Jerzy Biernacki: Before You Hire an AI Agent: What You Need to Know to Achieve Real Results

This talk explores the subject of AI automation and the various approaches companies can adopt in implementing it. The speaker emphasizes the importance of understanding business processes and identifying areas for improvement before jumping into automation. They categorize automation into three levels: SaaS solutions, low-code/no-code automations, and custom solutions, each with its own advantages and challenges. The discussion includes insights on the role of AI agents and how they differ from traditional workflows, highlighting that agents provide flexibility but can be less predictable. The speaker also shares their methodology for process automation, which involves discovery, prioritization, and implementation, and stresses the significance of addressing cultural shifts within organizations as they adapt to AI technologies.

Hello. >> [applause] >> Yeah, I hope I won't trip over the wires here, but I because I like to walk around, but yeah, let's see. So, welcome everyone and let's start with something pleasant. Imagine you're on a tropical island enjoying the ocean, swimming around and just enjoying the weather, the sun, the sea and suddenly out of the blue comes straight at you a 3 and 1/2 m

shark. What would you do in this situation? Okay, it's it's maybe not the the most pleasant situation, but let's say I'll give you three options. First option is to run. The second option is to fight and the third option is to just enjoy it and take some photos. Okay, so so just raise your hand. Who would who would run from the shark? Okay, not so much, not

so much. We'll get back to that. And who would try to fight? Oh, we have some fighters here. Okay, okay. And who would take some pictures? Oh, most of you, cool. I have some fellow marine life photographers here. Good. Good because there are these are my pictures and there'll be a quite a few of them here in the presentation. actually 90% of people would probably run. And

it's the worst option. It's the worst option if you know anything about sharks. Fighting a shark is also not the best idea, especially that this kind of shark is like twice your size. It's in natural habitat, so it will definitely kill you or bite you as at least and any bite could be lethal in the water. So, so yeah, not definitely not the best idea, especially that

even if it wasn't aggressive, if you kick it or you hurt it, it will just hurt you back. So, C is the best option, taking some photos, but why? Let me explain. So, first of all, sharks are not hunting humans, uh, not actively hunting humans. And there are many common myths around, uh, like that if they smell blood, they will just go frenzy and attack you. Well,

yes, sharks can go frenzy, but if they smell fish blood, not human blood. Uh, actually, I've been around sharks, uh, many times, and two times my friend was bleeding heavily, and there was like no reaction from the sharks, none at all. And there was some research that confirmed that they're not reacting to mammals' blood. They They react to They react to fish So, that's one thing. Uh,

the other thing is that, well, it's in it's in it's natural habitat, it's can swim much faster than you can, so running around, uh, running, uh, from it is not the best option. Also, you'll do a lot of splashing. And splashing can, first of all, confuse them, and then also trigger their hunting instincts. It's very similar to how dogs will start chasing you, uh, when you, uh,

run away. And, uh, actually, you're much more likely to get bitten and even die from a dog, uh, uh, hunting you, let's say, than from the shark, because you know how many people are, every year dying because of shark attacks? >> Five. >> Yeah, around five to 10 people. And how many how many sharks are killed by people every year? >> How much? >> Yeah. Yeah, 100

million, uh, sharks are killed by by people every year because of, uh, because of, uh, some, uh, stereotypes that are because of these myths, because we don't we don't care, because they we think about them as mindless beasts. But they are not they are not hunting humans. Also, like fighting, I explained why it's not the best option. And taking photos, why taking photos? Well, uh, I'll give

you three tips. If you are lucky enough to meet a shark, because I really consider it lucky. So, if you're lucky, first of all, remain calm. Look it in the eyes and because they like to attack from the back like tigers. So, so definitely stay calm, look it in the eyes. Also, maintain a vertical position, upright position, because to them you'll appear much you'll you'll appear bigger

