Welcome to Invisible Machines, a podcast where we explore the expanding world of AI agents and conversational technologies. I'm Josh Tyson, co-author of Age of Invisible Machines, a bestselling book all about the agentic approach to automation and the pursuit of organizational AGI. Joined each week by Rob Wilson, CEO and co-founder of OneReach.ai. Rob and I wrote Age of Invisible Machines over the course of many, many conversations, exploring his years of experience in designing and researching and developing and deploying and iterating on conversational technologies. And we decided to continue those conversations on this podcast in partnership with UX Magazine because experience design is critical to the proper deployment of AI agents, or as we come to call them, invisible machines. Quick note, this podcast is available on all podcast platforms, but for today's episode, you might consider jumping over to the Invisible Machines YouTube channel.
as this is an in-action episode. Similar to last season's AI Agents in Action episode, which was incredibly popular, today we are bringing you micro UIs in action. So to give you a little background, back in the third episode of season three, RJ Owen, who's a principal product designer at Slack, pointed us in the direction of China's WeChat, a mega app of sorts that lets users do a wide variety of things using micro UIs inside of a rich web chat experience. So WeChat users are sending friends private messages, interacting with brand accounts, paying for things online and offline, calling for rides. booking for airfare, getting hotel rooms, and using mini apps without having to download separate apps for their device and without having to leave the rich web chat experience. So according to Statista, at the end of 2023, WeChat had over 1.3 billion monthly active users. And to put that in context, last year, the Amazon app had over 197 million active monthly users and TikTok had 1.2 billion. And indeed, Elon Musk told Twitter employees back in 2022 that the app that has since become X should be more like WeChat, saying, quote, you basically live on WeChat in China. If we can recreate that with Twitter,
will be a great success. So micro UIs are here and they need to be a part of your design and deployment strategy where AI agents are concerned. In this episode, we're joined by Cole Gentile, a solutions designer at OneReach.ai, who's going to show us micro UIs in action. Cole designed most of the demos that are available at OneReach.ai and many of them feature micro UIs. So today we're going to see them do things like gather information from a slider and from radio buttons, generate forms and play audio. Cole is also going to show us a newer, more dynamic appliance repair journey.
that includes a vast array of micro UIs orchestrated as part of a seamless experience. Cole built these micro UIs in action using the agent builder tool on the OneReach.ai GSX platform, which takes a revolutionary approach to software creation. This tool makes it fast and easy to create AI agents and train them around specific objectives. By contrast, the programmatic approach to software involves giving machines very, very specific instructions, which is time consuming and also leads to brittle solutions. So the agentic approach seen here makes it far easier for organizations to orchestrate AI agents.
and pursue that state of organizational AGI that we've talked about for on this podcast. As you'll see, micro UIs are actually integral to this agentic approach to software and automation. So without further ado, let's dive into micro UIs in action. All right, well, Colt, welcome to the podcast. We're really excited to have you on the show to show us some micro UIs in action today. Thanks for having me. It's great to be here. Rob, pleasure is always seeing you. You too.
Yeah. So, I mean, we kind of established in the intro how important micro UIs really are in terms of designing the kind of conversations that are really going to drive business forward. It really opens up this whole new paradigm for communication. And so today we thought we'd go ahead and walk through some different applications. To set the stage a little bit, ever since I think the popularity of Windows, you know, as a interface, they just take for granted that Windows is the best way to interact with computers, the best language, the best interface for humans to talk to machines. Most people, it's all they've ever known.
You know, before Windows, we did a lot of, you know, had spaced interfaces. And Windows just dropped the learning curve heavily on being able to operate the machine because it had these graphical elements. You know, before you had to sort of memorize what the machine could or couldn't do. And now you had this sort of menu that was put in front of you. And then you clicked on the different menu items. And so that, like, really accelerated the adoption.
and lower the cognitive load of trying to operate these things. Also, just having pictures is more effective in a lot of cases like pie charts and things. And now we see this new interface that one could argue is even more impactful than Windows, which is natural language, one we're already familiar with. And I think when I say natural language, I think a lot of people imagine pure language. But really what I'm talking about is kind of a combination of graphical elements, maybe pictures, audio, video.
