The Model Is the Battery. Pick Your Poison.

[00:00] Dave Pengelley: you wanna cover as well? Um, what, what does your shirt say, Matt? Mine's pretty obvious. Mine says, "I drink scotch." Um, I can't read yours 'cause I **Matt:** don't know- I don't know what mine translates to in Italian. I've never looked it up, but it is La Marzocco. **Dave:** Oh, it's the coffee machine brand. **Matt:** Yep. **Dave:** Yeah, very good. That's... You're mirrored, so I can't- **Matt:** Oh, really? Oh, that's annoying. **Dave:** Yeah. You can change that- Yeah ... in, in Restream. You can go to the cog, and you can mirror your camera. **Matt:** Let's do that. **Dave:** Is mine mirrored for you, or can you read mine properly? **Matt:** Um- **Dave:** Mine's mirrored, too. That's ridiculous. **Matt:** Mirror off. There we go. **Dave:** Mirror- **Matt:** I like having, like, my own personal one mirrored, and then everyone else can just get the non-mirror version Yeah, I don't **Dave:** think Restream gives you that option. You just, you're in or you're out. **Matt:** Yeah, 'cause it weirds me out. I... It doesn't look like me. **Dave:** I, I've spent, I've spent so many cycles doing, like, OBS things, recording myself, that I'm used to seeing the, the, the backwards version of myself with the light. [01:00] Dave Pengelley: Like, my light's [01:00] Dave Pengelley: there, but that's the opposite for my mirroring, right? So I'm like... But yeah- Yeah ... now you can read that my shirt says, "I drink scotch." Yeah. **Matt:** A bit clearer. And, **Dave:** and I fix stuff, so that's two things. **Matt:** Oh, nice. I love it. Very good. **Dave:** I got that from, uh... Have you ever seen Llamas with Hats? **Matt:** Oh, I've heard of that. What is that? **Dave:** Uh, it's an old, uh, animation by the guys that did Charlie the Unicorn. **Matt:** Oh, okay. Yeah, that's probably why I've heard of **Dave:** it. From Pinkfong. Yeah. He's like, "What's wrong with you, Carl?" And he's like, "Well, I, I s- I stab people, and I eat hands, so that's two, two things." **Matt:** Charlie's good. Anyway, what are we gonna talk about today? I have a few ideas. I've got a few bits and pieces. I'm sure people that are tuning into this on the recording are gonna be interested to know what exactly we're gonna spill t- today, and it seems like it's just the two of us. **Dave:** Just the two of us, you and I. That's, uh, how that song goes? No, I'm not deep enough to get the "You and I." [02:00] Dave Pengelley: And- I don't know [02:00] Dave Pengelley: it ... to do the pitch shift and, and, and stay in tune, in key, it's very, very tricky to do that. I am nowhere near a good enough singer to really try that, but I gave it a red-hot go. Um, yeah, it's just two of us. I, I don't know where our friends have gone. We didn't, we didn't get any, uh, MIA emails. What it could be, they could have caught the Claude bug, and they're deep in the realms of Fable land. Fable land. Oh, my God, my mouse **Matt:** is down to 5%, **Dave:** my MX3. It's gonna wear out. My mouse is gonna die mid-show. Mouse buddies. Oh, and I don't have a USB-C cable that reaches across to where my mouse is, so yeah, I might be switching over here to my trackpad halfway through the show. **Matt:** Here you go. Oh, how wonderful. There's a spare. I'm not using this one. **Dave:** Oh, thank you. No, I mean, um- Where'd **Matt:** I put **Dave:** it? How do we get mine into the camera? [03:00] Dave Pengelley: Too short. That's not what she said. Okay, we're not that kind of show. Let's- Yeah, let's, [03:00] Dave Pengelley: let's, let's, let's, let's, let's start the show, man. Let's, let's do this You, uh, you talk about Fable. I, I, I heard that was a myth. Ha. **Matt:** Ha. Mm. Myth, MythOS. **Dave:** MythOS. **Matt:** Um, it's interesting. Hey, have you... Have... Do you still have any Claude access? **Dave:** I have it for a week, but given on the weekend I did a full clean hard reinstall of my Windows 11, I've not installed any stuff that I don't need. And so because Claude code for my sub expires in a few days, I'm like, "Maybe I just don't install the whole Claude co-work app and have all that legacy junk just cluttering up Windows," 'cause I've got a, such a clean, pristine install now. **Matt:** Yeah. **Dave:** Um, I've still got it on my Mac. I, I could switch over to my Mac and have a play with it over there, but everyone's saying it burns a million