AI Effects on Quality

In this talk, Buris Tranjif, the CEO of Qualifast, discusses the effects of artificial intelligence (AI) on software development and quality assurance, emphasizing its dual role as both a business engine and a facilitator in development processes. He highlights the importance of startups and the opportunities that arise when integrating AI into their workflows, particularly in creating structured requirements and improving communication. Tranjif warns against the pitfalls of over-reliance on AI, advocating for a balanced approach that prioritizes quality alongside speed in product development. The speaker also explores how different roles within the software development lifecycle can adapt to and leverage AI advancements, ensuring that quality remains essential despite the industry's urgency for fast-paced delivery.

Hello, welcome everyone. It's very exciting to be the last person to speak before the launch. I know that you probably would not have that much of patience and I'm I'm going to try to be entertaining. Obviously, I'm going to fail, but at least I'll attempt. So, uh as Yana said, uh I have diverse knowledge. Um I've been very diverse roles in my uh career and I do

believe that uh being positioned like that actually allows me to connect the dots and seek some kind of correlations uh and try to really speak of what we are faced with and try to speculate how we actually should react to it so that we are really the ones that are adapted and performing well under new circumstances. So uh my name is Buris Tranjif and I'm currently a

CEO of a company called Qualifast in which we focus on building startup solutions from scratch for clients of ours. Uh but what we actually do is employ as much of efficiency, automation and smartness as possible when we go along the process. I've uh been as Yana said involved in multiple uh communities and I'm proud of that. uh I've been also organizing some of them. Now I'm just

a plain member in couple of committees revolving around startups and also being a manager of big organizations. And uh the fact is that currently because of my position in Quadifas, I'm actually also acting as a CTO in multiple products in different stages of their lives. So today's presentation is going to deal about the the effects of AI uh to quality and uh everything you're going to see

actually I have a slide about this the presentation is structured in several sections. The first one is going to present the context of the the whole thing we are speaking about. Then we're going to speak about what are the things in modern times and then I'm going to also present what are the opportunities that uh are happening uh in this era of AI. So the presentation context

starts with uh who I am and what I do. The presentation context is going to be uh let's say mainly focused on uh start from scratch startups because I currently operate in in there and this is an exciting position to be because start from scratch startups actually allow you to start over and really think of the best technologies and approaches to use in a particular solution as

opposed to say well established solutions where you already have processes you already have certain foundation and there the changing the processes and changing the tooling the technology is a little bit harder. I would also speculate that uh many of the things I'm going to tell you are also generally applicable for software engineering not only in the start from scratch startups but still I I want to really

make stress the fact that the start from scratch startups are my main focus now and I would also mention that definitely um affected very much by the current VC ecosystem strategy and it's definitely very targeted towards AI Like if you are pitching a startup to a VC, the first question you're going to receive even before they ask you about your business idea or the name of your

startup would be where's the AI in it? So AI is very prominent there. And then uh to also present the presentation context a little bit further um I distinguish two separate sets of applications of AI in the context of IT development. And the first one is AI as a business engine meaning using the AI to build something for your business model for your business logic and whatever.

And then also AI as a facilitator of the development process. And uh this presentation will focus on the second one. We're not going to delve that much into detail about how AI can affect the business engines that you're building. So now AI and what is the current state of affairs? Currently AI is being pushed to be used for everything. You're using it for guiding you to the

d giving you the directions uh to where you want to get to for producing vaccines for producing some uh media content uh for uh speaking to humans instead of you and stuff like it's really ubutous everywhere. It's uh AI has been around for quite a long while but actually uh now it it became that common place when the uh W language models became available for everyone and

now everybody has an easy access to work with AI and see what it actually can produce. Then uh I also want to use this to mention that uh throughout the presentation I'm going to be speaking of AI but actually very often I'll be referring to the new things that are available through AOS. Obviously this is not the only tool that you can use uh in the AI

ecosystem. It's just the new thing that appeared and the the one that I'm contemplating on. And then um with the fact that AI became that common place and the fact that VCs are pushing very much for for AI u I I would say that certain organizations have taken the the A hype to the very extreme and they've probably even passed the let's say the the border of