graphical uis micro uis really like pie charts embedded in conversation um which is you know superior to graphical uis because even though graphical uis offer a menu um to all of us designers out there in the world that have designed graphical uis we realize they have a limitation to how many items become usable until you know trying to look at the number of items that you can access become unwieldy and then we need another app and another app and another app where feature bloat with software gets so many options that it gets antiquated and hard to use so there's this like limitation of scalability that comes along with graphical UIs that has put us in this world of, you know, a world of apps, an economy of apps, all of these apps for sale, all these different apps designed and focus thinking between apps is our life. We spend a lot of day of our day going, doing our work by jumping from app to app because of this scalability limitation and how many choices are practical in a graphical UI until they become unwieldy. With conversational UI, there is still a bit of a challenge on discovery. Like, how do I know what this thing can do?
But it's something that is being overcome with a combination of using natural language and graphical components together. So you can combine conversation with graphical elements, what we call micro UIs. And now you get the best of both worlds. You can see a menu of options and things that the system might be able to do. So you don't have to memorize what the system's capable of, but also you can find and go deep and there's no limit. And this opens up the idea that our software...
now could all be operated by one single powerful omni-UI, right? A single type of UI, conversational one. I think that even gets more exciting when you contemplate the idea that AI agents come into the mix and we have an agent that we're interacting with. So our conversation is talking to an agent that's acting on our behalf and it's operating the other software systems, our old software systems for us, so we don't have to memorize them anymore. Whether those are APIs or whether those are graphical UIs, that gets...
That gets really powerful. And now we see like, oh, wait, our new AI agent UIs and software can not only be a new way of accessing the new software, the new agentic and artificial intelligence software, but also a new way to access all of our old software as well that we don't have to learn it and we don't have to memorize it. And I think this idea of a hybrid language that is both language-based and contains micro UIs and multimodal elements. is what really unlocks the power of it. And then people might start to understand how, oh, wow, I get it. I won't use my email client. I'll just say, send Josh a message. And this AI agent will decide if an email is the appropriate best way to send it. Or maybe it's a text message or maybe it's Slack or Teams message. Like it'll decide how to get to it. And then at some point you have your AI agent and it just receives a message and lets you know what I said. And it doesn't care whether it received it.
on Slack or email. It's just watching all your inboxes. So this is all exciting. And I think micro UIs will sort of change the game and let's jump in, I guess. So as you said, micro UIs are like integral in quarter communication. It doesn't matter if it's peer to peer, peer to an existing AI, peer to a human agent. It's what allows us to communicate back and forth. The most common text input is also micro UI. It's a shame that most people stop just right there.
And of course, micro UIs isn't a new concept. It's not something like revolutionary of 2024. It's something we've been doing and everybody else has been for years. But the beauty and the glory of micro UIs now is how we can implement it into natural language models and the agent software. So just starting off with like a basic simple micro UI, we have a range slider and we can have radial buttons. And these are very useful for capturing the smallest pieces of information. Then on top of that, we have more complex UIs. And this IVR payment demo Annie Harshberger gave.
You can see in a mobile view, we have a very simple card details one. And of course, this is better than a text input because rather than just putting everything to a long line, here's my name, number, and everything, we have pre-designated areas in this sort of simple form layout. Yeah, and I think the important thing is people... like realizing there's clickbots, which appeared before LLMs, were just very strict logical trees. And you had to follow the logical tree. And it was a one size fits all UI for every experience. Every person got the same experience and it couldn't handle things like context switching or going back in the conversation or any of those things. Now what we're looking at is these UIs being conjured on the fly by an AI agent to just accelerate the input from the other side, from the human.
Kind of a big shift, yeah. And we'll get there in a second. So for right now, we're transitioning from the very simple micro UIs to more advanced ones. More advanced ones are the payment one I showed. We have this audio wheel where you can play different audio tracks. We can have upload abilities. We can also have dynamic ones, as you were saying. What if we don't want to construct the micro UI ahead of time? What if we want to gather some information and build it as we go? But what if we want something like more dynamic in style and design? What if we don't want to pre-designate all of these layouts beforehand and want it to be built as we go along?
