Between you and the other side there is almost always a system that collects, sorts, suggests or already decides. You do not have to accept that silently, you can name it. And before you go further, you settle three things: who you are speaking to, what the other side knows about you, and who is accountable if this goes wrong. It costs a minute, and it moves the information gap to the other side of the table.
Let us begin where it touches everyone: with the individual person who wants to buy, clarify or decide something. For the first time in history this person has an advisor who never sleeps, never sends an invoice, and has an instant opinion on any topic. That is the good news, and it is bigger than it first sounds.
Power shifts noticeably toward the private person. Whoever hires a tradesman, buys a car, reviews a contract or wants to make sense of a diagnosis no longer walks into the conversation naked. Studies show that people already trust AI more than classic advertising and influencers in purchase decisions, and that about half use AI to compare products [Ketchum/YouGov, DE, n≈1,070]. In Germany a quarter of people have already asked an AI for financial advice, and almost as many trust that advice more than a human advisor [Bitkom 2025]. Personal shopping and price agents that compare around the clock on the user's behalf, and even buy at a target price, are no longer a future prospect but already built into large retail platforms. The individual is, in short, equipped for the first time with an apparatus that shifts the information asymmetry to their side.
Bitkom 2025
But here the second, darker half begins. The apparatus that strengthens the individual makes them vulnerable in a new way at the same time. The AI hands them the confidently delivered average solution, and as we saw in Chapter 4, in doing so it raises their confidence more than their actual knowledge. The individual feels informed and is only half so. They anchor to a machine ballpark figure they take to be objective, and face the professional with a certainty their situation does not warrant. That is not strength; it is half-knowledge in confident disguise.
And there is a structural danger the individual sees least of all. As soon as it is no longer they themselves but their agent that negotiates, they enter a game whose rules they do not know. The research here is clear: negotiation between agents is an inherently imbalanced game in which stronger models exploit weaker ones [Stanford/Pei et al. 2025]. The well-resourced agent of a large provider can beat the simpler agent of a private person long before that person even notices a negotiation took place. The promise "my agent fights for me" therefore holds only if the place where it fights is fair. This is exactly why the individual needs the protocol most urgently of all four parties: they have the largest relative shift to gain and the greatest risk of being the weaker side in an unregulated agent market. Empowerment and exploitation are two sides of the same coin here, and which side lands up is decided not by the technology but by the rule of the game.
Stanford/Pei et al. 2025
Name what did the mediating
Most conversations that go wrong do not go wrong because someone is lying. They go wrong because you do not know who or what is answering you. On the other side there may be a person. There may be an assistant writing on that person's behalf. There may be an agent negotiating for a company. There may be an ordering that a model arranged for you before you saw anything at all.
You notice the mediation in small things. The answer arrives faster than anyone can type. It is complete in a smooth sort of way and produces numbers without hesitating. It picks up your last sentence but not your situation. It knows things about you that you did not say in this conversation. When you notice something like that, say so. "I would like to know who is answering me here" is not an attack, it is how a clean conversation starts.
Example: the test result
You received a result on paper, understood half of it and asked at the practice. What you knew in the end, you knew from a person who had looked at you.
You type the result into an assistant in the evening and get an interpretation that sounds certain. The next morning you sit in the consulting room and push back with sentences whose origin you do not know yourself. Nobody tells you that the practice scheduled your appointment by a machine estimate of urgency, because nobody asked.
Your insurer sends you an assessment before you ask for one, and your own assistant negotiates the appointment with the practice's system. Which data about you crossed the table in the process appears in neither conversation.
Example: refinancing the mortgage
You made two appointments at two banks, took two offers home and compared them at the kitchen table. Whoever sat across from you had a name, a branch and a superior.
You have the figure worked out in advance and walk into the conversation with a number that came from a model. On the other side a chat answers, reachable around the clock. The terms it shows you are tailored to you, and what they were tailored from it does not say.
The comparison runs between agents, and what reaches you is a result that has already been filtered. Whether your agent found the cheapest offer or the one that pays it best is not something you can read off the recommendation.
Three questions before you go further
Ask them at the start, not at the end. Anyone working cleanly answers them in two sentences.
- Who am I speaking to? Is this a person or an agent, and if it is an agent, who does it act for?
- What do you know about me, where did you get it, and what of it do you keep and for how long?
- Who is accountable if this goes wrong, and how do I reach that person?
If one of the questions stays unanswered, that is the answer. Then you decide nothing at this point that costs money or binds you.