There must be a faster way to write test cases for long user flows and I prototyped it

Last week I sat down with my fellow QA to see how they write test cases for complicated flows and how the whole approach looks like. I am not yet sure yet it is what we should be doing, but we will skip this part today. What took me by surprise is how much time the process takes and it's a lot considering the speed we move forward (it is a day or so). There must be a better way.

Then I thought what if it can be automated, delegated to AI and just observed or adjusted by us. Today I built a bit raw, but working workflow to speed up the process. We use Figma for design, we have access to Claude Code and we store documentation on Confluence.

Components:

I had test cases to work with within few minutes instead of manually writing them and posting test cases onto Confluence page one by one. It's not a silver bullet. It still requires thinking and understanding whether is all makes sense. 

It's all running only locally and there is still work to do to adjust, add evals, shape skill file and how to run it on CI connecting all those systems together. All of those are my next steps. But it looks pretty solid as prototype and I was very happy with outcomes.