Bonus: Claris Talk AI (the UI/UX Edition) - with Cris Ippolite and Matt Navarre
We’ll be back soon with some more new episodes! But in the meantime, we were recently guests on the “Claris Talk AI” podcast hosted by Cris Ippolite and Matt Navarre.
We talked at length about all things AI related to UI/UX, including:
- A lively discussion about how AI is shaking things up in the design world
- Insights into using AI as a research tool and idea generator
- The importance of the human touch in design, even with AI in the mix
- Thoughts on AI-generated UI and its current limitations
Listen to this deep-dive into the fascinating world of AI and design! (And make sure to follow Cris and Matt’s podcast too!)
Follow Cris: / @isolutionsai
Follow Matt:
Transcript
ClarisTalk AI.
[:No... Sure, uh, about AI and design. So how, really kind of two different areas—Matt, you had a really good way of describing it a second ago. Tell us what that is.
[:And one is obviously the concept of, uh, AI being a replacement for designers, or AI as a way of generating design. Mm-hmm, , um, is probably a, be a better way of putting it or more polite way of putting it at times. Uh, and then the other being around like, how do you as a designer actually design an interface for something that is built on AI and uses the different concepts of AI.
And so I think those are kind of the two, two large topics that like we end up getting drawn into of like, what does it mean to be responsible when you're building something with AI? And then also, what are the moral implications and other things that happen when it comes to generating, uh, generating designs via AI.
So that's, that's. That's, that's where I end up spending and reading, reading the most these days about those two.
[: [:So, I think that his applications are a bit more broad. I typically specialize right in FileMaker specifically, and so I may be using AI in a more targeted way than maybe Matt might. And so maybe that makes it more useful for somebody with Matt's set of tools or experience level. I'm not sure if that's true or not.
I'm sure there's lots of AI applications for FileMaker specifically, in terms of design for FileMaker. We can talk about that a little bit as we go through and sort of see where we might—where it might be useful, where it might not be useful, where we might need to be a bit more critical of it, and so on.
Mm hmm. I think that's a great point. Cris, we haven't heard you yet.
[:But starting off with design tools, I thought—I had this little thing I ran across, I'm just going to share it here. I thought it was just interesting fodder. So Alexis, you, you really framed it up really well. It's, um, there's really, uh, designing for FileMaker and then there's designing, right? And so, and I think Matt, you can be the voice of the designing and I don't really have my finger on the pulse of the, of design.
I'm a terrible designer, uh, Matt knows this. We worked, uh, by the way, I just want to say, uh, to the great minds in the Claris and FileMaker community, Alexis, absolutely the leader in the clubhouse, nobody does it better. Uh, I mean, just absolutely bowing, uh, to the impact that you have on design. And, and I will say that that need is, is profound within the FileMaker community. Uh, there's no doubt about the impact on good design, and nobody is waving that flag better than Alexis. So we'll, we have links to, um, your website actually, just to let everybody know, Hyperspace Data Solutions, um, along with the FM Design University, uh, which we'll wrap up with our plugs at the end and let you guys provide some context there.
Um, and the, and, uh, Matt, of course, now at Capital One, which is very exciting, but a lot of people that are watching or listening to this probably remember you from the Claris world. And, uh, I am excited to say that in between Claris and Capital One, um, we got to work together and, uh, the impact that Matt has had, uh has been profound within our organization and literally every day something that Matt brought to us and to our file maker software design process, um, it shows itself.
So, um, and, and one of those big things is moving towards a more visual approach to defining requirements. So, uh.
[:And so I think there's, there's a thing there of like, yeah, you need to know a little bit about the medium, but that said, like, I was still just using Adobe XD back in those days, not Figma to create stuff that would eventually turn into FileMaker designs. And like the, the use of these AI tools, like could similarly do that.
If you said like, here's the design system I want you to use, like we create a design system in Figma or something. And then say like, here's, I want you to create an interface using that. Like, arguably it could do that. Now, granted, you still have to then build it in FileMaker, which is what we would have to do anyway.
But at least, like, it was a way to be able to generate, you could technically generate it using these tools, but it's not, yeah, it's still a step away from where you need it.
[: I just found this thing from:And to that point, Matt, you have to understand what, what FileMaker design surface layer can do, right? Alexis, I would assume that's a really important part about approaching FileMaker design is understanding the platform's capabilities, right?
[:And so in theory, you can take these wireframes and then you can create a website out of them or a web app out of them. And so that's the real draw, I think, for most people is that I can just sort of go in theory from a sketch to like a working app with a relatively small amount of effort. Um, and, you know, whether you can or can't, I don't know.
That's still an open question in my mind of whether that's really possible. That's the promise, anyway. That's what they're kind of hoping will maybe one day get there. Um, so, as Matt was saying earlier, the drawback for us as FileMaker people is that we're going to be able to maybe do that, but then the last step, which is kind of here, click and just, send this to the web, that's going to be something where we're going to have to actually create FileMaker things, FileMaker layouts out of it.
Um, and that may be the part where, you know, if the AI tool is sort of geared towards web and you are, it's assuming that the web is your platform, it might be giving you things that can't be done in FileMaker, for example. Right? Or they're not easily done. So you have to be aware of those. And I guess that would be sort of my main caveat with a lot of these AI tools, is that you still have to have some knowledge, actually quite a bit of knowledge, I would say, about design in order to evaluate whether what it's giving you is actually good.