tokens. [04:00] Dave Pengelley: It's the crack dealer try before you buy kinda thing. Yeah. Let's get you hooked on it with your subscription, and then, "No, no, now it's extra usage, API pricing." It's like... **Matt:** It's 100% gonna be a rug pull, but there's no denying that it's actually good. Um, I, I haven't personally used it yet, but the, a lot of the people that I follow, um, you know, n- naturally first thing this morning I was on X scrolling, you know, reading what everyone's takes are. Uh, all the, all the Americans that have had it already for hours on end before we even wake up. Um, and yeah, it absolutely is a token burner, for sure. **Dave:** Mm-hmm. **Matt:** And that rug pull is pretty r- pretty brutal. The, the, the actual post said something like, you know, we're gonna let you use it, um, for, I think it's twice the burn rate of what Opus 4.8 currently does, and you can use it up until June 22nd. And then after that it's gonna be extra usage billing only. **Dave:** Mm. [05:00] Dave Pengelley: **Matt:** Which, you know, how fast is that gonna get used up? In a **Dave:** few weeks- I would... I would say we've been predicting that for weeks on this show, right? We've been saying, like, the, the, the token tokenomics of all this, um, is coming to an end. Like, the sub plans, especially for the, you know, the, the major US-based models, your ChatGPTs, they've s- they're a little more flexible, I think. But certainly Anthrope- Anthropic are pushing towards this whole pay-per-use kinda thing. I don't know what, um, X is gonna do with SuperGrok. They, they seem to be leaning more into subs now they've got the SuperGrok subs separate from your X subscription, so... And they're partnering up with Hermes. I don't know what's gonna happen on there. You got Nvidia and the Nemo clause stuff. That's all pretty much... They pay, they charge you per minute or something, don't they, per call? Like **Matt:** the- Oh, yeah, sure. ... **Dave:** Nvidia stuff. Like, there's all different versions of this. I think the Chinese labs, where they're running it on their cheaper hardware and cheaper power costs probably, because, you know, energy's pretty cheap in China 'cause they keep building coal, they, um- They're probably gonna stay on sub for a little while and they're gonna try and corner the market. **Matt:** Yeah. [06:00] Dave Pengelley: Um, yeah. I mean, I'm really- It seems like- ... happy with my Minimax sub so far ... **Matt:** the, the efficiency seems to all just be coming out of China right now. You know, like you just mentioned Minimax, like between them, uh, DeepSeek and Kimi, like a- all those labs, like even the Mimo labs, like the Xiaomi guys, they, they seem to just keep coming out with more and more efficiency gains. Like- Mm ... uh, the, the, the, the Xiaomi Mimo V2, whatever it is, um, is currently free like everywhere. Uh, I mean, yes, they're using it for training data, so it's- Yeah ... subsidized and whatever, but like to be able to offer that, that model for free for so long is crazy. **Dave:** Yeah, I don't know. I don't know. I, um, I think it's, it's still a race for the harness though. [07:00] Dave Pengelley: Everyone wants the harness. I saw Kimi have now got, like, sort of Kimi Co:Work. Mm. They, they had Kimi Claw. Like, if you logged into the Kimi website, they had a Kimi Claw where they were b- doing all the hosting with OpenClaw for [07:00] Dave Pengelley: you. Now they've evolved that into their own kind of thing, where they want you to pay their monthly sub and have access to sort of their Kimi, Kimi Work or Kimi Co:Work or whatever they're calling it. I don't know if Co:Work's- **Matt:** Yeah, I saw that too ... a thing. I can't remember what the name is. Yeah, it's just another desktop app type vibe. Um- **Dave:** Yeah ... **Matt:** which y- you're absolutely right. Like, everyone's gonna be doing it, and every individual software company is gonna be doing it as well. Um, I heard a really great analogy or, like, a, a phrasing for this kind of thing where, you know, last year it was, it was like X product with AI, you know? Yeah, yeah. It was like you're strapping it onto it, and it's kinda weird and hypey, and is it useful? No one really knows. And then the, what, what this person was saying is it, it's not gonna be X with AI. It's just gonna be AI's just in everything all the time. Yeah. And it's just inherent. It's like saying your light bulb