So I can have something like this where it'll make a form for me. I'll put in, I want a patient intake form that'll get the first name, last name, date of birth, and social security number. And it'll take out all of the important information for that and build me a dynamic form. But what if we want something even more dynamic than that? What if we want it to be even more complex and customizable? This is where the content area, micro UI, comes into play. In the My AI Demo Manage IDW knowledge, we have this fully customizable panel on the right where we can dynamically populate it and design it as we go along. So this, for example, is creating an FAQ for your website.
And we have different sections, different dropdowns. All of this is just a cohesive micro UI that can be built as we go along. Similarly, in the create a contract demo, we display on the right a editable contract. And as part of it, we can edit it both inline. We can change information here, type in different things. But using natural language understanding, we can also type in additional stuff in the text box and just have it update the contract for us. And so through that, it'll add a security clause. Now, all of this is good for building complexity, but it's still a relatively dynamic or relatively static, I should say, experience going on.
While the micro UI itself is dynamic, how we access it is static. I pre-designated the paths we go down. But what if we want the experience to be completely fluid? And that's the beauty of agents. Instead of building out the entire experience start to bottom, get information A using micro UI Z, et cetera, et cetera, I can instead teach an agent what a micro UI is and how to use it. And it can then dynamically decide, it'll figure out a use case and decide when should it use this UI and how should it populate it. So for this example, I have an appliance agent. On the left, we have the normal conversation window.
And the text box, you'll see the entire conversation over here. On the right, we have an internal dialogue panel using the content area I mentioned before. Right. Right now, it's kind of sort of a UX designer on fly, basically. Exactly. And so this panel right now, it shows the agent's goal, the agent's thoughts, the task, and the tool it thinks it should use, or the micro UI it should use to proceed helping the user. There'll be a couple more fields that we'll get to in a second. But right now, the appliance agent doesn't know what I want. So it's very basic and simple. And it says, how can I help with your appliances today? And so I will tell it that I need help with my dishwasher.
And so on the right, you can see everything has changed. It's now identified the core goal that it's going to try to pursue. It's going to try to help me fix my dishwasher. And you can see its thoughts of how it's going along. There's an issue with my dishwasher. And to identify anything past that needs to figure out what exactly the dishwasher is. This is then reflected in the current task, get the user's appliance information. And it has decided the best way to get information is to use a form UI, which we can see here. Yeah, this is like Windows, like reading the agent's mind. If this were a person, it'd be like reading their mind as you're talking to them.
Exactly. We wanted to show not only that the agent can do this, but why and how is it doing what it does. It's a little window into all of his thoughts. The last thing you see here, which wasn't there initially, is what I need to achieve my goal. And these are all of the pieces of information the agent has decided is necessary to repair my dishwasher. It needs to figure out what my appliance is, it needs to figure out what the issue is, and then identify replacement parts, shipping locations, delivery appointments, credit card, and contact information to round out the experience and successfully fix the dishwasher.
So to start out, it's decided to use the form micro UI. And so I need to put in a make and model for my dishwasher. Alternatively, if I didn't want to type it in, or maybe I don't know what dishwasher I have and the label wore off, I could tell it I wanted to upload an image to the appliance and take a picture of my dishwasher. And then it'll use vision models in order to extract the information and proceed from there. But I will tell it that I have a blue line auto release dishwasher. Yeah. And by conjuring up the dropdown, we saved the whole guessing what, I don't know, what dishwashers do you guys make? I forgot my model. Like just all of that just got skipped.
because they had a selection of available models and essentially doing the same thing as graphical. It's a menu to say, oh yeah, that's it, which is it in action. Because if it was pure language, you'd have to have memorized your model and spelled it correctly and all that fun stuff. And while this information could have just used a text input tool instead of using a form one, the agent identified that a form is better because it'll make the user provide all of the information. If it just had a text input and asked me to give that information, maybe I'll just say, oh, I have a blue line dishwasher or my dishwasher is from X year.
