How Will Businesses Put AI Vision to Work? [Ft Brad Chisum, EyePop.ai]

A video feed is a series of what, 24 to 30 screenshots per second. How is it possible that AI can process it that fast? You don't need to capture all 20 to 30 frames. You can understand what you see in an image or video really well and give that same capability to an AI tool and you just unlock all kinds of use cases. We've intentionally built it to be camera agnostic. Using your security camera to do operational improvement, and you're using the same infrastructure. You didn't have to add cameras. It's 2026. Let's fast forward to 2030. What does the world look like for AI vision? Welcome to AI 2030.

Today we have Brad Chisum, CEO of EyePop.ai. Welcome. Good to be here. Okay. So I'm excited about this chat because I think when people, at least in my world, a lot of times when we think about AI and the people we talk to at mid market businesses and all that, they think of AI as the chatbot, the thing that they interact with, the claw, the OpenAI, you know. But they don't think about how AI is kind of affecting us in the physical space, in our real world environment. Would you... Do you feel that in your... Yeah. I absolutely agree. And and you're and you're starting to see that change, you know, as as physical AI is becoming a bigger and bigger thing. But really, I mean, AI is AI understanding this physical space around us. Right? And so and when you think of applications like robotics, you know, that's an obvious one, but drones, but anything where AI needs to understand things in three dimensions, it's physical AI.

And that's, you guys have... Are calling it AI vision, and I'm sure that's a a broader term that's been around for a while. This AI vision, it is essentially the the understanding of how the physical space is interacting through cameras. Right? Because the only way it's going to see this is through another tool. Yeah. Exactly. And so, So AI vision, you know, you're taking visual data typically, though you can go outside of the visual spectrum. But you're usually taking visual data whether it's from, you know, camera from still images or video. We ourselves, we're particularly focused on video, that's what we do really well, but when you can understand that type of information with what AI can do with it, then it's really the same thing you would do as a person, Right?

You can understand what you see in an image or a video really well and give that same capability to an AI tool and you just unlock all kinds of use cases. Here's where I think, at least me not being in the space where I get, like, hung up is I'm like, okay, if I have a JPEG that I wanna makes AI to make sense of and I load it into a Claude or something like that, it's gonna do OCR on it, right, and it's gonna gonna look at it and try to make sense of it. Think of like Claude in Chrome. You ever watch that thing move? It's moving pretty slow because it's taking screenshots, trying to make sense of it, and then it's moving, you know, on. A video feed is a series of, what, 24 to 30 screenshots per second. Yeah. How is it possible that AI can process it that fast?

Great question, and this is really where... So you're seeing a big transition in our space. So the transition is, you know, last year, the bottleneck for these things was the model because how long does it take to capture whatever information you're looking for, like how long does it take to process that frame? Now the bottleneck is on orchestration. So is, you know, when you... So there's a lot of variables that you can adjust. So for instance, what... You don't need to capture all 20 to 30 frames. You know, if things are moving slowly, then maybe you only need to capture one frame or one frame even less often than that. So there's a... That's a knob you can turn toward optimization. What resolution are you trying to capture? Where in the spectrum are you trying to capture? You don't necessarily need to process the entire image if what you're looking for is only in a small percentage.

And so, the tools to be able to very efficiently and effectively get to that... All those optimal points are how you really are unlocking, you know, the the speed question that you're that you're asking. And then I think about, okay, every movie I've ever watched which has a cop, and it's... You could be looking at The Wire from 1990, and, like, they have a screenshot of a license plate, it's blurry, and then they, like, enhance it. And me coming from a creative background working with Photoshop for the past twenty five years, I'm like, you're not doing that. If it's pixelated, it's pixelated. You know? How close are we to AI being able to... And and maybe there's other technology that really can do that, but, like, can AI get to the point now where it kind of interprets based on...

