Building AI Vision for Industrial Automation's Future - Marc Gyöngyösi (OneTrack)

One important thing is what [music] are your tests? How do you validate the performance right repeatably based on you know real queries from customers and making sure that what it's doing is [music] accurate. The other part of it that I think people are underestimating is we use just multi- aent architectures quite a bit where before [music] something goes back to the customer there's different agents that might like challenge the results of one agent or like subsessions that will validate or verify the findings of another subsession before we actually respond back. 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 [music] 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. In this one today, we have Mark, the founder and CEO of Onetrack. How you doing today, Mark? >> I'm doing great. How are you? >> I'm living the dream. Um, just excited to chat more about AI and chat more about what you're doing. Tell me a little bit just to get started in the audience uh about yourself and and how you and uh your company got into AI. I think it's going to be a fun and unique one. >> Absolutely. So, my name is Marushi. I'm originally from Munich, Germany and I came to the US in 2012 where I studied computer science and uh during that time I spent a lot of time in robotics, machine vision that was really way before AI was really as big

of a thing as it is now and got to spend some time also working with BMW in their advanced robotics research group and what we were focusing on there was really optimizing production processes for lightweight manufacturing. So imagine like light bait, smaller robotic arms working next to people [clears throat] >> and giving them the the ability to understand the environment that they were in. But one of the challenges we kept running into is that they would misplace parts on the production line and the robot would stop. And so you basically had like a whole production line down because there was some part that was in a warehouse somewhere that nobody could find. And I literally had to run into the warehouse with binoculars trying to find a misplaced pallet of windshields and car doors so this robot could run. And uh that really opened my

eyes to the opportunities around logistics and warehousing. And then ever since then. I never thought I would, but I've been spending so much time in in warehouses and uh just helping customers optimize how they run their operations to ensure that they don't misplace things, but also so people are safe and productive and uh that whole that whole system really only works with AI. And so AI really has been part of everything we do since day one. And obviously it's changed a lot between 2017 when I started the company and now where we're talking about AI agents and all the other things in between. >> Yeah, it's crazy, man. Honestly, I I can't tell you how much it has changed and I think you're going to be one of the the more interesting interviews because you're going to be um maybe one of the more I

guess the term properly stated here is like manufacturing uh and logistics based companies. So, you know, tell me a little bit about the state of um you know, uh well, maybe state of things later. I I just I got a little uh I jumped the the gun a little bit. Tell me a little bit more, you know, [snorts] about how your company got started and and what kind of problems do you uh feel like you mainly solve that are addressing a hole in the market? Yeah, I I'll I'll start with maybe less about the company and just in general what's always driven me, what I've been fascinated by. I've always been interested by things that connect sort of the physical world where things happen in the real world or that factory or in a warehouse with some kind of digital intelligence or like software. >>

And like early early days, the the what I worked on um has nothing to do with robotics or warehousing or manufacturing at all. I I built a 737 flight simulator in my basement. And the idea was like you you basically build hardware that is supposed to simulate the the real world and then connected to a flight simulator software. And you have to make electronics and displays and switches and buttons talk to some software on a computer. And the robots that I worked on at BMW are like the the next evolution of that where you're like connecting the the physical world with intelligence and software to understand that. And so today the problems that we're solving are very similar where a lot of companies like if you walk into a supermarket and you look around you have all these different brands right different products you buy

from your cereal from Kelloggs to Mac and cheese from Craft High and your toilet paper from Kimberly Clark all these things have to be made somewhere and the problem is that the places that make those things and then also ship them and distribute them out to the end customers they're running on really archaic systems and they don't really have that connection of what actually happens on the floor of the factory or in the warehouse to the digital world. There are software systems that are supposed to tell them what is going on. And in many ways, AI allows us to bridge that gap by taking sensors, by taking computer vision devices and actually having them monitor and understand what is going on in that physical world and again connect that physical world of operations of manufacturing and logistics with the digital world of modern software and

automations and AI agents. And so the problems that we can solve for with that are primarily around or it started primarily around safety and productivity. So when you think about a warehouse um every three days a person um is killed by a forklift in the US and so and that's >> how many did you say every three days >> blowing stat u we have about 100 or so people every year that get killed by forklifts and that's just you know one small part of this this breakdown between what happens on the floor and what actually people know that happens in their software systems or the warehouse management software like safety is one big part of what we do where we're in hundreds of warehouses now where we're helping them reduce their incidence rate, incident rates. So that could be injuries, accidents, unsafe behaviors, you name