But it decided that a form would allow it to get all the information needed all at once. Right. Alternatively, an issue is a lot more of a flexible problem. And so it just decides that the Texan put us the best. Yep. Now for my dishwasher, I'll say, uh, my bottom dish rack is broken. And so it summarizes that into the very simple broken bottom dish rack. And now it's going to go into our internal catalog to figure out what the best replacement parts are. Since I was very vague with the issue, just a broken bottom dish rack, it's decided I just need an entire new dish rack. It's the best way to cover my problem.
But if I had a different issue, maybe I said, oh, the back wheel on my dish rack is broken, it would check our internal inventory and see, oh, we actually have dish rack wheels that are placed on parts as well. And so it can narrow and enlarge the scope as needed to cover the user's issue. So it knows we have dish racks at our appliance inventory. But of course, inventories vary based off of location. So it wants to find the nearest location to the user for shipping and delivery. And to get that, it's realized it needs the user's location. And for this, it uses the geolocation micro UI.
This opens a nice little map where I have the opportunity to either type in my address or I can zoom in and out in the map and drag it around until I find where I am. Just to cover the different input things you might want. For a dishwasher replacement part, of course, you know where your house is or your store is and you'd be able to type in the information. But beyond that, what if you were instead, you had a flat tire in your car and you don't know where you are. Right. You would be able to open up the map, drag it around and find roughly where you are on the street and send it in. Right, right. So I will share this location. So it'll take the location I fed it.
and then it'll go through our internal list of stores to find the closest one to where I am. And it's found the closest one is on Colfax Avenue. If I didn't like this one, or I wanted to order somewhere else, I have the opportunity to type in, no, that's too far away, or I want to go to this other one, maybe I have a sibling who can go help me with the part, pick it up, something like that. But for this, I will just order from this service location. So now that we have the appliance replacement part type and the service location, the agent can properly analyze our inventory to figure out what parts are available at this specific location.
And it's able to find that there are three different dish racks currently in stock at this location. And for this, as I said, the best UI is the cards UI due to its flexibility of selection. As part of the cards, we can show images of the product. We can have titles, pricing information, and we also add additional buttons to have more information, such as descriptions of the product itself, which is all, again, fully customizable, fully dynamic. And so for my dishwasher, I will go with the max load dish rack.
After part selection, it needs part quantity. I only have one broken right now, but maybe I'll have another one break in the future, so just to be safe, I will order two. So now that it's identified what I want, it needs to actually get it to me. And so I need to schedule a delivery window for my dish racks, and it's decided the best tool is the appointment tool. And so I can either just have the part delivered to me, or I can have the part delivered and installed.
Of course, a dish rack is pretty simple, I can install that myself, you just slide it in and out. But for more complex pieces, for compressors on refrigerators, or door latches on a dishwasher, that's a lot more technically intensive, and maybe you would want to schedule installation. So for this, I will choose installation, and I have time on Friday at 2 to have somebody come by. Can review my information, this all looks correct, and so I'll submit my appointment. So now that it has the exact part, the part type, part quantity, and I've scheduled when I want the delivery, it can now, of course, charge me for the product.
Right. And so we have, again, a fully dynamic micro UI window to both get the user's information as well as give them an invoice and show them what they're paying for. We have the dish rack, we have the quantity of two, we have the price per part, and since I selected installation, it has to add the installation fee. Also, this is like an iframed payment, so it's not going through the LLM, it's secure, fully secure as any other payment system would be secure. And so you're not...
passing to the underlying agent your actual information, just the same as you would a website, which is success or failure of the transaction itself, which is great from a security standpoint. Exactly. And for all of these micro UIs we've shown, the agent itself isn't running it. It's just being consulted on what should be run and used at any point. Right. So anytime someone enters information, if it's deemed sensitive, that information could bypass.
The LLM is just providing the UI element to input it, but it's not actually in line receiving the text's input content of it, which is great. So you have that security measure in place that's saying, yeah, the LLM's not receiving this data, which is fantastic. Exactly, and we'll see that in a second too. So now that I've completed my payment, we will see my credit card information populate on the right. But of course, this is an LLM, and we don't want it to have all of the credit card information. So instead...