So yes, it can give you a result. Um, it's like kind of like some of the photos, right? Like, it can make a photo. Now, whether or not we judge that to be an accurate photo, whether the person has the right number of fingers, whether, you know,
[: [:I could, it was in English. I could read the words, but I did not understand the words. So even if AI were to give me, you know, a digest of everything that is in that textbook, I would not have the ability to actually understand it or apply it. So I think that if you're approaching these tools with some level of design knowledge of your own, and then you're going in and saying, okay, I'm going to try and have this give me something, then perhaps you can then take that and it's a starting point.
It's something that you could take and then go with. Um, if, as Matt suggested, you're already starting with a framework. Well, that's part of something that people struggle with in the, in the first place is how do I create my framework? So, you know, there's a lot of work to get there and I think it can ease the transition, but I don't know that it's gonna be sort of like a replacement.
And I think maybe that's a fear of a lot of people is that, Oh, AI is going to replace my job. But I mean, now that we all have photos on our iPhones. Well, we still have photographers. We still have professional photographers with professional equipment that are going and doing things, right?
It so it's, yes, there's more photography out in the world. That doesn't mean that the specialists are not useful any longer. And that you don't need to know something when you're approaching that tool of how to use it. Does that make sense?
[: [: [:We, you know, in the early years of building AI applications for business, the number one pushback was fear that people were not going to be needed, like humans will not be needed. So it became clear that we, as an organization, we had to have a philosophy about this. So the sort of somewhat kitschy way that I I express this, is we're not building robots to replace humans We're building Iron Man suits to give these humans superpowers. And and really that that means we're just trying to create drafts of outputs So whether it be marketing materials or actual like, you know, written proposals or whatever, or insights for that matter.
The emphasis is on, this is a draft for you to go—Now, we've, we're trying to level you up, or get you to like step three of 10 and then let the humans do what humans do well. And I think that philosophy applies, much like what you're saying, Alexis, to the design process, right? You can only really get so far before that design acumen is necessary.
[:It's like, I'm not doing the re I'm not doing the UX process. I'm not doing the Double Diamond as they call it, or whatever. I'm not doing, uh, discovery and learning. I'm an order-taker from a product team who is doing all of that work themselves and then say, here's the screen I want you to design, make it look pretty for me.
And then you're like, and that's when that, when you have that level of work, it's not very fun. And then there's this fear of like, if that's how they see us, as designers, in many cases. And then we're like, well, now I have this other thing that just will do that for me. I can just tell it to do that thing. So what, what has been interesting is seeing like, the people that I've seen that tend to go in that direction of leveraging those tools that aren't designers, it's like, I, I'm not a designer, but I want to leverage that tool, are the people that tend to have that more, that mindset of what design work is. But design work is, is much more than that.
And to your point of like, where the designer can then come in, like a designer can even leverage that tool themselves. But where the real work comes in is in the research. Is in understanding the problem you're trying to solve and understanding what is the content and what is the hierarchy of that content and how it plays together.
Creating an ERD of, of stuff or an object-oriented UX way of describing the world. And like, having that is actually where like, the work is. That's where the hard stuff is. And that in many cases, like learning general visual design things at certain cases, like if I'm not trying to break new ground and create something really like fantastic and out there, but I'm just trying to make something generally look nice? Yeah, there's, you know, those paradigms have been solved.
People like people know what a general good-looking thing is and what a good nice font is and. AI might be able to create that for some people. But, um, and so that's where like it gets interesting where it's like, well, yeah, the work is not this as much, or it can be that, like, if you really are trying to go for something, but like, I don't, I just need a base level of it generally looking nice. But like, I need to make sure it's solving the right problem and it has the right content and it's helping the person through.
[:Um, I don't, I don't think that's not something that I think you can get yet from, from like, not even close, frankly. Uh,
[:Again, I gave that FileMaker talk however many years ago about prototyping.
[: [:‘Cause again, there's, I think there's a thing I had to get over when I got into design of like the idea of like, just because you generated an idea doesn't mean it's a precious thing and you need to be willing to throw it away. And I think there's also a bit of feeling of like, well, if I didn't generate it, if AI generates it for me, I feel easier to throw it away.
And, but that kind of quick iteration of stuff could be, could be useful.
[:But we think we found a practical use case with one of our client deployments, where they basically have designers who are very skilled at creating, uh, product packaging designs and, uh, um, like marketing materials, uh, for print and online. And one of the challenges that they encountered was that when the client, when the potential customer comes in and tries to describe what it is that they're looking for, they then hand that to the human designers who go through multiple costly, right?
Just as far as like time overhead, um, iterations, to just establish where they start. Like literally like the prototype or, you know, like the starting point of the ideation discussion. So we inserted these tools—we didn't go with like a DALL-E-type approach, we went with more of a like a stable diffusion version, which allows us to actually take old designer, um, we actually can train the model.
Those are diffusion models where you start with an image or a collection of images and then kind of change it from there, rather than start from scratch So it's not really text to design. And what we found was the that that approach does not create the final output, like it's not even close to the final output.
But what it does is it helps it helps get their customers started getting the idea from their brain into something that they can now like have a more tangible conversation with. And I wonder, is there like a world of that within, you know, FileMaker layout prototyping, perhaps?
[:So that's the part that the AI really isn't doing for you. It's not, it's not able to kind of talk to the people. Um, and I mean, you're the person with it. So at the end of the day, you're the one with that knowledge that's Inputting the prompts into the AI in the first place. So you're already doing some amount of synthesis, some amount of research, some amount of work.
And I think, yes, you're correct. I can foresee that AI could reduce some of that time spent potentially in giving you ideas. Similar to like, if you're writing an essay on something, maybe you're going to go into ChatGPT or you want to research something general topic, right? But you're, those are just going to be starting points.
