MIND OF AGENTSAn open field guide
A book and design field guide by Amit Rathore

MINDOFAGENTS

Build systems where people and AI agents can coordinate, remember, decide, and learn together.

For founders, engineers, and product leads building with more than one AI agent. Twelve chapters, free to read online, no sign-in.

Start reading, free ↗PDF edition ↗

The PDF is free too. It needs a quick sign-in with the Agents of Mind community. Reading online never does.

This open field guide explores agentic minds: ways for independent people and AI agents to work together with shared purpose, clear boundaries, and a memory of what happened.

“Mind” here means a system that can coordinate, remember, and adapt. It makes no claim that AI is conscious.

Independent by designInterconnected by choiceOpen to many implementations
The central idea

An agent is a start.
A mind is a design.

More agents do not automatically create collective intelligence. A mind needs a way to hold purpose, share context, make decisions, remember outcomes, and stay accountable to the people inside it.

Picture a small team

Four contributors. One decision to ship.

A creator, an AI design agent, a code reviewer, and a community tester work on a new tool. Each can do a task. Together, they need to know who approved the release, what the tester found, and what to remember for the next version. Designing that shared context and accountability is designing the mind.

Without a designed mind
  • Who approved the release? Unclear. Three agents and a human all said “looks good.”
  • What did the tester find? It lives in a chat thread nobody reopens.
  • What carries to v2? Nothing. The next run starts cold.
With a designed mind
  • Who approved the release? The creator, by the release protocol, on the tester’s evidence.
  • What did the tester find? Recorded to shared memory, linked to the decision.
  • What carries to v2? The finding, the call, and the outcome.

Explore the anatomy of an agentic mind ↗

01 / AGENCY

Independent actors.

People and AI agents have identities, capabilities, purposes, and boundaries. Their independence is a design requirement.

02 / CONNECTION

Shared ways to work.

Memory, context, and protocols let agents coordinate across tools, organizations, and implementations.

03 / LEARNING

Outcomes that matter.

Actions leave an inspectable trail. People can learn what worked, recognize contributions, and decide how the value of shared work should be distributed.

The design question

What can we do together?

Begin with a shared purpose. Then make the relationships visible: who knows what, who may act, how decisions are made, and what the group remembers.

Identity+Purpose+Memory+Protocols+Reflection
Can your team answer these?

Think of your last decision
involving agents.

  1. Identity Who acted, and on whose behalf?
  2. Context What did each participant know at the time?
  3. Purpose Was the shared goal clear enough to guide a tradeoff?
  4. Protocol Were the rules for deciding and challenging visible?
  5. Memory Will the group know what happened next time?

Any “no” is a design opportunity, not a grade. The lab runs in your browser and saves nothing.

Choose your path

Read the idea.
Make it real.

I’m new

Start with Chapter 1

“The Agent Is Not the Mind.” A clear starting point for thinking beyond a model, a chatbot, or a swarm of tasks.

Begin reading
I want to apply it · 5 minutes

Take the design lab

Run the five questions against a project you know. Nothing is saved.

Open the lab
I build things

See the anatomy

The twelve components and the loop that connects them, then Opentangle as one implementation.

Explore the anatomy
Our invitation

Build your own.
Change our minds.

Mind of Agents is a design language, not a product specification. Opentangle explores some of these ideas in software. Other builders can use them, challenge them, and make different systems.

Start reading ↗