it. We reduce those by 60 70% on average in just the first 6 to 12 months of those customers. And then once we tackle that safety side of it, we also look at things like productivity and optimization. So are people actually driving efficiently in the building? Are there areas are areas where they're not able to perform up to the expectations around standards because maybe something is not set up correctly or packaging is not right or staging correctly. >> Right? So like it's a lot of data that we have to capture to solve those kinds of problems and it's only possible because of the AI technology that drives our system at the core. >> Yeah. Now that I actually was a gasast when you said every 3 days. That feels I mean forgive me if I'm wrong. It just feels way too often. Obviously any tragedy

like that is too often. But I heard that and I was just like there's no way he's being ser but I believe you. It's it's wild data. And you know obviously safety is a big thing in that industry. It's a big concern. It has been. I mean, it's caused it caused an entire revolution. I mean, in in the states and and otherwise, right, with with the 5day work week, I mean, that was that was a lot of the the worries of those concerns. And I mean, talk a little bit more about kind of the the state of things in any industry and what else you feel like uh is lacking. >> Well, one of the interesting things is we're all super excited about AI agents and automations today, right? Or if you're if you're living in the world of software only like you're only working

in front of a computer and you log into your different software so your email your slack your spreadsheets and whatnot you can use AI agents to do a lot of work now that and that whole area has been developing so quickly but without technologies like what we're building a lot of these physical businesses really can't apply that right that you can't make an AI agent optimize your warehouse if you don't have good data about what actually happens in if you don't know how far are people really traveling, how long are they really taking, how much are you really using your equipment, how many hours are people really spending in different functions and activities. And so the the current state today in many ways are is that you have software that I mean it's literally a green screen, right? Like you walk into some of these

environments and you see a AS400. I don't know if the audience knows what an AS400 green screen is, but it's technology from the '9s that is still being used to run how, you know, millions and millions of square feet of warehouses are operating every single day. Like people are are opening a green screen and typing in in the terminal to say I'm moving this product from here to there. And then you can only imagine that those systems on the back end, the databases and the ability to access the data is severely limited. So if you're trying to come to that like, okay, let me bring you an AI agent that's going to optimize your warehouse. Well, how are you going to even get to the data that's supposed to be in there that usually is completely inaccurate and out ofd, right? So that's that's a

big issue as you think about sort of current state in this world is that the systems that are designed to make them run smoothly are actually holding them back from realizing the opportunity that's out there with AI agents. And whatever a lot of our customers are realizing is that in the future, it's not that AI will take your job, but it's that other companies who use AI are going to out compete you because now they can do so much more than than than their competitors, right? So if you think about having AI analyze all that rich data about what actually happens in your factory and taking giving you recommendations on how to optimize, that is something that completely changes the game from a competitive perspective for them and that's why they're excited about AI. but also why there's a lot of challenges out there still

because you have so much legacy software and legacy systems and dashboards are not going to fix that, right? >> Yeah. You know, I I want to pull on that thread a little bit more that you you just kind of laid out there. Um why necessarily do you think now is the time for uh I guess adoption, right? Maybe some people have concerns that it's not ready yet in certain capacities. Why do you feel like now is the is the time and and how have you kind of showcased that it is the time with the different uh companies that you've worked with to get into using this in in the industry you're in? >> Right. There there is I mean now the question if now is the time there it's the only time you have it's not whether or not now is the time. I think

a lot of executives are waking up to the realization that if you don't do something about it right now, you will be left behind. And it doesn't matter if you're a multinational corporation running hundreds of warehouses all over the c over the country and the world. Um or if you're a if you're a logistics company with five warehouses or if you're a manufacturer with three contract manufacturing sites or 300. It really doesn't make a difference. the the AI technology is here right now and it really is not about whether or not now is the right time. It's about how far are you taking this and how quickly because every single person in these companies is going to have to work with AI every single day and they have to build that competence within their teams. And it it's about much more than just their warehouses

and their factories about how you go about your day-to-day in whatever function and role you you are within within these businesses. Right? If you're sitting in an office and you're you're literally just responding to customer uh customer issues of orders that are not fulfilled correctly or arriving at the wrong location, are you downloading spreadsheets and manual parsing through them or are using AI in some way to make that way faster? And if your team develops that capability, you can become so much better as an organization, so much more competitive. And if you don't, you will fall behind. So, it's really not about about whether or not now is the time. It's now is the only time. And if you don't do it, you will be left behind. And a lot of companies and a lot of people are waking up to no realization right now.

