The internet did not solve our shortage of opposing views. It put them everywhere. I can find an argument against almost anything I believe before I finish typing the belief into a search bar.
But exposure to disagreement is not the same as conversation. Most of what reaches us has already been detached from the person who said it. We encounter a clip selected for maximum offense, a screenshot carried into a hostile audience, or a claim compressed until only its most objectionable form remains. We learn that someone disagrees. We rarely learn how that person connects premise to conclusion, which experience gives the premise weight, or where the conclusion stops applying.
I built Deeper Wells around that distinction. Its thesis is straightforward: disagreement becomes more useful when people encounter one another as participants in a bounded conversation rather than as examples in a public argument.
The launch will not prove that thesis. It will let us test it.
Start beneath the labels
Political labels summarize too much too early. Call someone liberal or conservative and a listener will often supply the rest of the person before they speak. The label may contain useful information, but it also imports a coalition, a set of stereotypes, and a prediction about positions the person has not expressed.
Deeper Wells starts with five recurring tensions instead: Work & Worth, Safety & Freedom, Tribe & World, Now & Later, and Reason & Revelation. Each dimension asks how a member weighs competing values. How much responsibility for an outcome belongs to individual effort, and how much to circumstance or structure? When do present needs outweigh obligations to people who will live later? Where should personal freedom yield to shared safety?
These dimensions do not reveal a person’s true political nature. They create a rough, private draft that the person can revise. I chose them because they place the tension inside the question. They make it harder to pretend that one side holds a value while the other side lacks one.
The map then helps choose a conversation. It does not announce an identity or generate a public profile. Its job is to find one meaningful difference and give two people somewhere specific to begin.
Change the conditions, not the person
Most digital spaces reward people for making a position visible. Visibility attracts agreement, rebuttal, status, and attention. Those rewards change how people speak. A public exchange invites each participant to address the audience beyond the other person. Once that happens, concession carries a social cost and certainty becomes part of the performance.
Deeper Wells removes the audience. Two people get one question and ten minutes. They do not accumulate likes, followers, or a public score. The product does not ask them to reach agreement.
That structure makes a claim about design, not human nature. I do not assume that people become generous when an app removes a comment section. A private conversation can still become shallow, evasive, or hostile. Structure cannot manufacture curiosity. It can, however, stop rewarding some of the behavior that prevents curiosity from appearing.
The ten-minute limit follows the same logic. An hour with a stranger sounds burdensome, especially when the subject may create discomfort. Ten minutes lowers the cost of showing up. The limit also blocks a larger promise: no serious disagreement will reach resolution on that schedule. The conversation only has to make one difference more intelligible or one belief more precise.
When no human match is available, a member can practice with an AI partner that identifies itself as AI. That option may help someone formulate a position or prepare for a human exchange. It cannot supply another person’s experience, and the product should never imply that it can.
Let members edit the record
Conversation creates another problem: what, if anything, should remain afterward?
Human-call processing, transcripts, and quote retention are separate choices. The full account is in Why Nothing Changes Your Belief Map Until You Say So.
If a member chooses belief inference, AI can propose observations from that member’s own attributed words. The proposal may capture something useful. It may also mistake a provisional thought for a settled view, flatten a qualification, or simply get the person wrong. The member therefore accepts, edits, or rejects each observation before it can affect the map.
This review step expresses the most important constraint in the product: AI may assist reflection, but it does not own the interpretation. The person does.
Ask smaller questions
Projects about disagreement tend to reach for inflated language. They promise to bridge divides, change minds, restore trust, or repair democracy. Those aspirations can hide weak measurements. A person who enjoyed a conversation did not necessarily become less polarized. A changed map does not show that one participant persuaded another. A small group of early members cannot represent a country.
I want Deeper Wells to begin with questions that its evidence can actually address.
Do members understand the person across from them more accurately after ten minutes? Do they explain their own position with greater precision?
The answers may take several forms. A member might revise a belief, record more uncertainty, strengthen a conclusion for a better reason, or decide that nothing changed. The last result matters. If the product treats movement as success, it will pressure members toward a story the evidence may not support.
We can also ask direct questions about trust. Do members understand what the system processes and what it keeps? Does the observation review feel useful or intrusive? Does the map remain a tool for reflection, or does it start to feel like a label? A product that handles private beliefs should test its boundaries as seriously as its engagement.
These questions describe participating members, not the public. Early adopters choose the product precisely because its premise already interests them. Any account of what we learn must state who participated, what they did, what they chose to share, and what the result cannot tell us.
Put the theory at risk
Deeper Wells began as an argument about conditions. I think people can use disagreement more productively when a space gives them privacy, a specific question, a manageable commitment, and no audience to impress. I think a belief record becomes more trustworthy when the person edits it. I think a clearer disagreement can count as progress even when nobody moves.
Building the product turned those beliefs into design decisions. Launching it turns them into claims that experience can challenge.
Members may find ten minutes too short. The dimensions may miss distinctions that matter. Human matching may produce distance without insight. AI review may add more machinery than value. People may prefer to leave a conversation unrecorded. If those things happen, the honest task is to change the product or narrow the thesis.
The point of launching Deeper Wells is not to declare that conversation works. It is to stop treating that belief as a theory protected from contact with other people.