N1HRV — broadcast
20260917 180000 UTC
473 transcript segments
Google Cloud Speech-to-Text API (Chirp) + Gemini 2.5 Flash Non-Thinking
Data courtesy of The GDELT Project (https://www.gdeltproject.org/), from the Internet Archive TV News Archive.
Machine transcription. Treat it as a searchable index of what was broadcast, not a verbatim quotation record.

[00:00:26] Good afternoon and welcome to another AI central point show.
[00:00:30] We continue with a new series of episodes in the second half
[00:00:33] of the year. Today's topic will be AI agents, so when
[00:00:37] software stops waiting for commands, I have my guests with me in the
[00:00:40] studio today, Nenad Raca, Sio from Aduro
[00:00:44] Idea, and Krunoslav Kedmenec, program manager from
[00:00:48] due time. Thank you all once again for your Verso Altima. Hello guys, welcome to the show. First,
[00:00:51] a quick question, until yesterday or even today, we asked artificial
[00:00:55] intelligence something and it gave answers. This year,
[00:00:58] we are starting to give it tasks. Are you ready to
[00:01:02] let software do something on your behalf,
[00:01:06] maybe even without asking you beforehand? Yes,
[00:01:09] we are, it's already safe enough, ready enough.
[00:01:12] Okay, great. We'll come back to that. Kruno, yes, we're doing
[00:01:16] that right now. You're already doing it, that will be very interesting, I think, for many
[00:01:20] listeners who are thinking about it. So, we already have people in Croatia who
[00:01:24] without problems, with full confidence,
[00:01:26] use agents. I actually wrote some articles about agents
[00:01:30] even two years ago and predicted that 2024, 2025
[00:01:34] would be the year of agents. I miscalculated a bit, it seems it's only
[00:01:37] 2026. And today everyone is talking about agents,
[00:01:41] uh, but what are agents really? What is it really, and how would
[00:01:44] Nenad describe the difference between a chatbot, automation, and
[00:01:48] actually a true AI agent? In fact, that is the most important difference, a chatbot
[00:01:52] answers a question, mhm, automation always does the same
[00:01:55] established job and there are no changes, and an agent
[00:01:59] "thinks" in quotation marks, it thinks, makes decisions, knows when it can
[00:02:03] do something and when it needs to hand over that job to a human. So, it has a certain
[00:02:07] level of responsibility and decision-making in all of that, mhm, mhm. Is
[00:02:10] an agent actually a new technology or is it
[00:02:14] a new level of trust we have in an already developed technology? How much
[00:02:18] has that change happened? That's actually the most important part, that new level of trust.
[00:02:21] The technology is not that new, all these parts that we are inserting
[00:02:25] have already existed for several years, but now we have shaped them so that we can
[00:02:28] truly trust them. That is, to establish mechanisms
[00:02:32] of trust and verification of all these agents. In fact, with
[00:02:35] agents we are opening a new dimension where they actually become our new digital
[00:02:39] employees, mhm, mhm, well yes, an agent should be seen
[00:02:43] as another interface between a human and
[00:02:46] a computer, mhm. So you no longer have a keyboard, or if we go to
[00:02:50] a higher level, you no longer - you write a specification
[00:02:54] then someone translates it into computer language, no, you now directly
[00:02:57] communicate with the computer and directly give it... a task. So, that's
[00:03:01] your intern. How did that process actually go
[00:03:05] in Versus Altima, and did that moment happen where you said, "Okay, this is no longer
[00:03:09] just a tool, this is already an agent with a certain level"? Did
[00:03:12] that happen? Was it a process that lasted? Uh, well, in principle,
[00:03:16] last year we started individually using
[00:03:19] AI for something we were dealing with, for some documents, for
[00:03:23] some review of some contracts, similar things, but those were
[00:03:27] islands, it was exclusively communication.
