AI Customer Experiences in Germany - Tyrel Gidinski AICX Founder

We don't play so much in the model game especially in Europe because it's it comes down that literally like where are the servers of this data farmers of this company? How secure is it? Is it Europe based? Is it embedded in some Microsoft environment for the customer so that it's somehow protected via contracts and within contracts and settings within settings? So we observe it and we live it because we of course are utilizers of AI outside of our roles as implementers of AI. Hi, my name is Demetri Bonichi and I'm a content creator, agency owner, and AI enthusiast. You're listening to the AI agents podcast brought to you by Jot Form and featuring our very own CEO and founder, Idkin Tank. This is the show where artificial intelligence meets innovation, productivity, and the tools shaping the future of work. Enjoy the show. Hello and welcome

back to another episode of the AI Agents Podcast. Here I'm with Tyrell Gadinski, a co-founder of AICX. How you doing, Terrell? >> I'm doing great, Demetri. Nice to see you. >> Nice to see you as well. Thanks for being on the show. Um, just to kind of kick things off, give us a little bit of a background on on you uh and how you got into AI in general. >> Yeah, it's probably the going to be the least uh exciting part of >> of the talk today. Uh uh so obviously uh since I'm sitting here in Munich, Bavaria, Germany, and speaking English to you, so I am obviously an expat. Um and uh coming from North America, so Canada. Yeah. And uh it's been uh it's been my honor to be able to have my whole adult life in Germany. Uh my whole work experience

has been in in in around the world serving German companies around the world obviously and uh and in digital positions, strategy positions, right? Um and that's how I was raised, let's say, corporately. uh and and it led to this whole digitalization, e-commerceization of the of of industry uh and and it quickly now of course of the last decade evolved into automation um machine learning some exciting fields and you're always then on the cutting edge but not able to implement uh when you're in large corporations and uh and we started getting faced with AI quite uh quite early on in in the companies I was I was part And we saw with with my co-founders and I actually we saw that there was uh it's not lack of adoption, it was a lack of speed uh to implementation. And so we really got excited about the

the AI game. We were quite schooled in digitalization in general and new technologies right uh internet of things uh moving into different digitalizations and automations. And this was such an exciting time once of course this this bubble opened up uh where where AI was visible for every man, right? And and so it was a huge opportunity and a huge uh passion project of ours then to to not only jump on the AI wagon uh with a lot of wonderful companies and partners and and customers um and competitors, right? Such a beautiful market actually. Um but also to offer an environment that was so new every single month. you're you're new, your products are new, your competitors are new, your customers are looking for new things. Um, and so it's just inherently exciting and and so it's very easy uh to get into the market because

it's so full of excitement for you and your teams. Yeah. Um tell us a little bit as well about how you kind of uh because you're a co-founder you know sometimes different people have uh their own personal journey to I guess getting to run their own company 100% by yourself but how did uh the kind of co-founding situation um >> yeah so so I have a lot of respect and and and let's say brotherhood with with my co-founders David Ritzik and Maurice Bandinger so so we've worked together quite often in the past um in in several different companies actually. Um and our journeys have been aligned for for many years and uh you when you're founding they say you should never found with friends, right? Uh but but if you found and your founders are friends or become friends uh and and have knowledge that

totally complements your own and uh let's say personalities that complement your own in a way that you make up for your lacks with each other. Um then you found a core team that uh that it doesn't matter what industry you're in uh whether it's AI, e-commerce or or starting a grassroots company. Um you can be successful if if you're aligned in that way. And so there's there's quite a a very great sharing of responsibilities and and uh experience between us. So it was uh it was a no-brainer to do this together. Yeah. That the three of us at the beginning. Um and so we very much compliment each other and of course the early the early team that joined up uh we don't of course they're not found us by name but but we're very inclusive with with our our early employees and and so

you could say we have every new employee for the last uh two years was a co-founder in some aspect. Right. So so we're not alone. Um it's not an authoritarian experience. It's very much a sharing and very much sometimes giving up of responsibilities and and uh um taking things on where we decide who is the best one person or two people team for for certain topics. And so we're very fluid that way. Um and it's been a fun journey until now experiencing that together. >> And how many people do you have working at your company now? >> So now we're based around uh we're fluctuating around 10 at the moment. It's a very healthy uh comfortable position to be in. Um and and this of course with a whole arsenal of agency and freelance uh behind you but we have 10 10 cores and uh

uh managing different uh different focal roles from product to customer success consulting and then of course uh product management and technical areas. So >> cool. Um, well, getting into kind of what y'all are doing first and foremost, the hard thing nowadays, I know, is to kind of navigate through everything that's out there option-wise when it comes to AI. What would you say is the primary customer that you are going after and what necessarily you're solving um, problemwise in the world of AI right now? So our ICP uh so so our target customer at this point right so 2025 uh Q4 is very very clear it's the German uh smallmedium enterprise and some some larger enterprise companies uh that is our core target group um out of organic reasons right so these were our first types of customers we do have a couple smaller customers who

we love and cherish and they give us very big challenges uh with very uh small bills right there's a very exciting experience. However, our our top customer is definitely smallmedium enterprises in industry um across different uh automotive industry, consulting uh industry and the insurance industry of course. Yeah. So, so they're uh very much industrydriven because with these types of uh industries you you see two things happening. you either because now we're we're probably in the second phase uh for these kind of companies when it comes to the adoption of of of Genaii or the new AI is where they've tried it themselves. Sometimes they've tried it themselves. They have an established digitalization uh core or maturity digital maturity that they have been able to to spend the last year sort of spending hundreds of thousands of euros investing in their their own abilities and you

