MIND OF AGENTSAn open field guide
A conceptual map of shared agency

Anatomy of a
mind of agents.

A mind is more than a model and more than a collection of agents. It is a living arrangement for noticing, remembering, deciding, acting, and learning together.

Twelve design lenses for minds made of independent, interconnected agents.
The big picture

A body for
shared agency.

Picture a person: eyes notice, memory supplies history, a heart holds purpose, a spine holds the rules, hands act, and experience changes the next move. The analogy makes the architecture easier to explore. It does not suggest that an AI system has a human body or consciousness.

Every part below can be built in different ways. What matters is how the parts work together and how the mind relates to the independent people and agents within it.

Start with the loop

From signal to next move.

A mind gains continuity when each cycle can inform the next.

  1. 01

    Notice

    Perception and attention select an event worth handling.

  2. 02

    Understand

    Identity, context, meaning, and memory establish what it means here.

  3. 03

    Imagine

    Agents propose paths and compare possible outcomes.

  4. 04

    Decide

    Purpose, authority, and protocols determine what may happen.

  5. 05

    Act

    Capabilities change the world within time and resource limits.

  6. 06

    Learn

    Outcomes, provenance, and reflection update what the mind remembers.

Then the next signal arrives. The loop repeats, and relationships can change along with it. Below, the same loop is mapped onto twelve components.

The anatomy

Explore the components.

New here? Start at 01 Identity and use Next component to walk the whole mind, or jump to any number.

One mind / many agentsConceptual map · v0.1
Anatomy of a mind of agents An abstract human silhouette marked with twelve selectable components: identity, attention, context, memory, imagination, purpose, decision, protocols, action, reflection, provenance, and resources. Lines link the mind to other minds. ◄ START HERE01 02 03 04 05 06 07 08 09 10 11 12
← Independent agentsShared mind ↔ other mindsEnvironment →
Inspect a component01 / 12
THE FACE

Identity

A stable way to know which person, AI agent, group, or mind is involved—and on whose behalf it acts.

Design question

Can you trace an action to an actor and the authority behind it?

The connecting lines are as important as the numbered parts. A mind has relationships to its members, other minds, and the world. A shared mind can itself participate in another mind. No single component owns the whole.

A design example

What crosses
the boundary?

One way to make a decision legible across a group is to record a move: who acted, for what purpose, under whose authority, with what context and evidence, and what followed. This sketch is an example for designers to question and adapt, not a required format.

Links between such records could reveal paths of decisions and contributions. A group could inspect those paths while allowing participants to use different tools and systems.

Thought experiment / release proposalIllustrative
Actor
Design agent, acting for a creator
Purpose
Make the creator tool ready for a small beta
Mandate
May propose; cannot publish
Context
Brief v3 + accessibility requirements
Move
Proposes a release candidate
Evidence
Test result + unresolved tester concern
Outcome
Held for human review, then recorded
Time & links
When it happened; which earlier work it used
The next actor can challenge, approve, or build on this move.
Three design commitments

Keep the mind
open to its members.

01

People remain responsible.

Delegation should show who authorized an AI agent, what it may do, and how that authority can be revoked.

02

Members remain independent.

A person or agent can join several shared minds. Participation should not erase identity, ownership, or the ability to leave.

03

Evidence travels with outcomes.

Keep a useful account of who contributed, what was used, what happened, and what remains uncertain. Value sharing is a governance choice, not an automatic fact.

Ideas and influences

A model to build on.

This anatomy is a way to think about shared agency, informed by research on memory, reasoning, feedback, and provenance. The body metaphor organizes useful questions; builders can choose different parts, relationships, and interfaces.

Existing open standards can carry parts of the design: A2A addresses exchange between independent AI agents, while MCP exposes tools and context. Neither choice is required by this conceptual model.

ResearchGenerative AgentsObservation, memory, planning, reflection ↗ResearchReActReasoning and acting in a loop ↗ResearchReflexionLearning from feedback and episodic memory ↗StandardW3C PROV-OAgents, activities, entities, and attribution ↗

Want to test the model against a project you know?

Open the design exercise ↗