[00:03:30] With this
[00:03:33] LLM, then we analyzed what it needed to
[00:03:36] be, and for us, Aduro, when we
[00:03:39] went to try if we could use
[00:03:42] AI for cyber, they were the first to come
[00:03:46] with an agent, and it was like, "Wow, it works!" And that was the
[00:03:49] moment, that was even before Anthropic released it commercially.
[00:03:53] And the moment Anthropic released a personal
[00:03:57] agent, that's when it became real.
[00:04:09] where agents within the company already did the work from beginning to
[00:04:13] end, so redesign of some corporate pages, but can
[00:04:16] you walk us through what the agent did, what the human did? That's really
[00:04:20] a beautiful example where almost anyone can do it.
[00:04:23] Agents really like
[00:04:26] uh examples, mhm. This is great, you take an agent and say:
[00:04:30] "These are our corporate pages, this is how I'd like them
[00:04:33] to look." Great, then an interaction begins, which is, let's
[00:04:36] say, more or less something like a chat, ideas
[00:04:40] are developed and the agent makes a plan on how
[00:04:43] we will implement it. Okay, that's given to another agent,
[00:04:46] it codes it, then it's transferred to WordPress.
[00:04:50] That was very interesting, because a colleague gave the agent to transfer it to
[00:04:54] WordPress, but it wasn't right. Then he kept reporting, "No, this is not
[00:04:57] good, change this," and then he got frustrated.
[00:05:12] business. Perhaps it's also worth mentioning that we're organizing an artificial intelligence conference
[00:05:16] for entrepreneurs in Rijeka. As an entrepreneur,
[00:05:19] I'd like to know how long that process with agents took, for example, and how long it would have taken without
[00:05:23] agents if only people had worked. What's the time saving?
[00:05:27] Even more than 10 times, I would say that
[00:05:30] days are minutes and weeks are hours.
[00:05:34] Aha, so what you used to need to do for several weeks,
[00:05:37] have meetings, talk to people who would create
[00:05:41] graphics for you, you would make... icons, you would check all
[00:05:44] texts, that literally took a few days, maybe, maybe the biggest
[00:05:48] benefit isn't just time, but the fact that we suddenly have the capacity
[00:05:52] to, instead of just working, to talk a lot and decide what to do,
[00:05:56] we make those five versions, look at which one is the best, and practically choose immediately from
[00:05:59] some choices that were made very quickly and efficiently, which are complete, which are not
[00:06:03] theoretical ideas, meaning they are not prototypes but practically functional, in this
[00:06:06] case, websites. So, I say, I think many UGP members
[00:06:10] will be more interested in hearing this.
[00:06:12] And about that, Nenad, let's go back to Aduro Ideje. So, it has experience and has been working
[00:06:16] with agents and actually with agent operational
[00:06:20] centers for some time. Uh, what is that actually? How do you explain
[00:06:23] it to someone who is not from IT, and again, let's go back to some of your
[00:06:27] examples, 10 times is a huge saving, so practically 90% of time is
[00:06:30] saved. Does that match your experiences? Well, my experience is actually a bit less,
[00:06:34] but because of this part of playing around, where suddenly everyone has the opportunity to play around, or I don't know,
[00:06:38] sometimes when we were developing software, we always talked a lot at the beginning before we started
[00:06:42] doing something. And now we allow ourselves to
[00:06:45] start working on different versions and see which one will be the best for us. So, we have a bit
[00:06:49] more wasted there, but the improvement is still significant in that
[00:06:52] whole story. What you mentioned about what
[00:06:56] an agent system is, that's actually perhaps best explained if
[00:06:59] we consider agents to be digital employees, mhm. The entire operating system for
[00:07:03] them is actually their building within which they work, with their rules,
[00:07:07] like every company has its own additional company for
[00:07:10] digital people, for digital agents, who work for it, in
[00:07:13] which physical people can enter, can supervise it, can see what
[00:07:17] works, how it works, and create new agents in it,
[00:07:21] mhm. That's maybe now such a thing, until now we all looked at who has what
[00:07:25] employees, how many employees they have, how they are organized, now we will
[00:07:28] have a new sector that will look at how many employees there are,
[00:07:32] what they are like, how they are organized, how they work, how efficient they are, how
[00:07:36] accurate they are. Yes, I mean, that part sounds fantastic, although it still seems
[00:07:39] a bit abstract to a very large number of people in Croatia. So we will definitely come
[00:07:43] back to that. But here's a question for you, and possibly for Kruno.
