Who Owns Your AI Assistant? The Lock-In Question Nobody Asks

Most people choose an AI tool by asking which one is smartest today. That's the wrong question, and it's the wrong question for a reason that only shows up a year later.

Here's what actually happens. You start using an assistant. You correct it — no, we always quote in writing first; that client prefers phone; never promise a date before checking stock. Month by month it gets genuinely useful, because it has absorbed how your business really works. That accumulated knowledge, not the AI itself, is the valuable thing you've built.

Then one of three things happens: the price goes up, a noticeably better model launches somewhere else, or the company changes direction. And you discover what you actually own.

The question to ask instead

"If I change the AI underneath this, what do I lose?"

For most tools the honest answer is everything that made it worth using. The preferences, the corrections, the examples of your voice, the connected accounts, the way you've set it up — all of it sits inside one AI company's system. Leaving means starting over, so you don't leave. You're not staying because it's the best option; you're staying because you can't afford the rebuild.

That's lock-in. It rarely arrives as a decision you make. It arrives as a decision that quietly stops being available.

Two ways to build the same product

There's a real architectural fork here, and it's invisible from the outside.

In the first design, the AI provider owns the assistant. Its memory, its instructions, its history — they live in the provider's system, in their format. The provider is the landlord and you're the tenant.

In the second, the application owns the assistant, and the AI is a swappable engine underneath it. Everything durable — who your teammates are, the skills you've taught them, what they remember, the accounts they're connected to, what each one is allowed to touch — lives in your workspace. The AI is asked to do the reasoning, hands back an answer, and holds nothing.

Same product on the surface. Completely different position for you when the market moves.

How to test it in five minutes

Ask any AI tool you're considering:

• Can I switch the underlying AI model? If no, the question is already answered.
• If I switch it, what carries over? You want: everything. Memory, training, connected accounts, permissions, history.
• Where is what it's learned about me actually stored? "In our system" and "in your workspace" are very different answers.
• Can I export it? Not to leave tomorrow — so that leaving stays possible.

The last one is the real test. Being able to leave is what makes it safe to stay.

Where we stand

We built My Headquarters on the second design, deliberately. Your teammates, their skills, your memory and documents, and your connected accounts belong to your office — not to an AI provider. You bring your own account from any major provider, you can run different teammates on different engines, and changing a teammate's AI is one choice that changes nothing else. Nothing to rebuild. Nothing to retrain.

We think it's like changing someone's glasses, not hiring a new person. Here's exactly what you own — and what it costs.