AI Vibes with Zowe

So, after my first experience with AI coding, I decided to try it in a real project and contribute to Zowe Explorer. I remembered that I still had a few tickets assigned to me that I had promised to finish, so now felt like the right time to do it.

For those of you who do not know what Zowe Explorer is, it is an open source Visual Studio Code extension that lets developers interact with mainframe systems in a more modern, user-friendly way.

I went ahead and chose a low priority “good first issue” ticket. Since I was testing AI, I thought starting with something simple would be better. It is a large and complex codebase, so I did not want to overwhelm either the AI or myself while reviewing the changes.

I chose issue 2037, which is about allowing a multi member filter in a single dataset search. What I like about this issue is that I already had an idea of how to solve it and even shared my approach back in 2024.

And I went back in 2025 and I’m still coding it! 🎉

So yeah, I felt somewhat responsible for fixing this issue, and this time I made sure I would finally create a pull request.

With the help of AI, I started by asking for a plan. I wanted to see if it could understand the codebase and whether its suggestions were close to mine. The codebase has changed a lot since then, so I was no longer fully confident in my original approach.

I used Open Code for this exercise. If you are not familiar with it, Open Code is an open source AI agent that helps with planning and code generation. You can install it as a VS Code extension and chat with it while coding.

So I asked it for a plan on how to fix the issue and AI came up with a plan. In theory, it looked good, but it had several gaps that you could either notice early by reading carefully or discover later during testing and debugging.

💭 Thoughts 💭

AI could be confidently incorrect! It can present a solution that looks well structured and convincing but still be incorrect. You either correct the assumptions early or go through a cycle of coding, debugging, and fixing as you go.

I tried to correct the assumptions from the very start, and the more details I gave it, the more incorrect it became. They did say that the devil is in the details. And AI devilishly added wrong assumptions, to the point that it was much faster if I coded this myself than if I asked AI to do it.

So, in the end, I told it to go ahead and implement the solution, and I would fix things during testing.

⚠️ Warning! ⚠️

Before you allow AI to make changes, set a rule that it cannot commit or push anything without your review. AI sometimes has the audacity to commit incorrect code and then apologize, saying that they did not know they were not supposed to do that.

Anyway, we eventually arrived at a solution that actually worked, and I was happy with the result. Then came the “un-fun” part, testing.

This is usually where I struggle when contributing code. Since Zowe Explorer is a large project, it has strict testing standards. So I asked the AI to handle that part.

I believe this is where AI really stood out. It was able to write tests, run them, and fix issues along the way. One of my biggest pain points, writing tests, became much easier with AI, and that alone made the experience worthwhile.

I also asked it to update the changelog, and then I was ready to create a pull request.

In the end, I was able to author four pull requests, with one more still pending because I felt like I was opening too many at once😆

Summary

In conclusion, I think AI is a good coding assistant. It boosts my confidence because it makes me feel like I can approach problems even though I have no idea how to solve them. AI can provide wrong solutions, but it also triggers ideas that helps you tackle challenges creatively. From my experience, your skills are being amplified because of AI.

Do I think AI will get better in the future? Definitely. I am currently using a free version, and it is already quite capable. I imagine that using paid tools like Copilot or Claude AI would make the experience even smoother.

What’s next?

That is it for now! In my next article, I will try creating a new application using Copilot. Stay tuned to see how it goes 😉

Are you interested in what I am currently doing and would like to hear more? Well, don’t forget to subscribe here on Medium or on LinkedIn. Just click that follow button and you’re good to go!

If you’re feeling generous and would like to support my experiments, you can also leave a tip here: https://ko-fi.com/jellypuno