You know, if you zoom in on something that is truly pixelated because it's capturing it, let's say, seven twenty, like, can it really enrich it or enhance it at this point, or are we not there yet? I mean, are some tools that that can certainly sharpen, right? But if the information isn't there to begin with, it's not going to be able to just conjure it up. Now, it can give you something that can be statistically likely, right? And so, it can do pretty well on some of these things. The other thing is, in those examples, quite often what you see when you are looking at a video screen of of the camera, that is not at full resolution. So there's... It's not an AI trick, but if you go back to the original data source, you can actually get to a higher resolution. So there's there's things like that that are involved, but, yeah, but the idea of, anytime it starts to flip into magic, then be skeptical.

So if you are trying to capture something in a scene, let's say you have a video that's looking at something that's kind of far away and you're trying to capture license plates, you're trying to capture people's faces, you pretty much have to be looking at it at that highest resolution, right, which is going to be more compute power per frame. It would unless what you are doing, this is one of the things that you can do with our platform, if you're trying to read that license plate at high resolution, well, first look for a vehicle, you found the vehicle, then crop down to just that portion of the vehicle where there's a license plate and then process that at full resolution. If you think about that, from the percentage of your whole frame, think of every pixel as something that takes compute power.

You just reduced it down to a very small percentage. You have to take one larger shot to determine where's the license plate, then zoom, and then just say, don't process anything else except for this small, you know, 200 by 400 pixel area or something. Right. And that matters... AI vision, this is really important because not all models are equal in terms of how much compute power they have. So for so for for models that have been under development for many years, like a vehicle detector, those can run pretty light. You know, if somebody that actually wants to read a license plate, that's a little heavier. If you want something that's going to go and tell you what make, model, or year of a vehicle, well, then that's a heavy one. So what you do is you you can use logic.

You say, okay, well, is there a vehicle? Yes. Then tell me what make, model, and year. Don't try to process make, model, and year across the whole thing if there's not even a vehicle there. I know that sounds almost obvious after you say it, but it is not how most people are doing it today. Well, it's also not how AI typically works. If anyone is using Claude and you you give it a complex task and it runs for fifteen minutes and it comes back and it... And you're like, wait. You went way further than I needed you to go. Right. Right? And you burned all that compute power and tokens and all that stuff, but, like, you weren't clear enough in your prompt. It's not the AI's fault. It's probabilistic model. It's going based on your prompt. Yeah. No. Exactly right. Yeah.

So, okay. Go back to the model orchestration. Explain that more to me. Yeah. So so if you think about any any model, you know, it's usually going to do a specific task. Some models are broad and then... You're talking about a model like Opus or like Fable, right, or we're talking about that? So, those would be included. Those would be the frontier models, right? And then, in computer vision, we have a category of models that we would call them a CNN based model. So, convolutional neural network is what CNN stands for, but these are very lightweight, simplistic models that are trained usually to do just one or, one thing or a small number of things. So person detector. That's all it's ever gonna do, is find a person. And and if it's an object detection model, then it's gonna find where...

It's gonna give you the coordinates of where that person is. And so you have this for common objects, you have this for things like vehicles and people, so those are very lightweight, right? And then if you have something, so a VLM, which would be closer to, you know, Opus or one of the... Now, Opus would be a whole lot more than... VLM, like video language model? Yeah, visual language model. And so, what these are capable of doing is giving a broad description of everything that it sees in that image or video. Those are more computationally heavy. But how you prompt them, and you were alluding to this before, how you prompt it really directs your results. So could say, describe everything in the scene, or you could say, what color clothing is the person wearing?

You can you can get specific, and when you get specific, it can do things in less processing time. It also is what allows you to take a lightweight model, like a person detector, and then only run the VLM when and where you want it to. So that's the order, right? Because I was thinking a bit more like, if the VLM would look at the video, determine that's a person, and then pull that smaller model to identify or do something about the person, but you're actually saying maybe it's flipped. It's better to say the person detection model is going to look at that frame and say, is there a person in here? Yes or no. If no, it's going use no more compute power after that. But if yes, then it pulls in this VLM that's going to do additional work on it. Yeah. And that's something that, only a few of the AI vision platforms enable that right now.