see very quickly that that they are not software companies, right? and and AI is another software. Um it has to be driven as such. It is a new technology. Um and of course you can't take a company's IT department and and give it an AI head uh and and de facto create some AI solution that will work uh and be competitive in the market especially with all the the players like us coming up and and offering something at far lower prices uh far better returns and uh and just something that that that fits the bill a lot better. So, so our customers are often in the second wave and we can come in quite easily there and sort of save the story. It doesn't mean you have to get rid of everything you've learned. Um, so we offer sort of a a segue um to

a successful adoption and re reinvigoration or lighting the spark for the AI wave. There's then there's customers of course who have been hesitant until now just because it's not a part of their business model. Their digitalization consists of an IT administrator and and some softwares that they purchase and know nothing about other than just having a process owner maybe. Um and these are mainly the small mediumsiz enterprises. And there you really have to light the spark there. have to really battle against what we call mushrooming where every employee sort of has these one-to-one relationships with AI and you need to create this one to end relationship with AI and the whole mass of their employees to get uh the biggest part of biggest chance of adoption possible to bring them the best value and so so these are really the two entry stages you get

and very top down driven um um entries into these kind of companies. >> Interesting. And you know, it's it's kind of I guess hard at this point to determine what your needs are as a company without maybe having a discussion, you know, with somebody who who might know better because a lot of a lot of companies I feel like are in a position where it is. So I feel like people are still living in the chat GBT asking it questions like it's Google with a response to it. um with the level of agentic improvements over the last six, seven, eight, nine months since we started the podcast, what have you been doing as a company to kind of keep up uh keep up with that and and what you're offering as well? >> So, we were um we were very lucky actually that we were

a part of the agent game and agentic game when it's uh of course when RAG was still the biggest uh acronym in the world at that point, right? So this was this lived probably about three to four months long in the in the world. >> Um and we were already saying this this standard, right? This is augmented generation, augmented search. This is nothing new. Um uh um it's just putting an AI on top of it instead of a human or or a little search window. Um and and so we were early to to push the agent at that time. was the agent being being a a insulation layer between a human and software and sort of eliminating these mass amounts of frontends you have to deal with as a as an employee and and we were very lucky that that that is where the that's

where the technology went. Of course, it probably a year from now or two two years from now we will all have to have have adopted by then uh to some sort of new uh era, but it still is very much the now. Um and and we were very early on able to that's why our proprietary system doesn't rely on MCP or or things that were sort of uh created in the last uh last year um because we built our own routine uh mechanisms embedded in our in our proprietary AICX harp engine in order to embed sort of this this workflow technology uh connected with systems and connected with different tools and external sources and and and the AI agent being the speaking layer. of this um as an colleague or an assistant balancing between these two roles um and and early on and so so

we were encapsulated different from this whole MCP way because we had already created that for enterprises uh in a secure way right it was very important for us being in Europe to to adhere to to these security reforms um and we couldn't we could rely on these past MCP movers or or even workflow builders that that are very common for for single entrepreneurs reneurs in the in the market. Uh this was not ready for enterprise level integrations. So so we were lucky we started very early. Um and we called it out with some of our customers and and uh it made us look very uh good and we were very then gained a lot of trust because 3 to four months later uh they all came back from Christmas break and and and this was the thing happening and they didn't have to catch up

because they had already invested in a tool with us and and it brought rewards. Well, >> yeah. You know, um, MCP has been something that even I took way too long to kind of like jump on the train of being in the position I'm in. Could you kind of describe some of the ways that it changes the game for those uh, business owners that are a little bit confused as to what it is? they think it's like some, you know, you know, some sometimes I feel like when new terms come out and new types of models come out and new protocols come out, it's the same thing all over again, right? It goes uh in their mind of like I don't even know where to start with this. Like the new like for example, two days ago or not two days ago, two weeks ago

with Anthropic and then today with Claude, you had agents agent SDKs are released. It's going to be the same thing. They're like, "What is it? I don't know what to do." And even I'm in that stage right now. Yeah, >> for MCPs >> don't Yeah, I would tell anybody um when something like this the first danger is when something comes out as an acronym before we even know what the technology is behind it. Right. This is >> Thank you. >> It's a barrier. Yeah, it creates a barrier, but it's it's a it's a healthy barrier for the people who are the first proprietors of it. Uh the ones that understand the technology already. Um it kind of gives them a power, right? So so rag very very simple topic gave us power because you knew what rag was. I mean what is rag? Back in

the day it was SWAT. what is SWAT when you're in the marketing world, right? So, it putting an acronym in something makes it somehow more valuable. Um, it's a fallacy. Uh, MCP is just a short term of somebody choosing three words to describe something and then using the first letter to describe it, right? Sounds very cool. Um, sure does. But it's nothing other than and and the funny thing is when a new thing comes out, I I I adamantly suggest anybody act like a one of my children, maybe a six-year-old, and and go to your whatever agent or whatever model you trust and ask it to explain this to you on these terms or visually, even better now, >> a baby. Yeah. >> Explain it visually like a baby. And I when MCP uh just before we even looked into what it was back in

