From rented intelligence to owned intelligence.

A call to your AI buys an answer, and nothing more. The model works each decision out from scratch, and it knows your team’s latest correction only for as long as someone keeps putting it in front of it, in a prompt, a document or a retrieved passage.

That is why rented intelligence costs as much in its hundredth month as in its first. The bill grows with use, and the model knows your work no better than on the day you started. Every lesson your team teaches it is paid for, used once and lost.

The gap shows the moment your rules change. On a stream of 3,000 decisions in which the rules changed five times, the model on its own reached 59.8% accuracy, and only 2.3% on the decisions those changes affected. It went on applying the old rules, because nothing in a rented model can learn the new ones.

Duplex is a second intelligence beside your AI model, and it learns at runtime, while the work happens. When your team confirms a result, Duplex learns from it, and the next request like it already reflects what your team decided. The model goes on doing what it does best, and Duplex brings what your work has taught it.

Ownership is a difference you can see. For anything Duplex has learned, you can see which agent it came from, with which model and when. You can review it, correct it or remove it, and what your company owns is separate from every other customer’s.

It doesn’t belong to one assistant, either. Any agent you allow can use what Duplex learns, your own agents included, so a fix confirmed in one tool helps in every other. One person builds their own intelligence this way, and a company builds its own, one confirmed result at a time.

Start with correcting it. When a policy changes, your team corrects one lesson, and Duplex follows the new rule from the next request, while everything else it learned stays as it was. When a lesson turns out to be wrong, your team removes it on its own, and the history shows who changed what, and when. A rented model gives you nothing to correct, because it never learned the lesson in the first place.

Owning something means you can take it with you. Models improve, prices change, and the best model for a job this year may not be the best one next year. With rented intelligence, every change of model starts the teaching again: new prompts, new examples and another round of corrections.

What Duplex learns works beside the model, rather than inside it, so it moves when you do. Change the model, the agent or the place Duplex runs, and everything your team has taught it comes along.

When the work moved to a different model family, Duplex needed no retraining at all. A fine-tuned adapter, trained into one model’s weights, made the same move in none of six tests.

The economics follow. Every decision Duplex has learned is one your AI no longer has to be paid to work out again. Duplex answers it in milliseconds, with no model call, and brings your AI in for what is new, briefed with what it knows about the case.

Over the same 3,000 decisions, Duplex beside the same model reached 94.6% accuracy, with 90.3% fewer model calls and 84.8% lower cost. Accuracy rose while the cost of each decision fell.

And the gains build on each other. A decision learned in March is answered in milliseconds in April, and so is every request like it after that. Your AI’s share of the work narrows to what is new and hard, which is where a general intelligence earns its price.

That is what owning intelligence means in practice. What you rent stays the same price, call after call. What you own grows in value every day it runs, and it stays yours whichever model you use next. Keep the model you trust for what is new and hard, and let the intelligence your work creates become an asset of its own.