sensibility. We'll see this in a little while. This is um a graph that probably many of you have seen. Um I speculate that the AI AOM adoption is currently a hype that is probably going to turn into a trend. And this this graph is actually depicting how both hypes and trends uh perform in our society. If something new appears and it seems like exciting, everybody are getting

very excited about it and the hype is piling up piling up piling up. But at certain moment the disillusionment comes and we start realizing that no this is not going to really solve all our problems and we are not going to really live in the paradise from now on. There are actually certain limits to the capabilities of whatever we're facing with and then in certain cases um

this turns out to be complete disillusionment and it turns out that the thing that we so firmly believed in is actually not having any real grounds in the world and it's just disappearing. I do believe that AIOS are not going to follow the hype trend. It's it's going to rather follow the the graph for the trends. Uh we're currently already passing the peak of the hype about

AIOM. There are certain people that start doubting the the big words that were spoken about it. But this is good uh because now we're heading to the place where we are going to start thinking practical about it and start thinking of what we can actually do sensible with this new tool that we are presented with and where can we use it so that we become better than

our competitors than our predecessors and stuff. And this is the graph that I promised to show you the over adoption of AI. I I do believe that in every single case be it software engineering or whatever uh AI AOM adoption as everything actually has certain saturation be between beyond it actually if you use AI AOS you're going to make your organization worse rather than better. I also

do believe that uh almost every single organization or task can facil be facilitated or effectivized a little with om and AI especially if you're do doing the things that really can be improved with this and I've drawn this graph just to show you that uh if you're not intending to use any AI probably you're not going to receive the greatest outcomes but then as you steadily start

adopting AI you are also about to pass a peak after which you're going to start diminishing the value of what you're doing and eventually you're probably going to doom your organization if you're so keen on having AI for the sake of it and not really thinking of where it is actually applicable or not. And this is actually deviating between organizations. The graph doesn't always go like that.

Sometimes it goes like that. So it it can be very little AI can contribute to you. Sometimes it might be a more but it's always you have uh to hit a limit and then beyond it uh it's you have to be sensible and and start thinking of other approaches that can actually improve your business. So um what are the AI effects on IT and software development? the

market demands um ever shorter times to delivery and we have uh very often the state of mind in which the product versions are now thought of as throwaways like uh I've met many people that think of let's build it now and then in two months we can just rebuild it just because the the building time is so short uh let's and it seems to me like we

are shipping PUC's now we are not really shipping products from time to time we are just shipping PUC's all the time this is I I call this the throwway way way of thinking and now everyone considers himself expert in everything. Um yesterday I was joking that uh this is exactly like me. I also do consider myself as expert in everything. Now I have a uh a peer

as the same as me. uh like uh AI really can advise you everywhere and now even if you speak with the people that don't have any technical background or whatever now they do believe they understand technological implementation they do believe they can challenge your assessment your approach or whatever and that goes with everything it it can go al go probably to doctors and surgeries everywhere in our

world it actually goes the same way and um probably you've already observed it today multiple times in the presentations but it's not the only conference all the conferences are now really really focused on speed I even had one conference in the in in June where the speed was uh literally uh pasted on the background of every presentation it had to be present on each and every slide

and we even had IET attending the the whole conference so that the speed was really emanated everywhere so speed is very important KPI so important that we uh and we also see the the words like I I've written it here like speed cutting edges uh reducing time to deliver and stuff like that we have to be very very fast eventual or fast so um speed is the

most prominent thing we can actually observe with AI why because things that we used to do completely manually AI is producing for ourselves and this is um kind of a reverse byproduct of the sensible way of thinking now because of VCs and Not only because of the hype I'm told I need to use AI if I'm using AI what should happen it should be faster so because

it should be faster with AI so being fast is actually my main goal and now you see because I was u projected the main goal of being fast now sorry main goal of using AI now I need to be fast and speed is like really my target and I have um my my popular uh opinion is But this is uh a kind of prank way of thinking.