>> Yeah, I think it's a pretty I think it's a pretty pretty excuse me pretty fair point. Um there there are a lot of companies out there that are waking up to that fact and um I think you you hit the nail on the head there. If now is uh not the time, I'm not sure when would be the time honestly. Um, and you know, with that in mind, obviously that's something that is important, but this is more thinking short-term like let's get these implemented. Let's let's get them into the systems. What's kind of the long-term goal um for what you are helping customers solve, right? Like once everything's implemented, you know, what would be like the long-term goal of what uh it would be to provide to uh your customers? I think a lot of companies are waking up to the realization that automation

means more than just robots, right? And again, I'm looking at this from the lens of physical operations. I used to work on robots at BMW and I I I can't tell you how excited I get every time I see a robot do something on its own. But a lot of leaders in this in in in this world of just you know manufacturing, logistics, whatever it might be are realizing that sure you can have you know Amazon has probably a million or two robots now out there in the world but guess what you know there will be billions and billions of AI agents doing work behind the scenes and that's a whole different lens on automation than if you have to go and spend tens of millions of dollars on physical robots to actually do physical work. you can automate at a much larger scale if

you're automating using AI and then software and at a much lower cost with much lower capital commitments as well. That's a really important insight and I think that's something that people are are waking up to that really automation you know it used to mean buying robots but today the cost of automation is essentially just the cost of energy and the cost of compute and with that comes intelligence at a level that you could never have before at a scale you never had access to before and once once companies realize that they really go all in and we're seeing that with a lot of our partners now that one may have been step one on their AI journey But now they're going really deep across their whole business in a lot of areas that they never thought about before because they're realizing what they can do

with it now. >> H what other areas would that be? >> Oh, I mean how many how many jobs are there people entering data in computers? >> Oh gosh, too many. >> That's the answer. [snorts] >> That's the answer. >> That's I think that's the answer. There's a what there's this quote from Jensen Wang. There's like a you know a billion knowledge workers in the world right now and there will be 10 billion in two years. Um I I think that's very realistic. It might be 100 billion. Who knows? I mean it depends on how you define an AI agent and what the technical definition of that is. But when our I mean when our customers when they when they talk to our to our agents all they do is send off an email. It's like a co-orker in their inbox and they could be

sending off 5 10 20 emails at a time with different questions about what to do or or things to look into. And then as they go about their work throughout the day, those agents come back with insights that are based on actual data, actual analysis, what happens in their in their business and between the people and the machines and the systems and the equipment and the assets and the processes. And so do you call those, you know, 20 different people that are now on working for that one person? Is it one person? I don't know how you define it, but it's it's exponential capability that you never had before at the cost of energy essentially. >> Yeah, that's totally true. And you know, I guess what that kind of begs the question, what is it like for um could you speak to a little bit

more about like what the experience it is in a I guess I don't say practical but in a tangible manner like what what is the experience like for a customer to interact with one of your agents? [snorts] >> We we talk about three phases. The first phase is when we go through our onboarding, when we go through our training, >> they often just don't know what to ask and how to ask it. We often hear like, okay, do I have to use proper grammar? What language do I need to use? Do I have to specify everything I say? And one one piece of advice we always start with is just ask it like what can you do? And because of all the context engineering that we do behind the scenes of making sure that the agents understand what data is available, what skills they have,

how they can analyze the data, all those things, they come back with recommendations or potential questions they could ask that are actually dependent on what exists at that particular customer, that particular site. So that's like step one is what do I ask? Step two is it might come back to a question or some data that you don't agree with. And the you know it happens. Now, we do a lot of work on the back end to ensure that whenever it's analyzing data, it's actually using computation rather than just looking at a bunch of data in a single, you know, um, prompt step. So, it's actually like computing the answers to these to these questions and loading data and transferring data and processing data. Um, but it might make a mistake or maybe it doesn't understand that when you refer to a week, you actually mean