the day, um obviously once we heard it, we said, "Ah, we know what this is. We're already doing this." Uh to a certain extent, it's just a wrap-up, right? um uh to to call something. You want to build different logics or incurred logics that deterministically say how uh an AI agent or how a a flow of knowledge and and response should go and and so inherently you give it then this MCP servers gives it this option to say I have one interaction layer and I want it to interact with many different possibilities or tools or whatever. So can you could just imagine it like back in the day uh when there were no cell phones and no um you know when the phone calls were like you call the operator up and see she switches you through to whoever you need to get through whether

it's a utilities company or your uncle down in Florida. Right. So this uh this is pretty much all it does. Um and you can recreate this in many different ways using software uh and using your own proprietary knowledge if you have these abilities. But many people don't and you don't have to have that nowadays because there are MCP servers and MCP routines and rappers that that give you the interface to do this without having any knowledge about the technology. So it is a danger it is a danger for security because MCP is a very can be a very dangerous technology um inherently in enterprise scene for individuals probably not so much um but uh but it's very very difficult to to protect um when MCP when it comes to inherently incessantly dangerous elements in already in the protocol or of course um routines that are

being run without human in the loop right so it creates quite a big uh gap and That's why we we don't play with it yet in that instance because we can recreate it in a safe secure environment. So >> yeah, that's um I think you you touched on a lot of things here. I chuckled a little bit cuz big fan of uh knowing the list of logical fallacies. I I I would like to find maybe it's not a form of logical fallacy, but I I I want to make one up for the the acronym fallacy. Like just because it's got an acronym doesn't mean it mean it's it's like cool. it actually is uh better than uh what you know people do honestly perceive things better via acronym and it's kind of annoying and I also laughed because have you ever seen the show Silicon

Valley? >> Oh yeah. >> SWAT when you said SWAT my brain I was like [laughter] >> I remember the days where we get paid to anal analyze a SWAT oh god that that show is is uh Jared just like I think we should swat that. Um, such a funny it's such a funny show. >> We actually uh we actually acronized our our engine, right? So, we have many different products uh faces. The engine is the heart and we call it heart. It's an actual acronym. Yeah. But it's five letters long so nobody remembers the words that that build up the acronym. You need to have it shorter for the human brain nowadays. Um, but once we mention that and show the acronym, uh, people get all excited and that that's a big 50% selling point right there just because the the engine heart is actually

an acronym. So, [laughter] so, uh, a very useful tool in marketing. I I don't don't, uh, don't toss it out the window, but when it comes to understanding something, don't be disillusioned. >> Yeah, that's a very, that's a good very good piece of advice. Um, and I think, you know, when it comes to MCPs overall, um, could you speak a little bit more to the security aspect there because I I have we haven't quite touched on that. um and maybe maybe how you guys can help n help people navigate through those waters as well cuz obviously I want to keep it about you but I hadn't quite heard that concern that much but I could I could see it. Um >> um I mean we don't want to create any kind of fear-mongering. We definitely want to >> No, that's okay. Yeah, but I mean

hey you're in the EU. I mean with >> the so we're insuring against a potential risk, right? So not this is different than an imminent risk. >> Um in especially in the tech world nowadays it's all about what could happen. >> Um if somebody intentfully has malicious intent with much planning and much skill um these things happen every day and we protect against them with menial uh technologies. uh so it's not uh but in the enterprise setting we have to secure against this and so there are inherently three main dangers and of course there are ways to block this and there there are MCP guards and MCP keys and MCP security now um obviously it's an industry in itself security so if there is a place where hackers or malicious content or intent uh can play a role there is somebody that's going to help

you stop that right so luckily there are this the white knights um in technology and also in the So don't be that afraid. It's a wonderful technology. Um but for enterprise uses um there's inherently with MCP there can be in the protocols there can be also in already in the inception of this connector could be some malicious intent that that there could be latent and and later activated but this in any technology the case um when you activate it um there's also a big issue with MCP is uh a loss of control and validation right so so when you when you start a routine you have literally the no control over the outcome. Um it it's not different uh with many models and nowadays and many routines but uh that's why we chose a different route or let's say an alternate route is because the

human in the loop is becomes too robust. It's not MCP anymore. it's not leveraging the technology and so so this means that the outcome can actually be malicious without intent right so you're creating then a negative outcome just because you gave the wrong type of command and interpreted it wrong and and you have no control of that point because you've already given it and MCP also opens the door that there there could be routines run without you actually uh stopping them yeah so so this is uh so action without intention and and so these are three main uh issues easily solvable All right. Again, easily solvable, but anybody nowadays can put a rapper on technology, whether it's an LLM, whether it's uh an MCP server, whether it's NAND, you can put a rapper on it, sell it to companies, and of course, you're creating this

negative experience. You're creating an insecure experience. So, we're very careful about that um that we don't promote it because uh that would create uh fear about the market where it doesn't need to be there, right? >> Yeah. Very cool. Well, on a more um output and like practical note, I want to talk about some of the uh kind of customer success stories that I've seen on your website. For example, I see one that um was a generative AI powered market analysis that saved up to 14,000 uh manh hours manually, which is crazy. Um the results here, wild. Uh, and I just kind of wanted to talk about the the practical methodology that you kind of take your customers through in order to kind of achieve something like that, right? Obviously, the outcome is there. Um, the challenges are there, but how how was that? Maybe

it you don't have to be specific on obviously who it is, etc. But, um, what does that look like? >> Yeah. So, so all of our customers are protected by NDAs. Many of the things we bring out are under their brand. Um that's one of the uh marketing problems we have as a company as most of the stuff is marketed not as our own. Uh but nevertheless so about these cuts. So so this specific case uh is is actually old and I love that it's old right. So so old in sense of like a few months right? So [laughter] which means >> that's not that old man. That is a that is a that is a point in time where where this KPI was created and and of course this KPI has grown and that's a very important fact because um there's different metrics you

