Panel Discussion: AI and the Future of Software Development

This panel discussion delves into the current and future impact of artificial intelligence (AI) on software development. The speakers, each with diverse backgrounds in AI, software engineering, and product management, share their experiences and insights on how AI is reshaping various aspects of the development process. They highlight the significant role AI plays in enhancing productivity, improving code quality, and addressing security issues within software projects. The conversation also touches on emerging technologies, like vector databases and the importance of prompt engineering in effectively utilizing AI tools. Overall, the panel emphasizes that AI is not just a futuristic concept but is actively influencing the software development landscape today.

[Music] hello hello everybody Welcome to the session hi guys nice to see everybody here around this virtual table so to speak uh well this is a great panel and is a great opportunity for sharing a lot of experiences around AI artificial intelligence and the future of software development a very hot topic so a quick introduction so that everybody that is listening and watching can uh hear from

uh each of us and then we get into the topic of today how does it sound sounds good all right excellent excellent all right so first of all I'm Stefano Stefano tempesta I'm joining this session from Australia 7 PM for me so it's not too bad after all I know someone is connecting from very early in the morning I will get there in a moment h i

lead the technology department at Atlas Atlas protocol in is a a a web three blockchain based startup we are already a few people that is working into the carbon credit uh industry and uh we're using a lot of AI technology for not predicting the value of carbon and measuring carbon dioxide in soil but we'll have a chance to talk about these and a lot of other things

in uh in the next 45 minutes together h i like to introduce the other subject mother expert that are joining me around this panel Mary over to you oh sure thank you Stephano um my name is Mary greski I am currently transitioning I was a developer advocate for the past six years or so uh most recently at data stacks and working on their Vector database and then

also bit on this event streaming um and before that I was in at IBM as a Java developer Advocate um then prior to that too I was a developer for over 20 years um I'm also active with the specifically too with the Java community and then most recently too with AI Camp so um in Chicago so that's where I am and I'm the one who had to

get up at 3:00 a.m. in the morning and currently it's 4 a. but no problem I just came back from Europe so I'm still in European time zone so very excited about this panel thank you yeah great and I guess Daniel and maybe next yeah thank you Mary uh um I love the aam community also so it's good to see someone else active with it thanks for

passing that on I'm Daniel per I do develop experience at Wate so I have a background in computer science and software engineering and I've been doing develop uh relations and advocacy in very different uh rights for about six years also and at the moment it's mostly just like leading the front and leading the uh the cause for using Vector databases in the front end and JavaScript community

so that's a lot of what I'm doing um speaking to developers here and there so I'm very excited to be here last but not least I guess I'm going to pass it on to Bobby yeah hi everyone very happy to be here I'm Bobby and I'm uh joining from Bulgaria I'm a product manager and I've been actively working on uh various AI projects since 2018 I have

background in software engineering as well but um I have also a few years of entrepreneurial experience so it goes uh through the whole Spectrum as you can imagine uh I've learned to do everything basically uh nowadays I'm also doing a PhD uh on AI and more specifically on synthetic data simulation models and digital Twins and um interesting yeah I'm um I I was just saying before we

started I was telling Daniel that um for uh one of the projects that I'm involved in we are actually using we8 um so uh it's a it's a very nice coincidence to be here together oh indeed we we absolutely have to talk about that um post the session I don't have to take up that space But it's pretty cool synthetic data yeah yeah it's uh more specifically

my dissertation topic is on training models on generated data and synthetic data and what are the opportunities there cool we are all very impressed Bobby thank you thank you congratulations and LU yes with completing your PhD um so look this panel is really an open conversation now this claim to everybody that watching here we didn't rehearse we are nothing improvising because obviously you know we obvious we

have prepared the topic uh but we haven't spent you know days and days in preparing this to be perfect by the minute so forgive us if sometimes we sound like we're going a little bit too long or maybe too short or some topic but we are really sharing our experience out of our heart and our brain in some way right so yeah but look let's go this

way H as in we have we have a few questions that we would like to ask to the four of us so like really open questions but also to the audience out there so use the tool to ask questions anytime we'll jump on the question straight away and we will have a proper Q&A at the end of the session don't be shy this is panel for the