We really need to focus on how we actually can do this better. So the very idea of qualifast was that um when you're building a startup, you cannot really afford only focusing on speed. You have to balance off between speed and the quality that you built in. Otherwise, even if you've been very fast to deliver the first version of what you have, you're doomed from then on

because you have to actually scratch out everything because it's complete crap and build it a new and then probably will be taken over by competitors. You really uh are going in a game in which you cannot win. Either your idea is never going to succeed or if it succeeds, you're not going to be able to keep up the pace. And I I really do believe in the

make haste swively proverb or uh fina lent in Latin. It means that you have to basically balance off urgency and diligence. You cannot just focus on one or the other. And it seems like in the AI era we completely forgot about this proverb and we're focusing only on the urgency. It's very very urgent but what about the diligence? What about the sustainability that David mentioned a little

bit earlier? So this graph is actually depicting um how I imagine uh software development and and not only it's basically a balance of the two qualities uh being fast and and uh producing quality thing and the green dot is depicting uh where you should actually strive to be operating. It's something of you have to combine somewhat of quality and somewhat of speed and you have to be

smart of what kind of combination mixture of the two are working for the particular case. I drew this graph by myself so I was not able to really animate it but imagine that this dot is also moving among the line uh because different cases are different. So now we go to the core of the presentation. We are now in the quality AI sorry in the AI era

and we have to think of what do we do with the quality? How do we compensate the urgency push? And I do believe that every certain situation, every new change of circumstances is creating opportunities. That's why I also placed the word opportunity. Let's try to really look up what are the opportunities that are given to us in the AI era. and could uh so sorry uh jumped

over a couple of slides um some uh sessions before Hamza was speaking of the uh industrial revolution and he was pres presenting um the AI situation from a different angle but I want to focus on something else related to industrial revolution throughout it we actually had the situation in which the AI uh sorry the design of the different things that were produced by people uh really got

diminished. It we had very degrading design. There was a lot of push against the industry of the revolution because there was a lot less human touch to everything that was produced. That's why the AI generated picture of the artisan compared to the manufacturing of uh plastic bottles. But actually uh with manufacturing we also were given a lot of opportunities and a lot of ways to produce even

further value for our community. And we have to be smart about what we're faced with. Some situations, you still have to have the human touch. In certain situations, you can probably allow yourself to really cut edges and and be faster in production. Many people need plastic bottles. Why shouldn't we produce plastic bottles when they're actually needed? And um this is also uh I'm just remembering a situation

in LinkedIn where uh there was a person who was also using an AI generated cartoon and there was a cartoonist below it. Stop using uh AI generated cartoons. Don't you see that the character is not looking in the eyes of the other character and it's a little bit um not straight or whatever. And I was okay this is your perception but I'm just a plain user and

I really don't see the difference in value between the two. The fact is that with the industrial revolution uh the whole society became a little bit more stupid. uh it's kind of a war uh its perception of quality and this is actually good for society because it seems like the spurious quality was never needed and probably that will also go with the AI revolution and we should

also be smart about our expectations and in certain situations were our quality to what is actually making sense and producing real value not something beyond that. So what are our opportunities? Um basically the opportunities that I'm going to present are going to uh start using the same weapon that created our dire situation or our crisis uh in a way that uh will also facilitate improved quality. And

you'll notice that the structure of this uh part of the presentation is uh quite matching what David just before me was using projecting from different pers perspectives of different roles but is going to be uh dealing with uh other aspects that I'm going to be focusing on. So we'll be looking at the opportunities uh for different roles in the software development uh life cycle. I've grouped uh

product people. This is actually encompassing also product design and and several other roles. developers and QAs. Let's delve into it. So, uh for the product uh people, we have several things that are improved or changed uh with the AI capabilities and can be and we should use these opportunities. We can create uh PC's using VIP coding. Uh we can use uh the VI the AI for uh

structuring our requirements. Uh we we can use tools that are facilitating our communication. uh we can use it as an inspiration source and we can get better insights using AI and a little bit better details about each and everyone. Um David just before me was speaking about VIP coding. I don't quite believe this is going to replace your software development and I don't believe in completely VIP