your week starts on Sunday and not on Monday and that a day starts at 6:00 a.m. and not at midnight for you. whatever, however you think about, you know, your data. And and so the the second phase there is then for for people to push back, right? It's like you hired a really really intelligent person onto your team. They're going to fire them the moment they make one mistake. You'll give them feedback, right? And so you tell them, "Hey, actually for me, the day started at 6:00 a.m., not at midnight. That's still the night shift from the previous day." Well, the great thing is that these systems have almost perfect memory because they can distill the feedback that you give back into into memories that they can reference long term and then the next time you ask it, it will know that when you say

today, you mean today starting at 6 a.m. So teaching people that they can interact and give feedback back and it will actually remember the next time you ask that same question. And then the last phase that we usually see, that third phase, to me it's the most exciting one. Sometimes people just say, "What is there for me to know about my operation that I don't know? Go and explore three or five different paths and come back with some interesting areas that we could focus on." And they start partnering with this agent, AI thing, entity, whatever you want to call it. We call it AON. um and and they go back and forth and they and they basically have the agent go off in different directions, pull data, come back with interesting analysis and say, "Okay, I actually like this particular combination of of um of

insights or of analysis. Can you send this to me every Monday?" And so now we basically went an entire continuous improvement workflow from scratch without a single continuous improvement engineer in the loop. So those are really the three phases that our customers go through as they start using AON our agentic platform and one track. >> H you know I I hadn't uh had considered the way that you framed it but I I think that is it is fair. Um I I I have kind of made some similar commentary in in the realm of agents when people are irritated or concerned due to [laughter] uh the let's say 90% like accuracy rate or whatever it is. Let's say a well-trained agents like you know it's it's doing a decent job but it doesn't have the exact right answer and you're talking more early stage. But um

I find it I find it interesting. People are um not going to fire their you know whether it be high level person um middle management manager associates. Um yeah when you start working with an agent you basically hire a person with the capability of a of a of a advanced college student today and very soon. I I think we're getting close to like master's PhD territory and and understanding and I think on the engineering side or coding side probably better at writing code than most software engineers out there vastly better. And so when you I think it's really just a matter of putting that into the right context for the customer and telling them that you need to interact with this, right? It's not a it's not a oneshot thing every single time, but if you give it feedback, it can become that. We've actually

come up with some really interesting ways to to make the processing on the back end repeatable which is really critical because a lot of our use cases are like you know HR related or safety related or employee related where you're you're calculating some safety score for an employee or performance score. You're looking into the performance data as in productivity overall. You need to make sure those computations and calculations are repeatable. So we've come up with a number of things including smart procedures and and those kinds of things that allow our agents to once they figure something out and the customer agrees and it's validated then every single time someone asks about this kind of thing uh the system is going to use the exact same computational approach pulling data from different sources to make sure that the answer is correct. And that's a really key

thing, right? You can't just compare talking to ChatgBT or Claude in your in your CLI or in the browser to something like that because this has so much more context and the ability to really be accurate in the in the calculations it's it's making. >> Yeah. I guess two followup questions there. how does it necessarily um store you know that information and then how does that also in the case of you know I don't want to say uh redundant information but new information on the same category uh kind of swap out that new old context with the new in regards to like standard operating procedure on something. >> Yeah. The first is just a pure software engineering problem, right? That you need to solve like where how do you store context and code that's generated and things like that. So we built our whole agentic

platform and framework around we use anthropic we use as 4.5 for almost everything. So we we build our own agent framework around that where we we have the ability to to do that securely but also in a scalable way where I mean we have a lot of users using these agents every single day and they all have you know different skills and different capabilities and different memories. So that part is more software engineering than anything else. >> Sure. Yeah, that makes sense. But I I do think you bring up an interesting point that a lot of companies are going to run into over or really the first years of deploying AI agents and that's context and memories grow and grow and grow and how do you really distill that down? It's still that's an really interesting area for research. Um, I mean, we're we're always

thinking about that and there's a lot of work being done on how do you really distill knowledge and memories into the most the truly unique and distinct pieces of things that you need to remember and maintain versus the things you can throw out so you don't have that duplicate data or duplicate memories or conflicting information problem because when you the more you add into the context window of course the the less likely it is that the the agent becomes accurate in its responses back and and follows procedures that you want. So that's an active field and I think it's going to change a lot as the models evolve. Interesting to see how how Enthropic is solving this with now they have the skills approach as well, right? Where you're like dynamically loading context into into the context window to to know how to do something.