comes with electricity. You know, like- Yep ... something like that. It was a really interesting- Yeah. You have a **Dave:** computer now with CPU. It's like, eh. [08:00] Matt Slager: Yeah. **Dave:** Yeah. **Matt:** A good one's actually probably car ads. I don't know if other countries car ads do this, but, you know, like, they'll say, like, "The new Mazda CX-5 w- with free air." And it meant, like, air conditioning, like, because that was an, an option that you would add, you know- Yep previously. But me as a kid, especially a little, you know, slightly asthmatic kid, I was like, "Mom, why does the car come with free air?" **Dave:** You used to have the, um, you had a AC, but then you'd have the, um, the econ mode. I don't know if you ever had, saw that, that in, in old cars where you had like, you know, it's like half the AC. So we're just, we're just gonna like not burn as much of the, the fluorocarbons or whatever. Like we're not gonna go full mode and then fully drain your energy. So if you know I'm gonna overtake this car, I better turn the e- AC into economy mode now so I can have the power to- **Matt:** Yeah ... **Dave:** to get around. **Matt:** Yeah, I, um, I was definitely from that car era where, you know, it, it, it was like Racer Boy, you know, high school. I had my Silvia S15. **Dave:** Yeah. [09:00] Matt Slager: **Matt:** You know, [09:00] Matt Slager: Japanese import, and I'd turn the air con off when I had to do like a pull in second 'cause otherwise you, you're losing too many kilowatts. **Dave:** Yeah. Yep. And sort of, um, reminded me of, uh, when I was playing, ah, what came out a few years ago, uh, X-Wing. Wasn't X-Wing. Was it called Star Wars Squadrons? **Matt:** Mm-hmm. **Dave:** The new sort of successor to the old X-Wing and TIE Fighter games. Uh, and that had, you know, you were constantly doing power management between your shields, your weapons, and your engines. And so, you know, you wanna blast away from people, so you sort of boost your engines and pull your weapons off. And then as you sort of loop back around to sort of shoot the TIE fighters, you'd sort of rebalance your energy, so you're constantly doing this energy management. I was playing this game all the time and then jumped in the car, and as I'm coming up to a corner, I'm like switching our car into sports mode for the corner and then back out. And I'm like, like doing dynamic power management in, in a modern car, fully automatic, but it's got like these sort of boost modes. And so I was playing around- That's cool ... with that as I drive along. Just- [10:00] Matt Slager: **Matt:** Uh, I... Were you, were you ever a Trekkie? Like you were part of that era. **Dave:** Yeah, I, I watched my fair share of Trek over the years. That's pretty d- dismal where it's gone over the last decade, but the old- Okay ... Trek I watched plenty of. You **Matt:** need, you have enough of an opinion then to, to, to meet what I'm about to say. Like- **Dave:** Okay ... **Matt:** I almost feel like I, I was born just too late to get into it. Like obviously there's people probably from my year, birth year and even younger than me that probably genuinely still got into it. Because **Dave:** for you, Next Generation was pretty much over by the time you were aware of it. Like it wasn't still coming out as a new show every week. **Matt:** Yeah. So I, I just- Next Gen was like- ... think I would've genuinely enjoyed it. Like, um, you know, I like Star Wars for what it is, you know, before it got weird. Yeah. Um, but then the, the, the Star Trek side of things, I think I genuinely would've enjoyed, so I kinda feel like I missed, I missed something important there. You **Dave:** haven't missed it. It's still out there. You can still- **Matt:** Well, that's true ... **Dave:** watch it. I mean, you just gotta deal with like janky graphics and sets for, in Next Gen. [11:00] Matt Slager: And you know, it's '87, so [11:00] Matt Slager: it's all four by three aspect ratio. So there's, it's all very standard definition. Um- **Matt:** I've gotta go get a, gotta g- got to go get an old CRT to watch it on. Yeah, **Dave:** yeah. That's, that, that's the go. Get that authentic, um, Star Trek feeling. **Matt:** Speaking of, um, janky graphics, like- The, apparently one thing that Fable has now- Yeah you know, brought back in, obviously Claude's always been fairly tasteful with design, um, it's apparently pretty next level. And