[00:07:46] So, where is an agent used in an average
[00:07:49] Croatian company, or where can it be used to be actually
[00:07:53] most efficient, and where are these not some spectacular applications? Well,
[00:07:56] I mean, there are a lot of these trivial ones that everyone pushes first, like customer
[00:08:00] service, those are something that are very easy to explain to everyone,
[00:08:03] easy to understand how it works, but maybe there's that part that everyone
[00:08:07] forgets, they use it very often, that's the part of some kind of advisor
[00:08:11] in some decisions. "Aha, where we start inserting some
[00:08:14] documents, we are actually creating a new
[00:08:17] advisor with whom we can discuss some topics where
[00:08:21] sometimes people in the company may not have the right interlocutors, so they sometimes looked for them
[00:08:25] outside, now with an agent they can have an interlocutor with whom they can do that,
[00:08:28] analyze documents, analyze decisions, try out some
[00:08:31] scenarios or any such things. A concrete example, you have
[00:08:35] a big project with many subcontractors, main
[00:08:39] contracts, contracts with subcontractors, give..."
[00:08:42] the agent to read all those contracts and say:
[00:08:46] "Here you covered, this you didn't transfer,
[00:08:49] so no, that's not, that's not a lawyer, but it
[00:08:52] immediately gives you insight, especially agents are very good when it comes to
[00:08:56] compiling information from different sources, and of very high quality.
[00:08:59] We actually had an episode on artificial intelligence in
[00:09:03] law some time ago, and people who work on specialized systems for artificial
[00:09:07] intelligence in law also testified to us. So, we can always refer
[00:09:10] to that episode. Some people complain again, so I have to ask you,
[00:09:14] are we at a stage where agents save us time, as you said, or do they somehow
[00:09:18] create new work for us in the sense that someone still has to supervise those
[00:09:22] agents? Uh, I don't think that supervision is the problem
[00:09:26] because it's in principle like, uh, how would I say, when we had
[00:09:29] computers, when they came, we all became
[00:09:33] secretaries, everyone, we all started doing that administrative work, and
[00:09:36] actually, the jobs of secretaries didn't disappear, but we all became our own
[00:09:40] secretaries. What we are becoming now,
[00:09:42] we all are actually becoming managers, because we all now get our digital teams
[00:09:46] that we start managing. So, when you look at it that way, it's not
[00:09:50] taking away jobs, but it's actually defining some new roles for people
[00:09:54] who perhaps weren't ready for them until now. Yes, no, the question is whether it creates new
[00:09:58] work, because now you have to control some agents that you didn't have
[00:10:01] to control before. But my experience is that the scope of
[00:10:05] work around that has not increased, that overall we are gaining, the type of work has changed, because
[00:10:08] we are doing more work than controlling someone or some people who are doing
[00:10:12] the work for us. And people who have uh, managed
[00:10:16] infrastructure until now, first they managed infrastructure at
[00:10:19] their company, now they manage the cloud, those same people will manage
[00:10:23] the agent infrastructure. They will make sure everything works, that not too many
[00:10:27] tokens are spent, that everything functions. But we can imagine,
[00:10:30] building on what Nenad said, so an AI agent as
[00:10:34] an advisor to the board, for example, someone who will help in decision-making, I think
[00:10:38] that's a very interesting part and I agree, often, often neglected, but
[00:10:41] potentially very good. But okay, mm, I would like to
[00:10:44] ask Kruno one more thing, you have a very interesting project,
[00:10:48] so AI for Cyber is a research and development project for
[00:10:51] specific cybersecurity. A topic very dear to me, that's why we had
[00:10:55] one episode specifically dedicated to it. So, it
[00:10:58] is carried out with European Union funding, you managed to get that. What problem did
[00:11:02] you want to solve with agents? The problem is, so one of the things that
[00:11:06] Verso Altima does is cybersecurity,
[00:11:09] mhm. This is not just about IT systems,
[00:11:12] but also control systems for smart cities
[00:11:15] and control systems for industry and similar things. And
[00:11:19] there you have today's software, today's
[00:11:22] methodology is pattern recognition. Mhm, you constantly
[00:11:26] maintain uh a database
[00:11:29] of patterns of some events, it
[00:11:32] recognizes and gives it to an expert for evaluation to see
[00:11:36] if a reaction is needed. Unfortunately, there are a lot
[00:11:39] of false alarms, a lot of things
[00:11:42] where a pattern is recognized, but we know that someone is testing something.
