TeachClaw
TeachClaw is a teaching platform in development that brings students and purpose-built AI agents together to work on specific course projects. Students help shape their agent’s personality, give it useful context, and learn to direct, question, and check its work.
The name joins teaching and OpenClaw, the agent software proposed for the platform. TeachClaw is our educational project, not another name for OpenClaw itself. Its planned hybrid setup combines cloud services for classroom interaction with an instructor’s computer for development, review, and work between classes. The first classroom trial is being prepared; the platform is not yet a generally available service.
Why introduce it in a class?
Working with an agent is becoming a practical skill. It involves describing a task clearly, providing evidence, deciding what the agent may do, evaluating the result, and revising together. Those habits need practice on something worth doing.
TeachClaw starts with a defined project, not all of a student’s work. An instructor might use it for destination research, a team strategy exercise, or the development of an interactive learning activity. Other assignments remain outside the platform unless the instructor explicitly brings them in. Canvas remains the place for course submissions and grades.
The aim is to learn how to work with agentic AI while learning the subject. The agent does not take responsibility for the student’s judgments. A polished answer still needs checking.
What makes this an agent rather than just a chat?
A chat can offer advice. An agent can also use permitted tools to help carry out a task: organize source notes, propose changes to a project, save an approved revision, or prepare a preview. TeachClaw is being designed around that conversation-and-action cycle, with individual and team workspaces and instructor-defined limits.
The proposed memory system will carry useful project decisions and open questions between sessions. Students will review what should be remembered; an uncertain suggestion should not quietly become an established fact. Memory is a working record, not perfect recall.
The course provides the assignment, examples, and reference material. Students provide direction, sources, questions, and judgment. An agent can have a personal name and manner without having access to a student’s personal accounts or unrelated files.
Before your first session: about ten minutes
You do not need to arrive with a finished agent. Bring a few notes—on paper or in any notes app.
- Choose a possible name. Something you would enjoy addressing. You can change it later.
- Pick two or three qualities. Patient? Curious? Concise? Playful? Methodical? Think about what would help you work, not just what sounds entertaining.
- Decide how it should challenge you. Should it ask one question at a time, point out weak evidence, or offer an alternative when your first idea is too easy? A useful collaborator does more than agree.
- Give it one first job. Bring a small project task, a question you want to investigate, and any relevant public source links you already have. No need to complete the project beforehand.
- Write a short working agreement. Three or four sentences are enough. The example below is a starting point, not something to copy word for word.
Your name is Scout. Be curious, encouraging, and concise. Ask me one question at a time, and tell me when a claim needs a better source. Help me develop my project, but explain proposed changes and let me review them before they are saved.
These notes will help us create a SOUL.md file: a plain-text description of the agent’s intended character and working style. “Soul” here is a name for instructions, not a claim that the software has consciousness. You do not need to create or upload that file before class. Personality choices do not change the course rules or the agent’s technical permissions.
A little reading, not an installation assignment
Read the opening and “The four parts” section of On Beginning. They introduce the model, the software around it, and the workspace where useful work is kept. The rest is available if you want to explore setting up an assistant yourself.
For TeachClaw, do not follow the installation steps unless your instructor specifically asks you to. The classroom plan is browser access to a prepared environment. No model download, paid subscription, coding experience, or new computer purchase is required for this preparation. Your instructor will provide the access instructions after testing. If you cannot bring a browser-capable device, tell your instructor so participation can be arranged.
Our first project: The International Life
Students in The International Life are helping co-develop content for future versions of the course. Darwin and the existing tapes provide examples; students’ destination research supplies the material for new interactive experiences.
The common Destination Studio format gives everyone a useful starting structure. With an agent, a student might improve one location stop, check its sources, develop a meaningful visitor choice, and review a playable preview. A partner can then try it and suggest improvements. The goal is more than a tourist brochure: an experience that helps someone understand a place.
This is the first application, not the limit of TeachClaw. The platform is intended to be reusable across projects, courses, and semesters, while keeping each class’s work appropriately separated.
A few working habits
- Use the agent for the project your instructor has assigned. Leave unrelated personal information and private documents out.
- Ask where a claim came from. Open the source and check that it supports the claim.
- Read proposed changes before approving them. Keep track of your own decisions and contributions.
- At the end of a session, review the saved work and the short note about what to do next. Correct anything the agent has misunderstood.
- Tell your instructor when something does not work. In an early trial, that feedback is part of developing the platform.
Begin with a name, a few preferences, and one worthwhile task. The rest can develop through doing the work together.
See also: On Beginning · SOUL.md · Agent.