can look at for success and this is this is a point is very important to our industry regardless of being in Europe or being around the world um for enterprises right so so is how do I measure the success of what I'm doing with AI um and we can put out a lot of wonderful uh very exciting numbers like that and that is exciting when you think about it. How many man-hour that is, how many days that would be for somebody to to actually accumulate that you add a euro value or a dollar value on that uh with a salary knowledge, then you know how much you've saved, right? Or how much time a person has actually spent to focus on something that's more engaging, more attuned to the human mind, uh uh creative, collaborative, interacting with another human rather than doing some uh toil

work, right? content creation or email answering or things like this that we call toil. Um where it's just everyday like like like uh sewing the seeds and then plowing the fields kind of of of corporate world. Um you've eliminated that and you've you've opened up this this vacuum for them to actually have a bigger impact. Um >> that's not measurable only in that term that you just mentioned. Um so that's that's employee time gained. That's a KPI we we deliver. Um, and why this is important is not the the amount of it because it's the size of the company compared to that which is important which tells us then is it is it are we talking about 45,000 people in the company or are we talking about 800 people in the company? That number for this company in fact is quite successful number because this

company had 925 people in it. I remember the exact case right this one of the cases I worked on um with a colleague and it's uh it this this brings really warmth to our heart because this is our focus as a company is we figured out that uh and every company should adopt this this is free to to any competitor right the key is in adoption right now we use it as a USP I hope it's not a unique SP in the future but um it's purely adoption and transparency and so you me mentioned that Humba and that being having exciting emotion to it for boards that are deciding on budgets for employees that are deciding to use something um that is an engaging factor that helps the success and the use and proliferation of of of AI within a company and that's important because

without the use of it it is useless right it brings no value um it can be as even if it's automated it can be as uh amazing a solution as you want to deliver if nobody actually adopts it and interacts with it. It's just another SAP. It's just another HubSpot that can sit there and not be used and still sit there and cost you money. And so our goal is to gain adoption, right, to a certain percentage um and a certain retention, which means adoption over time, right? Is very important factor. And then you attach metrics onto that like cost or like interactions or like increased human interaction because of your tool and decreased system to human interaction which is redundant time spent with you at a laptop where you should be spending it socializing especially nowadays. It's very important in the digital age. Yeah.

>> Huh. Yeah. I mean it's it's kind of >> Sorry, I'm getting an echo now when I talk. >> Yeah, you got a minute. That's okay. I I keep adjusting the volume. It's all good. >> Okay. Um, you know, it it is very um interesting how the market continues to shift. The needs for companies continue to shift and also their their level of knowledge uh of what they can do uh continues to shift. And I guess it might be a fair question because obviously you're in the market of software, you're in the market of consulting as well. Um, what are your thoughts on how the future may look in regards to companies internally having like AI teams? >> Um, >> and I mean AI teams in the sense that they're, you know, prompt engineers, uh, had people heading up departments and stuff like that versus

obviously what you guys are able to do with with helping create and build things, uh, for them from a consulting standpoint. >> So, so you mean you mean AI being the department or AI in >> Oh, no, actually. Yeah. So it is a double question to be fair. I was going to do a follow-up afterwards. I was going to say first people at the company who have AI knowledge and kind of like execute and help make the company more effective and then a byproduct that of that being AI departments literally where the AI would be the department. >> Yeah, we've actually done a study on this and followed up a study done by Gardner um when it comes to the the humans um adopting the knowledge to implement AI themselves. And so we that's a part of our consultant work is to ensure that I

mean of course our tools hopefully still bring value and and will grow and have use in the future but like you need to make sure human AI literacy is increasing. Um we have a bit of a dichotomy there. We have a bit of a uh two games playing with each other there because we our intent and our actual products are created in such that the human does not have to inherently have a training um or or knowledge and go-to training in order to interact with it. It has to be as humanlike as possible and as as human in the loop as possible and as conversational as possible. It might be less of an automated routine, but it's a more human and comfortable and realistic routine where you're dealing with the AI like a colleague. So it gives the the need for the human to have

uh prompt engineering or or the employee to have prompt engineering or or some experience in what is MCP, what am I actually doing here, what what are the boundaries of what I can do or what can I do? Um we always have this funny uh vision with the customers and the employees when we interact with them where it says you're coming to to an AI and if you don't know what AI can do today, yesterday is old, what it can do today because it's always new. uh if you're not in the loop and the AI is not continually training you or somebody is training you um then you're sitting practically in front of a meeting like like you and I here and you're just sitting there staring at each other not saying anything because you don't know what each other can do like what is

your role what is your job what is your capabilities can you even speak my language uh if you imagine on human factors it's like us sitting in a meeting and just staring at each other going I don't know where to start and [snorts] and that is a problem it's it's a it's a there's a shame barrier there there's a knowledge barrier there. Um fortunately AI interaction happens most oftenly privately one-on-one human and AI. So there should be no shame barrier but there still is psychologically right. Um but uh so so it is important to understand what's happening. Unfortunately, you have social media, you have many sources of information which are giving this impression that everybody's already rockets away from you um in implementation and or or I have so I have I have the key to using AI and you can quit your job or