code solutions. I don't believe this should be uh shipped to production. But on the other hand, I've faced multiple product people that uh in one situation or the other wanted to do a a PC that's going to uh place the whole uh concept, the whole product in a better position in front of VCs or whatever. And uh it used to be that the product people had to

go to the developers, they had to produce this PC and stuff. It was a bit cumbersome process. Now with VIP coding, you can produce a PC that will seemingly work or work in several cases and you'll be able to directly go and present it to the VCs without having any kind of a dependency on other people. The developers will be able to focus on the main stuff

that they are building and they would not need to bother about the PC that is either way going to be a throw away. It's just a segue to for example new investment or whatever. And uh the fact is that uh with VIP coding, David explained it in what further details, but uh when you build it, you are probably not going to really understand what's going in there.

Probably would not be able to fix all the problems and stuff, but it's when it's a throwaway PC, that's totally fine. Then uh requirement structure. uh to be frank probably um I'm one of the most extremely faced uh of the people that is faced with the most extremely not structured requirements of in the entire hall why because I'm speaking only with very early stage startupers people with

idea but people that don't know how to structure anything about this idea actually in most cases I'm just having a chat verbal chat with the people that have the idea and we start building stuff and it used to be that we We're basing our work mainly on verbal communication and probably certain several written steps. With the AI facility, we're actually able to take this to a lot

more structured, documented and tracked way. Uh you can use AI to transcript your conversation. You can use it to summarize it to structure it in tasks and then also to track the modifications that are happening. Now finally even startupers have some structure in their requirements and they can really start believing you that they changed their mind 5 minutes like in in the span of 5 minutes or

5 days and this is a cool application. Uh this is one of the main things we're experimenting building tools in. It's still not perfect but definitely giving us a lot of improvement compared to what it used to be. Then uh communication facilitation a couple of more things beyond the requirement structure. I as I put it a picture is uh communicating a thousand words. Now you can generate

a picture that's really communicating the the main line of thought that you wanted to say. And uh exactly like the requirement structuring, every kind of communication can be uh transcribed and summarized so that people that didn't attend a particular meeting or whatever can still catch up with what happened. And um another thing is that uh um it used to be that uh it is uh totally a

work of the designer to create mockups and wireframes. Now product people that don't have designer uh knowledge or maybe experience they can still try to create such just with verbal commands. And uh something very important when you're building a product uh you should not be thinking of it as a single instance and single iteration of uh work. It's actually iterative thing. You deploy to production. you monitor

what's going on and you're really planning what to do next so that you really match expectations and what people are doing. Uh with AI you're really able to explore a lot better what people are doing to get insights of interactions and stuff and really plan uh what is actually sensible for your product. So uh you finally know what's being built and it's very important about early stage

startups but actually I do believe this is spanning all over the the board even more complicated and more established organizations issues. Uh and you can have your own transcriber and librarian. Now these are automated. They're not real human beings. Uh in some cases you're independent and self-sufficient. This is with an asterisk just because these are in certain situations that you would not want to ship this to

the production or whatever and the decisions you take about the product or what better funded as as arguments. Oops. So now we have the developers perspective. Several things related to what the developers are seeing has changed in in the AI era. I'm going to tell you a little bit about how I do believe uh is the proper way to use AI tools. I'm going to try to

compare coding with an AI agent with pair programming with humans. I'm going to tell about the code inspections uh uh impacts uh the unit and integration testing and then uh the code to requirement tracibility is something that we already touched also on the production uh side the product management side. So I how I do believe AI should be used um the so let's go with the positive

and then we'll go to the to the negative. AI should be used to really automate some boring stuff that does not require too much of thinking. The greatest effect I've seen in AI is actually helping you doing stuff that you otherwise would have been reluctant to do. And I've faced this situation multiple times before and now I can really say there is an improvement. If I I've