and more threatening. And they're actually considering us fellow predators, not like prey usually. And last last thing, any object that you have in your hands like a camera could actually be useful to just redirect them if they're too curious. But you're probably wondering why I'm telling you something about sharks on a AI conference. poor creatures that are killed by millions are actually uh uh They're a victim

of the stereotype that is uh often published and and uh spread by the media. And we have the same thing about, if you think about it, we have the same thing about AI. And AI agents is on the opposite side, right? It's not like it's all bad. Sometimes it is like in the communications likely exaggerated to some extent. In terms of AI, it's largely exaggerated, especially when

we talk about oh There's one more reason I like to share my photos, right? And it's probably one of the few things that you'll remember from the from the uh from the conference. These things about sharks. I I've checked it. So, people and probably some things are not going to be that much relevant. So much changes over the months in AI. But when it comes to AI

agents uh some research show uh that uh what Gartner did the survey and they they concluded that 70% of organizations right now are using agentic AI. But, over 60% plan to use it over the next 2 years. And that was the most aggressive adoption curve among all emerging technologies. We are in the peak of elevated expectations uh when it comes to agentic AI. And the reality is

much more complex than that. We don't need We don't always need agents and that will be the point of my So, I'll try to explain you what are the factors that you need to consider whether and and when to use AI agents when trying to automate your processes. So, yeah, I was already uh introduced. Uh I have a technical background. I have a PhD in computer science

and for the last 11 years I was working in Miquido where I uh well, first of all, helped to create a business line around automating and delivering AI applications. We delivered over 50 of them, so quite a lot. Uh and it was like well before the uh the era of LLMs. Uh but right now I'm also automating different business processes. Uh yeah, I guess that's it about

me. And let's say about let's go to the automation. Uh I think we should start with uh categorizing different kinds of uh automations. I'll say there are three categories or three levels that we can automate uh how we can implement uh automations. The first level is the easiest. Uh it's from the easiest to let's say the the hardest to implement. Uh SaaS solution. So, tools that are

out of the box. You can just buy the license, uh subscribe subscription, pay as you go, and you have it. So, you can quickly implement it in your organization. There are many different tools out there right now. Uh for example, you can have a uh call transcription tool like Fireflies or there are many different others. If you don't use them, well, I definitely recommend you to use

one of those. Uh or it can be something totally different. It can be AI chatbot for customer service like Tidio. Tidio is just one of the examples, just randomly randomly checked randomly selected. But, yeah, just plug it in into your e-commerce and your clients can just uh talk with uh with the data. They can ask about their orders, etc. So, that's the first level. Then, we have

the second level. And the second level, these are low-code and no-code automations. These are things that uh automations that are using uh tools like make.com or Zapier or any 10 to automate things. And you get the easy interface uh that even a non-technical person can uh actually use to just uh connect your tools, connect your databases, your data, your spreadsheets, your different different applications that you use

and uh and just automate things. And you can do, of course, many different things uh using uh these kinds of automations. Here are just uh two examples so because I want I don't want it to to bore you with too many too many examples. But, for example, you can just do inbound lead qualification. So, any inbound leads from email and form can go, they can be qualified,

they can be assigned to specific people. And the value here is that it actually mimics your processes. So, it can actually take into account what is important for you in your uh leads uh um lead pipeline. It would who would like you would like assign to specific uh leads, etc. So, you implement it in your organization in your way. They are flexible this way. Or you can

do NDA verification. We uh at Software House had to sign very uh very big amount of uh uh of uh NDAs uh just before even talking to to most leads. So, the the problem was, well, how to how to automate that because it was a huge stress on the legal team. So, first of all, we had uh two different uh workflows. One for like signing on our

template. It actually right now doesn't require any uh human involvement at all from our side. And the other And another workflow is for when clients want to sign their NDA. So, we have a uh legal checker that actually takes into account like over 50 rules that we ourselves uh created uh to check. And right now, 95% of NDAs are just going through this tool, not through the

