// Field guide 001
Build Your First AI Teammate with Qwen Code
Start with a focused problem, give Qwen Code the context it needs and turn one useful idea into a working first iteration.

Qwen Code is an open-source coding agent that runs in your terminal. It can inspect a codebase, explain unfamiliar files, plan a change and help carry that change through implementation and testing.
The most useful way to begin is not to ask it to “build an entire app.” Start with one clear outcome that can be reviewed and tested.
1. Prepare a small, real project
Choose a repository you understand well enough to verify. A simple web page, command-line utility or small API is ideal. Before involving an agent, make sure the project already runs and that you know its expected starting behaviour.
Write down three things:
- the user problem;
- the smallest useful result;
- how you will know the result works.
For example: “Add an accessible theme toggle, preserve the selection locally and verify that keyboard users can operate it.”
2. Install and authenticate Qwen Code
The current Qwen Code documentation requires Node.js 22 or newer for the npm installation path.
npm install -g @qwen-code/qwen-code@latest
Open your project directory and start the interactive session:
cd your-project
qwen
Run /auth to choose a supported provider. Qwen OAuth's free tier ended in April 2026, so use an API key, Alibaba Cloud Coding Plan or another provider supported by the current Qwen Code release.
Always confirm the latest installation and authentication instructions in the official Qwen Code documentation before a workshop.
3. Ask for understanding before implementation
The first prompt should make the agent inspect the project and explain its intended change. A useful pattern is:
Read the relevant files, explain how this feature currently works, identify the smallest safe implementation and list the checks you will run. Do not edit anything yet.
This gives you a chance to correct assumptions before code changes begin.
4. Turn the plan into one testable slice
Once the explanation is accurate, ask the agent to implement only the agreed slice. Include constraints such as the framework, accessibility requirements, files that must remain untouched and the validation commands it should run.
Good prompts describe the outcome and boundaries. They do not micromanage every line of code.
5. Review the work like a teammate
An AI coding agent can move quickly, but it cannot own your product decisions. Review:
- the actual diff rather than only the summary;
- error states and empty states;
- keyboard and mobile behaviour;
- tests and build output;
- any new dependencies or external requests.
If the change is too large to review comfortably, reduce the scope and repeat.
A practical first-session checklist
- Start from a working repository.
- Define one user-facing result.
- Ask Qwen Code to inspect and explain first.
- Approve a small implementation plan.
- Review the diff and run the project yourself.
- Record what worked, what failed and what the next iteration should improve.
The goal of a first session is not maximum automation. It is learning how to collaborate with an agent while keeping human judgement, verification and responsibility at the centre of the build.
Explore the source and current release information in the official Qwen Code repository.
