Daria Tsion | Comments

Thank Brandy, for sharing your experience. That's such a classic feature flag story 😄 They add a layer of complexity that’s easy to underestimate and manage sometimes. And from my experience, such "unrelated flag" bugs usually come from FF dependencies in the system, which is also another topic to discuss. Did you have any strategies for managing or testing flag combinations? Would love to hear!

Thanks so much, Gary! 🙏 Feature flags can feel a bit "extra" at first. Still, once you start using them intentionally (for safer releases and experiments), they really change how you think about delivery. Curious if you’re considering introducing them in your workflow anytime soon?

@Gary, thanks so much for your kind words and feedback! I'm glad you have found something useful in my milestone!

Congrats! Well deserved!

Thank you, Pranav! This is a great way to frame it. That shift from execution to orchestration is exactly what I’ve been observing in practice. The more AI takes over repetitive or mechanical tasks, the more QA value moves into problem framing, defining constraints, and sense-making of the output. For me, prompting became less about “getting code faster” and more about making my own thinking explicit, especially around risks and assumptions. I’m glad that part came through in the article.

On keeping the prompt library separate: we discussed the same concern internally. We treated prompts as shared QA knowledge rather than implementation-specific assets, so a central repo made reuse and evolution across projects easier. For highly project-specific prompts, keeping them close to the test or codebase absolutely makes sense. In practice, we use a hybrid approach: central patterns + local adaptations. Consistency of structure mattered more than location. LMK please if you have any q!

Thanks, Nataliia, for a thoughtful comment. These are great questions. I'll respond to you in 2 separate comments^^ On role-based vs context-based prompting: I see them as complementary. Context defines what we are working with (system, feature, constraints), while the role shapes how the AI reasons about the problem. When I explicitly define a QA role, the output reflects testing heuristics, trade-offs, and structure much closer to real QA work. Context alone often isn’t enough for that depth.