Macromo

Technical Lead

Outreach

Senior Software Engineer

STRV

Frontend Developer

Dorsia.io

Frontend Consultant

Full Stack Developer

Web Bespokers

🚀 Technical Lead at Macromo - Leading innovations in the fascinating intersection of software and biotech.

💼 With years of experience as a Software Engineer, I've worn many hats – from coding to leading dynamic teams. At Macromo, we're pushing boundaries with DNA and Blood tests, and I'm at the forefront of that revolution.

🎙 Passionate about bridging the gap between IT, DNA, and career development. Through my posts, I aim to shed light on the latest tech.

🌱 Personal Growth Advocate - I believe in continuous learning. Over the years, I've seen how personal branding can redefine careers. Now, I'm on a journey to grow mine while helping others do the same.

Kornel Dubieniecki

I teach testers how to shift left. | Trainer | Coach | Workshops | Advisor | Speaker

I help frustrated testers move beyond checking boxes to become strategic partners in their teams and organisations.

My focus is on practical quality leadership and evidence-based advocacy. I write about:

• Reframing testing language so quality discussions land with engineers and business stakeholders
• Shifting QA from reactive task work to proactive risk and value conversations
• Test automation that scales, not just ticks boxes
• Evidence over opinion when shaping quality decisions

I pull insights from real DevOps and Agile environments and offer frameworks that make quality coaching concrete and repeatable.

If you want to talk about quality metrics that matter, meaningful automation, or how testers can influence without authority, feel free to send me a message.

Katja Obring

"Testing can't save us anymore." 
A claim in a post that crossed my feed last week.   

The author has 14 years in test, and their argument is that AI is generating code faster than we can validate it, the model is breaking, and the next evolution is "continuous reasoning about system alignment."                                                                       
The problems they describe are real. The picture of testing they're painting is not.
                                                                                                                                                                                      
Verifying function X returns Y, test cases living in spreadsheets, validation happening in sprint reviews after the fact: none of that has been considered good practice in the field for at least two decades. The context-driven testing movement challenged this framing in the early 2000s. BDD emerged specifically to connect business intent to automated verification. Specification by example exists precisely because "specs in Confluence, code in GitHub, decisions in Slack" is a documentation failure, and good testing practice was designed to surface it.

What they're calling "the next evolution," continuous reasoning about system alignment, is what quality engineers and test strategists have been advocating for, under various names, since long before AI entered the conversation.

The more interesting question is why so many organisations are still running testing as a late-stage checkbox two decades after the field moved on.

Training QA professionals to evaluate mandated AI | Keynote Speaker | Author

One attendee, answering what they'd do differently next time they evaluate an AI tool, wrote four words: "Set clearer success condition."

That answer came out of Deciding Fast, a four-hour workshop I ran this week for a software engineering team in Sweden. The workshop is built around one premise: most AI pilots fail not because the tool is bad, but because nobody defined what success looked like before they started. You can't pass or fail a vague hope. You can pass or fail a clearly defined question.

The delivery conditions weren't ideal. The client had everyone in the same physical room sharing computers, when the workshop is designed for fully remote with individual Miro access and breakout rooms. I adapted and ran it anyway, and six of eight people who responded to the feedback form rated it Good or Excellent.

The framework is Question, Uncertainty, Constraints, Action. It's for QA professionals and software engineers who are tired of running expensive pilots and ending up with "it was okay, I guess."

When you ask AI something and get a polished, confident response, the instinct is to take it and move on. Anthropic's new research suggests that instinct is the main thing separating effective AI users from less effective ones.

They're not an unbiased source, and I'd encourage you to read the methodology with that in mind. But the numbers on iteration are interesting. Users who kept refining, who asked follow-up questions and pushed back on parts that didn't quite fit, showed 2.67 additional fluency behaviours compared to those who didn't. They were 5.6 times more likely to question the model's reasoning.

The implication isn't "prompt better." It's simpler than that. Treat the first response the way you'd treat any first draft: as something that needs scrutiny before it becomes final.

For anyone who works in QA or testing, this isn't a new capability to develop. It's an existing one to redirect. The questions you ask when a build fails, when a requirement is ambiguous, when a bug report doesn't quite add up, those are the questions worth asking after an AI response. The same rigour in a different context.

The practice: ask one more question after you get your answer. Push back once. See what changes. You don't need a framework for this, you just need to stay in the conversation a little longer.

Full article in the comments.

Here is something from the survey I ran last month that I have not talked about yet.

18% of respondents said their biggest frustration with AI is not the output quality. It is that the tools have no sense of test strategy.

One person said AI generates tests "like a web crawler, not from a user perspective." Another said the tools "spit out test cases with zero regard for test strategy or code testability."

I keep coming back to this because it is a different problem from the "slop" issue I wrote about last week. Slop is about the output being wrong. This is about the output being irrelevant. The test cases might be technically valid. They just do not test anything that matters.

Think about how you approach a new feature as an experienced tester. You do not start by listing every possible input. You think about what is new, what is fragile, what changed recently, where the integrations are. You weigh risk. You make judgment calls about where to spend your time.

AI skips all of that. It goes straight to "here are 47 test cases" and treats a null check on a text field the same as a race condition in a payment flow.

The interesting question is whether you can get AI to stop doing that. Whether there is a way to bring your testing judgment into the conversation before the AI starts generating.

I think there is. More on that this week.

Tell the AI how you want it to behave, not just what you want it to produce.

That's the gist of the third finding in Anthropic's AI fluency research this week. Only 30 per cent of conversations include any explicit instructions about how the model should interact. The rest jump straight to the task and let the model fill in the blanks.

The model will fill them in. It defaults to being helpful, agreeable, and confident. If you don't ask it to push back, it probably won't. If you don't ask it to flag uncertainty, it will sound certain even when it isn't.

A handful of instructions that change how a conversation goes: "Tell me what you're uncertain about." "Push back if my assumptions are wrong." "Walk me through your reasoning before you give me the answer."

None of these are complicated. They work because they shift the model's defaults before the conversation develops its own momentum. Once you've had three exchanges where the model was agreeable and confident, it becomes harder to suddenly ask it to be critical.

Set it up at the start. It takes one line and changes the dynamic of everything that follows.

Anthropic publishes this kind of research with their own interests in mind. The habit itself is worth building regardless of who's recommending it.

Link in the comments.

taplio

The Battle

Kornel Dubieniecki

Katja Obring

Followers

Posts

Engagements

Likes

Performance ScoreKatja Obring's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Kornel Dubieniecki

Katja Obring

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Katja Obring's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreKatja Obring's LinkedIn Performance