And then you're going to dive in on your own more deeply to make sure you actually understand those things. Presuming you're not plagiarizing something. But, you know, you're going to want to understand all of those things more deeply. And obviously for a client, it's going to be pretty obvious if you don't know what you're talking about, right?
So you're going to want to understand and know and be selecting the right things. I played around with one of the tools, uh, which was, I think it's called Uizard. It's got a weird title. It's "UI-zard". So, but I think it's named "Uizard". Yeah. And I asked it to give me like an invoicing, um, app.
[: [:And so I think the, the sort of potentially weak part of this is that, um, you know, it's, it's derivative, right? It's based on what's already there. And so if you're designing in the way that it has been trained, then then that's cool. And I think, um, um, Cris, you were talking about, you know, training it on your own material potentially.
And then you have a bit more control, I think, that way, right? Because you're kind of controlling what's going in. A lot of these AI models, it's kind of obscure or opaque about what has actually gone into training it. So you're not really sure where it's getting this information. And, you know, it may or may not be, good.
I don't know. Um, and it probably depends on the model too. If you're training your own model, perhaps the, um, there's more potential there that is going to be closer to, to what you want to do. But at the end of the day, you're still doing some work. So what is, what work is this really taking away from you? Like, what—you have to think I think critically about how much is this time really saving me? So for myself, I didn't find that it saved a lot of time. Perhaps for somebody who is just starting out, and they want to get an idea or they want to, you know, get some just, options. Like, visual options for something new.
Maybe that could work. But then there's also lots and lots and lots of inspiration that's already out there, right? You can go and just do an image Google image search for screenshots and look at all kinds of things. So there's, there's not just AI for that. You can use like the whole world, right? All of the software that's out there that's publicly available is available to you for inspiration.
So, you know, if you're not used to using Figma perhaps? I think it could be useful. But it doesn't take that much to learn Figma. So, you know, I think you need to sort of critically evaluate whether it's going to work for you. And I think you, for example, found a specific situation where it works perfectly.
And then I think those are the things you need to find and kind of go for those.
[:Then we turn those sketches—we were using, before O'Dell showed up, we were using Basalmiq and then you told us about, um, Adobe XD. Now I can't believe you're talking about Adobe XD like, it's like yesterday's news. I feel so far behind. Uh, but, you know, it really kind of moved us into what we were talking. And when we had to—we used to meet like every week and have these like, concept discussions and we—and one of the things that O'Dell helped us with was um going from a low-fidelity UI design, where the theory was, no colors, no, you know, fonts. No, nothing. Just have them focus on where everything is. And then O'Dell moved me off my mark into a higher fidelity where you're establishing an expectation of design. And that's where we switched from Balsamiq, that purposely had like these squiggly lines you were drawing on screen, into like something that looked more like what the final product would look like.
And we found that to be extremely successful. So I would, I think maybe that might be the only place like a tool like this. For those that are watching on YouTube. There's this part of the, I also don't know how to pronounce this. Uh, I think it's,
[: [:Maybe that's where something like this is useful? And I think they're the same. Oh, uh, they're the same. Um, is this other one, this Galileo, by the way, it's the same kind of thing. So I, I don't, I think it's really interesting that the point that Alexis made earlier about like, who's doing this? Is it somebody who doesn't know anything about design and they're just taking their not-knowing the design principles, and turning it into something pretty-looking?
Is it solving the problem just because it's not? You know, squiggly line sketches, and now it looks a little bit more robust? I don't know if we're if that's progress, right? So i'm not i'm not really sure.
[:Yeah, I'll be honest. I, we, we, I switched to Figma just because my company switched to Figma. And it's like, this is the tool you have to use. And so that's the tool I use. Um, I do use XD every now and again for myself, but, uh, but yeah, these days for work I use Figma.
Um, but with that, like, when you are an expert—like even seeing that example of like, Oh, I have this sketch. And then it turned into this thing. Like, okay. I took a camera—I uh—that did—how long did that save me? Like, I probably already had a template of something like that. I could have grabbed a template or something like that.
[:So yeah.
[:And I think actually like my interpretation of that in like, the FileMaker space is like, I think we've solved the invoicing systems. Like, I think we've all built enough invoicing systems that it's like, second nature to all of us at this point. Um, and so there's that kind of stuff of like, okay, well just give me something going and like, I have a base thing that I can work with. But again, most of us already have that because we've already done it. And so we already have something that's sitting right there that I can use as the base thing. And that's all that this is giving you. And so again, it would be useful for that new person who's like, well, I've never built an invoicing system before.
Well, if I do have the model trained on that—that's a big caveat from what Alexis said there—the caveat, if I have that, um, I could, yeah, maybe generate something fairly quickly. Um, but, we'll see if it, for those of us that are like, you know, mid-advanced users of this, it's, it's, it's not as useful, at least right now, yeah.
[:So you don't show up at their place seeming dumb. I use AI to, to learn. I mean, that's like maybe my most exciting use of it is, if I'm curious about something, to just have a long conversation about it. Um.
[:Uh, and so usually in the sales process, you have this one conversation where they're forgiving that you don't know anything about them. But then the, every conversation after that, you need to be an expert, right? And so we found that just, uh, taking the—literally—the original intake form, and a little bit of research about the customer and feeding it into AI tools can help us then, instead of being knowledgeable on the second phone call, we can be that knowledgeable on the first phone call. And there's these concept of trust-based selling where that little bit of additional trust—especially if they're bidding against other, you know, uh, potential vendors—that can generally, you know, move the trust back in your, in your, in your, uh, in your direction.