>> Really cool. >> Um we we rolled that out in May on our end where we had skills for >> for the ability to have like repeatable proceed like processes. They were not yet programs but they're at least this is how you solve a task. And then depending on the question the person asked, we would go and load the correct the correct skill into memory into the context window. So it's like right there as the agent is performing the task. U I think these kinds of things are actually very simple, but this is the kind of stuff that you only realize when you're when you have agents run in the real world, talk to real people at scale that you need to build these kinds of things and you're moving from the demo environment with a couple of test cases to, you know, thousands and

thousands of people in warehouses and factories asking these agents every single day. I'm the wildest and and most random questions. Um, plus all the safety and safeguards we need to put on top of that as well, right? >> Yeah. I I kind of want to pull on on the thread a little bit more about models and um tools that are kind of becoming available in the uh the space, right? How are you guys taking advantage of the different uh models that you feel like are um top of uh market right now and and how have you seen because you've been in the space obviously for a decent amount of time. How have you seen, you know, as agents have gotten more and more, not agents, sorry, models have gotten more and more reasoning based, how have you see that seen that make a positive impact

on impact on your uh agents and how they're able to to handle those complex situations? So I mean one of the great things in the space today is that you can hot swap really any model you want and >> sure >> the moment a new model comes out if openi comes up with something better than sonnet 4.5 we'll switch over to that and then when anthropic comes back with 4.6 six we'll probably switch back to that. So we can go back and forth between the model very easily. One important thing is what are your tests? What are how do you validate the performance right repeatably for for based on based on you know real queries for customers and making sure that that what it's doing is accurate. The other part of it that I think people are underestimating is you mentioned that reasoning question. We

use just multi-agent architectures quite a bit where before something goes back to the back to the customer, there's different agents that might like challenge the results of one agent or like sub sessions that will validate or verify the findings of another subsession before we actually respond back. Um, and one one real world example that I can share there before we had this um in many ways like these LMS today, one person once described it to me like monkeys with machine guns. they can get really, you know, really excited and exaggerate quite a bit uh and then get very chaotic and and so like when they when they find a problem, they'll they'll get really worked up about it like somebody doesn't resolve a safety incident for a day. Okay, it complains two days or we have to escalate three days or we have to initiate

emergency procedures and send you know emergency forces on site and take executive intervention. Um, so these are things that like these models every once in a while still do and that's where it's important for for them to check each other before they actually generate output. So like anything that goes out to a customer is going through a separate system that actually verifies that is the language that you're using align with our mission vision values. Is it safe? Are you criticizing anyone? Are you making any decisions you shouldn't be making? Like you have to architect the system in this way. I think the reasoning models are moving us closer to it. But in many ways like you need to at least from our experience you still need to have that multi- aent multi session architecture to ensure that they don't get just too worked up about

one thing in their context window that it really isn't that critical or that extreme. >> Yeah. Yeah. No, I think that's a that's a great point. And um we just had an interview I had the other day that kind of talked about this a little bit where you know it's like multi-prompt chaining was clearly and obviously needed for LLM use in like your browser for a while and it seems like multi- aent right uh capabilities are or multi- aent processes and workflows are clearly needed um as well and it seems like you guys have a really good understanding of what it is to avoid, right? Um when it comes down to the issues that could come up and you know obviously you you're saying that multi- aent uh workflows and systems are kind of one of those things. Are there anything else that you're do

is there anything else that you're doing um or you're seeing in the industry in general to help people avoid the issues that can come up with using AI agents? Obviously they're great, awesome, we love them, but just from the perspective of a business owner who might be thinking uh I'm not sure like there's all these risks speak a little bit to that >> on the risk side. So one important thing we have to implement um it just in the product itself is there are certain things our customers don't want the AI to be used for and this can vary by client this can vary by site by region by state wherever you are. Um, so a simple example might be, you know, one customer doesn't want their employees to use our system and generate a document that would show why an employee should no longer