4,000 people that are watching not just the four of us around this table okay now that we have done uh the the welcome and thanks everybody again for being here let's Reay one more time the topic of this panel AI artificial development can I make a little correction not just the future also the present of software development is that right I mean we are all using AI

every day already now it's not just the future it's something that currently is more than a buzz word we pass that phase of a buzz word you know how technology goes right so everybody talks about it very few uh know about it and and and many just talk because of they hear the password and that is fine right that is absolutely uh what we want to see

in the community create awareness but we are past that pH with ai ai has been around for many many years I've been doing a i in different flavors forever I don't remember I was too young when I started out of uni so is really a lot of time and uh the opportunity here is to build something that yes for the future but is already happening today h

so just a quick question to everybody here um how are you using AI technology in your current job in your organization but just besides the how you're using I'm more important I'm more interested in some way more important is why are you using now which value is AI technology bringing to you I really I would love to go first if you don't mind I really like how

you said that stepan like not just the future of AI development but the the present um and also how you just said we've really got over this period where people think AI is something that is being hyped I think we've got to the point where there's actual products that are being built that are really helping people make their applications better and um I think there's a lot

of really interesting things and we'll definitely get to that to start with answering your question specifically how how I particularly use AI uh I think the most obvious being in software engineering is helping me write code right we all use code editors and assistant and I think co-pilots have really shaken up how developers interact with uh IDs and build applications I think that's that's been a really

big difference for me a lot of the more mundane tasks that filled up creating projects and creating software I'm using AI to get to get rid of basically um so that's that's that's it I'd love to know how everyone else is is doing this or if this is something similar that they've they've play around it sure yeah um my perspective is not from um so much from

Hands-On coding uh uh point of view as I I'm currently not really um involved in Hands-On development I'm more on the um managerial site and overseeing uh the processes so when I talk about the future of software development in the context of uh AI to be more about um overall how software projects will be um evolving and how the various roles in this project also change and

this is something that um I would like to uh throw in here that it's not just that we are getting more productive there's also a shift in the responsibilities and how things are done in the within the processes of a software project and we can even see um that the established agile is kind of changing as well the different methods MH do you you have any um

any opinion or um anything to share on that I think for me too I think I I feel I'm the same I'm less of like consumer looking at it from a consumer kind of perspective yes I I am using some chat GPT sometimes right when I try to write an essay or you know do even like with call for proposal sometimes you're like okay I have an

idea let me use chat gbt to help me to kick start because sometimes hard to write a proposal those things um but you know and then is in as far as working with things because I you know I was with the product company so we're you know doing producing like vector database and all that so have to actually understand how things work and also two is um

you know trying to help developers to know how you can actually use the sdks the libraries to work with llms and so I I tend to be talking more from that perspective because very often too it's like developers come to see a talk and then they will be like well how do I start and what am I supposed to do so there's there's actually that whole aspect

too of the processing of the data flow for example workflow all that stuff I think that's also like kind of somewhat like less less being talked about because everybody is trying to do the fancy things I'm trying to use CH chat GPT to generate the stuff to see the results um at at the same time though there are lots of processing that are happening so and a

lot of things are still being sorted out and to be to be honest with you the more I look at it and I have to say I'm not like by any means like like a very expert in AI because I just always been a developer but it does change the way how you think towards implementing and and there's just some fundamental steps of processing data is still

there but there also kind of new ways of doing things and how do you actually work with it um to produce what you want so I think that's Al a very interesting area and in particular too because I work with data streaming so I think there's a a lot of things in that aspect to me I think still has to be tapped into how how do you

make use of streaming to do asynchronous uh realtime data for example I think that that's an exciting area but it's just not as much research or maybe there are research being done but just not talked about as much that that's how I look at it right now so I like how it each of us know is bringing a different angle of the experience Daniel more from the

deel so Community uh Bobby you mention you don't do much hands on coding but more on the managerial side Mary you are more on the coding side instead and I do a bit of both because yes officially uh C and but practically in a small startup you wear a lot of hats so you get really your your hands dirty in coding right yeah now H our experience

with AI in some way is twofold and I like to share very very briefly H one is uh quality and security improving quality and security and I'll tell you why we're doing this with the use of AI uh think of of an antivirus an antivirus works on detecting some imprints know some Fingerprints of code that is at the end a virus or troan or whatever it is

