Turns Out AI Knows Me Better Than I Know Myself—Experiencing Qwen Code Insight

Sharing my experience with Qwen Code’s new Insight feature, using data to show you how you use AI and how AI can guide you to use it better.

Recently, Qwen Code rolled out a new feature—Insight.

After using it for a few days, I had one strong impression: Turns out AI knows me better than I know myself.

I used to think I was collaborating quite smoothly with AI, until Insight laid out all those “troubles I took for granted” in plain sight. What’s even more interesting is that—AI is teaching me how to use AI better. It doesn’t just tell you “what you did”, it also tells you “how you could do it better”.

Isn’t this just a timely review with a customized teacher?

Why Do We Need an “AI Usage Review”?

We all know the value of review—looking back, reflecting, improving. But honestly, how many people actively review “how they use AI”?

Most people’s state (including mine) is: ask AI when encountering a problem, use the answer AI gives, ask again when bugs appear. Day after day, we think we’re using it well, but in reality we might be repeating the same inefficient patterns and stepping into the same pitfalls.

What Insight does is automate this review process for you.

It analyzes all your conversation history with Qwen Code and generates a personalized usage report. It’s like your phone’s “screen time”, but instead of telling you “how many short videos you watched today”, it tells you “how much work AI helped you do today, and how you can get AI to help you do even more”.

It’s Like a Mirror: Helping You See Your “AI Usage Habits”

The most moving part about Insight is that it showed me patterns I wasn’t even aware of.

For example, my report showed that in 4 days I used shell commands 246 times, far exceeding all other tools. Based on this, Insight analyzed:

You don’t like to plan in detail ahead of time; you prefer to adjust as you go.

You prefer letting Qwen Code execute and verify, rather than just giving suggestions.

Reading this made me pause—it was so accurate. I am indeed the type of person who “acts first, thinks later”, but I never realized this habit would affect my collaboration efficiency with AI.

This is the value of review: It doesn’t tell you what you did wrong; it lets you see your own behavioral patterns, and then you’ll know where you can optimize.

Insight also analyzes your work content distribution, active hours, and request types, helping you understand from a macro perspective what you’re actually using AI for. I discovered that I was mainly working on documentation and skill development (19 sessions) these past few days, rather than writing code—turns out I’m a thorough “documentation person” (laughs).

It’s Like a Diagnostician: Helping You Find “Where You’re Stuck”

Review isn’t just about looking at data; more importantly, it’s about finding pain points.

Insight automatically identifies friction points in your collaboration with AI and analyzes them by category. For example, in my report, the top three pain points were:

• Build and dependency issues: npm conflicts, missing ffmpeg, permission errors—each time took several rounds to resolve
• File write failures: heredoc syntax, YAML frontmatter corruption—repeated errors when creating skill files
• Git workflow interruptions: PR creation failures, quote escaping issues—operations that should be one-step became multi-round debugging

Seeing this, I had a moment of clarity—these problems weren’t occasional; they were systematic. Every time I encountered build issues, I had to struggle through several rounds. It wasn’t an AI problem; my environment configuration was incomplete. If Insight hadn’t laid out this data, I might never have realized this.

It’s like going to a doctor: you might think “occasional headaches are no big deal”, but the checkup report tells you your blood pressure is high—the problem isn’t the single symptom, it’s that you haven’t seen the underlying pattern.

It’s Like a Coach: Giving You Concrete Improvement Plans

After discovering problems, what’s next? This is the most valuable part of Insight—It doesn’t just diagnose; it also prescribes.

Based on your pain points, Insight gives rule suggestions that can be directly copied into QWEN.md (Qwen Code’s personalized configuration file). For my situation, it suggested:

• Always verify and provide URL when creating PRs (because I encountered PRs not actually being created multiple times)
• Use direct file writing instead of heredoc for skill files (because heredoc repeatedly caused YAML corruption)
• Verify dependencies are installed before starting builds (because missing dependencies were my biggest time sink)

These aren’t generic “best practices”; they’re personalized recommendations distilled from your actual usage data. Like a coach watching your game tape saying: “Your third step is always half a beat slow; try adjusting it this way”—instead of giving you a generic training manual.

It also recommends features you might need but haven’t tried yet. For example, it noticed I was using shell commands as a workaround to create PRs, so it recommended I try MCP Servers, which allows Qwen Code to interact directly with GitHub, saving the command line hassle.

It Also Encourages You: Seeing Where You Did Well

A good review doesn’t just find problems; it also recognizes progress and highlights.

Insight tells you where you did well and in which aspects AI helped you the most. My report showed a 97% satisfaction rate, with 80% of tasks fully or mostly completed. The capability where Qwen Code helped me the most was “proactive assistance”—it proactively discovers problems and proposes solutions, rather than waiting for me to direct it step by step.

The report also has a future outlook module at the end, depicting the evolution direction of AI-assisted development—from “passively completing tasks” to “autonomous multi-agent workflows”. After reading it, you feel: The way we collaborate with AI now might just be a starting point.

An Interesting Easter Egg

The report recorded a little story: when I requested the local IP address, I made a typo and the input was garbled, but Qwen Code actually understood what I meant and successfully provided the IP address. This “telepathic” feeling was kind of cute, haha.

How to Use Insight?

Enter the /insights command in Qwen Code to generate your usage report.

The report includes seven modules:

Everyone’s report is unique because it’s entirely based on your own usage data. Your pain points, your habits, your improvement suggestions are all tailor-made.

If you’re also using Qwen Code, I highly recommend giving it a try. Not to look at data, but to see that version of yourself you didn’t notice—and then make the next collaboration with AI a bit smoother than the last one.

Final Thoughts

Insight is more than just a feature; it’s a new perspective—letting AI help you understand how you use AI.

In an era where AI-assisted programming is becoming increasingly popular, learning to use and collaborate efficiently may be more important than mastering a specific programming language. And Insight is your mirror, your diagnostician, your coach.

📌 Project: https://github.com/QwenLM/qwen-code 

If you find this useful, feel free to give the project a ⭐️ Star!

Have questions or want to share your use cases? Leave a comment below~

🔗 Links & Resources

• GitHub: github.com/QwenLM/qwen-code 
• Official Docs: qwenlm.github.io/qwen-code-docs 

Join Us!

Qwen Code cannot thrive without the co-creation of community developers, and it also needs feedback support from internal users. Our important features like the Java SDK, VS Code extension, and Chrome browser extension were all co-built by internal partners. If you’re interested in Qwen Code, welcome to join us in co-creation!

Lastly, we’re also hiring! The Qwen Code project is recruiting AI full-stack direction technical talent, all levels welcome to submit resumes! pomelo.lcw@alibaba-inc.com 

Embrace the new era of AI Coding, starting with Qwen Code in your terminal.