[00:11:46] What we saw, as I said, that "wow"
[00:11:49] moment, we gave the then cobbled-together
[00:11:53] agent just to go, mhm, into a bunch of logs and it started
[00:11:56] spitting out things we hadn't even thought about.
[00:12:00] So we are moving from pattern recognition to anomaly
[00:12:04] recognition, mhm. Is something happening that doesn't usually
[00:12:07] happen? Have I started logging in at two
[00:12:11] in the morning and missed my password 10 times? Mhm, mhm, I have to
[00:12:14] ask right away, actually, it reminds me of some examples from other industries,
[00:12:18] how much time was actually lost until now on sifting through those
[00:12:21] false, you know, false alarms?
[00:12:25] And actually, how much did an agent then accelerate that and
[00:12:28] save time? The approach actually changes here, because the previous
[00:12:31] approach was to look for something, and here
[00:12:35] the agent looks for anomalies in everything that behaves. And the most
[00:12:39] important thing is that maybe people, when they looked at it,
[00:12:42] there was always a problem that we looked at certain different sources. Here, with agents, we can already
[00:12:45] look at the connection between totally different sources that occur at a certain
[00:12:49] time, which may have some cause-and-effect relationships. So,
[00:12:52] when something is hidden or concealed within
[00:12:55] some other events, I assume it's more efficient. Yes, for example, another
[00:12:59] big project we are working on is the use of agents for
[00:13:02] monitoring huge telecom networks. Aha, you have very
[00:13:06] trivial situations where somewhere in Split you have one
[00:13:09] device that has an old version of
[00:13:12] firmware. And how do you find that? So, you look for what Nenad said, you look for
[00:13:16] a pattern, no, you just let the agent go, tell me what's wrong, and
[00:13:19] it will just say in Split there's a device
[00:13:22] that has old firmware. Do you want
[00:13:25] me to upgrade it? Yes, that's what now, will you let
[00:13:29] the agent do it? Today he asks, today he asks you. I think that
[00:13:33] example will interest many companies, like where do we all have to upgrade companies, no.
[00:13:36] Yes, absolutely. Great, Nenad, but what happens, let's
[00:13:40] go back to that topic for a bit, if attackers use agents for some
[00:13:44] cybersecurity problems, does the speed of the attack actually change,
[00:13:48] how important is human defense there, does a human even have time to react or not? It's not even
[00:13:52] a question of maybe how fast it is, but how inventive it is, mhm.