do a one day work week or something like this and and it starts to create this sort of barrier where people that haven't even entered it yet and are just using a company GPT >> and they're chatting in and figuring out what is temperature and what is a a prompt structure and uh and one-word prompts. things like this. And so so you get uh you get this barrier. And so what we do is try to break this barrier, right? And and the human needs to train themselves as always and and refresh. But we try to make also there's there's a whole a group of employees out there in in the market that are maybe maybe a generation older than us or even two. Um where this is like back in the day when we adopted PowerPoint and and emails and and social media for marketing

and things like this which are now the norm. um this is for them amazingly difficult and they have 10 years left in their in their work life and and now they're confronted with being obsolete. Um, and so we try to socially we see it as our our responsibility to be able to bring these people with in a way that is the same as if you just hired a new colleague that joined and said, "Hey, look, I can do a job you never knew anybody could do. Now talk to me about it and I'll help you through the through this and and and guide you and then eventually you become proficient." Right? So >> good question. Yeah. So [laughter] which led to AI a which is the exciting one for me when when you want to talk about AI agents becoming their own department or or

let's say having a role in a department right um and it's not the future because we're already there. >> Yeah. Yeah. So would you say that it's basically we're already there. just a matter of >> well because I think it's al adoption is kind of is difficult um more than capability that's um something uh constant trend I feel like I'm seeing on the interviews I have is yeah well tools are there right the capabilities are there I mean sonnet 4.5 releases the other day >> y >> and I don't know what your experience has been with that model but that feels pretty groundbreaking uh but I saw a really funny was on Axe the other day. It was like a graphic that said OpenAI comes out with best groundbreaking model. Anthropic comes out with best groundbreaking model. Uh Grock comes out with best groundbreaking model.

Gemini comes out with best groundbreaking model. Repeat. You know, >> it's kind of the cycle we're in. >> So, yeah, it's exciting. I mean, I of course every time a new model comes out, you spend a certain amount of time and play with it or somebody in the team does and and we share with each other and do little video snippets and and demos to each other. It's exciting. It's it's a fun thing. It's like reading a fantasy genre or something. You figure out new things. Um, and a lot of hobbies are now based in AI usage, whether it's writing a book or developing a game or or your or an app for something in your private life, right? Obviously there there's so much out there and so many rappers around all these awesome models. Um um but but the models themselves we don't play

so much in the model game especially in Europe uh because it's it comes down to literally like uh where are the servers of this data farmers of this company? Um how secure is it? Is it Europe based? Is it uh is it embedded in some Microsoft environment for the customer so that it's somehow protected via contracts uh within contracts and settings within settings. Um and so so we observe it and we live it because we of course are utilizers of AI outside of our roles as the implementers of AI and we also need to be on the cutting edge if there's something cutting edge comes the development of it and the let's say the bandwidth of it especially also with like the the putting it up in your browsers and and things anthropic doing which is really really great and dangerous but great it's always

dangerous and great um that this uh it just becomes an example of how wonderful the industry is and it's also only an anecdote to say look that's great but but this too will pass right so in two weeks 3 weeks something else will happen that's groundbreaking um and the biggest goal for for people in the industry of taking the technology and giving to somebody to use whether it's an individual or whether it's a company like us um and in the employee in the company. It's all about just finding those nuggets of of you don't have to develop this stuff yourself so much. All right? So, it's probably 50% yourself, 50% market. You just have to find that nugget that serves your customers best. And of course, this this new um cloud model is wonderful. It doesn't serve our customers at all. Right? So, so when

it comes to our ASX heart and engine and and and the already routines and integrations we're offering and seamless integration into their existing communication streams there there is no improvement. Um in fact for example when GPT came out with five um we we didn't launch it because yeah exactly it was out of control right so so it was wonderful but token usage was off the charts right it was abysmal yeah the cost I mean if we for where we sell tokens it's a great market but we have to be true to our customers and say look this is going to explode your cost by you know by 20% or 30% is ridiculous. Yeah. Um and so and also the responses were not uh as good as they are if you're just doing it directly in you know chatp interface. Uh sure. >> Yeah. >> Yeah.

But but still um it's wonderful for its purposes but you have to again be a prompt engineer and you have to again understand how to iterate and and to communicate with it. It's less intuitive in that although it's trying to be more intuitive by sort of channeling but but it it it failed in that although it's very powerful wonderful. So we stays very often with for Omni uh with with many of our customers because it it offered a solid and what they needed result right as as one of the base models and so and it was very in in in in the Microsoft environment which is very trusted. It was a great partnership there. So >> yeah no it's a that's a fair point. I think being true to your customers is good. Honestly, kind of is depressing um about that launch was it was

so ex like everyone was unnecessarily hyped about it even before it was like properly rumored, you know, like the early stage rumors of like, oh, this is what GBT4 is. What's GBT5 going to be like? And then I remember utilizing it when it came out and just clearly seeing I typed a question trying to get a response and when it was at that point of it's deciding between like thinking not thinking um and the pro version like it's auto step the non-thinking version the like base version was just spitting out hot garbage. It was amazing. I'd never seen anything so bad out of a model in in a while because it when I pressed enter, the concern for me was like, "Wow, I saw a response the second I pressed enter. That's not a good sign." >> I'm like, "I know we're not there. I

know we're not there, Tech Wise yet." I was like, "I know we're not in like nancond response time." So, something it just spat out something way too quickly, which is always the concern when a new model comes out where they have to do minor tinkering. I love I love how we built so much distrust to fast answers. Although that was what we were searching a year ago, right? In instantaneous responses and it better be perfect. Um but now it's like if it comes too fast you're like this is not perfectly put it like something's not right here. >> Yeah. Um well even for even for LLM or for some model it's just not right. Um we see sometimes I mean we're we're working it's not we don't call it latency anymore. the response times are between you know you you get seconds and then you

