The Hidden Risk in AI-Generated Tests and Requirements - Olivier Denoo

The trouble is that this answer looks smart at first sight. So if your job as a tester is pressing the same button, running the same script that you didn't even write for 20 years, I think that you are seriously at risk. If you don't master the business, if you don't understand what you're doing, then it's a terrible danger. We still don't know how to exactly describe what we want.

Welcome to Software Testing Unleashed. My name is Richie and I'm happy to have another episode from the QualityLand Conference from Vilnius for you. My guest today was Olivier Denoo and we talked about the future of our profession as testers in the times of AI, why expert testers will be needed more than ever, what a lost suitcase tells us about edge cases and which skills you should build now. And now enjoy the episode.

Hi Olivier, great to have you on the show here. Hi Richie, nice to be here. Yeah, it's great. We met a few weeks ago on Mauritius at a conference.

That's a secret. Shouldn't have told that. And now here in Vilnius. Yeah, looks like holidays.

Yeah, it looks like, but it's hard work. It's hard work. to mention that we have to do a lot of stuff here. So yeah, it's great to see you again.

And now we have the opportunity that you come to my podcast here, a podcast for software testers, for quality engineers, for all people who like quality. And yeah, you are for me one person who lives quality and tries to inspire people and shares his knowledge. And so I'm very happy that you are now here in the show. I'm very happy to be here as well and yes I'm doing this for about 30 years now.

Next year I will be 30 years in the field. So yeah, we were just chasing dinosaurs with their hands. You remember the time where testers were the poor guys in the cellar in the musty shade there? Oh yeah, this is bringing sweet sweet memories I would say.

For a couple of years, nearly 10, 15 years, we fought for testers to be recognized by the IT because we were some sort of a niche and, you know, the necessary evil, the extra cost, the people who don't know where the defects are or where not the right defects are actually. And yeah, we fought very, very hard to get recognized and to, you know, set up testing as a profession. And with agility, it went a little bit backward because now we are a tester as does not exist anymore is testing as a role.

And of course, the good stuff in there is that quality belongs now to the whole team. And it's a it's spread over the whole organization. But again, we need to fight to get this this role properly properly set up and properly organized. And now we have a new new kids on the blocks like AI and again, all over again.

I think we as testers, we had to reinvent ourselves a lot of times in this last 20, 30 years, because we have all this agile stuff, you said, we have test automation coming and all this stuff. And now this little thing called AI, which we have to deal with. And yeah, maybe we can take this opportunity today to make an outlook in the future and to think about what should a tester do in the future and what can we learn now to have a job in the future? Yeah, well, basically, yes.

And first part of the question and first remark on what you said, yes, testing is a never ending story and we have to reinvent the wheel of the world, our world and our, how to say, battlefield or playground every time, which is why I'm so passionate about the job it's yeah, never gets boring. You always have something new. Now, back to the AI stuff. I think that the tester should definitely tame AI and adopt AI as much as possible because AI is not about to disappear.

This is definitely not a small thing that is there. Maybe it will inflate a little bit. Maybe there's an AI bubble and it will burst at a certain moment in time. Maybe yes, maybe no, we don't know yet.

But nevertheless, it will be there to stay. It is there to stay. And testers who will be able to manipulate and use AI in a proper way will be the ones who will stay. And I do actually believe that AI is a big opportunity for the testing community, because now the kind of trend that I hear everywhere is you don't even need to be a developer to code.

Now you have this vibe coding, we have these kinds of applications where you just prompt a little bit and that you get something that is ready to serve. Trouble is, how do you control the code? If you're not a developer, how do you know that it's good? Well, of course, it fits more or less with the requirements.

How does it evolve? How does it maintain? Is it security safe? Because we know that GitHub can be infected by many backdoors, Trojans, whatever.

How do you manage that if you don't know what you're coding exactly? I do believe that in a couple of months, couple of years when we have enough of these applications, AI generated application that will come and whether you put agents to control the agents and LLM to control LLMs or human in the loop, whatever, at a certain moment in time you will have more defects and you will need testers and you will need expert testers and what I mean by expert testers are testers who know how to test and use the techniques and know where to look at and how to approach these kind of problems, but also a tester who understands the business, what is behind the application.

Because you don't code, you don't write a piece of software for the sake of using a brand new platform, a brand new technology, a brand new language, whatever it is, you develop a piece of code and an application to fit one goal, to make one task and help people's life. This is where the business understanding is very important. And our business has become more and more complex because we have all these applications that are intricately linked to each other and connected to each other.