Any ML engineer, if you're doing a a bespoke custom solution, they're absolutely going to pull out of that bag of tricks. But being able to do this orchestration on a platform level, and this matters because if you are a small to medium sized business, or even if you are a large business and you don't happen to have a team of ML engineers, then you need a way to be able to get the same level of orchestration that they would do, but with your in house existing team. And so there's only a few of us now that enable that. A year from now, you're probably going to see this much more prevalent because it's just the direction the industry is going because it unlocks that actual value that customers benefit from. Said back a different way, you're saying you've got the Teslas, the Apples, like you've had facial recognition on your iPhone for a long time, you've had a Tesla self driving for a while.

They've... They have this team of machine learning engineers who have figured out all of the stitching and all of the the orchestration of the models and all that, but they're doing it for a very specific purpose for them. There hasn't been, like, a true democratization of the technology for SMBs or mid market businesses to be able to tap into their video feeds to make sense of it or to take action on what it's seeing. Right. Right. And, and, you know, companies like Apple and Tesla, there's no there's no reason why they would make, everything they have available, and they're certainly not going to, you're certainly going to have a hard time affording their ML engineers, right? I mean, those guys are paid pretty well, but now, so what's really happened is, you know, the prevalence of high quality models has increased dramatically, right?

And this means that the bottleneck in terms of getting performance at a reasonable cost is no longer the model. So when Apple and Tesla and anybody else was getting started, the whole focus was on optimizing the model. Orchestration was a second thought, and then you just quickly did the orchestration afterwards, but they weren't trying to do it in a way that others could replicate, right? Like, you know, so... But now, you have all these models, they are available to anybody, but the orchestration isn't there. And so the bottleneck has moved from model to orchestration, and that's really where, you know, we're seeing, you know, our space, all of a sudden, you see all this activity of investments or acquisitions of companies that are making AI vision platforms like we are.

So, okay. You've got... The problem before was we've got these really advanced models. We've to train this one model on how to do it, and you've to have this big machine learning team. Now we've moved to a point where all these smaller models are available, specialized models, they're available to people to use. Without a machine learning team you're going to have a really hard time stitching these together, without burning a ton of compute power, or you know, paying a big amount of money to get someone in from Google or whatever that's gonna come in and work for you. So you've got these businesses that are probably on the larger side of mid market are the ones I would imagine you work with, if not maybe even public, but like they are looking for ways to make sense of their video footage, data that is coming through video feed, whether that be through robots or cameras and warehouses or on docks or anything like that.

That's right. Yeah. And they're trying to figure out how to do it and your platform allows them the ability to be able to do that. Is is it essentially you can connect just a video source to it and it is like plug and play where you're getting a file output that you can then... Okay. Yeah. It's a... We like to say we're camera agnostic, meaning that, you know... And because you... There are there are solutions out there that you have to use a very specific piece of hardware for for the rest of your tools to work, and and and and it doesn't have to be that way. So we've intentionally built it to be camera agnostic. There's a lot more upfront work to get it there because of all the different file types and and a lot of other technical details. But... So you can come in with any camera data that you have, and it can also be prerecorded.

It doesn't have to be livestreamed. But we we support both, meaning that you can process existing data and you can process a livestream data, and you are getting a real time... You're getting a live a live response back for what you're trying to do, and and some applications don't require that, but but quite a few do, especially if you're dealing with a live camera. We have a lot of clients that they build an AI tool and they're like, okay, we're gonna build it for a specific purpose. But man, the data we collect out of this, the data that the clients have to put into this in order to get, you know, like if we're building something for a client and they're going go use it, right, and they're like, wow, we could also potentially use that data.

Right? We can learn from it, we can sell it if the clients, you know, allow. This whole industry going to be an insane revenue generation machine for anybody who has a camera already set up? If there are cameras set up, couldn't you just start collecting an insane amount of data and using that to sell to companies that want to, I don't know, market to those people or learn about behavior? Yeah, I think so. For instance, imagine you're in a retail location, whether you're in a downtown environment, and businesses might want to know, if you're thinking about opening up a location for a new store, how many people are walking by that that corner today?