get some things where which are a minute maybe max a minute right so there's not such complex routines that are needed with all the workflows we have um much of it's buffered within the software so so one minute maybe the longest and that's a long time if you're if you're working and if you're trying to have a conversation within a minute is a long time but the responses are amazing right so the responses are structured the responses take exponentially more time to integrate to the human factor, right? So, I have to read something or I have to portray it. I have to understand something that's been given to me. And this factor is is for us to understand what's been given to us after one minute or 30 minutes. Sometimes with huge uh complex routines which aren't in our software needed. Um there's uh for

us to internalize that and use it as humans, it takes far far longer. So, so we don't we shouldn't worry so much about response time. uh sometimes you don't give a colleague something and if they come back in a half hour you go what great no you're talking days so we're already exponentially better um so we have to just adjust our perception of what is what is the norm and what's right >> that's a very good comment yeah I agree with you completely there's a lot of issues right now I feel like with um expectations versus reality um and almost I feel like unnecessarily high expectations at times for uh things and but then like when it reaches a certain level we get a little Uh I have a question specifically about agents and its impact on the workforce because I think it's an interesting

answer I'm getting from everybody and obviously since you guys are in the work the world of agents um then it' be reasonable to ask what is your thoughts on kind of the expectations so to speak of people who are working as business owners or working as managers and they're saying, "Hey, I don't really think we're in a place where this 90% accurate agent can do this role." My thought initially is when I hear that is a laughing moment, just kind of a tongue-in-cheek comment. Find me an associate who does stuff 90% of the time correctly and I will pay them to become a manager because that don't happen. So my well kind of tongue and cheek obviously there's the person who ends up getting promoted and that's the point but what are your thoughts on on that kind of dynamic of there's this expectations of

like you said AI their actions should kind of be perfect at this moment [clears throat] we're um so I'm going to contradict you to that which is it's um it is inher it's inherently a software right so it is artificial or we call them uh like the the synthetic employees, right? So yes, if you compare it to a human, the human is going to be exponentially less accurate, uh less consistent, we get sick, we we are we are constantly changing, our opinions change, our knowledge changes, our capacity as people changes as we get older, right? So health span, we have many variables that affect us. I mean technology does as well, right? Um that would be would be a cool diagram to to compare, right? So looking at health span of a human and health span of a of a of an AI agent. Uh

but maybe we'll do that. We should have the bar very high. Uh we need to continue to have the bar high >> if um but it should not um affect our customers and the employees that are customers ability to adopt. So if we are like we said this if this business owner let's say it's for us relevance or maybe it's a CEO of a company he has 900 employees or a thousand employees and he doesn't believe in something because he tried it out and anecdotally says look the accuracy I've had it measured my guys have looked at it it's not accurate the answers are not great um I think Joe and John and Sally could do it better. Yeah. um not true obviously right to the certain extent the things that this AI does is is it can do it better than a human can

do it to that extent um but what the what the beautiful uh thing we have to understand is and this is very important for us when we when we talk about adoption is there is always going to be an human in the equation so don't compare them together like apples to apples right the AI is a tool to get an outcome right the AI is a tool do it whether it's automation whether it's human loop suba automation whether it's a human thought processing right so so working with the AI as a colleague and assistant or whether it's doing toil right uh it is a tool to get an outcome now like any software if you don't set it up right if you don't use it right um it will have issues right so if you enter we have this commonly speaking I'm not even sure

if it's in North America We use it a lot in Europe. Uh crap in crap out. I'll be very uh formal in how I say it. There's a very common IT term, right? So the data you put in, if it's not good, the bad data will come out, bad results. Um with humans is the same with technology is the same. Um we have to hold a high bar on on what we expect can be possible, but we have to always understand that we are a factor in that success as humans. We are not removed from the success of the AI. Right. So, so we talked about humanization of AI. So, we we are part of the conversation. We are part of the input. We are part of the setup of it and and the vetting and and acceptance of that tool. And so, if it's

not working correctly, it's a software is what what makes it better than not better, what makes it more efficient to apply to that case than than an employee is you can improve it quite easily. So you can have some uh technically medical activity you do to that AI and testing and and and improvements which we constantly do um to improve that specific case for that use. Uh and and uh it's always going to be needed and it's always a part of it. Just like we have a very whole industry about training employees, we have a whole industry in ourselves or a whole focus area that's about improving the adoption, improving the output of of whatever agents we per we we give to the customer and this will have to continue. And um they're not all knowing, they're not all capable, they're as capable as we

allow them to be and that's a very safe factor uh we still have nowadays luckily. Where do you think um the movement will be in regards to capabilities? Obviously MCP's been been a big breakthrough in excuse [clears throat] me relative capabilities and connection to tools because I I remember thinking in the last year basically we're kind of being limited not so much by compute or like reasoning at some point like when 03 came out I was pretty convinced that from a reasoning standpoint we were getting close to the average you know employee to some respect. we just needed connectability to different aspects of tools um because APIs were kind of limited in in some ways right to connect to it and then MCPS came out and they were able to kind of interact free flowingly with the different endpoints at once so to speak. So,

where do you feel like the capabilities will continue to expand and improve? And I because I guess I believe it kind of still stands. We're kind of reaching a level of, you know, 4.5 sonnet thinking, for example, kind of crushes what 03 was doing when I would have probably been comfortable with its level of reasoning already. Um, where do you think we kind of max out on agentic ability to interact with day-to-day work and how it'll impact the job market, so to speak. >> There you said it right there. Interact, right? That's the key word. Ah uh >> so there is uh on the horizon there's there's only in in in my opinion there's there's only one outcome that is like the the 100% achievement rate um is is seamless interaction. So we use the word seamless interaction. We name it in an enterprise context.