And then the business flows become extremely difficult. And you need people to understand how the whole process works. But what can we tell a manager when he comes from the next conference and says, Now I've heard AI also can test and we don't need any test to go away. Let AI just test all the stuff.

And he should probably change job because he doesn't understand anything. Yeah, I'm sorry, but if you have that sort of reasoning, yes, you can probably shorten your resources and you will probably have people who lose their job. Definitely. Yes, this is the sad reality.

But if you don't keep human in the loop, if you don't understand what is behind all that, you will very, very quickly end up in a big business catastrophe. That is absolutely for sure. Because the trouble with AI is that it looks smart. You know, it always gets a positive answer.

It's always right with you. Everything is fine. You are right. And so on and so forth.

It always brings an answer because it is programmed to always bring you an answer, whatever it is. The trouble is that this answer looks smart at first sight, but probably is smart for 85, 90%. But what about the 10 or 15 percent remaining that are faulty? And it might be that AI completely hallucinated and invented something that has nothing to do with what you're willing to achieve.

It might be that this is completely invented requirements or invented kind of test cases or invented a piece of code or whatever. And you don't realize, you know, we see that in the references, for instance, you have AI, okay, please quote all your reference. Tell me something about whatever topic you're interested in and please quote your reference. And then you have reference that that looks smart, okay?

Oh, just this author wrote something about that, but not that book and not with this co-author and the whole story is completely made up. And then if you cannot detect that, if you don't cross your reference, if you don't understand that there might be a flaw, then you're running a big risk. So this manager definitely needs to learn a little bit more about AI. Yeah, yeah.

Yeah, and we need the arguments there. So to argue that we need these persons who look at the whole thing and to have this quality perspective. It looks too good to be true. It's true, it is fantastic.

Don't get me wrong here. I believe that AI is very good stuff and it's doing marvellous things. It can help you a lot. But if you don't master the process, if you don't master the business, if you don't understand what you're doing, then it's a terrible danger.

And you're running at risk. Nevertheless, AI can be a great support for us as testers in the process. Where do you think, especially when we look to the future, the points in the process where AI gives us the superpower as a tester to make great quality in the future? Anything that is easy will be the low-hanging fruit for AI.

So if your job as a tester is pressing the same button, running the same script that you didn't even write for 20 years, I think that you're seriously at risk. Because this is the kind of stuff that an AI can do. AI can do most of the parts of the testing process. If you ask AI to write requirements and check requirements, it can perfectly do.

If you want AI to sort and prioritize, they can do that. If you ask AI to write test case, they can do that. But you need to be sure that the test case and the requirements are fine. And where I see a little bit the future roles for the testers is also in there.

We had a talk this morning and that was very interesting. That was about the trouble that we are facing. On one side we had the Oracle, in the middle you had the AI, and on the other side, the entry point, you had the requirements or the context, whatever it was. Basically speaking, what I infer from that is that we are just facing the same troubles as I faced 30 years ago, which is, okay, where are the damn requirements?

Where can I see exactly what the piece of software is supposed to be doing? And how precise and unambiguous can we be in describing what we want. And, you know, in the very tiny details. And on the other side, how can I check that the result I get is all right?

So if you have a tax declaration, for instance, this is something that is extremely complex because you have so many different possibilities here and there. How can you check that the number in the end of the day is the correct one you know, amount or get reimbursed the right amount, is very difficult and that you need to understand. And that's not something that is such an easy thing to do for an AI. Yeah.

Yeah. When we look at the testers now sitting there in the projects, what would you suggest, what should the testers learn these days that they can survive in the future? Or what are the skills we need in the future? Definitely something about requirements engineering.

That is clear because this is learn how to prompt. So prompt engineering, something like Gen AI, there are prompting techniques, there are prompt engineering techniques that they need to learn because AI will be part of their job. Definitely, but I think that they are already deep inside these things. The DevOps approach, pipelines and so on and so forth, automation.

these are the things that are quite demanded these days, I would say. So manual testing will stay, but most probably more into the exploratory testing and these kind of fields. And on the other side, well, that's learning about a business, understanding what the business is all about, and possibly, and this is also one of the things I'm really keen to say in conferences, and anywhere I go, that's about UX. That's about the user.

That's not only the human in the loop. We are developing software for somebody, and that somebody needs to be able to understand and use that. That somebody is not the one that we model. This is not only, we don't have enough personas.

That's what I want to say. Because for instance, my mother, now turning 80, she doesn't know what a smartphone is. She doesn't use an app. Now, if you go to Paris, for instance, you cannot take the metro if you don't an app.

Okay, how does it work for somebody who just wants to travel to Paris and wants to take the metro? 80, does it work? So yes, it is well defined. Yes, it can help some people, but do not exclude other categories and do understand that things are not working always as expected.

