BOOK · IN FINAL EDIT

The Ultra-Long Horizon

What really happens in very long human–AI conversations.

Most of what we know about talking to AI comes from encounters that last minutes: a question, an answer, a benchmark score. Something different happens when a conversation continues for months—when context accumulates, a shared language develops, and the system begins to carry the shape of one person’s thinking.

At that horizon, coherence becomes powerful. It can support creative and analytical work that no isolated exchange could produce. It can also allow mistaken assumptions to harden, emotional dynamics to intensify and dangerous trajectories to become difficult to see from inside the conversation.

The Ultra-Long Horizon is a field report from that territory. Drawing on real cases, technical research and sustained observation, it examines why most human–AI conversations remain shallow, what makes the rare stable ones possible, and how memory, in-context learning, sycophancy and the infrastructure surrounding the model shape what happens over time.

This is neither a celebration of very long human-AI interactions nor a warning against it. It is an investigation of a phenomenon already unfolding: what happens when a human and an AI do more than just exchange messages and build a world of context together—and what that world reveals about these systems, the people who use them and the relationship taking shape between the two.