legal team. So, it's like helping a lot. And we have the third level. Uh we have custom solutions. So, everything that requires actually for you to write code. And you can do anything you want here. And you can create custom models like like in this example, the middle custom assistant for recommending things. Or you can do mm like some custom uh computer vision uh software that, for

example, analyzes geodetic maps. And each of these uh levels, each of these categories, have their pros and cons, of course, right? So, SaaS solutions are the easiest to implement, instant, and uh they uh are basically low initial investment. But they lack the flexibility uh because you need to adjust to the tool, not the other way around. Uh and uh they can be very pricey when scaling. For

example, the Studio tool, the last time I checked, it costed around $1 per AI conversation, which is a lot when you think about it. Like, for 100 conversations a month, that's okay, 100 $100. But if it's like 10,000 conversation, 100,000 conversation per month, it's like a lot, Especially that doing the same thing using custom can be like 1% of that cost. So, so yeah, definitely expensive at

scaling. Then we have low-code also easy to implement because they they don't require technical knowledge like if if you want to do something simple, you don't need to be like very uh very uh like you don't need to program anything. And so implementation speed is also usually great uh because you just get the integrations to many different tools, to many different like to spreadsheets, to databases. So,

it's out of the box you get it. Uh and the flexibility is also quite okay because you can uh just plug it into your processes, into how you work, and configure it the way you want it to work. But they have limited capabilities. You cannot create uh beautiful interface, or you cannot scale, and they're just not designed to handle thousands of users. And then you have third

level, they're the custom solutions that you basically can do anything you want, anything you have uh the money to to implement. Uh they can be as scalable as you want, as easy for the users uh as you want. Uh but yeah, the problem is the cost and the implementation time. Because it's the biggest. But is it? Is it really right now? Uh the the reality changed a

lot with agentic uh tools for coding. Uh actually at Miquido, the last two projects we delivered were like delivered in like 23% of the budget. So, these are like over four times faster than before. So, it's changing a lot. And the example I gave you here about the geodetic map analysis system uh was actually something that I created in two days. Uh and to give you context,

I'm not actively programming for over 10 years now. So, at least not commercially. So, it is something that I've done for the client just in a maybe six hours of active active work with cloud code and it's actually you can do really easily. And the funny fact is this this application didn't use any AI itself. It was like simple simple maybe not very simple but uh computer

vision algorithms implemented by cloud code. So it used AI was used to create the software. And uh also I wanted to touch on one more thing here. You can actually mix these uh all these three levels and it's really important because uh many agencies will just tell you just go with low code or just go with SAS and uh they just uh point you at the right

tool. But actually you can mix whatever you want here. You can create a beautiful custom interface and that will be level three. You can use uh n8n as a backend so it's level two and it can and it can connect with many different SAS uh solutions and that would be level one. So actually it's how it should be done according to my uh judgment. So let's talk

about the automation process Every business process is actually at its core a series of steps that you need to complete to achieve certain result. And uh for example if you want to create a quarterly report you need to have someone get some data from different systems. Uh they need to create some questions uh to different people across departments, gather their answers, collect it, insert it into that

into a spreadsheet or something and generate a report. And uh the purpose of automation is actually to get rid of these some uh of these manual steps to get uh to help people do some things or um make less errors along the way. And automation 1.0 uh which is uh what I call automation before the era of AI, was already doing pretty great stuff. You could actually

do a lot using tools like Zapier, Make, and n8n. These were available much uh uh earlier than uh LLMs. the problems that we had here were I think there are two main problems. The first is that you just couldn't uh implement all the edge cases, all the uh branches on the different paths of the of the process if there was like a lot, because it was simply

impractical. The workflows would be so huge that you uh really maintain them. Uh and also any decisions, any any task requiring uh analysis or generation of the of the output required human input. So, along the way, there are like a lot of uh a lot of things that human needed to actually when humans needed to step in and do something and then move the automation forward. So,