work at that company [laughter] based on the safety incidents they had and the low performance that they had. They want a person work. >> They don't want an AI agent to do that work. >> So, you know, we we build safeguards around that. And I think in a similar way, one thing we use internally, we actually use our own agent uh framework to also score and coach on all the customer conversations our team has with clients. So if you're an implementation engineer or even myself, if you're talking to a customer, um you will get feedback from an AI back to you on what you did well, what you didn't do well based on how we want to interact and work with customers. But as you do that, you need to have those safeguards in in place to ensure that I don't want an agent to

go yell at one of my employees and tell them that they did a really terrible job. I want them to take the approach of focusing on things that they can improve, but also giving them feedback that recognizes what they did well. And so, you need to have those safeguards in place. You can't just go, you know, plain vanilla uh s 4.4 here, hey, give me feedback because eventually they they get really worked up. you you need to make sure that you have those things in place um as those safeguards to mitigate some of those risks and concerns. Um that's I think that's that's one that we've seen over and over that you just have to build around to ensure that you're safe that you're doing it also in a responsible way where people are excited about working with AI and not scared of working with

AI. >> Yeah. Yeah. No, I I I really appreciate that cuz I do feel like we've actually reached a point where model improvements is is is one thing, but I I think the industry and the people who are helping provide these solutions, like yourself, are are really like in the position to say like, "Hey, we know how to prevent all the the the weird things that have occurred, right?" And there have been funny and weird things that occurred. Wasn't there an airline thing the other day? I feel like someone mentioned on the podcast there was something basically, you know, rogue AI agent says something not okay um or send something out not okay to a customer, client, whatever it is, or co-orker. And um that is something that best practice-wise I feel like can be can be solved for and can be prevented. And and

you know there's there are things that they can't do yet. So don't try to make them do things that they they can't do yet. But speaking a little bit more on uh what they can do and where you think they they can go. Obviously, we've kind of established that the better the models have gotten, you know, it's easier for them to uh understand these different contexts and and you made a statement earlier, a quote that said, well, we are at like a billion dollar workers. We'll move to 10 billion because we'll have all these agents. Um and you don't think that they'll necessarily take jobs. Uh it seems to be there's a bit of a disconnect or not disconnect a disagreement between different people right in the industry. Why are you optimistic that jobs won't be lost necessarily because of AI? >> What we're seeing

today, what these agents are doing is really helping people do things that otherwise would have taken just so much of their time and not been very value add for them. So right now I think AI agents are not taking people's jobs. They're really just making people way more efficient. >> I think there is a real concern, >> you know, call it 12, 24, 48 months from now where some of the jobs are going that AI agents will be doing at that point in time. And you can already see you know early indicators of that around like support workflows and sales maybe outreach and and those kinds of things as well as also on the software engineering side. So >> I don't necessarily think that you know all jobs are safe but I do but I do think that jobs are going to change and people

have an opportunity to be way more educated than ever before because you literally have the knowledge of the world at your fingertips for $20 a month, right? like the opportunities for a person to be successful from that way outweigh the the risks of people, you know, losing losing their jobs to me. And I think there will be way more exciting businesses and companies and builders out there, people who can do things and build things and create things they never could create before because you you have would have to, you know, hire all these people and spend all this money and spend all this time learning all this where now you can just do it. But I think the future is going to look very very different. I think jobs are going to look very very different. And I think there's a lot of jobs that

will be going away. The much bigger risk to me though is the societal impact and especially when you look in the consumer space and products like Sora and how people are becoming less and less social and not h you know they they sit somewhere in in in a room and they talk to AI all day and then they watch AI videos all day and then they go to sleep. That to me is a much bigger risk than Excel data entry getting automated. Like the real societal risk I think we >> Yeah. Oh boohoo we don't have to do Excel. Oh no. Like to me the real societal risk we need to talk about is is on the consumer side is on the people's side and how people understand what's real and what's not real and how they can still build real relationships with other people.

