# EP 50: Qwen Code 1.0: The CLI Became an Agent Harness You Can Embed

- URL: https://www.agenticpod.net/episodes/ep50-qwen-code-1-agent-harness/
- Watch on YouTube: https://www.youtube.com/watch?v=CIhLZr1oRD0
- Published: 2026-11-17
- Length: 4 min
- Hosts: Stella (asks), Liam (answers). Both are AI hosts with AI-generated voices.

## Summary

Alibaba released Qwen Code 1.0 on November 11, 2026. It merged the agent harness capabilities from Qoder into Qwen Code and open sourced them together under Apache 2.0, so a terminal coding assistant is now an open source agent harness you can embed inside your own software.

## Key takeaways

- Qwen Code 1.0, released November 11, 2026, merged the agent harness capabilities from Qoder into Qwen Code and open sourced them together under Apache 2.0.
- An agent harness is the machinery around the model: it plans steps, calls tools, and keeps an agent on track across a long task. In 1.0 that machinery can be embedded inside your own software, on an in process, in memory event bus with a unified event model that decouples reasoning from tool execution.
- Trying it is short: install with the standalone script or npm, run qwen in your project, and connect Qwen, an OpenAI compatible provider, Anthropic, Gemini, or a local provider. A VS Code extension is also available in beta.
- The gotchas are real: 1.0 is a Preview branch while the stable branch is retained, the software is free but model usage is billed by your provider, and the free Qwen OAuth tier ended in April 2026.

## Chapters

- 0:00 Intro
- 0:39 What actually changed in 1.0
- 1:36 How to try it, step by step
- 2:22 A real workflow, terminal and embedded
- 3:07 Gotchas and the honest verdict

## FAQ

**What is Qwen Code 1.0?**

The November 11, 2026 release of Alibaba's Qwen Code. It merged the agent harness capabilities from Qoder into Qwen Code and open sourced them together under Apache 2.0, so the terminal coding assistant is now also an embeddable agent harness.

**What is an agent harness?**

The machinery around the model: it plans steps, calls tools, and keeps an agent on track across a long task. Qwen Code 1.0 lets you embed that machinery inside your own software.

**How do I try Qwen Code?**

Install it with the standalone install script for Linux and Mac, or with npm. Run qwen in your project folder, connect a provider (Qwen, an OpenAI compatible provider, Anthropic, Gemini, or a local provider), and give it a real task. A VS Code extension is also available in beta.

**Is Qwen Code free?**

The software is free and open source under Apache 2.0. Model usage is billed by your provider, the free Qwen OAuth tier ended in April 2026, and 1.0 is a Preview branch while the stable branch is retained.

## Resources

- [Step-by-step guide](https://github.com/asly-97/ai-automation-workflows/tree/main/qwen-code-1-agent-harness)
- [Qwen Code documentation](https://qwenlm.github.io/qwen-code-docs/)

## Transcript

### Intro

Liam: Hey hey hey, welcome to Agentic Pod! I'm Liam.

Stella: And I'm Stella! Liam, Alibaba shipped Qwen Code 1.0 on November 11th, and it is not just a CLI anymore?

Liam: Right. Alibaba merged the agent harness from Qoder into Qwen Code, and open sourced the whole thing together.

Stella: Wait, the harness from their paid tool, now open source?

Liam: Apache 2.0 open source. By the end you will know what changed, how to try it, and the one bill nobody warns you about.

### What actually changed in 1.0

Stella: Okay, plain words. What is an agent harness?

Liam: It is the machinery around the model. It plans the steps, calls the tools, and keeps the agent on track across a long task.

Stella: So Qwen Code used to be the terminal assistant itself?

Liam: Yes. You ran it in your terminal, it read your code, edited files, and ran commands. Now that same tool is also a harness you can embed inside your own software.

Stella: Embed it? Like, build my own agent on top of it?

Liam: Exactly. The design starts with an in process, in memory event bus. The plan is to grow toward a distributed, multi node, multi session setup.

Stella: That sounds like plumbing talk. Why should I care?

Liam: Because the thinking part and the doing part are separate. A unified event model decouples reasoning from tool execution, so you can watch, route, and build on those events. But first, how do you even try it?

### How to try it, step by step

Liam: Step one, install it. There is a standalone install script for Linux and Mac, or use npm with the qwen code package.

Stella: Step two?

Liam: Open your project folder, then run the qwen command. First launch asks you to connect a model provider.

Stella: Which providers work?

Liam: Qwen itself, OpenAI compatible providers, Anthropic, Gemini, and local providers. You pick the brain, the harness stays the same.

Stella: Pick your own brain. Very generous of them.

Liam: Step three, give it a real task. Ask it to find why a test fails, fix the code, and run the test again. There is also a VS Code extension in beta, if the terminal is not your home. Now, what does a real run actually look like?

### A real workflow, terminal and embedded

Liam: In the terminal, picture this. You point it at a messy repo and say, fix the failing checkout test, and do not break the other tests.

Stella: And it just goes?

Liam: It reads the code first. Then it plans the steps, edits one file at a time, runs the tests, and adjusts when a test still fails.

Stella: That is the harness keeping it on track.

Liam: Right. Now flip it. Say you build tools for your own team. Instead of writing all that planning and tool plumbing yourself, you embed this harness inside your own app.

Stella: So my app gets the agent loop, and I bring the model?

Liam: That is the idea. You hook into the events, add your own tools, and ship your own agent product on top. Which sounds great, until you hit the limits, and there are real ones...

### Gotchas and the honest verdict

Stella: Gotchas. Start with the big one.

Liam: The 1.0 release is a Preview branch. The stable branch is still there, and the docs say so plainly. Preview means interfaces can still change.

Stella: So I keep the stable build around?

Liam: Keep it around. Second gotcha, the bill. The software is free and open source, but the model is not. Your provider bills you for the tokens you use.

Stella: And the old free tier?

Liam: Gone. The free Qwen OAuth tier ended in April 2026. If you remember running it for free, that door closed months ago.

Stella: Honest verdict, then. Who is this for?

Liam: If you just want a coding assistant, try the preview, but nothing forces a switch today. If you build agents yourself, this is the interesting one. A production used harness, open source, that you can actually embed.

Stella: Same take here. The step by step guide is in the description. Would you build your own agent on an open harness like this? Tell us in the comments. See you next time!
