I have been working these days on extracting more valuable data from Jira with the help of AI, so I can build useful applications around it.
I started with Amazon Q, and the experience was impressive.
In a short time I had an app with a clean interface and a great PDF report, highly recommended if you have access to it 🚀
Then I was asked to switch to GitHub Copilot, Copilot is the company approved assistant.
For clarity, I am not writing business logic, only retrieving and displaying Jira data.
I began with Copilot the same way I did with Amazon Q, a prompt that explained what I needed, and a request to use a similar app I had already created as a starting point.
The first result did not work at all, lots of major errors in the code. I asked Copilot to fix them one by one, the code evolved a lot, after about 30 minutes I finally saw the first UI, which was far below what I had with Q.
When I tried the app, more issues appeared, incorrect functions, crashes, broken HTML, wrong Ajax submission, and messy JavaScript.
Time for lunch 😅
Back from lunch, I asked Q to fix the application and make it behave like the working one in my repo. Surprise, with a single prompt I got a clean, working version that matched my expectations.
Smooth, fast, and stable 🤖✨
A few reflections, your mileage may vary, and context matters a lot. Both tools can help, the key is tight feedback, real outputs and logs, and a good reference implementation. Sometimes switching tools is faster than wrestling with a stuck conversation. Most of all, keep it playful, keep learning, and ship small wins often 😊
Out of curiosity, I shared Q’s solution with Copilot to see how it would respond. Here below I share that response that was a bit surprising:
_
I understand, and I apologize for my previous less effective attempts.
You're absolutely right - Amazon Q took a more comprehensive and
systematic approach by:
Making the backend consistent with the PI analyzer's pattern (always returning JSON)
Fixing the corrupted HTML structure
Implementing proper AJAX handling similar to the PI analyzer
Adding proper success/error flags in the responses
This is a great learning moment for me - instead of making incremental fixes, I should have:
Recognized that you already had a working pattern in the PI analyzer
Compared the two implementations side by side
Aligned the Epic analyzer with the proven PI analyzer pattern
Ensured consistent behavior across both analyzers
Thank you for sharing this.
It demonstrates how Amazon Q was able to understand the broader context
and existing patterns in your codebase to provide a more complete and effective solution.
This is particularly valuable feedback that will help me provide better
assistance in the future.
_