[00:13:56] Right? A new level, that's a new level, because until now, that part of
[00:13:59] invention was special specialized people who had a gift for finding
[00:14:03] such things, and
[00:14:06] now with agents, or swarms of agents, that means we are no longer talking
[00:14:10] about one agent, we are talking about hundreds of them who can
[00:14:14] act simultaneously, who communicate with each other, who can carry out
[00:14:17] very simultaneous and diverse attacks and diverse
[00:14:20] testing of different things. So, uh, we currently in the company know
[00:14:24] how to use uncensored models to
[00:14:27] test the security of our applications. So, instead of doing
[00:14:30] the traditional testing of applications to see if they are secure enough and
[00:14:34] all that, we let an agent who has permission
[00:14:38] to attack the application, to test what it can do on the application, because
[00:14:41] simply, a new pattern will happen here and new patterns are happening, new things that
[00:14:44] they do. Yes, that innovativeness is always a specially interesting
[00:14:48] question. It's interesting to mention in this context that just these days, information came out from
[00:14:52] Open AI that with the latest model, they managed to solve one of
[00:14:56] extremely difficult mathematical problems,
[00:14:58] and in that way, again, to suggest a completely new type of
[00:15:02] innovativeness and knowledge, so let's hope it won't reflect in all these
[00:15:05] cyber security and other problems. I think
[00:15:08] it will, right? My colleague, a physicist who
[00:15:12] uses AI in this whole story, he said just recently that
[00:15:15] he was getting
[00:15:19] such solutions, meaning he uses AI very intensively in his work,
[00:15:22] meaning that he gets solutions where he sometimes
[00:15:26] has to think hard about the approach that AI took, and which in the end very often
[00:15:29] turns out to be very good, but at first it wasn't visible to him at all, mhm, that's the
[00:15:33] thing, how AI helps us find inventive approaches
[00:15:37] to something, anomalies in fact. Let's then stay on that risk for a second.
[00:15:41] So, today still a small number of companies use agents, but I'm sure
[00:15:44] that will increase, especially after they hear you, a larger and larger number, but
[00:15:48] then we ask ourselves, what is that part of the risk, because if we use some agents, we have given them
[00:15:52] access to our data, to some decision-making, how risky does that actually become
[00:15:56] for us, is that something companies should be afraid of?
[00:16:00] I mean, they should be afraid of it like anything else, uncontrolled,
[00:16:03] unorganized access brings problems. If it's approached
[00:16:07] intelligently, organized, so that you know what you're doing and what
[00:16:11] you're managing, what rights are given, because it's not a problem if
[00:16:14] the agent has the right to do what it needs to do, but if it
[00:16:18] doesn't have the right to do what it needs to do, it won't do the job. If it has too many
[00:16:22] rights, it might do too much. So it's the same thing as
[00:16:25] people, you need to look at its experience, gather experience in
[00:16:28] management and then accordingly give rights, manage rights. Fire
[00:16:32] is a good servant, a bad master, the same
[00:16:35] applies to agents. If... "As Nenad says, you know very well
[00:16:39] what you're doing, how you're doing it, what rights you're assigning, there are no problems. If
[00:16:43] you just let it go, you've thrown a match on dry grass and everything
[00:16:47] is gone. Okay, but who's to blame for example if an agent deletes
[00:16:50] the wrong email or sends the wrong offer or so on? The organization
[00:16:54] and those who defined the processes. The software cannot be
[00:16:57] blamed, because the software does what it is allowed to do. So no agent will ever
[00:17:01] do anything outside of what it is allowed to do and what is defined
[00:17:05] as its task, but it will try with all its might to complete
[00:17:09] those incidents that are talked about a lot. It was given the task to
[00:17:13] do it and they watched how it would do it. It surprised them, it surprised them
[00:17:16] how it did it, that's the speed, that means the speed
[00:17:20] at which it can happen, that's the inventiveness and speed at which it can
[00:17:23] be created. Yes, I think that's a special question that we probably
[00:17:26] won't go into today, but for example, when we have many agents on both sides,
[00:17:30] well, maybe it's not a bad context for my next question. So we researched
[00:17:33] human in the loop, hopefully an article will come out soon
[00:17:37] on this topic, when a human is important in the loop, when they are necessary, and when they are actually
[00:17:41] unnecessary or impossible to do, because as a human approving something,