I was actually watching a few people this morning talk about it, actual graphics designers and, and I don't know what you call them, 3D rendering people, and- Yeah ... this one guy, um, he, he had two bits of, uh, video I guess. [12:00] Matt Slager: One was just ocean, like waves, you know, water textures, and the oth- the other one was the same with a few other landscape elements. And he's like, "Okay, so the one on the left was Fable, and the one on the right was me, what I do in my job." You know, I'm trying to just [12:00] Matt Slager: decide, i- is my job, you know- **Dave:** Yeah **Matt:** outmoded yet. And to my, like, fairly limited, naive, like, zero experience in graphic design kind of animation stuff, they looked great, both of them, and I couldn't pick which one was which until I read that left right. And then he's going on to say something about That the waves in real life are not made of domes, like this weird sort of way of describing it. But honestly, it looked fantastic. So unless you're an absolute professional in your field, you won't be able to necessarily tell the difference if people are starting to just build stuff with Fable now. **Dave:** Yeah. It's, um... I'm, I'm keen. I've, I haven't caught up with them lately, but watching Corridor Crew on YouTube. Have you ever watched th- those guys? **Matt:** I feel like I've heard of that. [13:00] Matt Slager: **Dave:** Yeah. So Corridor Crew are a, a VFX house. They do independent stuff. I, I think they do some paid contract work, but they also just seem to do a lot of their own stuff and, and make YouTube dollars. [13:00] Matt Slager: Uh, but they're really good and, and they'll do like the VFX artists react to CGI and- Okay just stuff. So you might have seen that- Yeah ... on YouTube. **Matt:** Where they're like, "Okay, watch this. This is really cool." **Dave:** Yeah, where they're s- Yeah ... sitting on a couch and they, they- Yep ... the three of them talk about it. So I'm, I'm sure if they haven't already, those guys are gonna be doing some, you know, reacting to AI stuff and- **Matt:** Yeah **Dave:** 'cause, and they're really good. And because they've got the eye for it, and 'cause they've spent time in, in Blender and all these things sort of doing these advanced Maya renderings and things, and they, they, they understand light and what's good, what captures in a lens versus what doesn't. Like, they actually understand the videography- Mm as an art form and a technology, as well as the compositing, uh, and what you do with the rendering. Uh, they're, they're pretty interesting to hear, like when they're looking at something and going, "Oh, but I see the way, the way that light hits the, the lens, that looks like a real artifact. That's really hard to do. That, you gotta do this and this and this to recreate that," and blah, blah, blah, blah, blah. **Matt:** Yeah. [14:00] Matt Slager: **Dave:** AI is shortening those steps, right? And it means people with less skill can, you [14:00] Matt Slager: know, facsimile some of that stuff. Uh, maybe sometimes better just because of modeling and simulation things. **Matt:** Mm. **Dave:** But still, I think, uh, and as Richard would say all the time here, that, uh, just it gives you kind of the average of all the outputs. Um, and it comes back to, to as we discussed, that human element of critical thinking, of taste, of what actually is meaningful and looks good- Mm ... versus just, oh, yeah, I can make a thing make waves. Yeah, okay. But are they... A- and, and for many things, like for this show, we use a lot of janky AI graphics on this show because it's good enough. We don't- Yeah ... need it to be super great. Um, but you know, if you're Peter Jackson making Lord of the Rings film or whatever it is- **Matt:** Mm ... [15:00] Matt Slager: **Dave:** you're gonna go to Weta VFX and get them to do full particle by particle simulation rendering of everything through their- Yeah ... advanced physics engines because you care about the detail, and you want it to look as precise as [15:00] Matt Slager: possible, not a vague essence of what it could be by an AI who's just copying other people that have actually done the hard work. **Matt:** Yeah, 100%. Uh, did you hear my dog just then? **Dave:** I did hear a growl. I thought maybe you were really hungry. **Matt:** Let me just have a quick spy out the window. N- no one's coming down the driveway **Dave:** While