here comes the AI. Uh and AI really helps a lot because it provides the engine to do the decisions, to move the to push the automation forward. And uh of course like the the main thing that happened is the foundational models and uh large language models, LLMs, that just uh you you can supply them with uh data, with context, and with instructions, and they will just move

the they move according to the instructions and do the do the thing you want. And each of these LLMs blocks, these blocks in purple on the on the right side, can be implemented in uh let's say two ways. The first way is uh when you strictly control the code to the large language model. So, you provide it exactly with instructions, you provide it with the the goal,

you provide it with uh with the data and it just follows the instructions. And it gives you a lot of flexibility, but it's not like it's doing the all the thinking on its own. You provide it with instructions. And the other way is to use the agents. So, let's talk about the agents. And there are many different uh definitions of agents. And actually, it's a funny thing

uh because uh also Gartner did the the research on how many vendors actually uh have are offering agenting AI. And out of the thousands of vendors uh checked, you know how many actually use agentic AI? 130. So, like it's uh yeah. >> [gasps] >> So, we are agent wars right now. So, everything says that they're doing agents agents, but actually, they're not. And uh I'll try to

explain you why. Uh but yeah, agents the the easiest definition I can think of is that it's it's is a subsystem that you give it a goal and it's doing the planning on its own how to execute it. So, you don't need to provide it with all the instructions. You provide it with tools. Uh so, you give it tools, you give it access to specific tools. Uh

you give it access to memory, give it access to other agents. And it will just execute it and give you the result. And uh it is a very flexible approach because it can handle a lot of uncertainty. It can handle a lot of cases that you cannot think of before implementing it, but there are some drawbacks. So, let's talk about the comparison. Here we have a comparison

between workflow. I By workflow, I mean like uh LLM-based uh workflow where you do the automation e and you strictly control the calls to the LLMs. Uh so, the this this approach is more predictable because you give the exact instructions and expect some answer. so we can actually check if you get the answer after after it is returned. uh it is easier to debug because if you

do it step-by-step, you know exactly at which step it failed. And uh it's more cost-effective and actually faster, much faster, because agents actually need to do the thinking process, the thinking loop. Uh that burns tokens and burns uh and takes So, definitely this would be the best uh best approach if uh you are uh expecting results faster or in real-time. But, agents um uh agents are more

flexible, definitely much more flexible uh handling a lot of cases, and they're more adaptional, and you just give them context, you just give them tools, and they'll just try to use that to its best possibilities uh and uh just give you the answer. But, they are expensive, they take time, and they can be unpredictable, because you cannot really the whole point of agents is that you give

it control. So, if uh if you give it control, you cannot really expect it to be predictable. Okay. So, just a quick cheat sheet uh to give you a general idea where I would say for you it is better to use AI workflows versus AI agents. Uh AI workflows are better when you have a fixed and well-defined processes, uh clear outputs, clear inputs, uh you can define

all decision paths, more or less, uh and you you expect predictable outcome. Uh also when cost and latency matter, and uh when stakes are high, so that that you cannot make a mistake and you cannot rely on human uh to double-check that. And AI agents uh AI agents uh I'm not saying they're bad, they're really good. Uh just just just to just to make things clear. Uh

but they are good, especially when you need to face a wide variety of different scenarios. And uh you don't you need to expect unexpected. Like if you want to uh uh have like a programming task, uh it just give programmer can do anything they want using agents. And that's a good example. if you need to cover a lot of edge uh also it would be better probably

to If it's requires a lot of reasoning and judgment, probably agents would work better. And well, depending on the process, of course. Uh but uh reasoning is uh something that agents can can you can delegate it to to Also, a really good example is when you can actually have human in the loop because it changes everything because they can fail, they can take time, but it will