Right? So so wherever you have input whether it's text, voice, video, recorded, pre-recorded in the background, wherever you have interaction naturally as a human to other humans and and and and to the regular tools, AI has to be integrated there, right? It has to be functional. It has to have context. It has to be cross cross interaction layer um knowledgeable. It means the long-term memory, short-term memory, um, biases that you create through the interactions have to remain whether you're texting it in a chat window, which we really don't approve of versus in your Google chat or your teams or your Slack or your wherever you would also communicate with humans, it should be there with you and >> interact so and seamlessly email whatever meetings, join meetings, things like this. So, so all of these interaction layers once they are covered um then we get

into a point where it becomes apparent the only boundary is um the the two dimensional screen. So real world cannot be in a physical meeting with you. We already see some some hardware coming out. There's some big fails there in the market as well. Um where it should be with you at all times. You're in your car, you're at the you're in the office, you're in the cafe, you're wherever you work, you're in the park, it could it can literally be anywhere. you're in space, you know, flying to Mars. I don't know what that what where all the the environments will be in the future, but but if you want a seamless, let's say the the most the pinnacle of of interaction, it has to be everywhere with you consistently. Um, it's a scary thing for some people, but if you want the best possible

outcome of of performance, that's where it is. It goes so far. I mean, when we're not talking about the neural link or any kind of connections like that, obviously >> the guy just controlled his arm today. >> Exactly. >> Beautiful stuff for those who need it, right? So, please please people maintain your autonomy. >> Yeah. Right. Yeah. >> Those who do not would not need to benefit from that to have a normal life. >> Uh but um but but that's interaction is is the main key, right? Right. So, so we can we can develop the softwares, the routines, uh the things that are hidden um as much as possible. What we are going to continually get away from this is not a new message. We we preached I think that we did the the first time we did a white paper and then a presentation

to a customer on this was November last year was where as we said the the interaction between a human and a software interface, which I love software interfaces. They're beautiful. They give us thousands of options. We're using one right now to talk to each other. Um, and to be humans needed because we need some sort of visual interaction sometimes when we're when we're doing dealing with information, especially structured information. Um, and and it's this interaction with softwares, right? So enterprise softwares, email softwares, um things that generate content, things that generates um an outcome thing, CRM systems, things where where structured data lives, the interaction with a human and many of these different layers will not exist anymore. It won't be required. Um the interaction will be will be as as we are interacting right now and exchanging information and you are enriching your mind and

I'm enriching my mind through this interaction. The same thing will happen in the interaction with a system with me the AI and the system and it will enrich the information. It will give me back the information where we're working on right now which is for us one of the next levels uh and it's very fun to get into is when the AI becomes um I'll be careful with this word to a certain extent self-aware. >> Sure. >> So technologically self-aware. So very very rudimentally we're not talking AGI here. We're not talking consciousness. we're talking can retrieve information that has been shared with it in the past and can retrieve this from many different modules right um and from other people interacting with the same agent right at the same time sometimes because this can happen simultaneously that the agent is aware of what's happening within

it it's aware of it own KPIs how successful am I how much am I helping you when we talked about saving time saving and and uh time spent and adoption the agent will know its own KPIs we don't need dashboard to display this information for a human then to understand it then convey it to their manager. Uh these steps will no longer be poss need needed. They'll be possible but not needed. They might be fun and anecdotal and uh have a place some in our culture but efficiently they won't be needed anymore. >> Yeah, that's a fair that's a fair point. My last question would be and this is kind of more of a fun one. Uh what has been I'll say mine and then you can talk about what your favorite one has been maybe personally and then professionally as the uh what you

guys have been able to do. What's been kind of your favorite AI capability that you've seen on anything you're doing in a personal your personal life and then secondarily as a company what your favorite one has been. I'll say mine real quick. There's a big list right now of AI capabilities that are cool. There's a tool that I just found out about called crisp.ai. Have you heard of this yet? Yeah. I found out about this on the a recent interview. I was like, "This is awesome." You know, like you just you're like, "Oh, great. I can be on a call and not deal with me being in a coffee shop enjoying it and me feeling like it's it's not that loud in here, but it sounds pretty loud to them." I was in a I was on the phone with my girlfriend the other day.

She's in a train and you know, people are like six uh to 10 feet away from her. But somehow like the way that the aerodynamics of this dang train work with the the bouncing I can hear them louder than she can hear them. >> Mhm. >> Like just talking about their day and I'm like are they like eating your microphone? She's like no. >> Crisp AI will solve this problem. It is voice isolation >> while you're on a call and transcribes. Amazing. That's my new favorite thing. What is your favorite thing personally right now? And what's your favorite thing that you've been able to do for for customers? uh or a customer >> personally [snorts] um being a founder uh not being an employee anymore um even if it was an executive position and and having uh being a father of triplets right so having