And then more than that, tell me something about them. So we... One of the things you can do with our platform is advanced demographics. So this is more than just, you know, perceived age and perceived gender. This is also, are you, what type of clothing are you wearing? Are you more of a punk type person? Are you wearing fancier clothing? Well, kind of information, if you were able to have that type of statistical data, could influence where you want to put your store because of the type of clients you intend to cater to. Another version of what you're talking about that we see that's pretty interesting, so a lot of retail stores will have surveillance systems for security, right? So they come to us and they're looking to do more advanced things like being able to identify, not just if a person's there, but are they shoplifting?

Are they putting something in a bag that they shouldn't be, right? And that's useful, and they can use us for that. But then, they learn about some of the other things they can do, which is, you know, is the line in front of the cash register getting too long? Using your security camera to do operational improvements. Is your store display that you had advertising a new whatever chips, did it get knocked over? Is there a spill on the ground? All of a sudden, those types of operational concerns can be addressed, and you're using the same infrastructure. You didn't have to add cameras. Which aisles are not being walked down as much. Exactly. Yeah. So, they change up the flow of where things get stocked. And you can do one thing that's... And this wasn't something that's unique to us, just doing these heat maps of, you know, the flow of people.

And then we also have where you can do that heat map, but you're going to specifically exclude all of your employees, because with our system you can actually have it where it will recognize anybody that actually works for your company. That is crazy. And you know, a while back I was in multi location healthcare, typically private equity backed companies, we would... So a lot of de novos popping up, a lot of new locations that we wanted to build. And I remember these platforms that we would work with companies, right, that would essentially tell you where the best place would be to set up your store, and it was so non scientific, right? It'd be like, well, here's a bunch of data on traffic and all of this, but there's a Whole Foods in this shopping center.

You should be in there. And it's like, okay. So that's what you're basing it on, like some... You know, this is where... Oh, you want this type of clientele, they go there. And so can you imagine, like like you were saying, like, anybody who has these security cameras, I mean, even the government, police have security cameras up. Couldn't they turn this into really powerful data that they could resell? I mean, it just seems like a a whole another opportunity for revenue from the people that already have the approved systems to have the cameras up. Yes. I agree. Now, there is there is a challenge, though, because what you have to do is you have to, you need to decide in advance what data is you're trying to pull out. Now, you can always change that over time, but but the reason is if you... When you set up a computer vision system, if you're gonna try to have it just see everything, then you're not gonna like your bill on the on the compute side.

When you can be smart about what it is you want to understand, now you can do that in in something that is, you know, in a budget friendly way, right? You can do this in a way where it would, where it makes sense, and so, so because it's it's, you know, we we do live in a day and age where already people have information overload, and so one of the things that we really talk a lot about with our customers is let's get very specific about the data that you want, the data that's actually going to move the needle for you, and you build in some of that business logic into the AI solution itself, so then that output is is is is valuable to you. Now, over time, if you want to expand on that, if you want to run some experiments, absolutely, you know, we would encourage that, but but really try to focus on what is going to move the needle for your business, and because, I mean, honestly, is tempting when you start to start doing the, well, what if we did this, and what if we did that, and you find yourself getting excited, and all of those things would be cool, You can't do everything at once.

You need to pick something and focus and then grow it over time. Honestly, feel like it comes back to a similar problem we have with ourselves, with everyone, when we say, let's just ingest all of our structured and unstructured data, and let's be able to ask AI whatever we want out of it. When you think about the ontology and the parsing and chunking and all the things that have to happen and the data modeling in order to understand which database to go get the data from, When you have a model like an Opus or Fable or even a Sonnet and you're asking it a question and it has all of these potential options of what it could digest and give you, asking a question like what's the most important thing for me to think about today Is an insanely complex answer that will almost definitely hit a context window and give you whatever answer you can come back with, you know, before getting to the end.

So what I think you're saying is when you have a... When you are grabbing whatever it is JSON or whatever's getting pulled out of this feed, right, that from your camera, you don't want to look at it as, let's just record it, and I wanna be able to process everything and be able to ask it whatever I want. You need to layer that and do the orchestration of the models in a way that actually serves the number one focus of what you're trying to accomplish, whether that's shoplifting or this or that. Exactly. Exactly. And so what are some of the... Because I would imagine with with this type of technology you are completely industry agnostic. Your TAM is one of the biggest out there, most likely. Almost every business could use this technology in some way.