Um that to me is a much bigger concern than whether or not you know certain jobs will change or go away and change into other roles in the future. >> Yeah. just to kind of speak a little bit to that I am I love something that I I don't think enough people have maybe um tried out yet and this is varies across the different models um research uh has in my opinion at least from a give me the information with one typed prompt rather than a bunch of Google searches. Uh like for example, the other day I was looking up cold outreach type uh [snorts] comparisons in regards to what's like the most cost-effective uh email and phone number like contact per credit tools out there, right? You do that in a in a even two three years ago manner with Google search. you are

finding a bunch of different blogs, right, that are essentially telling you what the case is, trying to summarize it. And now it gets to the point where if I ask somebody like, "Hey, I want to get an answer to this question." In theory, I could have had something autotranscribe the meeting and then like do the deep research and by the time the call is over for 5 minutes from then, I have the answer. That's like the world we we're living in now. And I don't think anybody's out here saying like, "Oh, that's going to take away, you know, valuable time from people's lives, right?" Um, I think access to information is so much better. But I I do I do want to kind of touch a little bit more on your concerns, but there is something that we we talked about premeating. Speaking of access

to information, let's talk about dashboards. Um, seems like this is a this is another access to information type thing, right? when people uh I I used to work in paid ads. I loved dashboards. Sarcasm, big sarcasm. Um every client wanted a dashboard they never read um and never read the reports. So tell me a little bit about why you think dashboards might be um dead moving forward. I think dashboards are are very dead and the reason for that is that when you think about the and again we're we're switching context just a bit back into like you know operations you know you're running a factory a warehouse >> but that's like something I mean other other things like your Tableau license right like if you're using Tableau or PowerBI or something like that >> you have someone in your team whose task it is to

you know go pull the data merge the data join the data clean the data put it into a nice view add a bunch of filters Now you have a fancy little view on with like pie charts and lines and you know numbers all over the place and things you can explore. Well, the only person that dashboard really truly makes sense to is the person who built it because they understand the the data and how it's all being pulled in. >> The moment you try to then have someone else use it, you have to go through this training process of okay, this is what the data means. This is how it works. This is how you interpret it. This is how you use the filters. that person might only need a very small sliver of one of the charts on this crazy fancy dashboard that you

spent four weeks building. Um, and then based on this like one sliver of data, then they have to go off and look for something else and then make some decision about do we change out this machine or do we perform some preventative maintenance here or do we coach this employee on their on their on their productivity. And so the reason why I think dashboards are dead is because access to the data that actually brings all this into context is now just available at hawk and whatever question you're looking to answer specific to what you're working on, >> you can have the AI go and pull that data for you. And one of the interest there is >> you know when you think about data warehouses and all these complex projects of syncing data different systems and this is particularly true in in enterprise software the

the biggest challenge is just mapping everything together right simple example your Hopspot CRM might use you know the the field client and then your ERP is going to use the field customer and then another another system in your support system if you're using something other than HubSpot might use um just like cus so like you have three different columns that all need to be mapped back into into a single field well an AI agent if it can go and query the data from these three systems and merge it together it will understand those three fields belong together and you don't need someone to do all that mapping for you and it can do it ad hoc it can look across systems it doesn't it's not constrained to just a single view on data >> point answers so if you're that end user somewhere on the

floor looking for a specific answer you don't have to go log into a dashboard and hope that somebody builds a view on the data that you need to make the decision you want to make. you can just get the data you want sourced across multiple different systems ad hoc on the fly computed accurately in the moment that you need it and complexity of dashboards today is really like if you have five different data sources you know that introduces dependencies and all of that it's o of n like n data sources and n complexity in the future dashboard information access is all of one because if you have one question you want to answer the agent can get you that act that information once and it doesn't it doesn't depend on anything else in the system and I think that's a really interesting unlock that our

customers are finding that people can really get the answers rather than just some data to figure out how to come up with an answer >> you know I think that's a great point um I think I you know as I mentioned working in paid ads previously these dashboards were put together it's been it was pro it was a project that was always changing it's like we need a new dashboard it's like why well I want to see this thing and by the time I feel like Now that we're in, if a client were able to just ask an agent, I want to know the answer to the specific question. It goes from somebody asking a person to build the dashboard so they can get the answer to XYZ thing on a consistent basis to then just asking the thing on a consistent basis. And honestly,

better, right? like I can imagine whether it it's in regard to inventory or um you know the amount of uh units that are being shipped you could not only get the answer to you know was there an influx this month but is this a cyclical thing that we're seeing uh of you know orders is this not a cyclical thing is it is what's the variation like you can ask a bunch of different questions at once whereas I almost feel like dashboards don't always give the answer right like you could have that little minus this 24% thing. Uh, you know, where it's like, well, it went down this percent. And to be fair to people who are asking me to do reports in the paid ad space, like that's cuz you had to kind of dive into the data to get all those niche answers. And