Matt does that, if you're, uh, if you're with us on the chat, drop a chat and say hello. Let us know you're here with us. If got any questions for us, it's just The two amigos this week **Matt:** I actually genuinely would be interested to hear if, like, you know, anyone listening in live, like, literally chuck your questions in. Doesn't matter what it is. You could ask what kind of beard oil that I use. Um- **Dave:** Use oil? No. I was like, "I didn't know you were gay." **Matt:** No, it's literally just whatever cream my partner buys, I chuck it on my face, and it just goes here, too. **Dave:** I love the, uh, the old, the old two-in-one shampoo conditioner- **Matt:** Yeah ... **Dave:** it just does everything. **Matt:** That's it. Yeah, you only need, you only need one soap for everything. Yeah. That's all I'll be talking [16:00] Matt Slager: **Dave:** about. Wait, [16:00] Matt Slager: wait, wait, wait, I... So during COVID, I grew out a ponytail, and I actually had, like, like, long hair. Um, and when it gets longer, the, the chemistry of your hair changes, and I actually had to do the whole, like, double shampoo thing. Oh. So like separate shampoo and conditioner. I had to do double shampoo and then condition. It was like, it was so much work having long hair. Yeah. Like, now, now I just, uh, spend all that money on hairspray. **Matt:** Yeah, perfect. **Dave:** Yeah. Does the job. **Matt:** Um- **Dave:** So, **Matt:** you know, back to models. I, I'm so curious, because I haven't actually re-subbed to Anthropic yet to test out this new Claude model. I, I don't think I will. I'm curious, but it's really just that FOMO thing, you know, just- Yeah the, you know, can I do the jobs that I'm currently doing any better, um, with it? [17:00] Matt Slager: **Dave:** I'm, I'm, I'm trying to do everything with M3 at the moment, and because I'm not doing, like, super advanced stuff. And, and for my software application, I still default back to my Codex sub, which is more than enough for the, the patches and changes I've [17:00] Matt Slager: got to do on that. Um, Dustin wants to know if you can solve one of the Rubikses. **Matt:** You know what, Dustin? I actually, um... It's gone from my head. So I had a crack at doing it the other day. You know, you just do your one side, and then you, you know, you do your, your white cross and then the white corners. **Dave:** Yeah. **Matt:** And then the next step is to flip it with that on the bottom, and then you start going through the center layer. Like, you know, you do these. **Dave:** Yeah, you... Yeah, you've got to do those, yep, which- **Matt:** And, and it was all gone from my brain. **Dave:** Yeah, 'cause, 'cause if, if you're like me and you followed a plan and just learned the muscle memory of you go left, right, right, right- Yeah ... left, right, right, twist, twist, and didn't actually learn the, "This is how I move this one to there," which, like, that's what, like, the, the, the geniuses at it, they actually can picture the cube and the movements, and they know to move that one to there and flip that round there. They just know how to do that. I just learned by rote. I looked up a how do you solve a Ru- Rubik's Cube in 12 steps and learned the moves. **Matt:** Yep. [18:00] Matt Slager: **Dave:** And I did it during COVID, and I'm the same. I've completely forgotten. I [18:00] Matt Slager: might, if I got lucky, be able to do the middle row, but the third row was always hard for me to remember and would be impossible now. **Matt:** Yeah, I, um, I remember two of the acronyms for, you know, like you have, you have front, up, down, left, right, and back. **Dave:** Yeah, **Matt:** yeah. So, like, those all translate to a letter. I remember subu, subu, and fudd letter. **Dave:** I remember fur- furrough. I remember furrough. Um, yeah. Yeah, anyway. Uh, we've got, uh, one of our LinkedIn viewers, uh, Dave has said, "Drops in to listen to AI convo, gets a hairdressing lesson." **Matt:** David, you know what? Speaking of hairdressing, I don't know why this comes up. Um, this is Dave in the comments by the way, not you Dave, so hold your horses. Okay. Um, I... During COVID, I actually had to... 