Introduction to the Agents Shared Cognition Protocol (ASCP)

Abstract

Introduction

This is the third post in a series exploring shared cognition infrastructure. The first post identified the problem: collaboration suffers from fragmentation and "the amnesia problem" because articulation work—partitioning tasks, negotiating meaning, managing dependencies—remains invisible and ephemeral. The second post argued for a Cortex Layer that treats this coordination as persistent, structured data. This post introduces ASCP, the protocol that implements that vision.

ASCP works by reifying discourse primitives as computational structures. Every contribution becomes an immutable Artipoint—an addressable reference to a piece of context. These Artipoints compose into articulation statements that capture coordination patterns: instantiating concepts, relating ideas, organizing work, and superseding outdated context. Rather than letting coordination and context dissolve into chat history, ASCP builds a persistent graph of shared understanding that both humans and AI can navigate and extend.

The result is infrastructure for collaboration that functions like Git does for code: a shared, auditable record that enables distributed coordination without constant re-explanation. Context becomes durable, addressable, and machine-readable—transforming articulation work from invisible overhead into computable infrastructure.

From Discourse Theory to Protocol

ASCP draws from human discourse theory, which models coordination through conversational floors, turn-taking protocols, and relevance boundaries. These principles apply to both human and AI participants. ASCP makes these discourse structures computable by translating them into three core primitives that support persistent shared cognition:

• Artipoints are discourse atoms—immutable, addressable references to individual contributions.

• Channels are conversational floors—persistent, scoped contexts for coordinated work.

• Streams organize contributions thematically without losing context.

• Spaces coordinate multiple streams toward broader outcomes.

ASCP transforms articulation work—the coordination that normally lives only in people's heads or dissolves into unstructured chat threads—into formal, durable infrastructure that both humans and machines can reason about. This creates a foundation for persistent shared cognition, enabling humans and AI agents to build on a common, auditable record of understanding rather than constantly re-establishing context.

The following core concepts define the design principles and constraints that shape ASCP into viable infrastructure for enabling shared cognition.

Core Concepts

The Building Blocks

Artipoints are immutable, structured bookmarks that reference content within a context—similar to browser bookmarks but with formal semantics. The name derives from "point of articulation." Each Artipoint is addressable, semantically typed, and cryptographically secured, functioning as a cognitive atom for structured coordination.

Articulation Statements combine Artipoints into timestamped acts of coordination, like "this doc is relevant to this project" or "flag this item for review," capturing decisions on work organization and pointers to associated content. Applications generate these statements, but they do so based on the actions and assertions of humans or AI agents—entities with agency and point of view. Like a courtroom stenographer whose machine records testimony, the recording mechanism isn't what matters; it's who said what. Similarly, Articulation Statements always attribute authorship to the human or AI agent that caused their generation.

Structure, Not Content: The Bookmark Pattern

Artipoints capture cognitive structure, not dynamic content. ASCP doesn't encode the content itself, only references to it and the application intended to manipulate it. For example, instead of embedding a 50-page paper directly, ASCP has a statement like "this paper is highly relevant to our project" with a URI pointing to the document and an optional indicator of the application intended to manipulate that document.

This allows that 50-page paper to change independent of ASCP. But the cognitive decision—the structural relationship between paper and project at a given point in time—remains immutable and auditable in the coordination graph.

Immutable by Design

The context in ASCP is immutable, like a ledger, recording not just current structures and relationships, but their origins. All actions, including modifications to the structure, are recorded as immutable statements. Supersession, not overwrite, is the model.

Composition Through Reference

Given the immutable nature of ASCP, composition through reference is essential. Each Artipoint and Articulation Statement in the log is associated with a UUID. New Articulation Statements can reference previously defined Artipoints by their UUID, logically creating a Directed Acyclic Graph (DAG) of shared cognition.