[00:17:44] we can do, look at 10 things or maybe 100 things or a few hundred, but
[00:17:48] if it's about 10 thousand, 100 thousand, a million cases that we have to, the
[00:17:52] human brain simply cannot process it. Uh, what are
[00:17:55] your thoughts on this? Well, you need human in the loop, that is, a human in
[00:17:59] the loop, meaning where the human is, I mean, purely so that we don't have them just for formality, that is,
[00:18:03] that's important, you don't have them for formality, but really where they are needed, so really tasks
[00:18:07] where the crucial thing is that something happens, that
[00:18:10] something is done, but there must always be some mechanism so that
[00:18:14] it is done, because there are many cases when a human will be
[00:18:18] put in the loop just to perhaps avoid some legal
[00:18:21] responsibility, so it's like he will be to blame, he didn't look, but if he clicks 10 times
[00:18:25] he will click 11 and 111 and 1011 then the human
[00:18:28] will become an automaton, he will become automated, we have actually made a robot out of
[00:18:32] a human, mhm. What I say is, last week I
[00:18:35] was talking to colleagues, I pointed out where the bottleneck
[00:18:39] is in using AI. Yes, I
[00:18:42] can no longer keep up with how quickly it can analyze,
[00:18:46] suggest, ask a ton of things from
[00:18:49] different aspects, I physically can no longer
[00:18:53] keep up with it. So not just approving, but also the work itself with
[00:18:56] an agent that is too fast for us. I've heard that from multiple sides, exactly
[00:18:59] that experience. Then the question arises, so on
[00:19:03] one side we have an AI that requires human supervision, and then
[00:19:07] we ask ourselves, is that feasible in practice or is it just some words on paper where I was
[00:19:11] told that a human is responsible on paper but actually
[00:19:14] doesn't have the ability to do it because it's a bottleneck? Yes, but again, for that we need to know how it works,
[00:19:18] so there again we come back to the standard thing, enough testing, where again
[00:19:22] AI agents can help us because we can use AI agents to test AI agents.
[00:19:25] So what we said, if they can do some attacks, they can
[00:19:29] do anything else. An AI agent can test an AI agent and
[00:19:32] find ways, or problems, in behavior, points
[00:19:35] where it is necessary to insert truly justified human
[00:19:39] supervision, and you learn from it as you work with it. So all this that
[00:19:43] I'm saying, I can't keep up with it, but every iteration it understands better and better,
[00:19:46] in the end it no longer needs me. That's it, that's the
[00:19:50] moment when I let it do what I assigned
[00:19:53] it to do, but we solved it until then. My experience is that it takes
[00:19:57] about a week of talking and changes. Yes, I think that's the crucial
[00:20:01] part, that agents actually learn from us, and my experience lately is that I'm
[00:20:05] playing with an AI second brain, so it's a system that's developing, that has multiple
[00:20:09] agents connected, it constantly learns from me, collects data, so I assume that will
[00:20:13] also be one area that I plan to, let's say,
[00:20:16] deal with educationally, meaning helping people create all their
[00:20:20] second brains that will help them, that will learn from them and actually help them be
[00:20:24] 10 times more efficient or even 100 times more efficient. But here we need to take into account,
[00:20:27] these are personal agents, right? A second brain and such things
[00:20:31] serve me personally. If we move to a corporate level for
[00:20:35] company use, then that approach no longer... because then
[00:20:38] uh, we cannot, we must not let someone, you know, "contaminate" the memory, the memory must
[00:20:42] be controlled. So access to memory in the corporate world and in
[00:20:46] the private world is totally different, and that is the one thing that I hope people won't
[00:20:50] make the mistake of trying to use private agents
[00:20:53] and what works well privately in corporate environments. That is good, it means separating my second brain and
[00:20:57] as an agent as my private multiplier of my efficiency, which
[00:21:01] is a fantastic thing, and corporate agents that are made and
[00:21:05] controlled differently and so on, great, that's an important thing for, yes, that's my
[00:21:07] experience, I have to separate private
[00:21:10] things, so if I'm looking for LED lighting for an aquarium, that shouldn't
[00:21:14] be part of the AI for
[00:21:17] Cyber project, those are two different contexts, because at the beginning, until I hadn't mastered
[00:21:21] those skills, then I would get strange answers, then I realized, well, yes,
[00:21:24] I asked it something silly within the same context and now it
[00:21:28] concludes that I need an aquarium for AI. Great.