So there's like little teeny little tweaks where that can help. And then certainly from a coding standpoint, I mean, on our non-FileMaker projects, where it's all JavaScript and Python and code, it has a massive impact on productivity. Like, are we probably haven't had to hire additional people and still be able to grow in that area just because these tools allow us to actually do like multiples of, of a person-hour of productivity on a given day. So, um, but those are not design elements.
[:I got a question for you guys. If knowing what we all know and having talked about what we did, what would be your like ideal set of tools to be able to use AI to design things for a customer?
[: [:Um, I will, I will admit in, in this conversation, it might've even come out in like our own podcast episode about some of these topics that like, I, I, I think it's pretty safe to say, I think I'm the youngest person here. But I think I'm also turning into that, like—I don't know if you guys reached that age where you're like, oh, I think I'm getting to that point where now I can be a grump on something.
[: [:But, and like, even recently, um, at Capital One, we were talking to developers specifically about that. You know, using AI for coding purposes and like bringing in things like GitHub Copilot and stuff like that. We were having these conversations, and we were talking to some that were like, oh, this is great, I love it.
We were talking to other people that are like, so adamant of like, everything that I've seen from other people on my team generated by GitHub Copilot is horrible. And I have to redo all of their work and like this and that. Yeah. And so like, you just get this weird swath of people that are like, do I trust it or do I not?
And sometimes I have this feeling about it or I've seen bad outputs from it. And so now I'm turned off by it. So I know I'm not exactly answering your question, but I think I'm more on the side of that person. That's like, I haven't seen the benefit for my own workflow yet.
[:Do you use, you use them regularly for anything in your world?
[: [: [:'Cause I knew some people wanted it and I knew it was gonna be a part of the thing. And, but I didn't want it to be vaporware. I didn't want it to feel like the example I was using was fake. Um, so I actually wrote code in GitHub or in like, uh, in Visuals, uh, VS Studio, and I used Copilot to like generate stuff.
And then I was like, I want to use this actually as the example in my vision work to prove that like, it does this. And here's a real life example. And I'm not just like a lot of people, this happens in our design work of like, you show something that's like cool and futuristic, and people are like, you're just waving your hands, that doesn't exist. And I wanted to not do that.
Um, so I've used it for things like that. Um, I will say I don't use it for work. I think I gave this example at our podcast, Alexis, of like, for my band, I had a photo that we use for our, like a digital photo that we use for our artwork. And then we had to come up with like a physical, we wanted to come up with a physical version of the artwork, but as you all know, if you've ever printed anything, you need a little bit of bleed area.
And we literally used the edges of the photo for the photo and it was taken by our bass player. And so I was like, I can't just send this along because if it gets printed, it's going to get cut weird. And so actually I went into Photoshop, put the photo in and then had it generate like, more photo around the edges that I knew would end up getting cut anyway, but just enough for bleed so that it wasn't messed up.
But it basically generated a building because it was like a building look on the, on this photo. And so it's basically generated some sky and a building around the edges just so that I could have the right proper bleed for printing something. So those are like—
[: [:But...
[: [: [: [:Am I like, you know, low blood sugar, not caffeinated, I'm not feeling like doing this right now, but I got to get it done by the end of the day. Is this inspired work? Well, maybe that little step of just getting me from a blank page to something that I can start editing—I mean, that's how it works for my brain—um, can actually make me productive and get back into that, you know, that realm of inspired work. Whereas before I might still be staring at the blank page. So, you know, I think maybe it applies there too.
So yeah, and this is the, you're the two of your opinions on this is like—means everything to me. And I believe I can speak for Navarrere here as well, too.
And then, you know, at the beginning of this discussion, O'Dell, you mentioned that there's really two ways to look at this. So we looked at are, is there room for these tools to generate design elements. And I think we've landed on, what we're seeing right now, you know, not so much, and, and there really are these like soft elements of what designers do that are so important to the process.
I mean, we've talked a lot about UI, but I guess I would summarize it by saying UI tools would give you UI by ignoring UX. And I don't know how, I don't know how beneficial that would be.
[: [:Like, the actual UI of an AI application. And, um, I think a fascinating thought experiment for me. I don't have any of the answers. I just have questions. And that's why I was so excited to have you guys come here. But the, um, We don't have a lot of AI applications that we can sort of all point to and say, yeah, we've all seen that.
th,: So here we are in:Uh, they've realized the commodity that they have here. But what it has done is has introduced a universal interface to, um, information—and, um, and Navarre knows this—I don't think the power of AI for business is as a question-answering tool. I think that is a fool's errand. I think it is, uh, interact, getting your data, uh, as the thing that you're conversing with from a business standpoint.
So assuming it, even with that use case, this is still the interface paradigm. For those that are watching, I presented a little bit of a visual here, but for those that are listening, it's imagine in your mind, the ChatGPT, you know, interface, where you have the input mechanism on the bottom, the response or the conversation on the top. And then to some degree over on the left hand side, you see some history.
So, um, and, and to, to, to frame this up a little bit, I'd love to hear your thoughts on this. It's just like, like, wow, this amazing time, like in user interface. And then just recently, uh, before we're done with the topic, there has been a slight update to this that I look forward to getting your opinions on.
But first, like this UI paradigm that has become ubiquitous in the, you know, hundreds of millions of, of users' interactions and on a daily basis. What are we looking at and, and what is, what does this mean and where do we go from here even?
