Every AI today shares the same brain.
We're building the one that develops its own.
Current AI stores knowledge the way software stores files — in databases, indexes, and retrieval pipelines. But the human brain doesn't search for memories. It reconstructs them. It consolidates them during sleep. It shapes what it keeps based on what it values.
Fast capture. Specific detail. The hippocampus encodes raw experience in a single pass — then replays it selectively during sleep.
Slow distillation. The neocortex gradually extracts patterns and schemas from replayed episodes — building a generative model of the world.
The missing piece. The neuromodulatory system that decides which experiences get consolidated. Shaped by culture, adversity, and institutional history. Not configured. Formed.
Dementia doesn't destroy memories. It destroys the pathways that weave them into meaning.
Insight from recent neurodegenerative research
Modelled on hippocampal-cortical consolidation dynamics. The innermost layer — the values modulator — is what makes each instance singular. It is not programmed. It is raised.
The current paradigm serves one model to millions. Every instance starts identical. Knowledge is injected at runtime and forgotten by the next session.
We are building the architecture for AI that is hired into a company, develops institutional knowledge specific to that organisation, and forms values through lived experience — like the best human employee you ever had, but one that never leaves.
This is a first-principles research programme.
We are speaking with a small number of aligned partners.