Our credit card UI runs the payment, authorizes it, sends it through, and then it submits a censored version to the agent, where it just has the last four digits and an expiration date. Awesome. And this is all the agent knows. It doesn't know what these other 12 digits are. It can't make any errors with the information if it doesn't have it. Yep. So it's very scoped and secure that way. Now that we've done everything. Sorry, you're off. No, no, I was just going to say, and even if it was an AI agent receiving this or a person.
The same would be true. It could be obscured. So if you had a human either observing this in the background or participating, they also could be abstracted from getting the data itself. So you got this UI that's collecting payment and not a human or AI agent ever seeing the data, but just the credit card provider itself getting the data. So that's cool. Exactly.
And so now that it's gotten all this other information and it's submitted the payment and scheduled the appointment, the last thing it needs, of course, is to send me my receipt as well as the confirmation. Nice. And it seems like this is easily adaptable. Like now you don't have to build one of these for every single product. This is a pattern that I think you said at transfer learning. This is a pattern that could easily be applied to different products very easily. It's not like you're rebuilding this whole experience because this experience is being built on the fly. So all you have to do is kind of provide the core elements. What is the product name? What are the parts? As long as there's access to that, then it can construct the whole experience on the fly. In a traditional app, you'd be having to create this experience pretty much for every single...
products, especially if you got into some like more diagnostics of, you know, is it squeaking or, you know, do you hear any sounds or what button is, you know, that stuff, you know, it would be pretty heavy and it changes over time with new models. So just like, it's, it's cool. It's crazy that this is better than talking to a person in my opinion, because it's got all these graphical elements. So I'd rather do this than be web chatting with someone. But also the fact that this is so easily easy to program and add and keep up to date on new products.
It just seems mind-blowing, in my opinion. Exactly. And I keep hammering this through, but the main focus of all of this, the reason why this matters is, as you said, the dynamic system of it all. I didn't build an extensive flow with very static conditions as they go on and on. All I had to do was just teach the agent how to use and look at tools. I didn't tell it, I don't have to give it an exact inventory of everything and manually search through it myself based off user information. I just need to teach the agent how to look at an inventory catalog and pull out an item. I don't need to, myself, program in doing all this of looking up a location.
and finding year ones for service. I just need to teach the agent how to do that. I didn't pre-build any cards or anything for the part selection. I just taught the agent how to pull out the pricing information of every part and it creates the cards for me. Right. And in a multi-agent experience, each one of those could have been a separate agent that could be called by others so that location and getting their location could be used for many use cases, not just product itself, but ordering the plants or finding the closest store or anything. That agent's reusable now by others in an orchestrated fashion where agents are talking to each other to get work done.
So it's a really compelling illustration too of something that we've talked about a lot on the podcast that when we speak about like the changing nature of software, this really shows you how a conversational interface can be connecting with like things that are being orchestrated behind the scenes. And here it's just surfacing relevant information as you need it. And it's also, you know, it's doing things deterministically and probabilistically. Yeah. It has a subjective of helping this person with the appliance, but then it has all these instructions, these criteria it knows it needs to meet. And then it's able to create an experience that it deems effective. And it doesn't have all this overhead with like designing, going through the product development road.
roadmap and deploying and all of that for each experience, you've got this just constantly continuous learning system that's learning the new products, changing it, reacting to feedback when it gets it wrong, self-improvement and learning, and then connecting to other agents that are doing other things to make it even more capable, which is, man, it's so cool. I just love this stuff. It's so different from software, from our old software world. I almost think of the old software world, you think about circular. product design where we look at, you know, reusability, you know, how do we make sure that the items we buy, we recycle and then can be, you know, are circular in nature versus linear where they start, you know, from raw products and end up in the landfill. You know, that's like a linear cycle versus a circular cycle. You think of software today as a linear cycle. We conjure up the idea of what process we want to automate.