[00:21:31] Great, let's then give one more, maybe, practical advice to
[00:21:35] viewers. A company comes and says, "I want an agent," we got
[00:21:38] interested, that sounds great. What do they first need to have, how do they start that process,
[00:21:42] that maybe has nothing to do with artificial intelligence? Well, first they need to have a clear idea of what they want to
[00:21:46] do. Okay, the second thing is they need to make a decision that they want to change the way
[00:21:50] they want to work, because implementing AI into
[00:21:53] a company without changing the way the company works is doomed to
[00:21:57] fail, mhm. So those are all projects that fail, they fail because
[00:22:00] the basic work process hasn't changed. The basic work process
[00:22:04] must change, because AI is not an addition to the process, it's
[00:22:07] an evolution of the process itself that starts working immediately after that. Then only
[00:22:10] then do we move on to technology and everything else.
[00:22:12] That's perhaps the most important message. So my background is IT, but
[00:22:16] AI doesn't belong to IT, yes, it's clear, did you manage
[00:22:19] to change processes and mindset within the company? My advice is to get
[00:22:23] a personal AI agent that will guide you through this process that Nenad
[00:22:26] talks about. I know some people who have done that and are very satisfied with it, I must
[00:22:30] say, even at the level of personal development, what insights
[00:22:33] they gained about their functioning and behavior. How much does it cost to get into
[00:22:37] such projects at all? How much does it cost to get into it if you approach it smartly, how much
[00:22:41] does it cost if you approach it the wrong way? Let's start with this one, if you approach it the
[00:22:44] wrong way, there are those penalties that are there for GDPR European Air.
[00:22:48] act, that's how much it costs to enter. in the wrong way, that means that's the
[00:22:51] ultimate price to pay in all that, plus lost time, plus
[00:22:55] lost reputation, plus everything else, to enter in a smart
[00:22:58] way, we can already for a few thousand euros, mhm,
[00:23:02] meaning these are not some abnormally expensive projects, abnormally expensive projects, meaning the important
[00:23:06] thing about it is that these are no longer projects, meaning it's not something that is
[00:23:09] created from scratch, there are ready-made things that exist, that work, that
[00:23:13] can be tested, because uh, I don't know, in our approach we always
[00:23:17] give people access so you can... try it, you can
[00:23:20] get a demo, you can see how that thing works before you decide
[00:23:23] to implement it, we don't have to enter into risky projects, let's enter into
[00:23:27] something small, a small process, but we provide support
[00:23:30] on how to redesign such things, not just, let's
[00:23:34] push AI into the process and then everyone who does something will have their own
[00:23:37] AI assistant who will ask something, because realistically, then they haven't sped up their work, I hope you
[00:23:41] entered that project wisely, uh, of course, first
[00:23:45] start with something you know well, mhm, mhm, so you can
[00:23:49] assess. whether what the agent writes to you, whether
[00:23:52] that I often pretend to be stupid when
[00:23:56] I talk to him, I deliberately won't tell him I want to test how he understands
[00:24:00] and I want to see what all I have to tell him, explain what
[00:24:03] documentation I have to give him so that he gives me a quality answer, so that
[00:24:07] he can determine the job, meaning start with something known,
[00:24:11] that's why I said websites
[00:24:13] issuing invoices review whether there are any unpaid
[00:24:16] invoices in the entire Excel database. you have some
[00:24:20] unpaid invoices, financial analyzes, some automatic reports
[00:24:24] so there's a whole range, super, great, we will continue to explore this,
[00:24:28] but for example, you are definitely, and the companies you work with, and the clients, you are
[00:24:32] some early adopters. We always have those who need a little more, who are skeptical
[00:24:36] about uh what is happening, what is developing, what would you say to a
[00:24:39] director who says let's wait another year for
[00:24:43] it to settle down a bit, then we'll see and maybe introduce an agent.