'Cause m- I, I used to go to a barber, like, fairly religiously. It was an awesome barber. They had a, a cafe attached to it. So the, the company was called Barber's Brew. So it was literally like you go for a haircut and a coffee. It was good. It was a good time. **Dave:** Yeah. [19:00] Matt Slager: **Matt:** And when I couldn't go to them [19:00] Matt Slager: anymore, I actually learned how to cut my hair myself. So, you know, judge it as you will, but I've been doing it now since like s- you know, for six years. So- **Dave:** Did, **Matt:** did **Dave:** you just buy the suck cut? **Matt:** The what? **Dave:** The suck cut. Have you seen Wayne's World? The **Matt:** old- Yes, I have. What... I don't remember that, though. **Dave:** Oh, that's one of the famous lines from, from Wayne's World where they've got this guy with this invention called the suck cut, and it's like a, a vacuum cleaner that cuts your hair. **Matt:** Oh, good. **Dave:** And so he's like trying to do it on Garth, and Garth's like, "Oh, oh, it's sucking my will to live." And then Wayne's like, "Well, **Matt:** Bob, it certainly does suck." I need to watch that again. That was, that was, that was peak. **Dave:** Yeah. Yeah, that was... Uh, again, you're probably a bit young. You're probably like b- more slightly pre-Wayne's World generation, but when they came out, that was- That one peak. Um, yeah, funny, funny stuff [20:00] Matt Slager: **Matt:** So I, I'm curious- Yeah, model ... I'm sure people who listen and, and v- [20:00] Matt Slager: tune into this are as well, you said M3 before. Yeah. That's Minimax, right? **Dave:** Yeah, yeah. **Matt:** So for- Yep ... people that, like, only know about ChatGPT or, you know, they're starting to hear about Claude now, like what is Minimax, and should they care about it? **Dave:** Well, you- you- you're the one that put me onto Minimax, so I appreciate you, uh, you trying to interview me and get me to, to share stuff. But I, um, it's, you know, it's another model. There's so many models out there, and one of the things I've really been trying to do over the last six months since I left ChatGPT as my core go-to thing, going to the website chatgpt.com and building everything in there, when I moved to Antigravity, when I moved everywhere, is trying to be a little bit more provider agnostic. [21:00] Matt Slager: Um, and I kind of went backwards a little bit with Claude. I went all in with Claude Cowork and was fully baked into their stuff, but then I was very stuck in the Anthropic world. Part of that was my, my exploration and discovery, trying to like find those point-and-click, easy-to-implement solutions for small businesses and what's [21:00] Matt Slager: productized and what works really well and what's mature, and I think Cowork ticks a lot of those boxes. You pay for it, but it ticks a lot of those boxes. Hermes is less polished and less mature in some ways, but a lot more powerful. You can do a lot more with it, and you're not tethered to a single provider. Um, you can mix and match models and, um, to your peril or, or to your advantage. Sometimes it becomes overwhelming that like with more power comes more complexity. Yeah. Um, and so my, my daily driver now is Hermes, and I can deep dive on the many complicated ways I've set that up a- and they're using that. But my primary model for that is Minimax M3 because I'm paying now 20 bucks a month, and I just have basically infinite agents. Um, like I'm just not hitting the limits on that. [22:00] Matt Slager: And so compared to anything else where you'd hit your five-hour limit regularly, you'd, you'd hit your weekly limit in, in four days and all this kind of stuff that was happening with the other models, and paying a [22:00] Matt Slager: fortune for it, I'm just getting a lot more usage. It's super functional, it's clever, it does my web research, it builds apps for me. It, um, it can manage my infrastructure and push things up to my Hostinger and do all this kind of stuff. It can build end-to-end workflows. It can- Write Python. Like, it does, does all the things I need it to do, um, so long as I sort of give it the right instructions. And so, yeah, I've been really happy with M3. Um, M2.7 was pretty good, and then I just moved to M3 because it was the new one, and it's, it's been great. **Matt:** Yeah. I think you came in right at the right time, 'cause 2.5 was actually pretty good, too. Okay. If you go back to 2.5 now, you would be like, "Nah, this is, this is not all right." You know, it, it's... y- you have to do a lot of work for it. [23:00] Matt Slager: Um, you know, a great analogy, talking about the Rubik's Cubes again, y- you know when, like, a