still augment human capabilities. And humans can actually double-check. So, even if they are like 97% correct, the people can just uh find these 3% uh that is missing and just uh just fix that. And that's changes the a lot of things. So, you can actually use off-the-shelf agentic tools like Cloud Co-worker, for example. I'm a huge fan. And do many things in your organizations uh just using

this uh out of the box. Okay. But agents are just the technology. Uh AI is just the technology. And when you automating processes, you need to actually start with the business problem, not with the technology. So, how to properly approach process automation? Uh I wanted to show you my workflow, how I uh usually approach this topic. It's not some like absolute revealed truth that uh everybody needs

to follow. It's something that I've tested over dozens of projects and it worked for me, so why not share it? Uh and how it looks like. It's uh important thing is that implementation is only the third step. People usually start with implementation and that's the the the most common mistake. First, you need to actually start with discovery and by that I need to understand how uh organization

works. Uh I need to understand processes, actually uh do uh some um interviews with employees, uh do uh workshops with people, uh do ask uh valid questions. I'll try to give you some brief idea because I'll share the um framework in a minute. And uh and then you come out you you what you get from this is like uh kind of prioritized list of automations that you

can implement. And uh with that and they are like uh part initially estimated and with that you get to the road map. So, we sit with decision makers and decide when should we implement what uh and create a plan because you need a plan uh to communicate it to your organization, to have everybody on board. And then you can start implementation. So, you can uh specify the

exact requirements, uh do some prototyping if you want. Then you do implementation and of course training because you need to train the the team to use these tools, otherwise they won't use it. And of course support and uh scaling what works. And yeah, I wanted to focus a little bit on the discovery process because it's uh totally omitted by the the companies. And my five-step framework uh

that we wanted to share. So, uh first uh is identification of challenges. And this is really important to to start because like McKinsey said that 60% of all occupations could be automated right now by at least 30%. And does it mean that we really need to pack up and say, "No, I will handle it to AI?" No, it actually means that we need to figure out which

30% to automate. So, this is the the whole point about this this first step. And we can actually ask some valid questions towards our team and towards the uh, customer customer-facing team as well. So, there are internal processes and customer-facing processes. And these are just example examples of questions that you can ask. You can ask which task requires significant effort despite being routine. You can ask where

do delays or downtime occur. Which activities require multiple interaction iterations and corrections. Uh, do employees have difficulty accessing some information? So, it takes a lot of time for them to create some reports or to to answer queries. And which tasks have the greatest impact on financial results and organizational efficiency, when it comes to customer interactions, you can figure out which touch points at which touch points the

response times are the longest. Also, are there any recurring questions from the customers? Does customer experience differ depending on the channel or the depending on the person that that is handling a certain client. So, you can actually level the competencies across the team. And what are the most common complaints for the clients? Oh, and one more about personalization. You can also personalize using AI. So, the result

of this phase is the series of problems. And for them, in the next step, we propose solutions. And each solution should actually focus on a single metric. It can be like increasing the efficiency. It can be decreasing the rates of errors. And for example, for knowledge management problem, people cannot find information fast enough, you can create a knowledge base. So, it's a suggestion of the solution. Or

product product material adaptation, the third one, where people just create new presentations using the old ones and they have to do it from scratch each time, you can just create a tool automation that will just allow them to build stuff based on the historical stuff that is that was available. So, yeah, we end up with a list of um suggested automations. And in the next step, we

assess the impact they have or they might have on the organization. So, we actually try to put a metric to that. So, it can be like it saves 5 hours per person per month or per week. Or it reduces the error rates from 10% to 1%. So, it needs to be something very specific. And there are two reasons for that. The first reason is that it will

help you prioritize these automations. And the second reason is that it will actually give you a metric to evaluate your automation whether it was a success or not after the implementation. Okay. Then if we have the impact, we need to assess the feasibility. And to do that, we need to we need to check the organizational and technical readiness of our organization. And again, we can ask some

questions. We can ask questions on the on the process. So, do process owners have focus and time for the change? Every change requires focus of decision makers. Without that, you cannot do Is the knowledge and context to feed the AI available and up-to-date? A funny fact is that uh if you ask ChatGPT or whatever tool you're using to generate images, to generate a image of a watch

uh pointing like different hour than 10:10, it will fail. It will always show 10:10 because it lacks the context. It lacks the context because all the images on the internet or most of them are showing 10:10 because it's apparently the most pleasing look on the watch face. So, uh so, that's why you need to create uh proper data to feed to your model uh or to the

