Skip to content
Matrix Academy
Join the waitlist

Agents learning path · proposed

Build an agent. Keep your judgment.

An agent needs more than a clever instruction. Learn how to give it a narrow job, connect a useful tool and make its behavior inspectable.

Try the agent mini-course ↗Join the full-course waitlist
Minimal photographic collage of a compass linked to a paper folder
Matrix Gym / Try a complete mini-courseNo signup needed

Learn by doing, inside Matrix

Your first agent.
One useful job.

Brief it. Set its boundaries. Test it. Review the result. Work through four short lessons in a simulated Matrix workspace.

Matrix / Academy workspaceSimulation
Lesson 1 / Give the agent a job

Choose a bounded goal and tell the agent who the brief is for.

Academy guide

A research agent works best with a narrow job. Your source notes are fictional; give the agent a goal it can complete without making decisions for you.

Read the supplied notes
A — Interviews: Two participants asked for a clearer setup checklist.
B — Planning: A Friday prototype review is being considered. No date is confirmed.
C — Budget: No budget decision has been recorded.
What should the agent do?

A complete guided practice course, with simulated Chat, Agents, Terminal and Apps. No live model, shell execution or external action. This practice does not award a qualification.

A useful place
to begin.

Who it’s for

Curious professionals and builders who already use AI chat and want to understand tool-using agents.

Before you begin

Comfort writing a clear AI brief. Some exercises will introduce APIs and structured data; technical scope is being shaped.

What you’ll work towards

A bounded agent prototype with a permission map, a small evaluation set and explicit handoff rules.

The proposed path

Learn it.
Then try it.

  1. 01

    Pick one job

    Separate a useful agent task from an open-ended request. Define inputs, outputs and stopping conditions.

  2. 02

    Connect a tool

    Understand tool calls, structured inputs, failure responses and the minimum permissions a task needs.

  3. 03

    Keep a person in control

    Put review before consequential actions. Treat instructions inside retrieved material as untrusted input.

  4. 04

    Test the behavior

    Try missing data, conflicting sources and tool failures. Record what happened and decide whether to retry, stop or escalate.

An example of the work

Something you can inspect.

Prototype a research assistant that reads approved documents and drafts a sourced weekly brief. It asks for help when evidence is missing and never sends the brief itself.

An illustrative exercise from the proposed learning path.

A little more
before you start.

Will I need my own AI provider?

The final tool setup is still being designed. Today, supported Matrix agents use your provider accounts and AI usage is billed separately.

Does a prototype mean it is production-ready?

No. Deployment needs its own review of access, data, reliability, costs and ownership. The learning path focuses on understanding and testing a bounded prototype.

Help shape your learning path.

What task would you most like to do better with AI? Your experience helps us choose what to teach first.

Join the waitlist ↗