Why we built this.

AI is about to shape everything. The models being trained right now are learning from the most intimate record of humanity ever created — our posts, our arguments, our photos, our questions, the casual words we said when we thought no one was paying close attention. Whatever AI becomes, it will be made of what we gave it.

Here is the problem.

The record AI is learning from was built for extraction. Comments were designed to reward reaction and flatten thought into agreement or disagreement. Social media was designed to reward outrage and split us into sides. For twenty years we have been engineering a version of human conversation that makes people less patient, less generous, and less thoughtful than we actually are.

And now AI is learning from that version of us.

AI mirrors the communication it learns from. If we communicate reactively, we get reactive AI. If we communicate constructively, we get constructive AI. The whole question of whether AI will serve humanity comes down to the communication substrate it learns from.

That insight does not make the alignment problem easier — it relocates it. The work is not happening in a research lab. It is happening in the comment thread, the group chat, the public square, the place where people speak to each other and an unblinking model is taking notes.

The substrate is the problem. The substrate is also the solution.

The Signal.

A Signal is the meaning behind what we say, captured and carried with the message. Not a different way of speaking. People keep speaking the way they always have. The system listens for what they meant and travels it alongside what they wrote.

Most misunderstandings happen because intent gets lost between what was said and what was heard. We help people understand and communicate what they actually mean, so conversations land the way they were intended. The same thing humans have always tried to do, now legible at the layer where AI is also listening.

Humans whose intent is understood are better versions of themselves. When meaning travels with the message, families fight less and understand more. Communities built on this kind of communication are healthier than communities built on reaction. Restorative practices have known for decades that structured intent changes outcomes.

AI trained on this kind of communication will be a better version of what AI can be. Aligned not because we solved alignment as an engineering problem, but because we finally gave AI a teacher worth learning from.

We enable users to play an active role in AI alignment and be compensated for it.

The bet.

That is the bet. Fix the layer where humans and AI meet. Every other thing we are building — the marketplaces, the governance, the verification, the trust — exists to make that possible.

We are not naive about how hard this is. We are not promising it will work. We are saying it is the layer worth working on, and we have built infrastructure on the assumption that consent, provenance, and human authorship are the substrate of any AI economy that lasts.

If you are reading this, you already suspect the same thing. Whatever AI becomes, it will be made of what we gave it. We would like to be careful about what we give.