A, how are you going to market? Like, how are you actually thinking about getting in front of you going use case driven and be like, we did this thing for one so let's go market this to every other shopping center or whatever, or are you thinking about this from a perspective of whatever comes in, we'll take? And then kind of follow-up is, what are some of the great use cases you have seen that you're excited about? So, yes, what you touched on in the beginning, that is our strength and our weakness is that, you know, if there's images or video, then we can have relevance, and that is basically every market vertical out there. It is... So that obviously is a strength from a total addressable market size. It is a weakness if you are trying to... You can't market to everybody, at least not effectively, at least we don't know how to market to everybody effectively.

So this is something that we are constantly talking about, and we're constantly evolving our approach. But what we have landed on right now is two different philosophies. So one is to target not by industry, but by the customer profile. So very often, it's a software developer that would be utilizing our tool regardless of what they're using it for, and so so you try to speak to that persona to let them know, hey, this is something you can do. You don't you don't need to outsource this. You don't need to go hire a new person. You can do this yourself today, And so, and so so that's that's one profile. Right? And we... And to go to speak that out... That audience, we do...

We, we try to do this jointly with others. So, for instance, our our chief product officer recently did a workshop with Qualcomm and InnoDisc, which is a... You know, makes these AI boxes where we showed them how you can use this hardware with our software and how you... And it's and it's all engineers in the room in their in their building, and they actually build something in real time, and it was really cool. So so that that's targeting by who the user actually is, and then you have the marketing approach where you go after the market vertical itself, and so one area that we focus on is the surveillance industry, And when you think of surveillance, you know, people typically go straight to security.

That's within surveillance, but surveillance is also a lot more than that, some of the operational stuff that we're talking about. Our internal definition of the surveillance market is anyone that uses a surveillance camera system. Now that matters because there's product details that you have to take into consideration. Every surveillance system uses a VMS, a video management system, and so we're integrated in with those video management systems so that the customer doesn't have to solve that problem, they can just flip a switch and turn us on, right? So there's some technical details on that go to market that you have to call it the table stakes of the features you have to have, but in that case, we are talking to customers about the capabilities that are relevant in that surveillance space, and even there, you have to pare it down by what you do for retail is different than financial institutions and is different than manufacturing, you market it that way.

If we had an unlimited budget, then we would immediately add the media space as one that we would market toward because there's a lot of stuff that you can do if you're trying to go from long form to short form or you're trying to understand what, you you have this 100 terabytes of B roll footage and you don't know anything, and you don't have any metadata on it, you know, lots of really cool use cases, or you can do one that is like the dream application that hopefully we get to be part of this at some point, but I have two kids, and so we care about the type of content that we're watching. Well, what's inappropriate to you may not be inappropriate to me and vice versa. I would like to filter videos that I watch by content that I care about as being appropriate or inappropriate, regardless of whatever the, you know, whoever does the ratings for PG are and all that.

I may have a different criteria than they do, and I want the criteria to match me and our preferences, and you can do that with a technology like ours. Okay. So just explain to me how that would work. You have a... Like, how would you... Because you... Because what we've always... What we've talked about so far is, like, retroactively looking at what happened on a video. But if you're talking about proactively filtering and not allowing your children to see something that you would deem inappropriate, how would you put that in the middle? How would you put the... How would you move the technology to upfront? So in that particular use case, think the way it would really get done is it would be post... You process it first, but you wouldn't have to.

Like, maybe you're watching something live, you know, and... Just shuts it off. Right, yeah, you could... And this would go beyond what our platform by itself does, but, you know, we've done things in the media space where we are looking for very specific events. So, for instance, we built the ability to identify when a soccer goal is scored. For traditional computer vision, that's very difficult. We would know as watching it, but those are actions that are spaced in time, and so that's where traditional computer vision approach is having that memory and that context, it's harder. So, but they wanted to do that so they they could make sure a certain ad was being displayed within the proximity of a goal. Alright. Well, we can do that in a in a live experience.