And then we go through the design phase and the build and the waterfall or agile, whatever you want to use, all the way to deploying and deploying akin to the landfill. Okay, we're done. And then we work on the next version. But in this case, this is just a self-learning system. It's always learning. Once you deploy it, it's iterating on itself and it's kind of growing. And as you deploy more agents, it's just all agents are becoming more capable. It's a whole new way of thinking about software. And when you add micro UIs into it, you realize that
You could easily create an employee onboarding, for example. But now that it's so effortless, you could check in with the employee every week. There's no onboarding, we're done. They're onboarded. You have now an agent that can just stay with them as an employee, as a personal concierge to make sure that their entire tenure at the company is great, that they're always getting what they need. There's no beginning and end to these experiences. There's just an ongoing, constant evolving agent that's working in the background.
so fun that makes it really exciting from a design perspective too right i mean it's you can you can look at a demo like this and there's all sorts of amazing use cases yeah designing experiences versus drop downs you're like let's design experience is not like should this be a drop down or not yeah and like the waterfall method that you described like that's a really uncomfortable constraint for designer because by the time whatever you designed actually reaches the public it's some weird version of it in a paradigm like this like Cole's pointing out, like these experiences are being designed on the fly, collaborating with machines. Like here's another micro UI tool you can use. And here are some ways you might want to use it. And then you can see how the system runs with that and then continue to fine tune. You can do it, if not in real time, you can do it very quickly. I don't know how you guys feel about this, but I like the idea of seeing the thinking, the logic on the right hand side, like not just in a demo, but just in general, I feel like it might just become a best practice that if I'm interacting with an agent, an AI agent.
that it's transparent about the logic because there's even practical value to it to see if it understood you or to agree or disagree with its approach. I think as humans, if we could show our logic and you could read our minds while we're talking, that would be helpful. I know for me, sometimes I'm not the best communicator. If you could read my mind at the same time, that would actually enhance the communication and make it more effective. And me seeing this on the right, it helps me with just confirmation that it's understanding me, that I'm not wasting my time.
That it's, you know, it's not going to take me down some wrong road or I don't know where I'm going with this thing. I sort of like that. I don't know what you guys think of that. Like having us as part of the experience versus like just a demo. Yeah, I can see that being valuable because especially as people are getting comfortable with these systems and wanting to know kind of what they're doing, what they're understanding, what they're working with. Like it's cool to have this window that shows you here are the steps it's trying to complete. Here's what it has so far. The other reason why this window is so good in my opinion is for this demo, I just showed the happy path because I wanted to highlight each micro UI individually.
But because this is an agent, which is thinking and deciding things on its own, we're not locked to just those micro UIs. Like, the agent decided that getting the appliance information through a form and getting part quantity through radial buttons was the best way to get it. But we don't have to do it that way. With the agent, we can flexibly bounce between any of these tasks in any order we want. For example, if I'd led with, I have a dishwasher with a broken bottom dish rack, and I want to order two new ones.
it would take that in, realize I want to repair my appliance, and populate a similar list as to this. But then it would understand what I said and pull out the dishwasher of the appliance type and pull out replacement parts and pull out the part quantities. Yeah. What's also cool is it doesn't have to be one or the other. You could do both. You could offer them a map and a text box, and they could go with both. If the agent saw that that was an option, that made sense. But also later, as you correct something, you can correct it in natural language. You can say, oh, actually, I gave you the wrong address. And then it drops.
drops that component in again. And then you can reselect, or you could just, I gave you the wrong address. You should have used this address, typed it actually in, and the system will then just use it. And so this concept of UIs, you know, not having to scroll back or hit the back button five times to go back to that component that you changed. You know, man, so much easier. It's just less linear. I love this. I just, great demo. Exactly. And...
The other reason why internal dialogue, this entire panel on the right side, is so great is because of what you said. If I've gotten to the part selection part of this ordering process and I realize, oh, that's the wrong service location, I'm not near there, I can say something like, oh, I want to do this service location instead, even though we've already gotten past that part. And while that is a functionality that's not too hard to support, the importance of this panel is that I can then see and make sure that it understood me. It did actively recognize my response was not about a part selection, it realized it was about an address, and I can see that it is updating properly. instead of just completely ignoring everything I said and just trying to refocus on part selection. The flexibility of inputs and the experience is just what makes this so, I feel like the best word is just natural. Yeah, and the micro UIs are just actions in a way, like put this drop down and populate it with these elements. But it could be like send a direct mail piece, send a postcard saying get well because you use this to say, hey, I'm going to be sick today, can you cover my shift? And it walks you through a process to have your shift covered and then sends you a get well postcard or trigger some other system to operate or some machine or to, you know.
