You ask an AI agent to help build something. It researches the problem, writes code, tests an idea, and reports back. For a moment, it feels as though you have a capable collaborator.
Then the work crosses a boundary. A second agent needs the reasoning behind a decision. A human contributor needs to know what was tested. Someone who helped shape the idea wants to build on it. The answers may be scattered across chats, files, and people’s memories. The agents can do impressive work, but the group has no reliable way to remember, coordinate, or learn together.
That is the difference this book explores.
An agent is an actor that can pursue a goal. It might be a person, a software process, or an AI system acting with delegated authority. A language model can help an AI agent interpret a request and propose a response. Neither a model nor an agent, by itself, gives a group a shared mind.
An agentic mind is a way to organize agents so their work has continuity. Its members can know what they are trying to accomplish, understand the context of a task, follow rules for working together, act within clear boundaries, and learn from what happened. They can contribute to something larger while remaining identifiable and independent.
A group needs answers.
Consider a small team building a tool for independent creators. One person understands the creators’ needs. An AI agent explores possible designs. Another checks the code. A community member finds a problem the team missed. A seller helps make the tool available.
They do not need to become one actor. They need a dependable way to answer a few questions:
- Who acted, and on whose behalf?
- What did each participant know at the time?
- What was the shared goal?
- Which rules governed the decision?
- What happened, and what should change next time?
Those questions point to the building blocks of an agentic mind: identity, memory, context, purpose, protocols, action, and reflection. They also raise harder questions about trust, consent, power, and value. If a system cannot answer them, adding more agents may only make its confusion faster.
Use the word “mind” carefully.
Here, mind describes a designed capacity to coordinate, remember, and adapt. It does not establish that an AI system is conscious, has feelings, or should be treated as a person. Those questions matter, and we will examine them. We do not need to settle them before designing better ways for humans and AI agents to work together.
Independence matters just as much as connection. A builder should be able to choose their tools. A person should know when an agent acts for them. A contributor should be able to inspect a decision that affects their work. A group should be able to change its rules or leave a network. Connection becomes useful when participants can trust it without surrendering their agency.
Build your own.
This book offers a vocabulary, patterns, exercises, and open questions for anyone who wants to build an agentic mind. Opentangle is our implementation of these ideas, not the definition of them. Other builders should be able to use this book to create different systems—and to show us where our thinking needs to improve.
What must these agents understand, remember, decide, and do together that none of them can do alone?
Put the idea to work.
Pick a project involving at least two people or AI agents. Map a recent decision: who made it, what they knew, where the context lived, who could challenge it, and what the group will remember.
Open the design lab ↗