So so now apply that to whatever it was that you didn't want to see, right? So what you're really gonna be doing is you're gonna put the video on a slight delay, so then then you could just either black out the screen or or, you know, whatever it is you want to to do to filter out or just skip over it, you know, be would be some of your options. So, okay, so that's censoring, but what about, you know, product placement? Yeah. Could you delay the video feed a little bit and put a Pepsi in a scene, you know, or or, you know, change what someone looks like? I don't... You know, there's so many cool capabilities. Guess, if if it could... If it... If you can do the processing that fast, you would then just need the video creation component to be just as fast to be able to insert something.

Right, and these days, especially if you have predefined what you want it to be, that is absolutely possible. Maybe... And you can do it by area. So, for instance, you're going to have a sporting event, and Coca Cola wants to get displayed. They bought California broadcast feed, so there's gonna be Cokes in everybody's hand, but in Texas, it's gonna be Mountain Dew or whatever it is, right? If you could segment your advertising that way, it would... And that's part of what's fun about this space is these use cases. You know, they really trigger your imagination, and it's really fun to get to be part of this journey, you know, without, you know, being the one that helps enable some of these things, and watching companies do these cool things. But we don't have an unlimited budget, so, you know, we do have some companies in the media space, but those are ones that they found us.

It wasn't something that... Because you don't have to change our product for that. You just, you know, but we just don't market toward it much today. So we talked about the cool applications. There's a... With AI, you always have to talk about the dystopian, dystopian, nefarious things that can be done with it. What are some of the things that you've heard over time where people will be like, cool technology, what about this? Is it like, is it privacy of people being on the street, or have people just said, you know what, if I'm out in public, I know there's cameras around, I understand, but like, what are the things that people are worried about with sub technology that you've heard? It's pretty much... I mean, 99% of the time, it's around privacy, and it's, you know, it's around... You know, people don't, and a lot of times, these are... Our customers, they don't want to be collecting personal, private information about companies and about people.

Like, so retail, they they... Some retail companies want to be able to recognize repeat customers, other ones don't want that information at all. Some of them don't even want to retain... They don't even want to retain the ability to see somebody's face, so they'll do face blurring. So... But that's where, on our side, we are... We've really locked everything down so that we're extremely private. We don't store customer data. We don't... And customer data is exclusive to those customers, so they can do whatever they want with it. There's nothing on our side that retains. We don't we don't... Like, when it comes to, you know, facial recognition, we have tools that can enable that, but it's all stored on the customer side and it's something that they have to enable.

So we're not the ones recognizing individual people, the customers are. You know, things like license plates being tied back to people are another example. That's personal information, and if I knew your license plate was present someplace, then I can probably track where you were at, and that's private information. And so it's an interesting landscape because these laws are continuously changing, and they're not the same. You go on the world stage, they're all over the place with Europe and Canada really locking a lot of this stuff down. But I mean, from the same perspective of, like, a company needs to understand, let's say, that a car is in their lot or whatever it is, maybe it's a car dealership or whatever.

Let's just say they want to... They're tracking the cars in their lot and you park as a customer and they may want to know, what kind of car did you pull up in? Because they wanna... They might wanna calculate the resale value or something of it in case you come in and be ready for you making it up, but they don't wanna store your license plate because any other time they could have a camera read the license plate, go look up the information and find out what kind of car you have, the make and model, but they are then storing information that is... I don't know if that would be considered PII, but it would be you can find out who the person is that owns the car by scanning the license plate. With your technology, you could also say, don't look at the license plate.

Look at the make and model of the car. Maybe look at any unique things on it, a dent here, these types of rims, and that does not necessarily tie that to the person, but it still allows them to accomplish what they want to accomplish. Yeah. Exactly. And we and we actually have a use case that's that's exactly that for a different purpose, but it's the Canadian ferry system. So BC Ferries, largest ferry system in the world. They are... You... We're going through the pilot process with them right now to do just that. Instead of capturing the license plates of cars as they go onto the ferries, record the make, model, and year, and then track them through their ferry experience. So, So, you're coming into the terminal, you get your ticket, you get on board the ship, you get off the ship, you leave the facility, and they want to be able to follow that journey, but because of some of the regulations in Canada, they don't want to do it by license plate.