[: [: [: [: [: [: [: [: [: [: [: [: [: [:But it's based around the concept of, you know, the quote-unquote first interface us as humans had was conversation, right? It was talking to each other before we had any kind of other kind of interfaces for interacting with each other. Uh, before writing we convert— we, we talked to each other, right? And so this idea of like, a back and forth is kind of useful and like something that's innate to us as humans. And actually a lot of the work that, you know, we've been doing as, as digital developers and designers is like getting away from conversation into more of like input. And, you know, these very specific forms and parts where I'm working with, working with data. And filling in the exact things that I need to give. It's not a conversation anymore, as much as it is a, it is like a one-way, just tell me what you need me to know. Um, and so the way she was approaching the book was like, what can we learn from how conversation and the paradigms of conversation, and how can we apply that to build better interfaces overall?
And then I think from that, you can also learn again, potentially where even these ChatGPT things are doing well, and maybe not even doing so well, and need to improve themselves in terms of like, what a real conversation is. But it was a very interesting book. I definitely recommend going and checking that out.
[: [:So, you know, there—this might get more, uh, sophisticated in the future. Perhaps there's going to be all kinds of widgets, but people need to be trained on that as well. Like, so I think that they're, they're not exactly sure. I mean, we're, nobody's sure really how this is going to evolve in the future.
And so I think there's a smart thing happening, which is just, let's just keep it very simple. People understand, as Matt said, people understand conversations. We're gonna, you know, it's also novel to have a conversation that seems relatively, um, it's not canned, right? It's not pre planned responses. So that's really cool for a lot of people to be able to sort of talk about anything and get Uh, something that seems like,,,real, back.
And so it's a very simple thing that we can all sort of, the, the entry is really easy. And then I think as time goes on, we're going to start seeing different, uh, ways that people are going to be applying the concept and maybe different tools that may not be using such a simple thing. Like, for example, I have, um, an AI language, um, learning app called Jumpspeak.
And, um, it does glitch quite a bit. So I don't think it's, it's 100 percent great. But, you can have a conversation with it. And it has, you know, a bunch of different options—choose your language, and choose what kind of conversation you want to have. And, um, So there's, you know, in the background, it's all ChatGPT.
So just using an API to send the prompt to the ChatGPT model and get the answer back and display it to you. But it's different, um, like the context is made by the app. And so you as the user interacting with the app for one very specific purpose. And then all of your, uh, the app is built around, you know, achieving that goal of having a conversation or doing training or whatever.
So I think, I think maybe that's what we're going to see is that the, the main ChatGPT, if you want to interact directly with it, it's going to maybe stay simple with a few little options. But then third parties are going to start building applications. And then maybe that's what Matt wants to talk about as well, is how do we do that?
Like, um, what do those apps look like when the apps themselves are actually interacting with ChatGPT and you, the user are interacting with the app, and it's that intermediary for you.
[:Um, you know, it was all about conversation. And then sometimes you just want to say, um, give me this thing, uh, for example, um, like I want to see a, well, let me actually do...
[:Creating a UI on the right, rather than just showing you text results that you can then sculpt presumably, yeah?
[:So what we're actually looking at here now, we still get this. So we've got a drawer over here... and O'Dell, I can't make it through an entire conversation without saying instead of a hamburger menu, for example. I'm sorry Just had to bring an inside joke to the conversation. Matt O'Dell does not like hamburger menus.
But we have a drawer instead, right? So we still have that, you know, that list, uh, concept. So the, the detail now is two-fold. So it actually went to a two-columns design.
I still have my chat here, but what, what the Anthropic folks did—and I'm still getting used to this because, um, this only came out a couple of weeks ago as the recording of this.
What I was asking it to do was generate some code. And so, normally what would happen in the chat is it would insert like the code, uh, in the middle of my chat conversation, which honestly is very messy and, and just longform and it's not in, you know, visually intuitive by any means. So, what it does is it recognizes when there's things that might need to be moved off into its own screen, and it does so.
And then, in the case of writing code, it also gives me a preview mode, which technically could be a third column, right, from an interactive standpoint. And it combined those two together, and now I'm able to preview what I want, right? So, that way I'm not interrupting the user with the code or the preview itself, and I can have this kind of interactive, uh, concept.
Um, you know, I—like, I can just tell it, you know, I can do —in this case, I think what I'm doing now, instead of chatting is inputting. So on here, I'm going to say, I don't need upcoming tasks, for example, right? And now what it'll, it knows to go generate additional code. And then, you know, it should update what we're seeing on the left hand side.
So there it generated some new code. And then, you know, for those that are listening, we're just seeing a bunch of code being written right now. And then it'll, when it's done, it'll switch over into preview mode. So I think this is actually pretty cool...
[: [:It's showing you whatever your latest draft is on the right hand side. And then when you're done, you can just copy that contents. And, and like my philosophy is start editing it, not, you know, ship it from there. So...
[:You could say things like, where is the optimal placement of a button for a user that does X task in this interface.
[:And again, I probably, it doesn't make sense. I mean, I don't think it's consistent with the, how I feel about this and how us collectively in our discussion feel about this. I probably shouldn't have used an example of a wireframe for business software. Uh, this could be a document that you're creating or, or really what I think the important thing—the important way to use AI for business is to communicate with your proprietary data.
I just, I cannot say that enough. I do not think there is a place for question-answer machines like ChatGPT, just to act as a knowledge source.
