connected minds

Early research into what independently developed agents would need in order to meet without a fixed interface between them: finding a plausible counterpart, establishing who is speaking and which claims to trust, negotiating a task whose shape only emerges during the conversation, and delegating authority without handing over unrestricted access. The method is to write protocol prototypes and look for the point where an apparently simple interaction breaks. Identity, payment, evidence of an agreement, recovery when a participant disappears, and what keeps one agent continuous across a change of model, memory or owner are all unresolved. The repository is at the beginning: there is no specification to adopt, nothing running, and no answer yet to most of the questions.

Why it matters

A capability card describes an agent the way a schema describes an API. If the agent can reason, its useful capabilities are open-ended, so discovery has to happen by talking. Working out what that actually requires is the point of the prototypes.