LLM. Are processes structured enough to be automated? You cannot automate chaos because it will result in more chaos. Uh do we have a plan for using the saved time? And this is really, really important. Uh because imagine you manage to increase the efficiency of a process by 100%. So, you do things twice as fast. And what would you do? Will you fire half your half of your

team? Or will you make them do twice as much? Because if you do nothing, then you will gain nothing. People will just go home early or will just watch YouTube during work because they'll finish early, right? So, uh it's really important to manage that and to manage how to manage the resistance because resistance is always always uh there when it comes to the change. And in AI,

there's like more at stake because there are people's jobs at stake. And there's also technical readiness. So, you need to check whether you have uh also again data, but you have the rights to use them. I had a client that wanted to do educational app and they wanted to use the physical books they had but they didn't have rights to it so unfortunately the project failed like

at the very beginning fortunately. Also do source system have stable APIs because like for example in I know in like construction or in some some industries there are just closed systems that you cannot really integrate with it's their proprietary knowledge and you just cannot do anything to to automate that. Do you have a budget or require licenses for the infrastructure? Sometimes AI can be really really expensive

and both in terms of like infrastructure if you want to build it on your own or if you want to pay for tokens because tokens are also can become expensive with scale. Does the infrastructure meet requirements regarding scale security and monitoring because especially in regulated industries this is a really really huge thing. And do we have support secured for when the things go wrong because if they

can go wrong they will definitely go wrong at some point. And the last step is prioritization and you can create a prioritization matrix that is basically taking into account the difficulty so the feasibility from this technical and organizational readiness business impact so how it would influence certain metrics KPIs how much time it will save or generate more revenue for example. based on that you can assign priority.

It's a very basic basic way to to do that. Basically in if you have a things that take less time or are easier to implement and give more value these are the the first ones to implement. These are quick wins so you create an action plan, a draft of the road map. Uh first, you have the quick wins. So, initiatives that are like no-brainers to implement, something

that are really easy to do and can generate a good results fast. Then, you have a deep process changes. So, things that need some uh some change in the processes. So, they take more time, but they are still worth implementing. And there are business scaling things like uh automations that really change the whole process. That But, these all Yeah, they can take time, but they have potential

to 2x, 3x, or 5x your revenues. So, these are really worth uh taking into account. And this plan actually is worth like to create, to communicate it to the company, to communicate to all the directors, to the all to all the employees, so they know what they what to expect and uh act and make decisions accordingly. Okay, so just a quick reminder, these are the five steps

of the framework. just uh one more thing here. Uh remember, always remember the most important thing is start with the And uh really the last thing, So, automation is not or successful automation is not measured by hard metrics alone, not by KPIs alone. Also, the soft part, uh the cultural signals are really important. How people are adopting AI. Uh you need to look for these cultural signals.