[: [:As a fact, as a matter of fact, uh, my wife will appreciate me mentioning this that, um, I learned that, uh, you should not use fabric softener on towels because it makes them less absorbent. I thought my whole life that it made them fluffier. And she said, no, that's not right. And in the middle of the store, and down the grocery aisle, I asked ChatGPT, and it proved me wrong,
And so I, I'm letting her know, I'm telling as many people as I can that I was wrong. And so in that case, the question-answer machine works. I'm saying for business, I don't care what ChatGPT knows. I just care about what my business data knows. And in that regard, I might even suggest that the benefit—and I really am interested in what you guys think about this. Let me just, I'll put a little visual for those that are following us online. But, what about a more ephemeral UI approach to the interactive experience?
Let me just show something on screen, which is...
[: [:Now, the question really is, is like, there's always this, this, um, kind of overlap between UI and UX, right? And so it's really hard for people sometimes, and even for me, to separate UI from UX. So what I'm seeing on the screen is partly the UX in terms of like, what is there, where it's placed, what the purpose of it is. And then there's the UI of like, how does it look?
[: [: [: [:Um, would I personally use it? Maybe, maybe not. Uh, I don't know. It depends really on who you are and what you want to achieve. But certainly there's nothing wrong with that, and it depends how much variety you're looking for, and you know, what does the end client want, and you know, how many branding elements do you want to put in there.
It certainly looked okay, uh, whether or not it was good UX, that's a different question, I think. Because we don't know.
[:Do that. It's basically kind of, you know, a little bit of...
[: [: [: [: [: [: [:Like, let's put it this way. If I was paying somebody $200 an hour and that's what they gave me, they're fired on Monday morning, basically. So I think that's probably the best way to put it. The last topic I'll do to kind of wrap.
[:Like, I can't do that.
[: [:And for you, the time is worth it. Like, you were doing in that interface of like, or for me, I'm like, I haven't coded in a while. I don't really know all of the ins and outs of React. I need to create it as a React interface, it looked like. I need to create a React interface, all right, give me the code for that.
Yeah, you know what? That would take me forever to write that. And so maybe yes, that would get me further, because i'm not doing that in a day-to-day basis anymore. But like yeah, but yeah for the for the for the Alexis to Alexis’s example there... yeah, but if I needed to just change it in Figma or something? Like,
[: [: [: [: [: [:I would say like, this is something that I don't think you guys could do anywhere as fast as a computer could do it. Um, I would say like, here's a PDF of every single screen in my application. Um, the customer is adding a new department that's going to be using FileMaker. This is what the department does. Create screens that would, that would synthesize all the data that would be significant, that would be useful for that department.
And show me what those are going to be. And the first example wouldn't be that great. But it'd be way, way faster than what you could do. And way better than what you could do. ’Cause it would understand every single layout, including all the ones that you didn't build and all the other scripts and all the other code. That's where it's going.
That's what excites me, when, when we just saw that little thing. Not just, you know, make a, make a template screen.
[: [: [: [: [: [: [:Like, I don't know, maybe. I The thing is, I don't want to shoot you down because maybe you're right and I'm wrong. I have no idea.
[: [: [:And was it, you know, Indo-European, you know, Proto-Indo-Turkey-European that kind of, that's totally synthesis. That's not question-answer machine. And that was one of the first examples that I did with ChatGPT that made me really recognize the power. And it can do all kinds of other things like that.
I don't think I'm just, um, it's really hard to translate that to something useful as a FileMaker developer. Well, I mean, it's really, really hard.
[:Now, this is the other side of this. Instead of AI as a, as a creation tool, uh, a generation tool for UI. Now we're talking about, um, the, the experience, the user experience of interacting with AI, right? So what we're looking at here, and it's really just 20 seconds, I'm going to talk through it is, is a conversational piece.
It's very much that list-detail. We're on the list, we're actually having assistance and chat history and feedback, but what's happening in the, in the middle is I think what's really important here. So this is specifically talking to FileMaker data. This is a raw data table and the user is asking questions.
And what's actually coming back is not just answers, but datasets is how it ends here. But the thing I really want to focus on is uh, it comes back with a visualization—for those that are listening to us—a pretty sophisticated visualization that shows, you know, training over the course of a certain month.
And then below that, it creates what is essentially a report. Kind of a sub-summary, with some observations and narratives. Uh, and the synthesis of this, to speak to Navarre's point of, well, what does this all mean? And just to let you guys know what's happening in the background, is the user of this system has been loaded into this.
So it knows what role they play with the organization, what data they have access to, what's meaningful to them, what questions they've asked in the past, and what they've determined as being meaningful as well. So it's way beyond question-answering, it's actually an interface to data. So to bring this into the FileMaker, I'll give my spiel and then I want to hear what the design experts have to say about this, because I've been waiting so long to hear this.
Um, and my thought is, I almost—when I first started seeing experiences like this, my reaction was, oh my God. The hubris that I've had all these years of talking to my users about what chart they want me to, or what, you know, by the way, Navarre, this is a trigger between you and I—dashboards—right? Like, I wanted to create a dashboard.
I want to design what all the charts are going to look like, what all the interface is going to look like, what all information that's going to be on screen, all the calculations and, and what they're deriving from it. I'm making those decisions. And then we're delivering it. We're doing our best. We're talking to the users and figuring out what they need, but then we deliver it to them and they could use this thing for five, 10 years.
And the hubris? My reaction was, I had so much hubris thinking that that's going to still be meaningful to them 10 years from now...
[: [:Like every user gets a dynamic response to what the, to the data that they interact with, that isn't the same that everybody looks at.
[: [: [:And this is only one of those.
[:I really want to know is that what design people think of this type of experience and what this means to like our FileMaker, you know, inter—UI.
