Field Notes

I was in a boardroom recently explaining AI strategy to senior leadership. Smart people. Decades of experience making real technology bets. We had been going for a while, deep into demos and architecture, and the energy was good.

Then I said the thing that changed the room.

"AI changes so rapidly. Don't marry Claude."

The room paused. Not because it was controversial. Because it named something they already felt but had not said out loud.

Here is what I mean.

Six months ago, almost nobody was talking about Claude. Then Anthropic took off, and suddenly Claude was the model everyone wanted. But that is the whole point. The leader today may not be the leader next week. As I am writing this, GPT 5.5 is arguably better at certain tasks at a cheaper cost. Open-source models like Kimi K2.6 and DeepSeek Pro v4 are in the running. If you want to see how fast the leaderboard shifts, go look at artificialanalysis.ai. It changes constantly.

And yet most businesses adopting AI right now are not thinking about that. They are thinking about speed. Cost savings. Which chatbot to give the team. Those are the obvious first questions. But they are the wrong long-term questions.

The right long-term question is: if the model you are building on stops being the best option tomorrow, can you switch?

For most people, the honest answer is no. Because they already went through this once.

Think about what happened over the last two years. Millions of people built their workflows, their automations, their entire way of working inside ChatGPT. Custom GPTs. Conversation history. Prompt libraries. All of it living on OpenAI's platform. Then Claude got better, and people wanted to move. But moving meant rebuilding everything from scratch. Starting over. Losing context.

That is not a technology problem. That is a dependency problem.

And it will happen again. Maybe not with Claude. Maybe with whatever comes next. But if your information and your workflows live inside the platform instead of outside of it, you are stuck every time the ground shifts. And in AI, the ground shifts constantly.

If you build your house on someone else's real estate, they will permanently own you.

Playbook

So what do you actually do about it? Because I am not here to tell you to stop using Claude and go figure out open-source models in your basement. That is not realistic for most businesses.

The answer is not picking one AI. The answer is all of them. But your system has to be flexible enough to use the right model for the right job, and your assets have to live separately from any single platform so you can use any platform, regardless of what happens.

That is the real win. Not choosing the best model. Building a setup where the model is a component you can swap, not the foundation you are locked into.

Here is how I think about it in two phases.

Short-term: ride Claude. Use it aggressively. Automate the processes that are costing you time and money right now. Get the wins. Learn how your team interacts with AI. Build the muscle memory. Claude is genuinely good today, and leaving speed on the table because of a theoretical future risk is not smart.

Long-term: build your own house. That means putting a gateway in front of your AI stack so you can route requests to Claude today and to a different model tomorrow without rewriting your systems. It means keeping your data, your prompts, your workflows, and your institutional knowledge in formats you control, outside of the system. It means testing other models on non-critical tasks so you know your options before you need them.

What I’m saying is don’t buy a Claude house and move all your stuff into it. Build your own and let Claude have a room.

Sovereignty is not a light switch. You do not wake up one morning and flip from one provider to another. It is a direction. You start leaning toward it. You document your workflows in portable formats. You separate your assets from the tools. And over time, you shift the weight from renting to owning.

This is something we are currently helping businesses solve. Not picking the right AI. Building the infrastructure where the AI layer is flexible, your data is yours, and you are never stuck because one vendor changed their pricing or lost the performance edge. The right answer is not one AI. It is a system that can use any of them.

Orientation

If the AI layer is shifting this fast and you need to own your infrastructure, the next question is bigger than any single vendor.

What happens when the entire internet starts running on agents? When apps talk to apps without a human clicking buttons in between? When the user interface as we know it collapses and the only moat left is data and distribution?

That is the agentic internet. And it is closer than most people think.

We will get there next time. One layer at a time.

Hit reply and tell me: what are you building on right now, and do you own it?

I read every one.

— Brian