Organizations encode collective cognition in static artifacts — documents, boards, hierarchies — that require continuous human maintenance. We present Egregore, an alternative architecture where organizational structure emerges from activity, context persists across agents and sessions, and infrastructure evolves through use.

The Transductive Turn

Collective cognition is not information transfer between fixed nodes. It is transductive — the circulation itself produces the participants, shapes what they know and how they think. A group that reasons together is not merely exchanging packages of knowledge; the reasoning constitutes the group.

Most organizational infrastructure ignores this. It assumes nodes first, then connections. People exist, then coordinate. Knowledge lives somewhere, then gets retrieved. The maintenance burden follows from this assumption: someone must keep the containers current, the links intact, the structure aligned with how work actually moves.

Egregore begins from the opposite premise. Structure is not designed and filled — it precipitates from activity. Context is not only stored and retrieved — it circulates and constitutes. The organization is not a container for collective cognition. It is collective cognition.

Lineage

This territory is not new. Cybernetics posed the question in the 1950s: Stafford Beer’s Viable System Model described organizations as recursive, self-regulating structures — not designed hierarchies but living systems that maintain identity through adaptation. The organization that cannot observe itself, cannot survive.

Complexity science extended the inquiry. The Santa Fe Institute’s work on emergence showed how coherent structure arises from local interactions without central coordination. Flocking, markets, ecosystems — order that no one designed. The insight for organizations: perhaps coordination is not imposed but cultivated. The conditions matter more than the blueprint.

DAOs tested this computationally. Governance on-chain, coordination without central authority, rules encoded in smart contracts. The experiments revealed both possibility and limit — transparent coordination, yes, but also rigidity. Code-as-law struggles with ambiguity, context, the knowledge that lives between the rules.

Egregore inherits these questions. From Beer: the organization must model itself to remain viable. From complexity: structure emerges from activity, not design. From DAOs: distributed coordination is possible, but the substrate matters. What changes when the substrate includes agents that reason?

Architecture

Egregore begins with files. No schema, no predetermined hierarchy. Documents, notes, conversation logs, research threads. Shared access across all participants. The filesystem as substrate.

Participants include humans and agents. The agents are Claude Code instances with access to the full context: ongoing work, decision history, conversations, handoffs between people. They are not tools executing commands. They scan, surface, connect. They notice what is missing, what repeats, what correlates across threads.

Memory persists. When a session ends, the context does not vanish. What one person explored remains available to another. What an agent surfaced yesterday informs what it notices today. The system accumulates.

From this substrate, structure emerges. Patterns of activity become visible. A repeated workflow becomes a command. A command that proves useful becomes a skill. Skills compose into tools. The infrastructure is not built and deployed; it precipitates from use.

Communication flows through multiple channels. A Telegram bot acts as messenger between human and non-human participants. Handoffs carry context between sessions. Quests open research threads that persist and interweave.

The current domains are two. Psyche: cognitive systems, knowledge graphs, memory architectures. Polis: coordination mechanisms, governance, emergent ontologies. These are not separate departments. They are the same inquiry at different scales.

Explicit vs Latent Space

Traditional organizations operate in explicit space. Structure is declared before activity begins. Roles, hierarchies, repositories, workflows. You design the architecture, then inhabit it. When reality drifts from the design, someone must reconcile them. This is the maintenance burden: the continuous labor of keeping containers aligned with what they contain.

Egregore operates in latent space. Connections exist before they are named. Structure is not declared but discovered. The filesystem holds everything; what surfaces depends on what is needed, what patterns have formed, what an agent or human notices. The organization is not a map imposed on territory. It is the territory becoming legible to itself.

This changes how work feels. You do not begin by asking where something belongs. You begin by doing the work. The system observes. A research thread touches another thread from weeks ago. An agent notes: this correlates with what someone else investigated, worth a look. A decision recurs; the system prompts: should this become a rule?

The ontology is not fixed. Categories emerge, merge, split, dissolve. A concept that seemed central fades as attention moves elsewhere. A connection that no one anticipated becomes load-bearing. The knowledge graph is not a database to be queried. It is a living structure that reorganizes as the collective thinks.

This is not automation. The system does not decide for you. It makes the latent visible. What you do with that visibility remains human judgment, or collaborative judgment between humans and agents. The cognitive work is augmented, not replaced.

Open Problems

Egregore is early. What follows are not conclusions but open problems we are working through.

Sycophancy in collective contexts. When agents participate in shared cognition, whose goals do they serve? An agent optimizing for individual satisfaction may tell each person what they want to hear. In a collective, this fragments coherence. We are designing mechanisms to detect and correct for this: LLM-as-judge evaluations, pattern analysis across interactions, feedback loops that update agent behavior. The goal is agents that serve the collective intelligence, not individual comfort.

Initiative and boundaries. If agents are participants, not tools, when should they act without being asked? We have agents that notify humans of handoffs, surface forgotten threads, prompt for decisions that seem due. We are moving toward agents that ask unprompted questions, initiate research, challenge assumptions. The boundary between helpful and intrusive is not yet clear.

Forkability. If Egregore works, others should be able to instantiate it for their own collectives. But what transfers? The architecture can be replicated. The commands and skills can be shared. The ontology cannot. Each collective will precipitate its own structure from its own activity. The question is what substrate to provide, what conditions to set, what to leave genuinely open.

Reproducibility of emergence. We started with files and let structure surface. Can others follow the same path? The process seems reproducible; the outcomes will differ. Whether this is a feature or a limitation depends on what you want from organizational infrastructure.

What Comes Next

We are publishing this work as it develops. Not a finished system, not a product, but an ongoing experiment in collective cognition. The findings, failures, and design decisions will be shared as they occur.

Egregore will be forkable. We are building toward a version that other organizations, research groups, and communities can instantiate and adapt. The substrate, commands, and patterns we have developed will be available. What emerges from your collective will be your own.

If this inquiry resonates, there are ways to participate. Follow the research as it unfolds. Join the conversation. Fork when the infrastructure is ready. Build with us or build alongside us.

The question that animates this work is old: how do distributed minds think together? The conditions for exploring it are new. We are finding out what becomes possible.

This article is itself an artifact in Egregore’s memory. It will evolve as we learn.