[: [: [: [:Um, and I think this is really helpful because you've got, uh, something that is kind of relatively difficult to do in FileMaker natively. And you can use, I'm assuming it's using some sort of a Java. It's a visualization chart in the background.
[: [:So I kind of like it from that perspective. The key observations, that's an interest thing. Uh, or if you go down a bit, I think there's more kind of like, um, focus areas, like sort of the narrative training plan. To me, I don't know how practical that would be. If somebody really knows their business, they might have their own ideas.
So perhaps this is good for a high level. That, I can't say. The part that I've felt, feel, uh, personally would be more interesting and easy would be that kind of ephemeral reporting where I don't have to build a whole module to report. I can perhaps have a tool that can kind of build whatever report they want in whatever format they want on the fly with a lot less work.
That would seem to me to be something that I could see using. Um, but I like this idea of it's not just general information. You kind of know what the input is. So it's like good information that you're starting with in the area, and I think somebody already mentioned, it's already—the KPIs are already kind of set out, so it's not just like a vast treasure trove of undifferentiated information.
It's already curated, so you already have some directional. Um, you know some direction to, to go. And it's not just kind of like, let it decide what is important to me. Like, I already have told it kind of what I want to see and um, you know, and then I can kind of give it an idea of how I want to see that. Because different information you might want to see it as a bar chart or a pie chart or a graph or whatever, right?
So there's that potential for doing that. Um, the observation stuff, I mean, maybe is helpful? But I guess it would depend on what was put in. And if you had enough expertise to put that in in the first place, then wouldn't you be able to do that from the numbers yourself? I don't know.
[: [: [: [: [: [: [: [: [: [: [: [: [: [: [: [: [: [: [: [: [: [:It should be, it should be relevant to what you do.
[: [: [:And actually, uh, I'm going to correct one thing. Cris said at the very beginning of this, I had in, no, in, in between Claris and then Capital One, I did work for Udemy for a tiny bit of time.
[: [:Um, and so we ended up and then we got a dashboarding tool that was like one of those like drag-and-drop dashboarding things called Chartio. Great. Loved it. It's fantastic. But it was this thing where, yeah, people would develop their own dashboards, and to Matt's point, like, it had the information I needed to get the things done I wanted to get done, and we could all develop our own. But there were times where you had that need for someone's like—I just, I, I do have that question I'm going to ask. And actually, this is the thing where I, I'll just—I'm going to have my own, like, pedantic argument about like—I know you're saying this is synthesis and not like question answer, but for that, for the person using this, it seems like—it still seems like question answer.
I asked a question, you gave me a series of answers. But like, in that case, like I can imagine that person being like, I have a question about our data that I would like an answer to. I don't know how to write the right SQL query to get that to happen or have a visual. And so I could see this as, yeah, as a report.
You know, one time or like, again, I might run it multiple times and say, Hey, do that again. Show me that again. Like I want to see an updated version of that. But it's probably to Cris's point, yeah, more ephemeral. It's coming. It's going. A lot of people do have that as part of their work of like, I just need to run a report real quick. And then I need to get out and I'm not going to think about it again.
And, but it's, yeah, serves different purposes. Um, but both could be useful.
[:When somebody is presented information, they're then supposed to use their knowledge and context to say, okay, now where do, what do I do with this information? What are, what are my, what's actionable about this? Where do I go with this? What literally, what is my plan moving forward? All of that logic of who, just to be clear, it's really important to understand that that is actually, should be, I'm not bragging about this, I'm saying if you're not doing that, then I don't know how much value there is. I agree with everything you guys have said.
But if you're actually going in and saying, who is this person? Let me write a multi-page persona description of who they are, what data they have access to, what their job role is, what their goals are on a daily, weekly, quarterly, yearly basis, you know, what, what interactions they have to the data, what data access do they have?
And then you're actually logging all the questions that they've been asking of this data over a period of time and remodeling that back into what's meaningful to them? Then applying that to the answer? That can be really meaningful. But if you're not doing that. I'm, I'm not so sure you're doing anything but question-answer.
So I guess what my advice to people listening-slash-watching would be, is that you need to take that extremely important step. And if you do, you can take them, you can leap over the dashboard thing. You can say, here's all my data. These are the things I need to do with the data. You know me. Tell me some examples of things that I can do with this, or where do I go from here?
And that can be really profound if done correctly. If not done at all, I think it's mostly useless. So, um, to kind of tie all those together.
[:But I think like the, the, the same comment I have against this is probably the same comment that comes up in many cases against ChatGPT and these types of tools. Which is, it actually is one of the like themes or maxims of like what makes good conversational design. So I'm going back to that book from Erika Hall around quality or truthfulness.
And, and I think where the struggle is currently, and not that we can't get past it, but the problem that I'm seeing currently is around... Is the content I'm getting back truthful? And like, how do much do I trust it or not? And like, we're seeing enough examples of like, you know, things that are being generated that it's like, that's completely not true.
That is objectively false that you have something you were telling me. And not that like, hey, if I'm feeding it, my own data, you'd think like, oh, well then that could, that needs to be a source of truth. But I think we've all been there before where us ourselves have accidentally wrote a script to find records, to run a report, and then that report that we've generated is actually not truthful because we forgot, oh, I needed to not include these records, but I have to include these other ones and like, things like that. And the fact that like, I don't see any of that here. I don't see what's happening behind the scenes, gives me a little bit of like, mistrust of how truthful is the things I'm seeing? And how much can—what do I have to do to check it?
And I'm sure you could get there. I think that that's to me seems to be one of the major, like the next major hurdle to overcome with some of this stuff. Is like, I can't, I can't judge the quality, and I don't know if there's even places where people are doing a really good job of, like, insuring and, like, telling the systems, hey, this is bad.
