Your entire research workflow lives inside one platform.
They change their pricing.
You rebuild from scratch — or you pay.
You didn’t buy a workflow.
You rented one.
The next moat is not the model. It’s the stack.
For the first phase of the AI wave, the model was the product. GPT-4, Claude, Gemini — which model you used determined most of what you got. The model was the moat.
That phase is ending. The models are converging in capability. The new competition is happening at the layer between the model and your workflow: retrieval systems, memory, orchestration, context management, output routing. Who controls the stack between the model and the work is the next strategic question.
Composable architectures — where you can swap the model, the retrieval layer, and the interface independently — are winning the architecture debate. The reason is simple: they give you capability without lock-in.
The stack around it is becoming the moat.
Own your workflow or rent it. That is the decision.
Every time you build a critical workflow inside a single vendor’s closed platform, you create a dependency. The workflow works until the platform changes something. Then you face the lock-in cost.
Composable systems have a higher setup cost and a lower lock-in cost. Monolithic platforms have a lower setup cost and a higher lock-in cost. The decision you make now is about which cost you want to pay later.
For workflows that are nice to have, lock-in risk is tolerable. For workflows that are core to how you operate — the ones you would rebuild from scratch if the platform disappeared — the composability question is existential.
Lock-in cost is paid forever.
Map your lock-in before it becomes a crisis.
The practical discipline is a lock-in audit for your most important AI workflows. For each one, ask: if this platform disappeared tomorrow, what specifically would break? How long would it take to rebuild? What data would you lose?
That is your lock-in cost — expressed in time, money, and data loss. Once you see it clearly, you can make an informed decision about whether it is worth the setup cost to reduce it.
You do not need to eliminate lock-in. Sometimes a monolithic platform is the right choice. The goal is to make the choice consciously, with full visibility of the cost — rather than discovering it when the platform changes something you cannot control.
Map one AI tool you depend on. Ask: if this tool disappeared tomorrow, what specifically would break and how long would it take to rebuild? That is your lock-in cost. Decide whether it is worth paying the setup cost to reduce it — or identify which single component you could make portable first.