And and with us, that's absolutely possible. Interesting. Okay. So talked a lot about what's possible today. It's 2026. Let's fast forward to 2030. What does the world look like for AI vision? Yeah. Well, it's already moving fast. I mean, it's really fast, and so it's just going to continue to move faster. So one of the trends I think you're going to see is, and you're already seeing this now, but it's just going to accelerate, won't, right now, a lot of companies that have any kind of notable AI vision solution, especially if it's a big company, they built it in house. They've got an in house team, or maybe they outsourced it to a team, but they're owning that product development piece, that is not efficient.

What is more efficient is to use a platform where you're not rebuilding everything from scratch all the time, and you're seeing that transition happen now. And so, what I think is going to happen is that's going to accelerate and you're just going to, you know, you're going to see that nobody's going to you're going have developers that are building on top of platforms like ours and doing amazing things, but you... And and this is important because what is... What's what's really hard or what has been really hard is there's two different there's two different expertises required to build these computer vision, AI vision solutions. One is on the AI vision side. Another is on the problem you're trying to solve.

So, when you can put the person that really understands the problem you're trying to solve, whether it's in security or operational efficiency or whatever it is, when you can give them full access to the technology, now you can unlock some really cool things. So you're so you're gonna see that in what's... It's a... And by 2030, I think it'll it'll absolutely be there. The other thing is, you're going to see more things happening on premise. Right now, the vast majority of everything is processed in the cloud, but the hardware that's coming out these days is really, really good combined with the fact that the computational efficiency is improving. Now standards are also going up, it's little two opposing forces there, but when you can move all the processing to be on premise, you solve a lot of problems, and one of them is, I mean, the Internet bandwidth to get all these data streams to the cloud.

That's a bit hard. When you can, this phase you're in of competing for, you know, server time at these large data centers, you know, that's going to still be there in 2030, and so having the ability to run things on premise is a big unlock. It also goes to privacy. It solves a lot of things. There is more upfront work to get the on premise setup happen, but that's where the advancements there are really accelerating. It's really cool. So if we have a ton of people... A ton of businesses right now hiring people to, like, build AI in their business and AI optimize the business and put their... Get their data in a data lake and all of that, we might see a shift to the internal teams being more focused on do we figure out...

We don't have to worry so much about the models and the AI enablement and all that because that's gonna be solved. I mean, every business will be using it. It's like the adoption of the Internet. Can't imagine running a business without the Internet right now, yet when it first started, the first ten years were a lot of businesses, and people were just using it for personal. Right? People were still on paper and fax and all. So you're gonna... Fast forward to 2030 because of the adoption rate is so much faster with AI than it was with the Internet. You're thinking we might see more teams that are dedicated internally just to figuring out how are they gonna do all this processing locally without... Like, on a local server versus, like, allowing it to be out in the cloud and being constricted by that. You will see that, but then you'll also see that the process of do it becomes much easier because, like, with our platform, you can deploy on on basically any hardware you want within reason.

Right? Like, if it's if you're running a, Intel, NVIDIA, AMD, Qualcomm based processor, then you build one solution, it can deploy across all of those. You're not rebuilding for each one. That's been historically one of the problems for on premise is, like, you have to customize to the hardware. That customize for the hardware, going away, and we're part of that transition. So what you're going to have though are people that really understand the problem they're trying to solve and they can customize what they want that solution to do on their own. They don't have to build the pipeline from their video source to where it's getting processed. They don't have to do any of that kind of configuration.

That's already pre built. So if you think about it, like almost every industry goes through this curve. I mean, just to date myself a little bit, when the internet, when your web pages were first becoming a thing, every business knew they needed to have one, but people were coding, they just were direct coding a website from scratch every single time. Who does that now? And there was a period in there where, you know, you were looked down upon if you started using one of these tools. I mean, obviously software's gone through this. I mean, who doesn't use a coding tool now? And so the same thing is happening right now within computer vision. If you're doing things from scratch, you're doing it wrong.

Right? You need to leverage what is out there. You'll move far more efficiently. Love it. Alright. I'm Keith Jensen, president of CADRA AI. We have Brad Chisholm, CEO of iPOP. Thanks so much for coming in. Appreciate your time. Great to be here.