Two AIs debate before a costly decision.
Two AI models argue the idea, review each other and check every fact. You get a document with what has been verified and what remains open. You decide.
The problem
“The idea was validated by the person who proposed it.” “I asked an AI and it agreed with me.” “Nobody checked the fact the decision rested on.”
It happens to anyone who has to decide something costly, technical or commercial: a single opinion, from a person or from a model, tends to confirm what was already believed.
How it is done today
An important idea is validated with a single opinion, sometimes that of the person proposing it. The facts it rests on are taken as good. If anyone checks them, it is after the decision.
How we solve it
- You ask for the debate in one sentence, with the idea you want to validate.
- Two different AI models argue the idea, each with its own arguments.
- They review each other: what one claims, the other examines.
- They check every fact against primary sources.
- The result is a document with what has been verified and what remains open.
Result
It is how we work every day: before a costly decision, two models argue it and a person decides.
What it takes
The idea, in writing
What you want to decide and why it matters. The more specific the question, the more useful the debate.
The starting data
The documents and figures the idea rests on. Without them, the debate can only check what is public.
What counts as a source
Which sources you accept as primary in your sector: regulation, official documentation, your own data. We put it in writing with you.
A person who decides
Whoever answers for the decision. The document reaches them with the verified and the open kept apart.
How it is built
It is adversarial validation between two models from different providers: Claude and ChatGPT. Each one defends or attacks the idea, and nothing one says is taken as good until the other has reviewed it.
Every fact is checked against primary sources. What cannot be checked is not presented as true: it goes into the section on what remains open.
The result is a document, not a chat reply: it sets out what has been verified and what remains open. The decision does not come out of the debate; a person makes it.
Frequently asked questions
Why two models and not one?
Because a single opinion validates itself. With two, each reviews what the other claims and the facts are checked against primary sources.
Who makes the decision in the end?
A person, with the verified facts in front of them. The debate prepares the decision; it does not make it.
What happens with what cannot be checked?
It stays in the document as open. That way you know which part of the decision rests on verified facts and which does not.
Related automations
Code review
Four agents review every change before production. A person decides whether it goes in.
Knowledge per client
Each project's knowledge, indexed and isolated per client. A person decides what goes in and who has access.
Part of AI agent development