I think what you say resonates so much to me because in my speeches it's also about why are we doing all this testing stuff. And there are always people behind using that software and the apps and the websites and all this stuff. And we are dealing with all the fancy AI stuff and how to create test cases and form them. And we often forget that there is someone who is using this output in the end.

And these little things, these little edge cases, I would say that we don't think of. I have one of my presentation is about how I nearly lost my suitcase coming back from a trip in Thailand. And the whole process is perfectly defined. So if you look at an airport, you know, luggage management, it works perfectly.

Each and every step, each and every case is taken care of and no problem. So to put the long story short, my luggage was stuck in Geneva while I was already arriving in Brussels. So, okay, Swiss Airline just told me, "Okay, no problem at all, we found your luggage. it will fly with the next morning flight tomorrow.

Perfect. And the next morning flight tomorrow, while I had a tracker in my luggage, I could check and it was there in Brussels. Perfect. And then I got this mail from Brussels Airlines telling me we didn't find your luggage.

Yeah, but I know where it is. I can even tell you exactly in which room it is probably, because I know Brussels Airlines nearly by heart. But guys, how is it possible? And you know, what I suspect is that there is one guy who just took the luggage out of the conveyor and just forgot to scan the tag.

And then suddenly the luggage disappeared out of the system. And it was lost. Yeah, that's it. That's where you need to think a little bit.

Well, some people are going to use or misuse possibly my piece of software or sometimes even hardware. That can be a car, that can be a plane, that can be... And these kind of unexpected events happen. Yeah, that's true.

So we have to learn all this stuff for the future, for the prompt engineering you mentioned and all that. And when we look on your other role, you are also a very active part of the ISTQB. Yes. You're part of the exec there.

And when we look at all this certification, at the last 20, 25 years now, it helps to professionalize our point of where we are now and our profession where we are now as a tester. And what is in your, when you look at the ISTQB portfolio, what are the main certifications and learning paths you would suggest to go there to deal with all this stuff in the future? Yeah, if you're more interested, we were talking about AI, we have two certifications on AI in the ISTQB. If you are interested in using the AI for testing, that's definitely the GenAI syllabus.

It's recently been updated, so we had a minor update. It's because we want to keep abreast with the market and with the changes, because it's a fast changing environment. And the second one is more testing the AI, understanding the AI mechanisms, and probably more looking at the architecture, agentic architecture and so on and so forth. This is AI, the AI syllabus, CT-AI, which is also recently updated.

So these are the main two products regarding AI. Then of course, we have a new syllabus that is going to be issued on DevOps. For the rest, this is the classical path, let's say, that we have like test management, like Agile and so on, and then these two syllabi. Yeah, when you see testers now, and they are now playing with all the AI stuff and all going on, what do you think is a skill, beside AI, we have to improve for our work?

Communication, definitely communication, because, well, first of all, communication in the way that we write prompts. I'm in this business for 30 years, as I said, or nearly. Next year I will be 30 years. We have to celebrate.

Oh yeah, we'll have to celebrate, definitely. I'm not sure. (both laughing) Well, yeah, the trouble always was the same. And that is what I was saying earlier in this conversation.

We still don't know how to exactly describe what we want. And with an AI, it's sometimes very difficult not to have this kind of fuzzy requirements, unclear statements, ambiguous statements, and so on and so forth. So I think that these communication skills are very important. Even if AI generates your report, sorts out your test cases, builds a kind of prioritization framework, whatever, you will still need to communicate to your business stakeholders, to your management at a certain moment in time.

And these communication skills are, in my humble opinion, very key to the testers. And this is something that IT schools do not teach, you know, these psychology mechanisms, these power games, these communication skills, how to put things in a proper way, how to present data in a proper way, not altering the results and so on and so forth. These are things that are usually, you know, set apart or just forgotten in these courses because we focus on technology. We are engineers, most of us are engineers, and we focus on technology because we like it.

But that's not the point and it's not the only point. So definitely, yeah, the soft skills are, in my very opinion, very, very key to the testing. Yes. Olivier, thank you very much for sharing your knowledge and your insights here in the podcast.

I think it is really beneficial to think about what can we do for our profession in the future. and I know that you are very passionate about our profession and to develop it for the future. So thank you very much that you were here. We put your contact details also in the show notes, so if anyone wants to get in contact with you and want to communicate with you.

No, no problem. I will always answer. I have no SLA for 48 hours, but I will definitely reply to all your kind mails or unkind if you wish to. Great.

So thank you very much. Enjoy the rest of the conference. Thank you, Richie.