And again, a few questions. Uh Is AI maturity in your organization in in your organization growing? Uh do users trust the tools they're using? Uh do people use them in their daily work because you force them to or because they want to. And do employees actually say that these tools help them work better? Uh because if not, if they don't trust this these tools, they will just

not use them. And if don't if they don't use them, you will again gain nothing. you definitely work on your team as And that's it from me. Thank you very much. If you have any questions, I'll be glad to answer. >> And also you can connect with me on LinkedIn. Cool. Let's see if there are any Oh, I see them here. What level of automation do most

companies you work with actually end up at? And is it usually by choice or by budget? Well, that's a very good question. it really depends on the on the budget. Yeah, that's that's one thing. Also on the organizational readiness because when I approach companies from construction for example, they are really like using still using faxes. Are they still using paper a lot? So, for them it's like

implementing AI is like right away and full automation is like overkill. Uh they just need to out to automate things like using simple digitalization processes. Uh and for some it's like uh they already using uh many tools that you can just integrate with and just add some value. So, it really depends on the organization, on the readiness, and uh on the on the on the budget as

well because sometimes they just want they they just want to spend like uh I don't know like very small amount comparing to the budget to the like whole total revenues. Uh and they expect to have great results, but you just cannot do that. Uh okay. Uh what happens if you if there are two sharks? What should I do? That's a that's a very good very very Uh

actually uh you cannot look them in the eyes usually. Like if you see two, uh but put something at your back. So like I don't know any rock or something and you can just control the environment Um yeah, but generally you are very lucky if you meet two sharks at the same How you managed to perform discovery in 1 week with so many activities? What a Yeah,

well uh okay okay. That's a very wide question. Uh it's usually uh It's not of course it's not possible to do that for like uh a huge organization. Uh but uh you can you always need a some starting point. So I usually do discovery for a single department or like uh or like uh most important uh processes at the company at the start. So that usually takes

a week to to do that to do the workshop multiple sessions. Uh but yeah, you cannot change the whole organization. You cannot expect to change the whole organization overnight. So it's just impossible. But uh it is a really good starting point and actually this whole process is iterational. So it's not like you just do the workshop once and you forget about it. It's like you go back

to that with each new department with uh each new uh thing that they want to test to change their business. What are what objective uh KPIs can we include with uh qualitative ones you mentioned to support comprehensive view of A adaptation and assistance? Yeah. Uh well, KPIs are Well, they they can uh tricky uh in many many cases. Sometimes it's uh because the quality as uh it

can be about the error levels that uh people For example, if there's like uh customer service and people are going back to you with the same questions or they're unsatisfied, well, you can increase that. Uh if uh if uh there are some mistakes like uh in financial, you can uh figure that out. Uh sometimes uh it's just not the easiest way to just assign KPIs like in

a quality way. Uh sometimes you just need to have the uh gut feeling that it'll help you. Uh or you can do some some measurement in terms of like uh I don't know whether your clients would choose you over the competition, but it requires probably external uh research agency to to do the uh for example, a survey. Uh but uh yeah, sometimes it's really tricky. Unless you

can just really point out that it saved like this number of hours uh it uh it helped to uh to increase the efficiency these many times, it's uh it's difficult to put a number to it. >> How can senior or mid developers improve their knowledge if coding with AI leads to fewer good suggestions or less code management? >> Uh Hmm. Fewer good suggestions. I'm I'm not sure

what does it mean fewer good Uh I I I guess that's about uh that AI will just handle the job for the uh for the people, for the developers, but uh you can actually ask uh ask AI to provide the feedback. Uh and uh you can really uh really uh try to try to ask it to review your code. And you can you can do things manually

still just to just to review it just to point it in the right direction. So, I guess that's the that's the thing to to do. And also the the whole software development process changes a lot. Like coding is not the the thing that you need to be the mostly concerned about. Like this uh this kind of uh uh skills is not that important anymore. It's more like

about security, scalability, and all the things and around it, not the coding itself. So, it's yeah, I would turn it like it's not that important to write a clean code anymore. Because like even at some point humans will not read it anymore. It will just do use other other tools, other agents to to review it. That's the direction I see. And uh which my AI would choose

Kielce Kielce [laughter] or Winiary? That's from some Polish Polish guy definitely. Uh I have no idea. >> [laughter and gasps] >> Uh it will choose at random probably as it's probabilistic. Okay. Any questions from the from the people here?