examples that have proven themselves. The structured decisions about how we talk to customers in vertical X. All of that can live in a flat markdown file or a private repository or a folder you own. Costs nothing. Takes an afternoon to set up. Means that when the platform changes, you have the recipe. You don’t have the platform’s memory, but you have the bones.
The memory part is harder. The model’s accumulated learning from six months of feedback inside one platform is genuinely not portable. That part is lease. The question is whether the lease is worth what it’s costing.
Most teams won’t even have that conversation, because they haven’t named that the lease is being paid. The first move is naming. The second move is pricing. The third move is deciding whether to keep paying or start hedging.
The Monday Move
Pick one AI workflow your team has invested significant time building. A custom GPT, a Claude project, a specific system prompt that has been refined over months, an agent your ops team uses daily. Pick the one that would hurt the most to rebuild from scratch.
Open a doc. Two columns.
Column one. What I have a copy of in my own filing cabinet. The system prompt. The project description. The instructions. The examples you’ve trained against. Anything you could paste into a flat file today and own.
Column two. What’s only in the platform’s memory. The platform-specific project workspace. The conversation history. The accumulated tuning the model has done based on your feedback. The integration layer wiring your tools together inside the platform.
Look at column two.
That’s your accrued rent on this one workflow. That’s what you’d lose if the platform changed pricing tomorrow, or sunsetted a feature, or got acquired by a competitor you don’t want sitting on your team’s customer-prep notes.
Don’t fix it this week. Just see it. Decide whether the rent you’re paying matches the value you’re getting. If yes, keep going on purpose. If no, the portability question moves from invisible to live, and the next move is to start parallel-storing what’s recoverable.
Most operators have never run this exercise. The exercise is the move.
So.
Your context is capital. You just didn’t put it in your bank.
The platforms aren’t trying to trap you. They’re competing on stickiness. Every feature that makes the tool better also makes it harder to leave. That’s not a sneaky play. That’s the structure. The structure is fine if you’re aware of it. It’s expensive when you’re not.
The Context ingredient in the Professional Recipe got cheap to build last year. That’s the win. The same win has a quiet B-side. Cheap to build and easy to own are different claims, and most operators have been treating them as the same claim.
Price the lease. Decide on purpose. Keep what matters in a file you own. The kitchen still runs on the platform. The recipe still lives in your bank.
Source influences: external commentary on AI memory features and platform stickiness as the new lock-in vector. Distilled and operator-translated.
Framework spine: The Professional Recipe, Ingredient #2: Context. Read the full framework. Companion piece: Mike’s Head Is Your Wiki (the build-it-cheap side of the same ingredient).
~ source material · Source influences: external commentary on AI memory features and platform stickiness as the new lock-in vector. Distilled and operator-translated.