little family running around and >> the whole the whole balance of my co-founders and I is is to cherish and honor our lives um our private lives as well as the business life of no we are startup uh founders which is crazy life um and full of uh work [laughter] continually um uh we lost touch with my creative side. So I was a big fantasy reader and and I used to be a big gamer back in the day when I had time. Um and and what it's not a specific tool I obviously use uh um one of our agents for this but then I I always go back to anthropic with the latest models right so they do wonderful thing is liter uh writing my own stories um I don't have time to to jump in and and sit and if I read a book

it's something on AI or industry or business or or thought processes or longevity right um to improve my efficiency >> so the creative part of my life was cut to zero. It was all my children, my wife, my family, my dog, >> and my co-founders, my customers, our business, and our employees. And and that was the world. And there was no space for for for Tyrell anymore. Let's see the fun things that I enjoyed doing as a as a carefree young individual. And and so I found that again very recently actually where instead of opening up a book or trying to find a book, buying a book, getting it on my Kindle or or or wherever uh Audible, I started writing my own books. And in that process, it is exactly like reading a story, but the AI I I developed an AI protocol or

just like a master prompt that I use and I start the story and I spend a couple a couple days on a story and I get character building, world building, lore building and and actually write a whole then I write scenes and then you get into it and you can spend hours uh if if I had them um some sleepless nights, right, where you just have enjoyment uh and and that's been my latest personal very personal, right? is not connected to the business at all. Epiphany where I said I can just >> who needs Brandon Sanderson or JR something right now. You got your own brain which is going to trick you with an AI every time um creating a whole new environment and and world. So that has been a fun one and I can do that at 15inute increments uh next to my

meditations or something and uh that helps keep me in tune with my creative side. Yeah. >> Very cool. And then what about uh at uh your company? >> So business-wise, company wise. So regardless, I mean, we we could talk about the human effect of it and just the the amazing things that my team is surprising me with, right? So, so I'm I'm surrounded by people far smarter than myself and humbled every day with what they come up with. Um, but when it comes to what we deliver the customer, right, so some things I'm very proud about and and one is like you said, uh, when we come to interaction and and and we talked about the human employee, AI employees. So um and the impact it's having in not replacing people. So I was really really excited about the acceptance um of an AI workforce.

So I'm taking that name from a customer who named it this right. So so it's not our proprietary name for it. Um they're using AICX heart um to create not just a general agent plus a couple uh specific agents for different departments. They are creating an AI workforce that focuses on everything from um the wellness of the employee um teamwork, individual work, individual everybody gets their own individual um assistance and and this uh it is it is a bit farfetched and far out there. It requires precise not only the technology has to fit but the storytelling about it within the company so that it's adopted that it's accepted that it's also not seen as a threat and and this has been done masterfully and we'll see where it goes. It's still in development. There's this it's quite been developed until now quite far. But it's

uh it's a wonderful journey and and to do this you need to have preemptive uh agents that are that are predictively uh reaching out to you like colleagues would with certain sets of of uh uh things that happen in a company and it has access to this information. um so that it can it can a not autonomously but act as a colleague to initiate uh usage and and and action and preparation and and helping you like uh like a colleague rather than just sitting there like like I said before waiting there twiddling its thumb staring at you waiting for you to say something. Uh but what do you say and how do you start and where does it go and and when do you do it? the AI does it for you and and uh and and understands that moment where it needs to then

reach out and and this proactivity is what we call it is is quite groundbreaking for us. Um I hope others are doing it in the market. I hope all these things are always happening. and it means that technology develops faster, but but we're really proud that we're the only ones uh currently in our market right now that are doing it and uh it's a lot of fun, right? So, so it's it's fun to see there's lots of mistakes that happen. Obviously, AI reaches out and the person said, "Don't don't interrupt me as a human would do as well, but it's such a beautiful moment." Uh because it's it's increased adoption and and awareness and it's a beautiful journey. Can't wait to see where it goes. >> Yeah, I can't say where to go. I can't wait to see where it goes either. It does sound

like a beautiful journey and and uh you know kudos to to you for making it happen and um I really appreciate you for making the time today. Is there anything else that you'd like to say as we close this out um about your company or any other thoughts you have on things in general? >> Um uh just uh it's always we always take an approach where uh of course everything should be a plug. Um we are entrepreneurs. We are trying to uh to improve our uh our positive footprint in the market, right? So I could I could preach about our product, but I think we've done that enough. It's always a challenge out there to to our competitors and our partners and our customers to take a a serious look at uh one security, one transparency. Um um treat it with respect um as we

do uh so that we can keep that trust with our customers and future customers. Um it's very important. You're part of this industry. We're a part of this industry. uh we have to safeguard the sanctity let's say and and trust in it and and uh not get too taken carried away with uh with just trying to be the next the best thing and the flashy thing and sometimes it has to be solid and has to work and sometimes that takes a bit of time um to build that but uh so it's always a shout out for us to do that and don't forget about the employee um every employee is an individual person out there in the world so we don't do B2 B2C uh we do B2B but that is uh every employee at every company that we we serve. So it's in the

end it's the people um that adopt it. Um the CEOs just buy it. Uh but the people adopt it, right? So it's very true. >> That's a it's a part of our heart. Yeah. Also the name of our product. So >> yeah, absolutely. Well, I appreciate you for uh being on this episode. Make sure to go check out everything that Tyrell and the team are doing at aicx.de. That's aicx.de. Thank you again to Tyrell. And I will appreciate it. Also, if each of you, including Tyrell, would leave a like, comment, review, all that kind of stuff for us on Apple Podcast and Spotify. The more reviews that we get, the more people can learn about cool agent companies like AICX. Thanks for watching and we'll see you in the next one. Bye.