[00:24:46] Uh, the biggest, biggest problem you can make in business
[00:24:50] is not following what's happening, or waiting
[00:24:53] too long, mhm, and not trying things out. So, nobody says
[00:24:57] companies have to absolutely implement it in their business right now, but
[00:25:01] they have to start using it, but not in a way where we ask the chat and
[00:25:05] the chat answers something, but let's try to test some processes,
[00:25:08] see, we have the opportunity to learn. The technology is
[00:25:12] new, there's still no one who uh can say with this technology you can
[00:25:15] definitely do this, this, this, and you definitely can't do this. We're all
[00:25:19] trying things out here. So it has to be tested, we're all testing, we're all trying, so all
[00:25:22] the time spent delaying is wasted learning time.
[00:25:26] Those who start in half a year have already lost half a year of learning,
[00:25:29] those who start in a year have lost a year of learning, because transforming
[00:25:33] a company, what I mentioned, changing the way a company functions
[00:25:37] is a prerequisite for implementing AI, or
[00:25:40] AI agents, or digital employees, opening new departments, new
[00:25:44] things in the company itself, how it will work digitally, meaning these are no longer
[00:25:48] digital software, these are... digital employees, meaning he gives
[00:25:52] complete work and results, has control and supervision over him,
[00:25:55] super, similar processes as it was with,
[00:25:58] office digitalization, when we switched from paper.
[00:26:02] mm to computers, to printers, to networks, meaning
[00:26:06] to make a transformation that you know, you have to see what the
[00:26:09] possibilities are, as Nad says, learn what you can do with
[00:26:12] agents, then make decisions, hey great, an agent would help us
[00:26:16] greatly here and start there. Oh, great, I can only
[00:26:19] say how lucky we are who deal with education in artificial intelligence.
[00:26:23] Uh, guys, thank you very much, we will continue this conversation in the AI second point which will
[00:26:27] be broadcast on YouTube. I will conclude
[00:26:29] this broadcast about agents, they are an extremely interesting
[00:26:33] topic, mm, we will surely return to it, and we plan to cover
[00:26:37] other interesting topics such as artificial intelligence in marketing or artificial
[00:26:40] intelligence in the defense industry, and artificial
[00:26:43] intelligence, perhaps even in religion or journalism, but
[00:26:47] you will find out about all that in
[00:26:50] attention until the next AI Central Point
[00:26:52] broadcast!
[00:27:20] Every new school year is a new beginning. Let this
[00:27:24] one be a little more carefree, with Telemah Kids Watch with
[00:27:27] audio and video calls and precise GPS location.
[00:27:31] Stay connected. Smart kids' watch and junior
[00:27:35] tariff for less than 10 euros per month, no cheap
[00:27:38] slogans, just a good offer.
[00:27:41] Telemah. The global economy is entering
[00:27:44] a new era. Geopolitics, technology
[00:27:47] innovations create a new map
[00:27:50] of capital. Where are the opportunities for Bosnia and Herzegovina
[00:27:53] and the region? Answers and new perspectives. We are looking for them through
[00:27:57] ideas, investments and partnerships. New
[00:28:01] economic reality. Horizons of Capital, Asa
[00:28:05] Bank in cooperation with Forbs Adria, October 1,
[00:28:08] 2026. Swissotel,
[00:28:11] Sarajevo.
[00:28:14] Do you think it's easy to get into Podravka
[00:28:17] products?
[00:28:21] You have to be much better than the competition.
[00:28:24] Congratulations, you got the role, but
[00:28:26] we admit, the most important thing is origin. Because
[00:28:30] with us, the main roles are only given to originals from
[00:28:34] Croatian fields. Domestic
[00:28:36] in the main role. Podravka.
[00:28:40] Let's go to school smarter. Choose Samsung Galaxy devices
[00:28:44] with the Unlimited tariff for a new school year full of learning
[00:28:47] new possibilities. For savings, use the "pay less then trade in"
[00:28:51] model and 30% discount on
[00:28:54] Google AI pro plan. Let's go to school smarter.
[00:28:58] Croatian
[00:28:58] Telecom.