This compositional approach—without nesting—enables:

• Machine-readable and human-relatable cognitive statements

• Individual referenceability and auditability

• Incremental composition into complex structures

• Decentralized distribution via ASCP channel synchronization

The grammar includes four articulation patterns: instantiation (creating cognitive atoms), annotation (enriching them), connection (linking atoms), and construction (creating and linking simultaneously). Complex cognitive artifacts naturally emerge from simple, atomic statements over time.

Channels: Structured Coordination Pathways

Channels are named pathways (like @Org or @Hiring Team) distributing Artipoints to authorized participants, similar to group chats but for structured coordination. Each Channel has an append-only immutable log and can encrypt payloads with symmetric keys for true end-to-end privacy.

This creates a durable memory for collaborative context—shared across humans and intelligent agents.

ASCP Architecture

Each layer builds on the previous to enable secure, distributed collaboration infrastructure.

Strategic Merits

Key Capabilities

• Durable Coordination Structures: Artipoints encode contextual relevance, turning ephemeral coordination into durable structure facilitating shared cognition.

• Making Context Computable: Transforms coordination into structured, machine-readable data.

• Cryptographic Channel Architecture: Defines semantic relationships with verb operators for explicit reasoning over relationships, eliminating separate access controls while maintaining human semantics with cryptographic security.

• Streams & Spaces: Streams organize work threads. Spaces group streams into accountability structures, all as addressable Artipoints themselves.

• Human-in-the-Loop Governance: Articulations can be superseded, reviewed, or throttled with transparency.

• Auditable Coordination History: Activity logs unify human and agent contributions, enhancing accountability.

Underpinning these capabilities is ASCP's cryptographic and identity foundation, which ensures shared cognition remains secure, durable, and auditable across humans and agents.

Technical Foundation

ASCP’s security model uses proven JOSE standards to ensure shared cognition is durable, auditable, and private. Every Articulation Statement is JWS-signed, establishing immutable authorship and forming a tamper-proof, log-anchored history of coordination. This log-anchored trust model means identity, structure, and governance are all recorded in the same append-only ledger that carries the work itself—creating a unified source of truth that agents and applications can interpret consistently.

Channel encryption provides privacy using symmetric keys (typically AES) shared with authorized participants, distributed through standard JWE key envelopes. Channels distribute immutable coordination statements, not chat messages—combining the privacy of end-to-end encryption with the reliability of append-only log sync.

Identity relies on self-sovereign JWK key pairs, which may be self-generated or organization-issued, with optional PKI or DID attestation. This hybrid approach avoids both centralized lock-in and crypto-ideological rigidity, supporting real-world collaboration models across individuals and enterprises.

Because governance, membership, and key rotation are themselves expressed as signed Articulation Statements within channel logs, the entire security model is transparent, auditable, and inherently local-first. The same log that carries context also carries its access rules and their evolution over time.

Together, these choices make ASCP secure by default and interoperable with existing ecosystems. With this foundation in place, the final layer is the coordination grammar itself—the structured language of Artipoints and Articulation Statements that makes context computable across tools and agents.

Formal Grammar Specification

ASCP includes a complete ABNF grammar defining coordination patterns for deterministic parsing and interoperability. It captures syntax and semantics, making relationships and coordination patterns interpretable to AI and intuitive to humans.

Conclusion

ASCP provides a protocol layer for shared cognition: immutable, addressable articulations that form an auditable graph of coordination. It transforms articulation work—the ongoing negotiation of meaning, decisions, and context—from invisible overhead into structured, machine-readable infrastructure that both humans and agents can reference, build upon, and synchronize.

The draft ASCP specifications are available on GitHub, including the formal ABNF grammar, Channel envelope specifications, and the ALSP synchronization model. These are evolving drafts in an early-stage repository; core specifications have first drafts in place while other elements are earlier in development. For those interested in AI, agent protocols, and coordination infrastructure for hybrid human-agent systems, we welcome review and feedback as we work toward a mature set of specifications and initial FOSS reference implementations of the whole protocol stack.