to do whatever to create some sort of action within a system that might be appropriate for whatever is going on in this interaction. So like dropdowns and using these micro UI elements are great, but they can really extend beyond just micro UIs too. Like physical things. We can see that right now even. Since it's completed all of the information it needs, it itself has already identified that it needs to send the receipt and the appointment confirmation to an email. And it's going to enact an action to send that email. Yeah. And if it's got information on you already, if it already knows your dishwasher, then it just grabs that at the top. It does that first. The agent would say, well, let me see if I already know this. And then skips all of this because it's already recognized that it has it. Just a person would. Like, oh, wait, I already got this info. All I need is your address. And you just skip this whole part.
Where people have to go fill it out anyway, like how many times do we hate that we know it? We know the system knows our address, but here we are typing it in every time And Cole, you mentioned the virtue of this being very natural. And this comes up on the podcast, but I don't think we brought it up here in that micro UIs also are reflective of the way we actually communicate in our daily lives. If I'm explaining to Rob, you know, where this shawarma shop is or something like, it's probably going to be faster for me to show them on a map in some instances than it would be for me to try to describe it. And then, you know, we talk about in the book, the idea of kind of the beyond human experience that you were hinting at, Rob. And this side panel opens up an interesting thought there too, because of what you said about, you know, being able to read someone's mind. Like this actually shows an element of...
sort of context that doesn't exist in our daily conversations sometimes we say something to someone and we don't get exact confirmation that they understood us but here is a side panel that is kind of providing that assurance which which is an interesting yeah it does get interesting in the self-learning side if you design it right so imagine it's you know you're in saying hey my dishwasher like is is broken um and you're like hey can you is there an error light yes can you send us a picture um then the system analyzes the picture realizes oh that error you know is due to x y and z um and that's using as you as you alluded to cool using imagery as a way to communicate versus text but it can also go in the reverse where you where you say hey is this the light are you seeing this light and sending a picture of a dishwasher with the light on to confirm instead of trying to like describe it in words by saying do you see a light at the top of the dishwasher three lights to the right you know that's great like you know it's just easier to just send a picture um and so if the system is making those decisions on the fly we're just getting ultimately a custom hyper personalized experience within each conversation it's like like it's writing its own application ephemeral application for every interaction it's having with someone i can't count the number of times when i've been driving my car and a warning light pops up and i need to google and flip through a manual for like 30 minutes trying to figure out what it is yeah or my i have an internet issue and there's a blade code on my router but i don't know what the blade code is just being able to upload images of all of this and having just taking a picture of your dashboard so nice all right you take a picture of your dashboard and send it in send it to the system and it says oh
it's either this this or that it can tell what the light means you don't even have to like ask it what the light is you send a picture and it's like oh it looks like you're having an issue with tire pressure and the reverse like you said where oh i'm having this issue and i don't know what it is and it sends back a image of a dash panel are these lights on are you having these issues you can even pass an audio file and be like is this what your car sounds like right now yeah i think that in health care like oh my god you know like awesome awesome so well this is great yeah fantastic Yeah, it was a lot of fun. Thanks for having me on here. Yeah, thanks for sharing. I look forward to more of this. We're going to do many more inaction. I think these inactions are great for people. Yeah, we'll keep them coming. People enjoy them as much as I enjoy doing them with you guys. Thanks for all.
Thanks for having me. All right, well, thanks for hanging out with us again here on Invisible Machines. Don't forget to like and subscribe wherever you get your podcasts. Always remember that you can watch this podcast, too, on our Invisible Machines YouTube channel. Head over to uxmag.com. We have a podcast page there where you can explore our back catalog, and you can look at episodes centered around popular topics. Many, many thanks to our incredible production team here in the U.S. and in Ukraine, Elias Parker, Michael Liptonov, Allison Harshberger, Ola Slessarenko, Vera Pakhodkov, and Darina Moskovchuk. Till next week, Owaqi.