seen something in the system and I I certainly uh suddenly realize that this is not properly implemented and I have a better concept. This concept usually just goes down the trench and uh I never realize it just because I realize how much things needs to be changed in different places in the code. Now with the AI tools, if I can explain my concept well to the AI

tool, probably this concept will be um propagated not only to the new stuff that I contribute but also into the legacy code that used to be there. And now my code becomes more consistent and probably with better structure and probably more robust. Um very important thing about the AI, it's actually not only about developers, it's about every kind of a role that wants to really accommodate this

era of AI. We need to make AI really accessible and needs to be only a single click Every kind of evolution is always faced with a lot of reluctance and push back from every human being. We human beings don't want to be changed. We don't like being put in new circumstances. The first step is always the hardest to take and everybody needs to make sure that this

step is as close to you as possible. in if you're for example manager please do help your people to really know what they need to do so they utilize AI and this might mean even helping them setting up their ID so they are AI enabled or telling them about the tools that are available with this people would really try and experiment and see what the results are

and now we go to the don'ts the most important don't is never use AI to solve a task that you don't know how to validate the result of and David also touched this uh but the fact is that AI is never producing anything of a value if you hope it's going to do the stuff and you are just going to ship it without ever understanding what it

did this failed me all the times not me because I've never done it obviously I'm a smart guy other people that are hoping on on AI building stuff for them shipping something that they actually never understood the assignment of for example building stuff in technology that that they never care to understand or uh building a requirement that never they never understood the actual requirement of so if

you don't know how to validate the task you definitely should ask until you understand it used to be the case now it's still the case you cannot rely on the AI to segue this one also AI is not an excuse for not learning and not getting more knowledge you should have This knowledge AI is not going to be replacement of you getting more intelligent and more expertise

expert in the things that you're doing. Now the the pair programming uh metaphor that appeared in my mind when I was presenting uh preparing the presentation actually it feels like pair programming but it's not the if you compare the attributes and the characteristics of human to human pair programming to human to AI pair programming you'll see that actually it's completely exhibiting different attributes. The only thing that

you benefit uh when you're working with an uh AI pair peer or whatever is uh increased velocity on the quality side, AI probably would know more features of the frameworks that you're using than you and it might suggest new solutions that might be useful for you. On the other hand, uh the every time that you I've been faced with AI produced modules, I'll say that the architecture

is subpar compared to real expert architects. So these two kind of balance off each other. The fewer books uh it has an asterisk just because it depends on how you actually use the AI. Uh if you rely on the on the AI to be the main coder and you're the observer and putting in some inputs, there are going to be more bugs. But if you are the

main coder and use AI to monitor what you're doing and suggest where problems are, this is actually going to improve the quality of what you do. So this one depends on your approach to using AI for AI facilitated coding. And the two maybe most important thing about uh things about pair programming. This is the knowledge sharing and transfer of knowledge around the the organization. These are completely

nonpresent when you're pair programming with an AI agent because not a single other member of your team is involved in the thing and this needs to be stressful and and work on code inspection. This is uh basically a practice in which you're improving uh the quality of what you do and AI facilitates in multiple ways. It can help you orientate in the code, find problems, patterns and

stuff and unit and integration testing. I'm developer and I can tell you that developers are very reluctant to write tests only if they're pressed to they do and now AI helps me do the thing that I was reluctant to do basically bootstrap test suits create some test cases obviously I still need to revise but at least uh I'm having some partner to help me start off with

this one and the requirement traceability we already covered in the product management also it it can basically track everything that you change also hot fixes and you have better tracibility of all changes. So uh for the dev summary we saw what's the right approach to using AI. By the way this is not only for developers it's for any kind of a role. Uh and then uh we

mentioned several other areas that AI can improve. And the last perspective the Q perspective um we have the market misconceptions and couple of um areas of testing that are affected by AI. And please note the AI generated picture here. When I was generating it, I wanted to really represent that AI facilitates testing. But then I realized no, this is depicting already QA even before AI, right? Because

uh automation is native for QAS. So the market mis obsessions and this is a thing that I'm faced with every day. uh because of the speed uh push for the speed uh people were very often uh uh really forced to cut edges and very often they tell me we don't need to include QA also because of the throwaway kind of thinking they're telling me but this is

going to be thrown away either way in two months why should we QA at all we don't need to test and then the expert in everything doesn't help us either they tell me now I can QA myself because I'm expert in everything with this to be frank Although the AI era creates the greatest possible need for queuing, there is the greatest the lowest possible demand for it.

And this actually uh forces me to very often hide the queue in different roles, different kind of perspectives just so that I know they're needed, but they really cannot be sold like they used to be before. Now several QA practices I covered uh with what uh AI can actually help when we're speaking of functional testing uh only about blackbox testing AI can help us identify better selectors

help us uh do some kind of testing that used to be a lot harder for example canvas image comparison insertions it can help us do a responsive UI testing better because of better understanding of semantics and we can also have uh AI generating test cases for us and but this is still exhibiting the same things like developer code being generated by AI. The architecture is a little

bit poor and it probably would not be that easy to maintain and reuse. If you built in the proper uh architectural patterns this can be significantly improved on the performance wall testing. David already mentioned it just before me. uh performance testing is one of the topics that actually is increased in interest just because of the AI era. Uh business owners are very often considerate of the fact

that if you're a coding your solution probably it would not be performing that well. So this one they're ready to accommodate. The fact is that what testing and performance testing you can do a little bit better uh just because AI facilitation helps you uh mimic the user behavior more closely. And then you can write even more tests for the wall testing. The security penetration tests are the

the topic that are of the highest interest after the AI started being introduced uh for coding solutions. And the fact is that it can also benefit from AI. You can track uh you can basically do blackbox penetration testing even better with AI and you can even use AI to extract adversary user behavior based on walks and what's going on. uh a lot better wipe testing. Uh David

presentation just before me already proves that this is just emerging term and it doesn't quite uh have a single definition yet. The understanding of VIP testing according to me is testing that's still using the VIP coding techniques but is focused on the perception and the emotional uh experience for our users. meaning uh feeling of smoothness uh feeling of uh this be feeling right and consistent with everything

on our system and this is just emerging term there are a couple of tools related to that especially using kms that are going to help you understand what's the vibe that your users are using uh feeling sorry and obviously the report and uh defect tracking uh this is uh quite a necessity when you're doing queing if If you don't track what you've done, you basically didn't do

anything. But it's also cumbersome, takes time. Uh with AI, you can actually automate this very often. You can just uh with couple of force instruct certain bot to create the proper defects with all the facets needed for for the defect trace to requirements and stuff. It's cool that this kind of a boring stuff is So the summary for the QA uh we can uh facilitate many practices

related to testing. Um uh the fact that uh automation is native to QA I already mentioned when we saw the picture and uh the fact is that the bridge between the product and development requirements are is even uh narrower now when we have AI being fast. Sorry in the end of the presentation seeing that my time is uh going away the conclusion quality the AI away. So

um in order to have uh the the to adapt to the era you have to really understand that nothing should be considered out of outside your capability you cannot just put up with the fact you cannot understand AI and not use it. Basically with this you'll be dragging behind the other people. You have to always keep in keep an eye on what can be done better and

where you can actually improve with the new tooling that you have and AI is one example of this. You have to always be clever as of how you balance off um urgency and diligence and you you have to always think of opportunities opportunities opportunities and right here I should also confess that I was not able to produce the picture that I wanted to to AI generate. I

wanted to also demon and then let's see whether you'll be able to imagine it. I wanted to also show you that the developers, product people and QAS are now brought closer together. They can speak uh like closer to each other's language and they can understand each other better. And you already saw that there are practices in the different rows that allow you to speak uh and do

some some things that were uh attributed to the different roles before. And the end of the presentation actually goes with this one. The title is actually not exactly the the proper one. It should be the satisfied with wife attitude. So it's true about this presentation, but it's true about everything you have in life. You have to be ready to adopt the new stuff. You have to not

be scared of anything and you have to accommodate and seek opportunities, opportunities, opportunities. Thank you.