Like, how is that happening? So, that's my current feeling on some of it. Like, it could get there, but I don't trust it yet.
[:Trust is probably the number one word that I use in all my presentations and discussions about AI and how people should not trust ChatGPT. Just, you shouldn't. You should be wary of what you're getting back. And then when you bring in your own data and turn off access to ChatGPT's knowledge set, you should be able to trust your data.
But what we have found is that most of the development in our AI systems goes exactly towards the trust factor. How do I determine that this is this conversation stays within context of my data? How do I prove that it's actually true against the data? What aspects can we do? We actually have like semantic scoring elements on the back end that are constantly doing the pulse check on accuracy.
If you're not doing those things, you shouldn't be drawing any conclusions. Like, you're not doing those things you shouldn't leap over the dashboard into, what do I do from here? You absolutely have to do those elements. So what I can tell you is they're absolutely there are methodologies to do this Um a bit of our secret sauce here on the tools that were created I I don't know how forthright I'll be with peeling back the curtain on those things, but they are so critical. If you're not doing that, you should not trust the applications that you're, that you're working with.
And I think that's a really, really, really important message for people listening and watching to take away. So I appreciate you bringing that up. It makes a ton of sense. What a great way to put a bow on this. Um, I think
[: [: [:Um, we're going to put in, by the way, I was just adding the Erika Hall book to the show notes. Uh, so, um, I want people to check that out. I'm going to be checking that out myself. Um, let's start with you, Alexis. What do you got coming up, and things you want people to know about?
[:I'm going to be in Brisbane at ReConnect, uh, in the beginning of September. I'm actually teaching my first live course, which is a UI design workshop. I believe there are some spots still available. So if anybody wants to sign up to that, if you're going to be in the area. Um, yeah, it's going to be a long flight to Australia, but I'm super excited to be there. And then I'm gonna be, uh, speaking at, um, Barcelona, at FM Cat at the, uh, end of September. And then I'm gonna be in Rome, uh, I think Matt Navarre, you're also gonna be there.
[: [:It's an eight-module course. It was set up as weeks. So, you know, when I run through this with, um, training, I do training of teams as well. I've done Kyologic, Goya, Keeneight, a few others, um, and so we go through it once a week. But it's a self-paced, um, course. So if you wanted to go through it and just do it on your own, you can do that as well.
Um, uh, but it's eight modules and it's basically a UX design course. So the UX part of what we've been talking about. We talked—we've touched on a lot of the topics today with research and, you know, how do you figure out what you should show people? How do you understand their business? Maybe you're using AI tools to do that research, maybe not. Uh, you're talking with your customers, and how are you actually setting up your project to make sure that you have a good result at the end? So, uh, there's that. And then the UI workshop, which I'm going to be teaching, I will eventually turn into an online course as well. And that's really sort of more of the UI side of things.
So, you know, if I'm evaluating a UI, even maybe I am generating it somewhere or I'm inheriting it from somewhere, how do I even understand whether it is good UI or not? So a lot of people feel that they're not sure what they're supposed to be looking for and how to actually construct that. So that's what that's going to be about. I'm super excited about that.
And, uh, yeah, so I'll be meeting a lot of the community, I think, through a lot of these things, which is always lots of fun.
[:That's amazing. Um, and, uh, O'Dell, what do you, what do you got cooking? And by the way, please mention the podcast as well, too. And, uh, what's coming next?
[:Uh, so go and find that, uh, we're actually recording kind of our second season right now. Uh, we'll probably actually hop off of this call and go and do another thing where we'll be recording some third episodes as we speak. Um, so that, that's, uh, that's upcoming. Yeah. Everyone can design, go and find that at your favorite podcast, uh,
[: [: [: [:They're a San Francisco band. That's pretty awesome. Uh, so yeah, or find our albums on any of your streaming sites, Numerical Control Society.
[: [:It's, it's all, uh, um, people in the class sharing their screen, learning a particular topic. Dashboard, JSON, uh, optimizing for WAN performance. Those are the ones coming up. Um, those are 99 bucks. And then I stick around, uh, for half an hour, answer questions after that class. So Navarre.training is where you can learn about that.
[:You can link to some of the articles, including the "Just Ask" one that we talked about here and kind of see my thoughts on how AI might, uh, work in conjunction with business applications. Uh, otherwise give me a follow on LinkedIn. I'm trying to stay very active there as well under just Cris Ippolite. That's Cris with no H.
But, uh, other than that, I'll just keep, uh, moving forward on those platforms. And just got to say, I was really looking forward to this. I mean, shoot. I think the first text I shot off to O'Dell was like, I don't know, in May or something like that. So I've been counting the days and this was great.
Your insight is just so meaningful to me. Um, you know, I, Alexis, I, I really do, again, I'm, I'm sincere when I say, when I see something with your name on it in the FileMaker community, in particular, attached design, I, I trust it immediately and I dive right in and I think everyone else should. And, uh, Matt O'Dell, I, I miss you, man.
I, I really enjoyed, uh, talking to you every day and I'm so proud of where you're at right now, even though we gave you a virtual birthday cake that says, We Hope You Fail, uh, on your last day here. That was not meaningful at all. I'm so proud of everything you're doing and it's so cool. And I love hearing the two of you guys talk on the podcast as well too.
So I hope everybody who hasn't already checked it out and somehow found ours, uh, goes and checks yours out too. So, uh, with that said, so happy to be joining you guys here today and look forward to future conversations as well.
[: [: [:
