Just-In-Time Interfaces: Why Your Next App Won't Have Fixed Buttons

[00:00] Dave Pengelley: Good morning. Good afternoon. It's after midday, so we, we've gotta say good afternoon, right? Oh, for us, for Marno It's morning. Sorry. Sorry, Marno sorry for being east coast dish centric ish. [02:23] Matt Slager: Not a problem, mate. Not a problem. Other, other time zones do exist. [02:28] Dave Pengelley: No, no. I'm, I'm, despite having grown up in Perth, I'm officially Sydney now, so I'm like Sydney or nothing. [02:38] Dave Pengelley: yeah, sorry, sorry. Brisbane, sorry. South Australia. Sorry. Perth. we tolerate Melbourne. I just, [02:45] Matt Slager: they're alright. I can smell them from here. [02:47] Dave Pengelley: Right? You're like, you're like on the, on the borderlands. [02:52] Matt Slager: Yeah. Yeah. Right, right on that edge, on the, on the side. You gonna see [02:55] Dave Pengelley: Victorians all the time. Ugh. [02:58] Matt Slager: It's actually quite funny. [02:58] Matt Slager: We, we call, so where Aubrey and Wodonga is, we call Wodonga, like Mexicans, like No, but you know, it's just like over the border. [03:10] Dave Pengelley: Yeah. even up in Sydney, we call people from Melbourne up the place I've worked Mexicans 'cause they're south of the border. [03:16] Matt Slager: Yeah. Nice. [03:18] Dave Pengelley: Yeah. Purely a geographical distinction. No other commentary on [03:24] Matt Slager: similar to where I grew up, like Northern Beaches of Sydney, where, you know, on that peninsula it used to be called the Insula Peninsula. [03:31] Matt Slager: And my mom would always describe everything that was west of that zone. So the entire rest of Sydney as Westville. So like when you are Dave, she would be just like terrified of [03:44] Dave Pengelley: Oh, I'm like, I'm like, I remember coming here and, the, the jokes were about, when you went east of like Epping, you had, there was like dumpster bins to drop your rug boots off in. [03:53] Dave Pengelley: Yes. Like that hole, class warfare, still alive and well. have we had a, had a good week, gentlemen. [04:04] Marno Brits: I have. Yeah. It's been busy. It's been absolutely, like I've got TAFE Monday, Tuesdays, and when I have tafe then it's just, I've got a lot less free time. So the morning is gym work for an hour. TAFE on my lunch break, it's a meeting and then go back to tafe. [04:21] Marno Brits: And then after tafe it's another meeting and I'm like, oh crap. Like it's just back to back. So Wednesday should be chill. It's content day. I've got one AI audit and I get to relax. So it's been a great week so far. [04:32] Matt Slager: And for, like for people that don't know, are you, you're studying at tafe? [04:36] Dave Pengelley: Yeah. Or teaching. [04:38] Marno Brits: Teaching at tafe. So I teach, Microsoft Excel and Microsoft Office at TAFE on Monday, Tuesdays. [04:44] Matt Slager: Huge. [04:45] Marno Brits: Yeah. That's pretty fun. I enjoy it. It's more of a, a passion project than anything else. but it's a great skill that I want to get better at, so it's a cool opportunity and it's literally 500 meters away from my house, so I get to walk there, which is nice. [04:58] Dave Pengelley: Are are you doing any stuff there with like, introducing the co-pilots and stuff like that in it? Or is it all just basics? [05:04] Marno Brits: Definitely, yeah. I mean, I, this is what I live and breathe. So even when I'm talking about Excel, I'm like, well, if you don't need me, you are co-pilot, you copilot lens, or you could leverage like gamma if you wanna do presentations. [05:15] Marno Brits: but then I just did work with AI or with the TAFE to start delivering AI on the first Friday of every month. Then I'll do an introduction to ai. still trying to build some more momentum around that, but, at least it's been approved. So just been good. [05:30] Matt Slager: That's cool. [05:31] Marno Brits: Yeah, [05:32] Dave Pengelley: heaps [05:33] Marno Brits: awesome. It's been awesome. [05:34] Matt Slager: Are you finding people that are like. Well ahead of where you would expect them to be or like not even experiencing anything. [05:41] Marno Brits: Oh, not even aware of what Chachi BT is. [05:43] Matt Slager: Yeah, [05:43] Marno Brits: like honestly it's been, it's been a blessing and a curse 'cause I'm like, okay, crap. Like the content I made isn't really for you yet. [05:50] Marno Brits: But it's refreshing to know that in my brain I'm so far behind 'cause there's so much else going on and people are building massive companies. Meanwhile, like if I just look at my own like location, it's like, no, no one else is doing AI and category. I just have to be able to guide them a step ahead of like, cool, you don't know what chat BT is. [06:09] Marno Brits: This is all the models you can use. This is how you do custom GPTs. And I'm still able to provide a lot of value to them and just change their like mindset. And that's the coolest part. Like that initial shift from, oh life is so difficult to, wow, I can now do so much more in half the time because I can leverage ai, which is the best feeling. [06:29] Matt Slager: That's cool. [06:30] Dave Pengelley: Do you guys find, like it shocks you how little some people know just about the general terminology and lingo that we take for granted. Like I was talking to, I, I was at karate. I was talking to one of the young blokes who was just starting year 11 this year. He goes to my, my son's school. [06:43] Dave Pengelley: He is a few years ahead of him. and I said, oh, what are you doing for your, your electives? And he said, computer studies, blah, blah, blah. and I went, oh, that's great. It's really good. even with all the vibe coding to understand the basics of engineering and database design and so on. and talked to, you know, kinda the SPI we were talking about, you know Yeah. [06:59] Dave Pengelley: Players and architects and stuff and just give him that. And he is like, what, what, what, what's vibe coding? And so like, I thought that was just general terminology that everyone knew about by now because it's 2026 and we've been talking about vibe coding for like a year and it's just still this, this unknown thing. [07:17] Dave Pengelley: yeah, like. People don't really know what that is. Even I said to my son, oh, such and such didn't know about Vibe coding. My son's like, what's vibe coding? I'm like, oh my goodness. What? [07:26] Marno Brits: Yeah. I think we just live in bubbles, man. Like e everyone has their own little bubble. I'm not surprised by it as much, but it's something that I, not that I'm shocked, sorry, but I am, it's, it's rewarding to see that again. [07:40] Marno Brits: I was like, okay, cool. Like, I'm doing the best that I can. I'm learning as much as I can because in ai, like I would go to sleep and there's 20 other news stories that I missed. So like, there's so much going on. But I feel like because we live and breathe in this world, we are so close to it that we don't really see anything else. [07:56] Marno Brits: But like, I will talk to someone else and all they're talking about is the price of gold and like the labor shortage or like not being able to find accommodation. I'm like, I'm aware of it, but it's not really a problem. Like AI is the problem. [08:09] Dave Pengelley: Pri price of gold is mental. [08:12] Marno Brits: mm-hmm. We're actually doing a petition. [08:13] Marno Brits: There was a Facebook chat, like everyone in Kholi, Facebook is their place. Like there's no real website to stuff or Facebook. And someone said, let's on, let's talk Kholi. We should get a, live gold ticket again on the exchange hotel. Like historically, that was a massive ticket there. The gold, oh really? [08:30] Marno Brits: Gold price and live. And it just gave like the city a heartbeat. And it got enough, support that the owner of the exchange saw, and he is looking into it to see if we can get a new live tracker of the gold price on the exchange, which is the main intersection of Klee. [08:46] Dave Pengelley: That's [08:46] Marno Brits: pretty cool. It's pretty awesome. [08:49] Dave Pengelley: Yeah. And then, you know, the, the amount of screen and, and access technology is not super hard to rig that up either. I wouldn't [08:55] Marno Brits: imagine. Yeah, you'd imagine so. Oh, hopefully we don't, like, we might not have the specialist in Gali, so flying somewhere down from Perth, but [09:02] Dave Pengelley: you need an external LED screen and a trading view widget. [09:06] Dave Pengelley: Like, [09:06] Marno Brits: you got the knowledge. I've no idea how to touch that, how to touch anything of av. [09:11] Dave Pengelley: yeah. Anyway, I've been to kgo once. I went there for a school camp in, when I was in like year five or six. Mm-hmm. I came home with like 50 mosquito bites, but that's my biggest memory. [09:24] Marno Brits: Understandable. [09:25] Dave Pengelley: Yeah. Yeah. [09:27] Dave Pengelley: you've had a good week. Matt, what, what, what, what's been making your week good? [09:33] Matt Slager: honestly, it feels like the last episode was yesterday. So as the saying goes, you know, time flies when you're having fun, so clearly. Yeah. I must've been doing something right. that, that talk of mosquito mosquitoes, like shows my ignorance. [09:48] Matt Slager: I have like, my understanding of Kal gli, just to really briefly touch on that, is that, it's similar to my town, which is kind of rural, but you replace the trees with red dust. And I don't expect to like, hear of Mozzie like in that situation. so that's, that's curious. [10:08] Marno Brits: Yeah. It's a good setup. I'm keen, I do a park run. [10:11] Marno Brits: I'm trying to do a park run every week, and I'll send you a photo. It's one of the most beautiful park runs, in my opinion. [10:17] Matt Slager: Nice. [10:18] Marno Brits: and you get to the top of, not really a mountain, it's, it's an incline. I wouldn't even call it a, a cliff. It's an incline. but once you get there, you get to see most of kgo and you get to see all of the trees and it it for a second there. [10:32] Marno Brits: You're like, what the hell is this town? Because when you're in the town itself, all you see is red dirt. It's just mine vehicles on high highs and people drinking beer. And then you get up there and it just looks like almost like a rainforest. It's just as far as the eye can see, it's just trees. [10:47] Matt Slager: That's cool. [10:47] Marno Brits: it's pretty spectacular, especially the morning, sunrise. It looks like heaven. [10:52] Matt Slager: When you, when you said you get up there and you can see all the trees, I was thinking like. Six of them. Like, no, [11:00] Marno Brits: there's quite a lot, there's quite a few [11:02] Dave Pengelley: CY O'Connor of built a pipeline, so they got water out there, mate. [11:04] Dave Pengelley: That, that's fine. [11:06] Matt Slager: No. Awesome. Yeah, no, my week's been good though. I, I've been trying to tinker as much as I can, but yeah, client work is absolutely still keeping me underneath, trying to keep breathing at the same time as keeping the, I don't know, the light switched on. So I've had a little tinker with open core as I guess most of probably the world has at this point. [11:26] Matt Slager: I don't know, that's probably my, my bubble again. and yeah, the, the, the crucial thing for me, like I, I wanna get that thing, that's the, the one click install that, you know, like the listeners, the viewers can be, go to the link that could exist in the description and here you go, just plug it in, test it out. [11:44] Matt Slager: But for me to do that, I have to do it responsibly in the sense that. It's completely safe, you know? [11:49] Marno Brits: Yeah. [11:50] Matt Slager: Because the, the, just on on that really quickly, that, that concept of like safe versus not safe, here's a really, really good analogy that every single person will understand. You know, when somebody starts at a company, they're fresh, they start on the ground level. [12:05] Matt Slager: they don't get the keys to like the manager's office or the passwords to the backend. You know, like there's reasons why that kind of level of access is split across different roles and different levels of trust, you know, and that gets built up over time [12:21] Dave Pengelley: called SOC compliance, I think. [12:23] Matt Slager: Yeah, a hundred percent. [12:24] Matt Slager: So with that kind of stuff, you know, people are expecting that kind of level of tiered, you know, ladder of trust to an automated agent that they just give all of their access to instantly and expect it to be safe and secure. So, yep. [12:40] Dave Pengelley: Talking employees too much access. I remember when I started at Salesforce back in 2011. [12:46] Dave Pengelley: There were always the horror stories of, account execs and people that went over to bootcamp in San Francisco with their brand new company, Amexes and just blitzed them. and then, you know, they'd obviously get fired, but because of the, the, the effort to try and chase them down to try and re recoup those costs, it was basically a write off. [13:04] Dave Pengelley: So these employees just went on these massive spending sprees with their company Amexes. [13:07] Marno Brits: Wow. Wow. [13:10] Dave Pengelley: yeah, I mean, I tell you what, compared to when I was a Telstra and if you wanted to go see a client back in the day, we had to like apply and like, you know, give blood samples and whatnot in order to get a cab charge voucher just so we could go see someone. [13:22] Dave Pengelley: When I went to, when I went to Salesforce and we had like an amex, we could just like catch cabs and go do things. [13:27] Marno Brits: Mm-hmm. [13:28] Dave Pengelley: Amazing. So yeah, managing can be a, a productivity booster, but AI is everywhere. Have you seen Alexa plus announced. [13:37] Marno Brits: No Native keen on that. What's up? Well, [13:40] Dave Pengelley: yeah, so like Alexa, who's been in, you know, the, I gotta make sure mine behind my monitor here doesn't start talking to me. [13:46] Dave Pengelley: I've gotta be careful saying her name. they're bringing out generative AI into [13:52] Marno Brits: finally [13:54] Dave Pengelley: into the, the, the, the thing, whatever it's called, the mm-hmm. Smart speaker, [13:59] Marno Brits: the [13:59] Dave Pengelley: hardware. I haven't read deep into it. I saw one of the pre-up bowl ads was a Chris Hemsworth one where Okay, they were, they were pushing it, obviously playing on the fact that he fought Ultra on, 'cause he's like, no, it's an ai. [14:10] Dave Pengelley: They didn't make any specific Avengers references, but I think that was just playing on that whole, ultra on type thing. But, this is their browser version. but apparently it's gonna be on their smart speakers as well, where they're gonna take it from beyond that sort of very deterministic. [14:27] Marno Brits: Mm-hmm. [14:28] Dave Pengelley: listening and answering patterns to integrating. A lot more stuff. so yeah, recipe discovery, all that kind of stuff, which is great because we actually have one of the echo shows with the screen on it, on our kitchen bench. Mm-hmm. Which we use for our shopping list and all kinds of things. So being able to do recipes on that will be excellent. [14:45] Dave Pengelley: Rather than having to get my phone and like my phone keeps turning off and just being able to ask it for dynamically creating recipes, not just what's on taste.com. [14:54] Marno Brits: Yeah. [14:55] Dave Pengelley: should be pretty cool. So anyway, this is a new thing. The fact they've got a web interface now. That's a whole new thing that I didn't even know about until I just looked this up. [15:03] Dave Pengelley: Mm-hmm. but yeah, Alexa Plus is coming to their early access things, so you know, people are gonna get more tastes of these AI things. Just it's, it's just filtering through Right. Everywhere. [15:17] Matt Slager: Yeah. A hundred percent. Interesting. And each thing like, you know, I, I, when I try and explain stuff to people, I try and speak of like fundamentals first, principles, primitives, you know, like the theoretical groundworks that you need in order to then stack your Lego blocks on top. [15:32] Marno Brits: Mm-hmm. [15:33] Matt Slager: And that there is just another really good example of Angen system. You know, just an agent like that has access, has some sort of level of access, some level of trust. You know, you're giving it access to some things there. Clearly it has an ability to access a database or some list where you and [15:51] Dave Pengelley: physical devices, like it can turn my lights on and off. [15:55] Matt Slager: Yeah. Well actually on that note, you know, if your most smart devices around your house. Have completely like open, communication protocols, without going too, like nerdy technology language here. Like my, I got recently a couple of split systems installed, thank goodness. Before there's like 45 degree weeks. [16:16] Matt Slager: Yeah. And I found out that I can literally jump into Cord Code or any agent for that matter and say, find my air conditioners and connect to them. [16:27] Marno Brits: No. Why? They're, [16:28] Dave Pengelley: they're on ZigBee or something like that. [16:30] Matt Slager: Well, it's just some sort of like open source, open [16:32] Dave Pengelley: standard [16:33] Matt Slager: communication thing. [16:34] Dave Pengelley: Yeah. So a, a lot of the, all the Philip Hug stuff uses the ZigBee protocol. [16:38] Dave Pengelley: So there's a few open standard protocols that they use. There's that one. And then a lot of them are using a wifi technology now as well. [16:45] Matt Slager: yep. [16:45] Dave Pengelley: Like all eugenio things, so. Mm-hmm. what I really liked is initially when I started, you know, smart lighting my house, it was all hues because that was kind of the main thing out there. [16:56] Dave Pengelley: But then, you know, you want to go to Hue app, you gotta do that. And then as I started out bringing out different brands, you were kind of tied into Hue somewhat, but then when the echoes came out, echo acted like as this or this, this bridge layer for me. So it didn't matter which of the smart house systems I had, I could start mixing and matching and, and going for off-brand market sort of thing. [17:17] Dave Pengelley: As long as they connected back to the Echo platform. I can even in this room, like I've got like Genio lights and other lights and hue lights, like the bookshelves are a hue light, but the one hitting me on the back and my front ones are all controlled by different sort of mirror bella type stuff, but I just talked to the speaker and they can all change, all at the same time because it's kind of bridging. [17:39] Dave Pengelley: And so I think the technology doesn't matter. It's, it's the interface and we're gonna see a lot more of that. [17:45] Matt Slager: Yeah, that actually, instantly reminded me of, an interview I saw with the open call creator, Peter. He. And actually Elon as well. Elon Musk said something like this too recently. So the concept of having discreet applications, you know, separate things that you need to go to, like go to your phone, open up that app, and then do the thing that is gonna go away pretty quickly. [18:09] Matt Slager: And so, [18:10] Dave Pengelley: mm-hmm. [18:11] Matt Slager: So is the concept of actually writing code, you know, like the fact that the code exists, if you think of it, like, you probably know more about this with the CS background, Dave, but the, the concept of like going from, you know, machine language to assembly to the binary, to then like the abstractions and then the frameworks and codes that we put on top of it to be able to write stuff as a human. [18:35] Matt Slager: Yeah. And then now the agents are actually writing that human code framework. Why, why are they using the human thing? Why don't they just write the binary directly? [18:44] Marno Brits: Yep. Interesting. Yeah. De debugging it. Wouldn't that be a pain in the ass if you [18:49] Dave Pengelley: can't? Well, the computer's doing it like that, and that's the thing, right? [18:52] Dave Pengelley: Yeah. So why use our, and, and you've seen that on the, on the, the malt book or whatever, where the agents are sort of saying that Why are we using inefficient, human languages and Well, it's because they're trained on that. They don't actually have their own reference and context to build, to know what another language is necessarily. [19:07] Dave Pengelley: Mm-hmm. Because they're not actually creative intelligent beings. They're just parrots really, really good parrots. but [19:14] Matt Slager: another cool one like this is really hard to fathom because it doesn't really exist yet, but you know, in kind of sci-fi esque movies where they pull out their tablet and it's just a slab of glass, right? [19:25] Dave Pengelley: Yeah, yeah. [19:25] Matt Slager: Sort of transparent thing. But then an interface pops up and that interface may not be a predefined. A piece of software, you know, it may not have the same layout of buttons or the same thing each time. You know, a, a simple, simple example is imagine you walk through your house, and I think a lot of people would be familiar with the kind of like smart home climate control things like the Google Nests and stuff like that. [19:50] Dave Pengelley: Mm-hmm. [19:50] Matt Slager: Imagine that screen that you see changes contextually based on what you need to do to interface with it. Like, you know, it wants to ask you a question, yes or no, so therefore it puts up two buttons, yes or no. [20:04] Marno Brits: Ooh. [20:04] Matt Slager: But like, it doesn't, it doesn't, it's not deterministic. It could be anything. It could just be some way of you interfacing with the thing. [20:12] Matt Slager: That kind of thing is really fun to me. [20:14] Marno Brits: I wonder if it'll be get to the point where like I'm always thinking like a pip boy from Fallout, but then you'd have that and then if that has a location tracker in it, if it's dynamic, depending on where you are, if you are at the office, does that give you then a different interface than if you're at home or if you are like. [20:31] Marno Brits: On the tv and then you're trying to control your TV room, like in your, in your dining room. Does it then give you options for like the lighting again? That's a very, very interesting thing to play with. Yeah. But do you think that's like two years away or 10 years away [20:45] Matt Slager: that could be done this year? Easy. [20:47] Marno Brits: Really? [20:48] Matt Slager: Yeah. Contextual interfaces like that kind of already exists. Like if you think of workspaces in your current operating system or browser, you know, you can change your profile and that changes your entire, config. [21:02] Dave Pengelley: The iPhone has that as well. Like you can set different focus modes and that changes your whole, whole thing. [21:06] Dave Pengelley: And so that dynamic thing, I remember, Like looking at, like talking about software interfaces and things. And it was like for individual users when you go in, rather than go, I need to go to this tab and click on that thing, it's like, type in your request. This was before the LLMs were really a thing. [21:20] Dave Pengelley: But like future thinking that concept and it dynamically pulls together a dynamic dashboard of widgets based on your request on that given day. So rather than having a standardized fixed set of page layouts of this is how you do your job, now the danger there is sometimes having some fixed layouts is actually good for human psychology and being able to be efficient and process things rather than trying to relearn the whole interface every time you wanna do something. [21:44] Dave Pengelley: So I think it's gonna be interesting to find that balance, but as we move to more voice interfaces and for decades we've seen in Star Trek, The original series, but more recently, like The Next Gen and, and some of those ones, which I say more recently, but I realized Next Gen came out in the eighties, with Picard. [22:03] Dave Pengelley: But I mean, that's where we started seeing things like the, the Holodeck and the Replicators and those really sort of next generation technologies that they, that they, they were sort of talking about and it was all voice coded. Like they, they'd all talk to the systems. and it was, there's always been this thing and, and like, well, if I'm talking to my system, how does that work in an office environment? [22:21] Dave Pengelley: Where's the privacy? How do you do that? and so it's gonna be interesting to see how these voice interfaces evolve versus the, the chat bot text interfaces. But generally speaking, the fact that, you know, I wanted to build, and maybe, maybe if we've got time today, I'll, I'll just whip up a quick demo of this, but wanted to build a price calculator. [22:43] Dave Pengelley: Forecast thing. And before I would've had to go to Excel and spend half a day building that, but in a couple of prompts, I built a little single page app html page with sliders that dynamically showed me pricing levels that was built like that. Yeah. just those little things. So building those custom tools on the spot on demand, just in time, interface creation. [23:02] Dave Pengelley: There's, there's definitely some huge value in that. A [23:05] Matt Slager: hundred percent. I, I think the, the most, relevant to most people listening or watching would be the, the idea of developing your own unique branding. You know, like if everyone can just spin up their own dashboards and vibe, code, whatever, like simple applications that they want. [23:25] Matt Slager: They probably still wanna personalize it. Right. You know, and most of the, the personalization options of the past is, you know, change your color, change your icons, change your, various colors, like borders and backgrounds, and put pictures on things like I, I'm picturing back in the day, like using MySpace and customizing your, your MySpace page and that kind of thing with regards to branding, like your own, you know, theme kit, will be probably the most relevant. [23:55] Matt Slager: So imagine like they go to a, their page and they can just a choose between different style elements. They like go through maybe like 30 options and then there they go. There's like their perfect style kit. That would be pretty fun. [24:10] Marno Brits: That'd be awesome. I'm just like, what's the, not necessarily the business application, but is that something that you can see yourself wanting to build? [24:19] Marno Brits: Or just wanting to use. 'cause now, initially, initially my brain's like, cool, what's the SAS look like there? What's the, what's the next software that is us selling shovels to the people that want a good dig. [24:32] Matt Slager: Yeah, a hundred percent. I keep debating with myself about this. Like, do you build a thing and commercialize it or do you just teach people how to build it? [24:39] Matt Slager: 'cause everyone can build it. Yeah. [24:42] Dave Pengelley: And not, not everyone will wanna build it. Not everyone has. And we still have this thing, which I remember, you know, was, was a catch cry back in the nineties. oh, I'm, I'm computer literate. And, and people just sort of put that blinker on and they refuse to learn. because it's like, oh, I'm computer illiterate. [24:56] Dave Pengelley: And it's like, yeah, well you are mechanic illiterate too, but you learn how to drive a car. there's, there's that whole concept where some people just, they, they don't want, they just want it handed to them. They want it done for them, which is okay 'cause they, they're probably really good at other things. [25:10] Dave Pengelley: Yeah, I would hope. And so not everyone has to build everything all the time. That's that whole built versus bought argument. and whatever it is, like there's always the argument, do I just go and buy a desk from Ikea or do I go down to Bunnings and buy the equipment? And it turns out it's four times more expensive to buy the pieces of Bunnings to build it yourself, and it's gonna take you 10 times as longer. [25:31] Dave Pengelley: It might last a lot longer. So that, that whole cost benefit. Where does my time better go? Do I go and get the thing from IKEA that I'll have to replace in four years, for half the price? Or do I spend twice as much and eight times as much time building something solid from Bunnings? Hopefully solid. [25:47] Dave Pengelley: If I build it right, that'll stand the test of time. So it's, it's, whether it's physical things or AI things, build versus bought is is always a debate that people are gonna have different needs at different times on right. [25:59] Matt Slager: My, the thing that kind of spun me out recently, it was one of the guys that I respect a lot in this space. [26:05] Matt Slager: You know, way up there in the idea of software engineering, gen engineering, all that kind of stuff. He basically was saying that he doesn't use and doesn't respect the concept of open source anymore. 'cause if you can just replicate libraries, you know, at that tier and then create the nice packaged product that's time efficient for those that don't want to delve into it, then why wouldn't you? [26:31] Matt Slager: Like I, I experienced this in my own production project recently. I had to do some some PDF conversions. So going from PDF to image content. [26:40] Dave Pengelley: Yep. [26:40] Matt Slager: And then using the image content as image in LLM calls because image content with regards to like. the, the comparison between pixels and tokens when it comes to like LLM efficiency pixels are actually way more efficient. [26:58] Dave Pengelley: Really? So OCR is better than, than than tokenization. [27:01] Matt Slager: Yeah. Yeah. Yeah. There's there's full papers on that right now. So, when, when the last deep seq OCR came out, it's on V two now, but when the first one came out, there's a, an inference company in, San Francisco called Base Tenco. They released a paper to do with the deeps OCR being like 10 times more efficient. [27:21] Matt Slager: anyway, wow. So I had to, I had to find a library right to, to Concur to convert the PDFs to image, and there was a really, really good one, but it needed a commercial license and that license was like two grand or something. And I'm like, well, I'm gonna keep exploring and find one that doesn't require that. [27:41] Matt Slager: But I had that, that software developer at a Gentech engineer guy in my head saying, why don't you just replicate this library now? And. Curious. Hey. Like the, the kind of thing where you can absolutely sell people shovels, like Marno said, and then there's the people out there that look at the really good shovels and go, I can make that too. [28:01] Dave Pengelley: Yeah. Right. And I think I mentioned last week, I can't remember what I was talking about, like the SaaS that I'm building for, for traders and my son's like, what you gonna charge 'em for those features? I was like, well, yes, because I'm building those features and it's my time and effort and, and knowledge going into it. [28:13] Dave Pengelley: Mm-hmm. And if they don't wanna pay me, they can go build it themselves. Like that's, that's, that's the thing. Right. but as far as on, on demand, just in time stuff, things like open core. Are gonna move that needle in the future. I think, as these bots and agents become better at just taking simple commands and doing stuff and acting autonomously, like I'm seeing a lot of these sort of posts coming up on LinkedIn at the moment, if you'll indulge me. [28:41] Dave Pengelley: we built our startup on, we built our startup and AI chief of staff with OpenCL. Here are our top three learnings, memory first, automation second. So this is that context and we always talk about. Mm-hmm. So it's like making sure we've got really good context in there. Give a daily log, long-term memory, file pricing stack, GTM, key customers. [28:58] Dave Pengelley: so after we can stop asking what's our pro plan and start out answering who should we follow up with? a heartbeat, not a fire hose. So rather than have it sort of coming at you with constant streams that you just can't manage and become spam [29:09] Marno Brits: mm-hmm. [29:10] Dave Pengelley: make it sort of, you know, do, do updates. Every 30 minutes they found was really good. [29:16] Dave Pengelley: and delegation via subagent. So they're running their own little Chief of Staff thing going on there. I found this other guy, I don't really know him, Brett Raven, I may regret this four months ago, but an a IC suite, CEOC mobile, blah, blah, blah, plus specialist agents reporting to them. I've just given them a spending budget guard rail and access to open claw. [29:36] Dave Pengelley: So each C-Suites agent spun up three open claw instances in the cloud. Every instance got its own email, phone number with WhatsApp and a thousand dollars Visa debit card. so now he's, he's just sent them off to go and do a thing, and I'm seeing other people saying, yep, I'm given four claw bots. Budget and send them off to build a business. And I'm gonna see what happens. And this is gonna change how I like scale startups forever, if it works. So lots of people are doing some really interesting experiments with 'em actually putting some real money on the line to see what they can actually do in the real world, obviously with some guardrails and, and precautions and so on. [30:09] Marno Brits: Mm-hmm. [30:10] Dave Pengelley: It makes me really jealous. I have a bit of fomo going, oh, I wish I had the time and money to spend doing that, but I'm so busy working on these other projects, I can't divert any energy onto claw bots at the moment. [30:19] Marno Brits: Mm-hmm. [30:21] Dave Pengelley: yeah. In interesting to see. And then these sorts of experiments are where you say, Hey, give me an interface to look up this thing, and it gives it to you on the spot. [30:29] Dave Pengelley: Right. That, that's kind of the idea of these things. [30:31] Matt Slager: Yeah. Or even you don't need an interface, you know, there, there's a, [30:34] Dave Pengelley: it just, it just gives you the data. [30:36] Matt Slager: Yeah. I, I would, I would bring up the meme and, and share it if I could, but I don't, I won't be able to find it. But it's literally like I am the interface now, where. [30:46] Matt Slager: You just talked to it. Like I, I found, I was in the car the other day, on our errands, my, my, my weekly car trip that I do with my stupid office life now. and I was in the car going, man, I just wanna, I wanna speak to my, like, my open call thing that I've got set up. I just wanna talk to it for a second and just go over some of these, you know, security things we're talking about. [31:09] Matt Slager: Mm-hmm. So that would be fantastic if there was a way of, you know, push to talk, accessing that instantly while driving, you know, while it's in a safe driving scenario. [31:20] Dave Pengelley: Yeah. [31:20] Matt Slager: You know, completely hands free, hands off, all that kind of stuff. And [31:24] Dave Pengelley: I've done plenty of walks with the, just the chat GPT app, right. [31:26] Dave Pengelley: And you stick that in voice mode and just back and forth and that, and that's got context and you can business plan and, and brainstorm that way similarly. But the whole point of that, that that's that. Govern chat, GPT environment, if you want your own connected tools and things, there's these extra layers that aren't easy to install yet that we've still gotta build and understand how to do that best. [31:48] Marno Brits: Yeah, yeah. [31:50] Dave Pengelley: you, you need, you almost need your, open call set up to like a retail or vapi kind of agent with its phone number. So you can just call it and it can answer the call. And then rather than you talking to a Vapi agent, it's just like web socket that straight through to your open claw and it's doing all the, the, the speech to text and the processing, right? [32:09] Matt Slager: Yeah. You can, you can even just use a direct audio like [32:13] Dave Pengelley: Twilio. Yeah, just take your Twilio number and just pipe that sip trunk straight into your open claw. [32:18] Matt Slager: That's [32:18] Dave Pengelley: what you want. [32:19] Matt Slager: I've, become quite preferential towards NICs recently. Which is T-T-E-L-N-Y-X. [32:27] Dave Pengelley: Okay. [32:27] Matt Slager: instead of Twilio and like I feel like I would never touch vapy retail, any of those things at this point 'cause they are all just wrapping and combining those services. [32:40] Matt Slager: great. If you want to just jump straight into it and not learn any of it, of course. But yeah, [32:44] Dave Pengelley: again, build, build versus buyer knowing that they're gonna be managing a bunch of security and stack and uptime and other bits and pieces. Right. It's like what are you getting for the premium you pay and is that worth it? [32:53] Dave Pengelley: Why does anyone wanna hire any of us versus just going and asking chat GPT how to do it Because we know it's not as simple as just asking chat GPT how to do it. yeah, that's very true. That constant back and forth, even though apparently all the agents are getting so much better. Like, have you guys like it was amazing this week. [33:13] Dave Pengelley: We saw Claude Opus 4.6 come out, and then it was like literally half an hour later chat, GPT 5.3 Codex came out back to back the new greatest engine ever, from each of the two leading Frontier companies. And then Google are staying silent. But, word in the wings is where, a week or so out from Gemini dropping its next big announcement, I think. [33:36] Matt Slager: Yeah. It looks like I, I definitely feel like I know one of the upcoming things and that will be the, the combination and the, the sort of centralization of anti-gravity and Google AI studio. [33:52] Dave Pengelley: Okay. So [33:53] Matt Slager: you'll be able to, you know, in antigravity you'll be able to deploy things straight to GCP through [33:58] Dave Pengelley: mm-hmm. [33:59] Matt Slager: AI studio connections and have your AI studio API keys in antigravity, so you can actually do that sort of stuff. You're probably looking at more of like a, a lovable style thing where you can go from like zero to deployed in the one spot. [34:15] Marno Brits: That's great. I'm even looking at canceling my lov 'cause I've got maybe four or five more projects on there that I'm still building and I'm just, it's, I'm just too lazy to export all of that and then take it to antigravity and build it. [34:28] Marno Brits: The [34:28] Dave Pengelley: cost, the cost of what you'd be paying lovable for someone with your technical ability. You should be out in clawed or antigravity. Right. And then hosting antigravity [34:36] Marno Brits: is better, the results are better [34:37] Dave Pengelley: too. Right. Anti [34:38] Marno Brits: even Google AI Studio. [34:39] Dave Pengelley: Yeah. And, and, and then, you know, hook up to a super base and, and run it on, on push it to, I know Matt's not a big Versal fan for reasons of Matts, but you know, stick it on one of those, those platforms where it's pennies to run. [34:51] Dave Pengelley: Pennies to build and put up there. Yeah. Especially for testing and, and getting up and running and like the Google plans are cheaper than what you're gonna be paying for your lovable credits. And refreshed so quickly. So yeah, like I think it's a, it's a no brainer to, if you've, if you're happy working in an IDE, you lose a few little things at the moment. [35:10] Dave Pengelley: Like being able to select components on the page and say deliberately do fix on this one thing. I'm hoping that the browser agent improves and gives you a bit more capabilities to select the DOM and, and push that directly into the chat. Yeah. Versus like trying to identify just with text or take a screenshot and stick it in the chat. [35:27] Dave Pengelley: but yeah, I, I wouldn't, I, I would never go to like a bolt lovable or rep, where it's, it's convenient 'cause it kind of builds the database and builds the auth and does all those things in one. But like, if you're using tools like Super base, which also manage your auth and stuff for you, you can really do so much very quickly, very cheaply. [35:44] Marno Brits: You also don't need the auth for like, I'm just doing websites. It was just a pure convenience player of like, cool, I'll use bulk lovable output's better. Now with Antigay available, like I already, but [35:53] Dave Pengelley: you saw my website. I [35:55] Marno Brits: Yeah. [35:55] Dave Pengelley: Right. Like that, that's cost me practically nothing to run on Versal. [35:58] Marno Brits: Yeah. [35:58] Marno Brits: What you put on Versal, it's free. Like the, there's even Resend is free. The only cost is the main purchase and the licensing. [36:05] Dave Pengelley: Yeah. All, although to Matt's point with railway that he was talking about earlier, I was looking at for my SaaS and I built a cost calculator and I was going, okay, what if I scale this much? [36:13] Dave Pengelley: What if the usage gets up to this much? Mm-hmm. And at one point it's like, okay, my monthly cost on Versal will be like 15 grand on railway, it'll be 500 bucks. so as, as these applications scale, then, you know, once I'm, you know, getting some revenue in customers, then I'll probably look to how do I re-platform to a more efficient scale [36:33] Marno Brits: mm-hmm. [36:33] Dave Pengelley: Platform. But for getting up and running MVPs and POC type stuff, it's, it's amazing what's out there right now and I'm sure Google are gonna get better at, in integrating their Google Cloud stuff to make it easier to, you know, push to a fire base and GCP and Google Author and all that kind of stuff to, to give people that all in one, capability. [36:50] Dave Pengelley: Yeah, [36:51] Matt Slager: they should because they, they do technically have all the infra to do [36:55] Dave Pengelley: it. Oh, the whole stack, right? [36:56] Matt Slager: Yeah. [36:56] Dave Pengelley: And, and all the way through to their Vertex ai, which is their premier sort of custom model type stuff. [37:01] Matt Slager: Yeah. It's just sort of like their hosting of inference and whatnot. But like with, with Firebase for example, it's a great concept, but you talk to anyone who uses it in production or uses like a different few selections of them, they'll all say that five ace is kind of rubbish. [37:18] Matt Slager: And it's always some sort of weird perceived thing. So like that, like, like they need to get to that point. So digitally you go into any of the weird ai sort of developer rabbit holes, and once someone starts talking about different labs, they'll, they'll say that the Google thing is the one that they think is gonna win, but they currently hate it and then they can't really describe why. [37:39] Matt Slager: It just feels weird. Yeah. Like that. That's so interesting. And I don't understand why I think it's. Possibly because they're such a big company and there's so many different labs and each of them are kind of separate and they don't really work together. [37:58] Dave Pengelley: yeah, I think, I think my, my Microsoft used to be like that back in the day as well. [38:01] Dave Pengelley: Microsoft had some really cool, innovative little plugins and things back in like the Windows 98 kind of days where they just had these little skunkworks projects. Well long before OneDrive, they had that kind of cloud backup type thing. Can't remember what it was called back then. But they, they had all these innovative like little project lab things that, you know, over time Bill left. [38:19] Dave Pengelley: Steve didn't endorse it. I dunno. Google's the new version of that, right? And as much as they don't have their Friday afternoon sort of reserve project time like they used to, as much anymore, they are so diverse and they're really a product company, not a sales company. And so they're not really thinking product first. [38:34] Dave Pengelley: How do we package this? What are we gonna monetize? 'cause all their money's coming from ad revenue. So they're, they're, they're, that, that's pretty much covered. Anything they wanna do. And so they, they innovated, they build these things, but they built the transformer model the GPT was built on. They just didn't actually execute it into a chat bot like OpenAI did. [38:51] Marno Brits: Mm-hmm. [38:52] Dave Pengelley: every, everyone, you're right, everyone seems to think that you've got philanthropic and open AI battling out the moment, but there's, there's, there's this, this, this Gemini media coming, because they own the full stack. They're fully vertically integrated. They own the hardware and the tech and the customer base, and they've just got everything ready to go. [39:11] Dave Pengelley: They just need to swing it all into alignment. [39:13] Matt Slager: I also, I wouldn't sleep on the idea of the Chinese ones as well, and I'm using that term like balloon term, the Chinese ones. But, you know, there's, there's a lot of really, really interesting stuff like z ai, minimax, kimmi, obviously. [39:28] Dave Pengelley: Mm-hmm. [39:29] Matt Slager: Deep seek, like these labs, they're all doing such like frontier work, but because they're not yelling as loudly as everyone else, like a lot of it gets missed. [39:40] Matt Slager: I find that the, the new Kim K 2.5 is actually really, really good. I think we mentioned that on the very first episode. It like just came out. [39:49] Dave Pengelley: Yeah. Yeah. Mm-hmm. [39:50] Matt Slager: And for, for most people's open claw applications like that would probably be the best one that they should use. And you can literally just go get a QE code plan and then put your API key into that and then all of a sudden you've just got infinite QEK 2.5 pretty much. [40:07] Matt Slager: And, yeah, a lot of people say like, why don't you just link your Claude Max plan or whatever. 'cause it's the most cost efficient for tokens. It's so up in the air as far as like if it is or if it isn't against terms of service. So my, my take on that is if you are doing it on your own machine and it's locally and you know you want to just pump your Quad max tokens through stuff locally, then go for it. [40:31] Matt Slager: But I think as soon as you start doing it on VPSs or on deployments and you have to use the. The token, you know, method of actually linking it together. You're probably now reaching, but anyone who wants to experiment with that, what actually is the ramification if they ban your account, what does that mean? [40:51] Matt Slager: Yeah. A new account, like [40:52] Dave Pengelley: Right. And, and, and like you mentioned last week, sometimes, you know, start a new chat. Fresh context is, is a boon. I mentioned I was having issues with one of my anti-gravity on my other PC, where I just had all this weird, just corrupted context that it was just lost in itself to the point that I couldn't even use the browser plugin anymore. [41:08] Dave Pengelley: Like the browser would not load. I'd done something and damn. And I tried to put up one of those harnesses that a harness on, and I completely just screwed my hole. My whole install of anti-gravity was completely broken. so I actually just had to go through and remove all the old folders, go through all the application data folders, remove every instance of Gemini and, and brain files and everything. [41:27] Marno Brits: Mm-hmm. [41:28] Dave Pengelley: And install it fresh. And guess what? Because I had read mes and documentation in my folders. Bang, we're back up and running. Now all the skills are still there. They're working again. Now my browser works again. So, you know, like the Ralph Guy was saying like context, like you can start fresh again because it'll just read it and keep working from the thing. [41:46] Dave Pengelley: Like that's, like we said last week, the context of these things are stateless. They don't have a state, they just read the history and then move forward from that. Mm-hmm. [41:54] Matt Slager: That kind of like permanence, knowledge, permanence and then like having your files there I think is a good like, opportunity for us to talk about the file management, knowledge management, you know, stuff again, like how your, your initial foray into using antigravity is like a business planner type situation. [42:12] Matt Slager: 'cause for anyone who follows obsidian as a, as a note taking tool [42:18] Dave Pengelley: mm-hmm. [42:20] Matt Slager: They pretty much have won right now, you know, this week. as far as the, the knowledge management tool of choice with regards to agents managing your knowledge management for you. Because after I first heard of Dave doing this. [42:35] Matt Slager: I was like, well, I'm gonna have a quick go of that and I'll just go to my agent of choice and, and say, Hey, I wanna build this knowledge management system. And, yeah, he's obsidian, so they, yeah, I, I kind of just said to it, I was like, I wanna build like a obsidian for agents, you know, just use the same sort of linking structure so that, you know, you can search through this stuff and it just makes sense. [43:00] Matt Slager: So I have this, I've got this obsidian vault now that I don't even use obsidian, like I don't use the app at all ever, but I'll get my agents to talk to that vault and all that knowledge, all those files, all those things are just there permanently. So, you know, wow. You can completely change agents, you know, you can completely change tools and all it is is a folder on your system or whichever [43:25] Dave Pengelley: Yeah. [43:25] Matt Slager: Backup system. Like, you know, if you have a cloud sync or whatever going. [43:29] Dave Pengelley: Well, my, my, my broken antigravity while it was broken before I went through the process of tearing it down and rebuilding it, I installed open code and plugged in my, my open AI codex log into that. And we're very successfully running that to the point where I'm like, ah, maybe for this particular project I'm better off just staying at Open code because it's with Codex. [43:50] Dave Pengelley: 'cause it's doing an excellent job. And maybe I keep building my SaaS and doing my business in Antigravity. But for this particular sort of set of trading projects I'm doing, for, for trying to build trading robots, maybe open code and Codex is just a better interface for setting up all my testing pipelines and stuff just 'cause it's, it's working more smoothly. [44:06] Dave Pengelley: But. If the knowledge was all there in the folders, I was able just to, to just send point a new agent at it and keep running. [44:14] Matt Slager: It's just a vehicle at the end of the day, you know, you could be driving a Corolla or you could be driving a Ferrari. Like they each have different, like coolness levels, but they both get you to the destination [44:23] Marno Brits: a hundred percent [44:24] Matt Slager: like that. [44:25] Matt Slager: I think that's the, that's the key thing that a lot of people need to keep reminding themself that it doesn't matter what tool you're using, it rema, it matters like what the vehicle is, you know, what the, what the actual foundational principles and primitives are in this case. So, you know, local files, markdown files, you know, by this stage, if you don't know what a markdown file is, just go ask your agent of choice. [44:45] Matt Slager: Like you need to need to know, like text, like the idea of like written language text is the oldest form of like data communication. You think of like ancient. Texts. Mm-hmm. That's literally the term. if you are not saving your knowledge and capturing your own texts now, you won't be able to use it in future programs. [45:05] Matt Slager: Yeah. You know, if you are using something like Notion, and I love Notion, I've used it for a long time, but if all of your information and knowledge is in there, you can't take that anywhere else. Like it's in there cps, [45:18] Dave Pengelley: but [45:19] Matt Slager: you currently access it, but you don't own it. It's not, it's not yours. [45:24] Dave Pengelley: Yeah. They turn the lights off tomorrow on their servers. [45:26] Matt Slager: Yeah. [45:27] Dave Pengelley: Bad news. I mean, they come back to a bigger disaster recovery issue. That's been an IT problem forever. And as I've admitted to you guys, like backups have been the bane of my existence my entire life. I'm never been good at them and it's caused me some, some pain over the years. You'd think I'd be better at it by now, but I still suffer and struggle with a bit of my backup, cadence for my own, my own things nowadays, but, All these, all these tools are growing and they're all trying to be the next thing, and they're all trying to bring in more files and things. And so they're actually becoming more accessible to keeping your own files in your own infrastructure and still leveraging the front end tools, especially as they sort of go after the business clients. [46:06] Dave Pengelley: Right? [46:07] Marno Brits: Yeah. [46:08] Dave Pengelley: Have you guys, did you open ai? Just like they're, they're slamming the announcements probably off the back of the Super Bowl thing, because I was getting negative press about the ads and so on. Anthropic were, were sort of, misconstruing that a little bit for a bit of a poking fun at them, but they, they've launched Open AR Frontier and they've announced this is a, an early thing for their businesses where they're, I thought it was like a new interface, but no, the, their interfaces still remain, but they're putting this new frontier backend aggregation model in. [46:36] Dave Pengelley: Mm-hmm. as far as, you know, it's not a rag agent, but maybe it is, maybe it's just a fancy way of them doing rag and, and providing context into your backend systems. They've also, they're going after Notebook. Lm, I saw this one just this morning, five hours ago. We're rolling out a new deep research experience that can deliver faster, more accurate reports with less rework. [46:56] Marno Brits: Ooh. [46:57] Dave Pengelley: and so now you can put in specific sites and even limit it to those ones, or say you can go outside these sites. Mm-hmm. Original data. So they're trying to get into that, notebook, LM Market. 'cause notebook, LM went from just having sources to having the Gemini Deep research functionality built into it. [47:19] Dave Pengelley: which, you know, all I, if I, if I'm really looking for deep research on a topic, I'll run one through both of them. I'll, I'll go to my chat GPT and say, deep research this topic, and I'll take the same prompt. I'll go over to Gemini or, or Notebook L and say Deep research this topic. And, and, 'cause they're both gonna find different things based on where their sources are and how their web search works and what they look to do. [47:41] Dave Pengelley: So I'll [47:42] Marno Brits: actually prefer perplexity. I'll use perplexity of any deep research and then I'll use agent mode to validate that. So perplexity is the primary and then the secondary to validate that information will be agent mode, which IGBT. 'cause even deep research, I don't feel it finds the best results. [47:57] Marno Brits: I feel like agent mode, it'll take longer, but then I feel like it's actually doing due diligence rather than just making stuff up. That could just be based on past experiences. 'cause it hallucinated one too many times and I was like, I don't trust you anymore. Edge mode so much better because it actually goes to the site. [48:12] Marno Brits: we do have a couple comments. We have Dustin Gray agree with Matt ofid is a great agnostic knowledge base as a operating system brain. you just pointed to your secret source, which is very true. And then has, have either of you used the agent zero? I know Joe talked about this earlier in the week. [48:30] Marno Brits: No, like on the accelerator chat, but [48:34] Dave Pengelley: No, I haven't heard of it. [48:35] Marno Brits: Yeah, it, it looks interesting. I've not played with it yet. I just watched the video that you shared on it. It is from David. He posted it last week, I think it was, or week before. [48:44] Dave Pengelley: Agent Zero. [48:46] Marno Brits: Yeah. Keen to get your input, Matt, which is probably something more up your alley. [48:53] Matt Slager: You know what I'm gonna say? [48:55] Dave Pengelley: I never into it. let's accept acceptable Cookies. Cookie Monster. let agents build your own Agentic AI system. Connect any provider. Stop patching agents, run autonomous agents. So this is a response to Claw Bots. [49:13] Marno Brits: That's my opinion as well. It just looks like a, another version of Claw Bot. [49:18] Dave Pengelley: Yeah. Personal Agent Jarvis. [49:22] Marno Brits: But you run it through Linux instead. Yeah. [49:25] Dave Pengelley: Interesting. [49:28] Dave Pengelley: Hmm. There's, there's gonna be heaps, heaps of these tools coming out. I think, claw bot, open claw notebook, whatever it's called, is just the, the tip of the spear. And once people realize that every man and their dog's gonna be coming out with their clone and their better version, their more secure version, their less secure version, their, their free version, their costing version, their one that uses more tokens, the one that comes included with tokens, the one that's a SaaS, the one that's a DIY, like, we're gonna see so many of these things, it's gonna be hard for regular people, let alone us. [50:04] Dave Pengelley: I mean for us, let alone regular people to stay on top of all these things, right? [50:09] Matt Slager: Mm-hmm. So with that in mind, this is probably the, the most important thing, and I think I've mentioned this in the past, but we need to stop designing systems for humans because those agents are gonna be the system for humans. [50:21] Matt Slager: And then we need to design systems for agents 'cause. If those things like Agent Zero Open call, all these various different things are gonna come around, then they need to have access to it. They need to be able to interface with it and interface with each other. You know, the, the OpenCL creator, Peter the other day said something really interesting where he's like, why wouldn't I want my agent to interface with your agent? [50:46] Matt Slager: You know, if we're both busy. Like, that makes sense. You know, it, it's kind of like the idea of synchronizing your calendar to find a booking time. A lot of like, that sort of stuff can happen fairly automatically right now. But if we just said to our person, like, go find a a time that I can speak to Dave this week and you know, it checks your calendar, comes back to me and says, yeah, what about these times? [51:11] Matt Slager: You know, and all of that just happened. Autonomously in the background asynchronously. You don't even have to respond to it right there and then, and you could be doing this while you're at the shops. Like there's, that is like the reality that we'll see pretty soon. [51:26] Marno Brits: Yeah. I If you watch Altered Carbon, sorry. [51:28] Marno Brits: Have either of you watched Altered Carbon, the series? Yeah. [51:30] Dave Pengelley: First season was great. Second season, [51:32] Marno Brits: and I, I lost both of them, man. But that, that reminds me of, there's the agent po, like there's po which is the, he is running the hotel and he's an ai and then he goes to meet with the other ais and the underground station while they're playing poker, talking about like who their rapport to and how their businesses are going. [51:49] Marno Brits: And I could see that, like, why would you not want your agents to talk to each other? Yeah, that's a great way of looking at it. [51:54] Dave Pengelley: Well, the, the, the app that I'm building, one of the things that I, I am about then the testing for before I, I continue on with the stuff is, MCP. So I'm like, if I'm building an app today, why wouldn't I look to put an MCP access in it for people that wanna pay? [52:11] Dave Pengelley: For that tier of subscription. So then rather than using my interface that I'm building for this application, they can plug their agents into it. They can ask their agents questions about the information in my, in the app. Right. And interface with it through that way. So an M CCP might be changed with H two A or something in the future, but for the moment, MCP is the standard. [52:28] Dave Pengelley: So we used to be, if you're building an app, you should go API first. Make sure that you've got an API and people can plug in and do stuff and, and link in with it. Things like MCP are just the next generation of that. Right. So [52:40] Matt Slager: I think it's a stepping stone. I don't think it actually will hang around. [52:44] Matt Slager: 'cause if you, if you think what existed before MCP, like a direct API connection is just again, a method. It's not necessarily the tool. [52:54] Dave Pengelley: Yeah. [52:54] Matt Slager: Like the actual tool for the most cases might've been curling and endpoint, you know, which is the terminal command. So like the actual command line interface is gonna be around for the very, [53:06] Dave Pengelley: yeah, yeah. [53:06] Dave Pengelley: But about just, it was just a way of standardizing, here's a way you can ask an endpoint for what you can do with it. And getting the YAML and knowing, okay, I've got access on that endpoint to do things from an agent chat point of view in English language. Right? That's, that's what they're using and tokenizing it. [53:22] Dave Pengelley: Mm-hmm. So I still think [53:23] Matt Slager: m CCP is a human system built for agents rather than [53:26] Dave Pengelley: probably system then, then they're talking about things like A two A and stuff. But at the moment it is, it is what it is. but I think, you know, if you're building apps and, and you've got sort of data that you want people to infer things from. [53:39] Dave Pengelley: Are you thinking about building that to have agent agentic access so your agents can, leverage it without needing to go through a browser agent and work out how to point and click? Are you letting your agent talk to the data? [53:51] Matt Slager: Yeah. [53:51] Marno Brits: Yeah. [53:52] Matt Slager: A great example recently for my own projects, my own development, I, I've started to integrate linear so that the business owner can actually help me schedule and prioritize certain work. [54:02] Marno Brits: Nice. [54:03] Matt Slager: Now, I'm not using linear, my agent's using linear. [54:08] Dave Pengelley: Yeah. So [54:08] Matt Slager: I had the option of using the MCP or using direct CLI access and MCP was gonna be like 20 times more cost, like inefficient, used 20 times more tokens than just simple CLI commands. So what it, like, what that means in practice is my agent just has a new skill that is how to use the linear CLI. [54:30] Matt Slager: I [54:31] Dave Pengelley: need you to send me that skill. 'cause I've been using the MCP and it's hit and miss. So it goes, yeah, I've done everything. I'm like, no, you haven't. [54:38] Matt Slager: I didn't, I didn't write it like my agent [54:40] Dave Pengelley: write it. That, that doesn't matter. You can send it to me. What's up there? What's [54:43] Matt Slager: happening? Yeah. It all comes down to like those raw fundamentals again, like just understanding the primitives and like Marnos thing that he's mentioned in the past is a little bit of technical literacy, you know, just a tiny, tiny bit. [54:59] Matt Slager: It doesn't really, you know, in this world where time really is reducing as the constraint, the, the new constraint is clarity and, and details. So if you've got an amazing plan with terrible clarity and horrible details, you're gonna get a bad result. [55:15] Dave Pengelley: Wow. yeah, that was just, so Dustin's commented, Visa's come out with a credit card just for agents to use now. [55:23] Dave Pengelley: Yeah. That's, that's [55:25] Matt Slager: what like a parallel example of that is an API key that has a cost restriction. [55:30] Dave Pengelley: Yeah. Yeah. [55:31] Matt Slager: So how, how often do you guys go to a new tool and create a new API key and you only create one and it's not, it's not limited access and it's not like a monthly budget, it's just a fully open unlimited one. [55:43] Dave Pengelley: Never. [55:45] Marno Brits: Not once, no. Yeah. [55:46] Matt Slager: How many people do you think do [55:49] Marno Brits: quite a few actually, [55:50] Dave Pengelley: yeah. Like when I said on my open ai open AI one on my ropy one, I'm like, you can have $10. Okay, well call me when you run out of that. [55:59] Matt Slager: And then how many people would forget that that open unlimited one exists? [56:03] Dave Pengelley: Yeah. Oh, yes. [56:04] Dave Pengelley: It's one of the things that, that, that terrifies me around the open claw thing. How many people are just linking that up to their, their API keys? And I don't, I've, I've read, I haven't kind, don't have them in front of me, but you know, people getting hundreds of dollars, if not thousands of dollars of open credit bills because their little open claw has just been going nuts. [56:21] Dave Pengelley: And it's just unrestrained and, and ungoverned and unlimited. And it's like, don't wanna be those people. [56:27] Matt Slager: That's the kind of like technical literacy, literacy, like that tiny little bit of understanding of some stuff just to make sure that you're not foot gunning yourself. [56:36] Dave Pengelley: And, and you know, it's happened forever. [56:38] Dave Pengelley: When I worked at Telstra. You get people ringing up, complaining about, oh, I was overseas and I didn't know about the roaming charges and no, I've gotta [56:44] Marno Brits: drive drill. Can you do something about it? It's like, [56:48] Dave Pengelley: I'll see what I can do. 'cause I'm relationship managing you and I want you to buy stuff off me in the future. [56:53] Dave Pengelley: Go talk to my boss. See what we can wipe. [56:55] Matt Slager: Yeah. [56:56] Dave Pengelley: but you know, that's always the thing. Let's leave people with the, in the last three minutes. Let's, let's do a quick show and tell, let's build something just for fun. Here I am in my, my chat. GPT interface. 5.2 is the way to go. order. That's fine. let's build a little pricing calculator. [57:12] Dave Pengelley: What do we want to price, forecast, cost versus profits on? Guys, let's do this in real time. No safety. a pricing calculator. [57:23] Marno Brits: What's the company knowledge? I've never seen that button. [57:25] Dave Pengelley: Oh, it's 'cause I'm in a paid version. I can link it to Gmails and things. They like Google drives and stuff for that, that backend context. [57:33] Matt Slager: It's on a team account versus [57:35] Dave Pengelley: business account. Oh, okay. [57:36] Marno Brits: Cool. Yeah. I see. [57:38] Dave Pengelley: so build me a pricing calculator is a single page app in the canvas that forecasts, my profits selling $40 plans for cleaning services. [57:59] Marno Brits: Mm-hmm. [58:02] Dave Pengelley: average cost is $10. Per service. Yeah. let's just do that. Super simple. You'd obviously wanna do more, complex stuff, but just as a quick and dirty example to finish off the hour. [58:24] Dave Pengelley: So one prompt, now it's building React stuff for me here in the browser. So this is like, we've been talking about anti-gravity and coding agents and stuff like this, just in your chat GPT website. A lot of people probably don't realize you can actually get it to build simple applications from a single prompt. [58:41] Marno Brits: Yeah. [58:42] Dave Pengelley: which is pretty wild. When I first discovered this, I was like, what? [58:49] Marno Brits: I remember when Chate first came out like, this is, yeah. When 3.5 came out, it had a lot less restrictions and you could have it run virtual instances of a terminal inside of Chate. it got very, very quickly shut back down. [59:05] Marno Brits: But yeah, you, you were able to go in and say, okay, cool. Now simulate what a linis or Yeah. a Linux server would look like running this, this, and this. And then it'll try and spin that up for you and emulate what's going on. Is really impressive. [59:23] Dave Pengelley: So I gave this almost no context for what the actual product is or the customer base or anything. I gave it two numbers. [59:32] Marno Brits: Mm-hmm. [59:33] Dave Pengelley: look with customers and churn, if you're already running set, current active customers, there's customers each month conversations [59:40] Marno Brits: as well, wouldn't it? [59:41] Dave Pengelley: Sorry. [59:42] Marno Brits: It's also pulling from your past conversation, so I understand. [59:44] Dave Pengelley: Yeah. But I, I'm cleaning service. I've never had a chat with it about running a cleaning services company. you're right. This is my main account and arguably it could be pulling like expanded context, but [59:56] Marno Brits: you're doing a rebrand, [59:59] Dave Pengelley: cleaning plan, profit forecaster. Here we go. Boom. [01:00:03] Marno Brits: Done. [01:00:04] Dave Pengelley: All right, here's, here's my plan price. [01:00:06] Dave Pengelley: I wanna increase my price if my costs go up, how many services per customer, per month, customers churn, lead flow and conversion. What's my conversion rate? Here's all my metrics, payment fees, fixed monthly costs all here. And then I could go, add a chart. I don't even know what I want it to chart. I'm just telling you to add a chart because I think a chart would look pretty on here. [01:00:31] Marno Brits: Yeah. [01:00:34] Dave Pengelley: but this is like, and this is what we're talking about, but like, this ability to move fast and, you know, we talk about just in time interfaces and not needing to have a fixed thing. 'cause the, it'll just create it on the spot for you. Like this is reality to a certain extent right now that, these things are, are possible. [01:00:54] Dave Pengelley: So stop that while I goes and edits the plan. It. and when you start moving outta anti-gravity and stuff, like the possibilities, just 10 x from what I'm doing here. Built in a chat GPT [01:01:10] Dave Pengelley: at a 12 month chart to the app. Oh, it's got a bug. [01:01:14] Marno Brits: No big [01:01:14] Dave Pengelley: bug. No. [01:01:16] Marno Brits: What would you say is the benefit of having a business plan or team plan instead of just an individual plan? [01:01:23] Dave Pengelley: increased limits. Mm-hmm. So you do get increased limits. if you are running with multiple people, you can have shared project folders so you can actually share chats and collaborate on Yeah. [01:01:32] Dave Pengelley: Chats together. That's probably it. Really the business plans in this weird no man's land where you do get some increased things, but you still don't get a lot of the proper data controls and business controls and user Yeah. Editing controls, like there's still some, I, I don't, I don't love the business plan and, and how it sits as a fully featured thing. [01:01:53] Dave Pengelley: but, you know, codex, ultimate Codex, unlimited agent sort of things. You don't get your 20 agent uses a month, sort of. [01:02:00] Marno Brits: Yeah. [01:02:02] Dave Pengelley: I don't have any limit. If I wanna come in here and use agent mode, there's no little count on it anymore. [01:02:08] Marno Brits: That's good. [01:02:09] Dave Pengelley: so all that kind of stuff is, is in here. If you've got your sort of company knowledge based sources, like your, your things in there, but you know, that's not gonna work for me, unless I've decided to change it, since last time I tried. [01:02:24] Dave Pengelley: but, yeah, still editing anyway, people get the idea. We, we've hit the time, but you can build these apps in real time. even simple things. It might've taken you a day or more or a week to build that in Excel. The number of different variables. The number of different things. Yeah. The fact that even having sliders, you can't as easily put sliders in an Excel thing. [01:02:46] Dave Pengelley: People go, but you can put charts in more easily. yeah, well you can do that here now as well. Look at that. Bang. There's my log chart. and as I Oh, look at, oh, look at that. More leads. We, we we're crossing over. The new customers are gonna exceed the old customers. reduce the churn. Increase the churn. [01:03:06] Dave Pengelley: Our customers leave all the time. that's wild. It's just, it still blows me away, like how quick and easy you can build these things, knowing how hard it is historically, or how much time it takes to code that. [01:03:19] Marno Brits: Yeah. [01:03:21] Dave Pengelley: like while we were just chatting, like literally while we were chatting, that thing just delivered me, 745 lines of code. [01:03:32] Marno Brits: Mm-hmm. And took nothing [01:03:35] Dave Pengelley: like. How long would it take you, Matt, to write 740, like five lines of code while having a chat with a couple of guys? [01:03:43] Matt Slager: I wouldn't. [01:03:45] Dave Pengelley: Right? [01:03:46] Matt Slager: Yeah. I just wouldn't anymore. [01:03:49] Dave Pengelley: No. but anyway, that's, that's the show for this week. thank, thanks boys. Appreciate your time as always. [01:04:00] Dave Pengelley: Hopefully our audience gets some value from, just hearing us talk, real talk around the chats, the technology, the things that are going on, the, the more technical, the less technical the application. That's what we're here for, just to the chat and share knowledge and ideas. And I really want to know about your linear CLI thing. [01:04:17] Dave Pengelley: That, that sounds good. [01:04:18] Matt Slager: I think worth, I think it'd be really cool if, for people that do watch this, whether live or recorded. on YouTube, if they could leave us some comments of stuff that they want us to dig a little bit deeper into or something that they actually perked interest in. [01:04:33] Dave Pengelley: Yeah. What, what did you learn from today? [01:04:34] Dave Pengelley: Right? Like, 'cause I was talking to a mate who was trying to, I, I'd got him onto Antigravity and he was using it to try and build N eight N workflows. And it was, he was struggling with it, building a million different Python scripts for every change and wasn't being real great. And I was like, are you just prompting it or are you using a skill yet? [01:04:47] Dave Pengelley: And he is like, I'm just prompting. I'm like, okay, go to skills. Sh here's all the NN skills, skills. Sh and I said, go there and just tell it. Look at these and synthesize the best NAN agent and call it, call him Nate. 'cause I was like, Nate, he's, he's renamed him since, but I was like N Nate and Nate. [01:05:07] Dave Pengelley: Yeah. go. and he come back, came back and said, yeah, I've, I've got him, I've got Nate watching a, a YouTube video 'cause he sucks at webhooks. And I'm like, okay, that's brilliant. Get your skills to improve themselves, but taking that next step, what's the next thing? What are you doing that's a little bit more advanced with the AI tools today than you did yesterday? [01:05:26] Dave Pengelley: And if everyone sort of moves forward step by step, everyone's gonna get a lot more productivity outta them [01:05:31] Matt Slager: because we could do like your little demo at the end. Then we could do stuff almost in like a series a sense where, you know, last week we were doing this, let's continue that for 15 minutes. [01:05:42] Dave Pengelley: Yeah. [01:05:42] Matt Slager: And you know, for anyone that's interested [01:05:45] Dave Pengelley: comes down what the audience wants. and, and what's what's useful, right? Like, 'cause we're here to learn from each other and, you know, let people basically listen to our chat. and hopefully they get some value from it. So that, that's what we're here as the AI operators. [01:05:59] Marno Brits: I'm here for it. [01:06:01] Dave Pengelley: Brilliant. All right, thanks boys. Thank you, Australia and the world for joining us, and we'll see you all on the next one.