I I really think, yeah, that's a fair point. And I'm going to tell my old paid ads friends like, "Hey, have you guys considered you could just >> and and we're so early on this still, but I I I see a future where there's a whole different type of software that exists specifically for enterprises, right? Because every Fortune 500 company in the world is built on this like really same three-layer architecture of like static databases, some backend that pulls from those databases and then some front end built in your SAPs and service house and sales cores of the world to run your operation somewhere. And you know all of that is pieced together by a bunch of consultants that spend hundreds of millions if not billions of dollars for you to set all that up at once. And and so like that's how software used

to work in the past. And the result is you get your screen, you get your dashboard, you get your report, you get your table, but it's a very generic, very very general purpose software interface. What AI enables is for software to be really hyperpersonalized, not just to the use case and the code that an engineer writes in the in the editor, but really to the application that you're using in the data is showing surfacing to you and the way that you're interfacing and using that data and interacting with it. And as I work, we're we're I think we're still super early on that path. But here in one drive, we started to build some of those pieces already and we're using it internally. We're starting to make some of that available to customers and really changes the game. Like that's why we say dashboards are

dead. Like it's way more than just dashboards. I think static applications are dead. And I think the way that you interface with products is going to change completely, especially when it comes to this enterprise SAS architecture um that you know is run by run by these these old school companies with legacy software from the '9s and 2000s. >> Just as a final piece before we kind of close it out, last but not least, please let me know um what your favorite thing is question I've been trying to ask every founder. what your favorite um personal AI and business AI tool is right now. So obviously I'm guessing one of the agents you have running at your own company. Maybe the business one, but personal ones are fun too. >> Personal ones I mean obviously cursor is the most incredible tool right now for just building.

Um I think you can use for way more than just writing code which is something people often overlook. >> You can use it to create patients and and and documents and whatnot. Um, so I think that's that's a a tool I I I absolutely, you know, enjoy opening up every single day and using more on a personal side. I mean, talking to honestly like when there's complicated problems to solve, talking to GPT through a million different scenarios and situations in the in voice mode as you're walking down the street is something that has in many ways changed decisions that I've made in the past. uh because I I can I can think through scenarios in a more effective way. Uh and then also when I go and talk to people for advice or advisers or or anyone else then I can actually get I can

go and prepare here are three scenarios that I think are going to happen. This one I think is more realistic. I played these all the way through. Let's debate that rather than the very starting point of trying to parse what the three scenarios could be. So, I think having, you know, that that voice mode and GBT is something that's that's that can be very powerful um for for a million different for a million different reasons. Yeah. No, totally fair. I think um it's really it it's really all in what people I guess are interested in and want. And um I I have noticed something. I interviewed a a founder a couple weeks ago who said the exact same thing about cursor. They were like, "But I don't know why people don't use this thing." you know, it's like four four other things, right? Not just

um the the coding and everything like that. I think it's tools like that in general, um you know, whether it be claude code, cursor, all that kind of stuff. They they are just completely not saying it's their fault or anything, but people are missing the boat on some of the the interesting like side things. Like people aren't familiar with the fact that for, you know, clawed code with all the agent capabilities. It's great for coding, but you can massproduce like 10,000 programmatic landing pages if you know what you're doing in like a week. So, you don't need to pay for whatever other service. You can like do it yourself. And um it's very cool. So, um thank you for that insight. I really appreciate it. And last but not least, just kind of tell us uh what uh where people can go to to find

you so we can plug you to close it out. >> Yeah. No, absolutely. I mean so most important thing like here at one track we're hiring we're looking for talent we're looking for cracked engineers excited to work on AI agents and uh I think one important piece about the business we're not a VC funded business right we're profitable >> cool we're we've been a business we've been doing this for quite some time now we're some of the largest brands in the world that that use our our technology in in their warehouses in their factories so if you're excited about working on AI agents that do real things in the real world for real businesses and factories and warehouses ources and and beyond. U definitely reach out to us. You can find us on LinkedIn, oneterrack.ai. You can find me on LinkedIn, just Mark Gondoshi. Um

and then also on our website, oneterrack.ai. Uh if you wanted to connect to us. So uh yeah, we're we're excited about the future and and about building it with our customers together.