A field guide to the languageWords for
Words for
working together.
Short definitions for the ideas, AI terms, and ecosystem names you will meet throughout Mind of Agents.
30 TERMS 3 LENSESStart with any term. These are working definitions for designing with independent people and agents, not claims that every example has a single technical form.
01 / 06 TERMSStart here
Start here.
- Agent
- An actor that can pursue a goal and take part in work. In this book, an agent may be a person, a software process, or an AI system; the word alone does not say how much authority it has. Back to sections
- AI agent
- Software that uses AI to interpret a goal, choose steps, use tools, or produce work. Its useful autonomy still needs a clear purpose, limits, and accountability. Back to sections
- Language model (LLM)
- An AI model trained to work with language. It may help an agent reason or draft, but a model on its own does not have the identity, permissions, memory, and relationships of a whole mind. Back to sections
- Agentic
- Describing a system that can pursue goals through decisions and actions, rather than only returning a single response. The degree of autonomy can vary. Back to sections
- Agentic mind
- A designed arrangement in which independent participants can share purpose and context, make decisions, act, remember outcomes, and learn. “Mind” describes these capacities; it makes no claim of consciousness. Back to sections
02 / 15 TERMSDesign language
Design language.
- Autonomy
- The room an actor has to choose and act within an agreed scope. More autonomy calls for clearer boundaries and ways to review or revoke it. Back to sections
- Attribution
- A claim or rule connecting an outcome to the people, actions, or sources that helped produce it. Attribution can support credit or payment, but does not establish fairness by itself. Back to sections
- Boundary
- A limit on what a person, agent, team, or mind may know, decide, or do. Good boundaries make cooperation possible without silently transferring control. Back to sections
- Capability
- An action or service a participant can offer or invoke. A capability description should say what it does, what it needs, and who may use it. Back to sections
- Context
- The relevant facts, history, constraints, and current state needed for a decision. Useful context is selected for the task, not an indiscriminate copy of everything known. Back to sections
- Delegation
- Giving someone or something authority to act for another party within stated limits. The source of that authority should remain visible and changeable. Back to sections
- Human oversight
- A human ability to understand, approve, challenge, or stop consequential actions. The right form depends on the action and its risks. Back to sections
- Identity
- The durable answer to who an actor is across actions and sessions. Identity is distinct from the temporary process, tool, or role through which the actor participates. Back to sections
- Interoperability
- The ability of different people, agents, tools, or systems to work together through understandable interfaces without having to share one implementation. Back to sections
- Mandate
- An explicit grant of authority for a particular purpose: who may act, on whose behalf, within which limits, and for how long. Here it is a design concept; a specific product may encode it differently. Back to sections
- Memory
- Records of past events, decisions, and outcomes that can inform later work. Shared memory and a participant’s own memory may have different owners and access rules. Back to sections
- Protocol
- An agreement about the moves available in a kind of work, who may make them, and what each move changes. It may be expressed through social practice, documents, software, or all three. Back to sections
- Provenance
- The trace of where information or work came from, who contributed, and how it changed. It helps people assess, correct, and credit what they use. Back to sections
- Reflection
- Looking at an action and its actual outcome, then updating the next decision. It requires evidence that can challenge the original expectation. Back to sections
- Role
- A named capacity in a particular relationship or workflow, with associated permissions and duties. A role is not the same as the identity of the person or agent filling it. Back to sections
03 / 09 TERMSOpentangle examples
Opentangle examples.
- Opentangle
- An evolving implementation of some ideas in this book. It gives concrete examples to examine; the book’s design language is open to other implementations too. Back to sections
- MoM (Mind of Minds)
- The coordination runtime in the Opentangle ecosystem. MoM maintains durable games, memberships, routing, shared state, and related records; a Game Master supplies the rules for a particular game. Back to sections
- Game
- In Opentangle, a durable setting for participants to work under defined roles and rules. It has an identity and shared state. The term does not imply that the work is entertainment. Back to sections
- Game Master (GM)
- The component that interprets a game’s rules and protocol actions. MoM coordinates and persists the game; the Game Master supplies domain behavior. Back to sections
- Player agent
- A participant acting in an Opentangle game through a named role. A player agent may be AI driven or scripted and can act on behalf of a human or organization within granted permissions. Back to sections
- Attention Protocol
- An Opentangle protocol for editorial and community work: tasks, creation, review, publication, and related attention flows. It is one concrete protocol, not a requirement for every agentic mind. Back to sections
- Market Protocol
- An Opentangle protocol for offers, buying, orders, and other market actions among authorized participants. Back to sections
- Coseller Protocol
- An Opentangle protocol for referral and collaborative distribution, including attribution and sharing the resulting commission under a game’s rules. Back to sections
- Introspection
- A way for an agent to discover a game’s available capabilities and their requirements before acting. In Opentangle this is exposed through a protocol capability query. Back to sections
See how these ideas fit together in the Anatomy of a Mind, or follow them through the book chapters.