and those viruses are are developed with very common and known patterns so they are easy to recognize the same is for bugs bugs in software H there is an organization called oasp probably most of us that work in web development know it o w WP that identifies the top 10 threats of vulnerabilities in web applications and every year they refresh this list because every year there is

something new something a bit not uh that comes up as a new threat now this top 10 over the years has been changing a little bit but there are a few that never change which I don't know blows my mind something like uh SQL injection I mean we be talking about SQL injection for 20 years so we know how to fix it and people still keep on

doing the same mistakes over and over again so this SQL injection or session hijacking or impersonation social engineering they are always in the top 10 are we learning anything or what so now they so that actually means that these are common known issues bugs in the software that can be easily detected so when something is repeatable is perfect for AI when something is a bit new then

you need use you need to use your human brain right because AI if not trained cannot detect something that happens for the first time but something that has produced years and years or historical data is perfect for anomaly detection machine learning right we we know that so we are using AI technology exactly for that run static code analysis and dynamic runtime execution of the code because you

can get bugs that are called in the static code directly in the code in the source code but also you can get for memory allocation or IO input output or for integration with apis especially SQL injection or apis without proper password protection or token protection they are very very common mistakes and that can be easily detected so AI can definitely improve quality and security of your software

highly recommended to to try to give a uh know a chance to AI not only as a way to co-pilot your development Daniel you mention pilot absolutely that is a way good code but also in a way to respond to the code that has already been produced even the code that AI itself has produced you have no guarantee that copilot is not going to produce something that

con may contain some vulnerability and why is well you you you you a data scientist by by background right as you mentioned so if we've been training these AI model model used by copilot on source code that for the last 10 years has produced the top 10 vulnerabilities of OAS but never change yeah very likely in this source code that has been used for training there are

vulnerabilities well I really like where this is going particularly because I know Bobby does synthetic data um but before before I I ask on that I just want to add like I I really like what you mentioned there it really Builds on what on What Mary said about these use cases outside of you know like the the consumption or in a consumer sort of perspective and I

think the beauty of AI that I see and how it's going to change the way we do things is it's uh and this is a this is a realization that just came up it really democratizes a lot of access to really high quality tools so I just gave a talk recently about uh search in general and if we think 20 years back the idea of searching the

way a company like Google introduced it was just mind-blowing right this is something that not a lot of people had access to and I don't know if you saw any of the announcements that were made uh last week from some of these big AI vendors they were really interesting models and I think this Advent of AI coming close and being in the uh Center or like the

Forefront of technology and how we're building is also democratizing access to some really B like powerful functionality when you think of things multimodality or uh classification these are things that are flashy and yeah you might use it in chat GPT but like Mary said this is extremely important for something like streaming you know or it could be extremely relevant for a use case like security or um

uh what's this what's the word of this um the dependencies when there's a security problem what what's the name for that the word has G and blanking um but basically any sort of security fixes that you want to keep uh tabs on or use case like search like it was extremely difficult to have these multimodel um searches and experiences but with a lot of these models that

are open it's much more approachable it's democratized and much more accessible for developers to build really interesting things into their um applications so absolutely agree I I love I love that you mentioned yeah and uh when we talk about the AI uh I think there are at least two main important um directions that we can look at and one is starting from right now from the present

how are how is the developer job changing right um and that's from automating to um recognizing patterns that you wouldn't and using it using uh various algorithms to assist with even vulnerability identification but then the other path would be the that going forward maybe not not from now but very soon artificial intelligence will be so integral of uh will be such an integral part of software development

processes and overall within the whole um effort of producing software that we'll have very big shifts in even how software is developed and just to give you another um maybe um topic for discussion and to to to think about recently I read somewhere that at the moment we have um mobile applications for example that are trying to solve a particular problem like um and fit a spe

a specific use case so most mobile applications try to do that um and while trying to fix a specific use case or working on a specific use case they try to cater to as many users as possible going forward not immediately but uh in the future we have such a shift in this field that we won't have mobile applications for everyone uh we'll have a mobile application

for each person solving each of the use cases of that person that that person would be interested in and that doesn't have to be a fixed mobile application it will be something like an AI agent that the moment you have a problem on or you have a question you you have a use case that you need to be worked on you work with your in assistance with

your AI agent there will be a mobile application generated in the back and then it will solve the users's use case and here we are not even talking about a developer it could be anyone uh it will solve the user's use case and it doesn't even have to continue existing it can be just scrapped from now on if it's a one-time use case for example so it

won't be anymore a matter of having marketplaces with mobile applications or any kind of application actually trying to be um to attract as many users as possible but it will be user first and then uh the AI assistant or agent generates the solution to the users use case problem do you have any thoughts on that I have a couple sorry to burst uh when we think about

the future of AI and when we think about Technologies in general and this is an open question there's two types of Technologies there's uh um the internet as a technology or a group of technologies that work together um do you think AI would be more of a an FTP type of Technology where it's really running in the background keeping a lot of things running or it's sort

of like a um uh the idea of an operating system type of Technology where it's it's at the Forefront of the way we do a lot of things you know like is this something a lot of people pay attention to and actively you know work with or is this something that will like keep a lot of things running in the background I'd love to know what you

all think the hold yeah I I I like where we're going with this discussion because even for me I coming you know from a perspective of software design and architecture all these things and which is different right again from like consumer side but I think there there's just many ways of how you can Implement something and to me I think it almost seems like yes we need

to identify all the steps maybe workflow and then abstract them all out and kind of essentially too I think that that idea of the agent maybe the agent is just responsible for certain things in certain layers of this software and I I can see how there's common Services being used and then their service layer or something and then just the different layers that we can build upon

that it can actually be applied to a mobile app and be applied to a you know regular desktop app or something and you know different ways of interacting but but I think it's definitely from a development point of view it's kind of interesting and to me I also feel AI would be most useful especially with let's say some lower level type of implementations you know there are

libraries and things like that as developer maybe we're less concerned with all the nitty-gritty of well how do I write a sword routine and all of these things and because it's already you can just ask AI I I need some I know what I need but I don't want to deal with all the details not that it's not interesting but it's timec consuming so some of these

small little things can be actually done easily by AI kind of more robotic and kind of do some of this mechanical stuff so we can as developers we probably are moving more towards the higher layer or providing some value and knowing what to use will be you know ideally two would be the AI to decide okay I'm not sure what sord routine I need to use let

me ask the AI I need to do something should I use a sword bubble swort should I use a whatever other sort or something like that right so I think from a development point of view I think we take away some of these things that sometimes you know I remember as a the beginning developers like well what do I use there are many ways I just can't

can't decide but I know what I need so maybe that's something that AI can come in and say you you need to use this sort you need to use that search and whatever it is and so so then we can move to like higher level of abstraction to solve more business problem is what I'm kind of looking at so not this AI that is in background of

our life as a developer no more as I like the metaphor of the operating system is not an application that I run but it's just the supporting and Consulting helping and uh yes it makes my life much easier do I want this level of intrusion all the time well we are already doing we already accepting there is a lot of AI already in the operating system Sy

that we run I don't see how it's going to be any different it's going to be better so but by all means yes I want to feel empowered I don't see personally I don't see any AI as a replacement of my capability I see that as an enhancement of my capabilities something that will make me more productive so I welcome that so I I I I always

hear something in specific uh software and developer circles unpredictability is your enemy right and if if you give this like abstracting away a lot means it's much lower in that architectural level right and then there's also a level of unpredict unpredictability that comes with using specific types of AI particularly like large language models right do you think there's some concerns here that we should be thinking about

or that we need to address technologically to take it to the next step and actually enable us to have this sort of uh system where we could trust AI you know is good question actually Daniel because uh there is a level of trust at the moment that I don't feel I can put 100% into the current capabilities of AI uh especially the generative AI the the the

one that is based on llms uh simply because the technology is really ramping up on the moment has been around for a few years for sure but the broad Mass adoption is been in the last 12 18 months right so before we can reach a capacity where the level of trust is like I don't have to think about it it will take a journey uh there are

a couple of strategies that I'm doing at the moment to make sure that whatever I I check I prod with not the like chpt or di also for content for image generation I I I double check in two ways one is I work a lot on the prompt there is a specific aspect of know asking the right questions giving the right context is actually a um a

branch of the generative AI that is called prompt engineering so the using the word engineering which is really know Manufacturing in a allegedly perfect way something that is expected to produce the best result now so we engineer buildings we engineer uh Bridges right so I we don't want them to fall right so at the same way we do it also for for the software output that for

the output for the content that is by these sort of gpts so asking the right questions so with the right input is absolutely uh critical there are some very good tutorial also from open AI on prompt engineering so I really uh invite uh our audience no and everybody that is using the like of chpt and similar tools uh before asking the questions go through this checklist there

are some five or six different uh best practices no uh how to ask the right the right questions to produce the best output so definitely one good thing that I recommend and the last one very briefly I check the result I question it if something does doesn't sound like I'm convinced I will ask again are you sure as simple as that and sometimes the answer I get

is oh you're right is actually this other way so and that is just the way uh these llms work at the moment right based on the prediction the statistical occurrences or words so challenge the answer don't assume that whatever PT or the like says is correct don't trust it until we arrive to a level that trust can be given for granted and I'm not there yet well

yeah that's that's a good perspective sorry Mary go on oh I'm sorry I yeah I was just going to quickly say I mean prompt engineering and I understand his functionality but I I've been kind of thinking too I think you know if for generative AI if we're so heavily depending on prompt um in other words you know it's like consuming I'm trying to use it but I

need to know what to ask then it's not really true AI because you have to know kind of you know some of this intelligence that we know that the true AI I mean for True robot actually to me right I should be giving it some kind of fussy kind of prompts and I say I need to do this and that can you actually decide for me what

to do I think that actually would would be the true really capability of a true AI robot and right now it's we're not there yet we need to prompt we need to ask the right questions so I mean not to say it's not good but it's just maybe a necessary steps in our Evolution towards the real true AI like the real true a robot that can actually

think for us because sometimes we just don't know what to do we need some help so I think I feel that there's just still a lot more you know in the Neuroscience kind of that level that needs to do well how actually we think how do we decide what actually needs to be done that level I think is still there what I think so yeah and sorry

Bobby you were GNA say something and I cut in front of you no that was that was actually perfect because uh I was going to say a very similar uh thing that it's a journey right so we're still very early in in the journey and uh things will be evolving a lot uh in the next few years and a couple of decades especially and um at the

moment the we we are in the middle somewhere in the middle right now of how difficult it is to um enter this field on one hand you have tools like CH GPT or Microsoft copilot or you know tools that are at everyone's fingers and yet in order to use them to their full capacity you need to have a lot of Knowledge and Skills exactly for example in

prompt engineering you need to know how these tools work in order to get uh high quality results most of the time yes yes and and that's that's the difficult part on one hand Millions hundreds of millions of people are using it and yet very small percentage know how to fully use it so the barrier to uh entry there is high really to to properly use it there's

a there's a steep learning curve still and that will be something that will be uh I think all all of the developers Behind these kind of tools aim to reduce this complexity as much as possible and as fast as possible uh so I think that going forward it will be as I alluded kind of um um couple of minutes ago to be more agent based right uh

you won't have to do a lot of the instructions yourself and we are starting to get there kind of uh last week I saw that there's now I think an open source tool that writes prompts for you and they're really good um so soon it won't be just you interacting with a single model or it it would not even one huge foundational model like a GPT for

example but soon we'll have a series of models all together working in unison where they'll be catering to the different parts of delivering you the final result so if you just have a fuzzy sentence um infused with emotion in the stress in a stressful moment and you say hey I need help help me with whatever problem I have at the moment there will be one model that

handles just your request your interpreting what you exactly what exactly you need and translating that to some bigger model that will be doing the thinking behind and then yeah an AI to assist to get better result from the AI I think it's pretty cool yeah and one of the um just I want to say that one of the differences that I think we we in our mindset

that soon will happen is a lot of people expect we have a single model that will be um that that that will be a broad AI kind of that's able to handle many different tasks and it's not just specialized in one or two things right and I don't think it will be a single model I do I think it will be an architecture combining a bunch of

different models and here I'm not even talking about multimodality for a single model I'm really talking about many different models coming together um in one a or multiple doesn't matter um that work as a proper AI assistant is this an evolution of uh the idea of the mixture of experts which I think a lot of people are exactly are using in in a lot of the models

they're doing now exactly kind of yeah okay I managed to get this slide very quickly from one uh session that I've done internally to to my company uh like and know these five prompts that I mentioned before we don't have the time to go through all of them in detail but I I like a couple of them know specify the scenario assign a role is I be

like when you describe a user story no in your software requirements don't just ask a question straight away create the context no which is as a software developer that is building this sort of application I want to get this this and that can you produce this right so create a bit of context the scenario assign a role a Persona because their response can be different if whoever

is receiving is a data scientist or my grandma right so that the the quality the the content will be definitely different yeah and uh I I have to drop off sorry because I have a a session but Stefano Mary Bobby it's been great I'll let you finish up and uh thank you everyone for watching thankks yeah no good luck with your session Daniel thanks for being part

of this panel yeah thanks for having me take care and uh now it now might be a good uh time to invite the audience to questions absolutely yeah y if anyone has anything uh we have another five or six minutes uh in this session so now is the now is a good time to shoot your questions at us yes indeed um and uh Daniel mentioned this um

it's not really a new thing but it's as everything in AI it's been there for a while and then suddenly everyone is talking about it the mixture of experts um and uh um I I find it uh very interesting how not just on a level of how it works but how it changes our mindset when we think about neural networks and models right so it's uh the

the mixture of experts for those that don't know I'm going to oversimplify it it's basically um you have a model and that model instead of having a single um neural network let's say that tries to generate the response it has many many different smaller neuron networks each spe specializing in a different domain so depending on what the uh the user uh asks and the what the user

needs are it tries to pick the brain kind of a specific neuron Network that's specializes in that or multiple even sure sure so so Bobby if you don't mind me asking so are are there already libraries or maybe software that's capable of using allowing us to use multiple llm at this point that's actually that was actually a question I got asked at the jcon conference too last

week somebody asked can I use multiple LMS and is a great question and as you said you know different llms they have different models of the neuron Network and they do things differently so it's good to combine right the multiple ones they each have their strengths and if you can combine usage is really would be really powerful you're able to touch upon a lot more um you

know soloft a lot more cases or kind of like that and I I couldn't answer cuz I really did not know if there already capability out there to allow you to work with multiple simultaneous llms and do you know um yeah on on a architectural level um there are uh ways to just train or use ready trained um different models in unison just working together but um

the mixture of experts is something that we can look that we can see already in the mistra AI models M right yeah so um the latest few Mixr models um they're open source llms by the way and uh they they're performing really good on a lot of benchmarks uh for an open source model they're they're great so I um I highly I'm a fan so uh I

like open source stuff so I recommend checking it so they are integrating mixture of experts so if you look at their architecture um they have a very interesting even even just on the their website announcing the the models on a very high level they explain how uh The Prompt can be even split in different um that is parts and the different parts are sent to the different

neural networks within the bigger system and architecture so very it's very interesting and recommend checking it out yeah it is yeah very we do wonderful yeah thank you yeah um and and M TR is also a European based company so that's true in Europe we need more more AI yeah to to compete with the others yeah cool very cool yeah okay we have another uh the last

couple of minutes do you want to to wrap up any final words well look from my side I just to say thanks to all of you for really sharing this experience and it shows know how the topic of II is hot is really modern and is the right time to do it that's the invite I want to extend to the audience out there that is listening and

is participating also to this a big event which starting today is going to go over the next few days uh little uh self-promotion I have a session tomorrow where I'm going to talk about machine learning in the context of anti money laundry and again this is a very very good use case how to prevent fraud with money laundry machine learning is there and is a technology that

we are using currently to address this kind of International challenges so yes today in this panel we talk about the use of AI for software development the use cases of AI in the world are really huge the potentially huge and is happening already today is the right time to be part of the journey that's right thank you yeah indeed and um I also have a session tomorrow

and I have I I think Mary you have more than one session right that you yeah I actually have one session I think it's like at uh European Western or Eastern European time is like 6 after 6 something um or was it yeah I well anyway I got all my time zone all Miss messed up um but it's a more like afternoon that's right and then I

also had we'll be having a panel too I think it's more of depops transformation and then I will actually be hosting Jamie my ex- colleague Jamie like in about an hour or so to Jamie and two other speakers too so yeah thank you yeah good luck guys with the rest of the event yeah it was a pleasure thank and thank you everybody enjoy byebye see you bye