human looks at a Rubik's Cube, even if you don't know how to solve it- Mm-hmm ... you know what the goal of it is. Yeah. You know, it's to get all the colors on each side. Y- you may not know [23:00] Matt Slager: how to get there. You know, we talked about knowing those maneuvers by rote and muscle memory- Yeah um, not because you understand the technicality of what's going on, like, mathematically, but you kind of memorized that thing muscle memory. Yeah. **Dave:** I know, I know a formula that gets me where I've gotta go. **Matt:** So I think Rubik's Cube solving is such a perfect analogy of AI coding- Mm ... and us being the navigators of the intention and the goal, but not necessarily needing to understand- How do you exactly move that component to be where it needs to be? You know, if you've come from that land of extreme developer awareness, you know- Yeah ... you understand the languages, the frameworks, all the various bits and pieces to do it by hand, you know, wrote- Yeah ... off memory. Um, great. But y- you can definitely use a lot of that logic and transfer it more into like a navigational sense. [24:00] Matt Slager: Like [24:00] Matt Slager: you've just been saying with your Minimax through Hermes, you're navigating and you're orchestrating, turning things from an idea into reality. **Dave:** Yeah. Well, I mean, and I, I grew up by and cut my teeth sort of doing sort of procedural programming. Like, not even OO, like it was, you know, C. Uh, like old, old school C and, and Pa- Pascal and then sort of these things. Then Java came in and we had to learn like a bit of OO, and uh, did a bit of, you know, visual C, um, and visual C\+\+ which was object oriented, and then sort of started doing a little bit of that in the late '90s, uh, early 2000s. But then you fast-forward now to React and these, all these TSX files and routes and everything, and I'm like, I just wouldn't even know where to start building one of those. [25:00] Matt Slager: I mean, if you give me a, you know, a, a proc or a function main and, and then I can define some other functions and do some calls to it, I can, I can hack my way through that. But as soon as you... Like these modern web frameworks are just... The syntactically and [25:00] Matt Slager: the architecturally I do not have the understanding and I haven't done the homework to do that, but I know what I want and I know the end goal I'm trying to achieve. And so if I can get the robots who understand or can pattern match the syntax to make it work, and it might not be the most cleanest code. If, you know, an, an expert, a React dev looked at it and went, "Why have you got that there? And you're duplicating that and, and, and you could have, should have packaged that into there and refactored that as that other route." Like maybe, but is it getting the job done? Yes. **Matt:** Yeah. And, and realistically, like, you know, what are the downsides of having messy code? **Dave:** Yeah. **Matt:** You know, the... Like your, your agents will tell you that, um, it's hard to maintain. I- if you have- Yeah ... other people coming in and working on the code base, that'll take too long to understand what it is and... But how many people are you gonna have come work on your code base? [26:00] Matt Slager: **Dave:** Yeah. So as long as it's not a performance issue, as long as it's not taking five seconds to load instead of 500 milliseconds, in which case you tell the robots to do better and keep performing and, and, and [26:00] Matt Slager: refactoring and taking big blocks of code and splitting them out or taking multi scattered code and putting it together, what- whatever it takes. But it's like the Rubik's cube, right? If you know the sh- the actual mechanics of the Rubik's cube, and you know that I can move that to there in three steps, that's quicker than me. That's just gonna go, okay, well, I'm just gonna go through my motions and took me 15 steps to do the same move. Um, still get the result, but I was way slower and way less efficient 'cause I didn't actually know the underlying principles of how to do it. [27:00] Matt Slager: Yeah. And I think that's probably true with some of these models. Even, you know, M3 is good, and I get unlimited use out of it, but it... I can clearly see sometimes when it does, does the thinking, and even as it reviews its own context, it has to re-reason things over and over again, where I think, you know, your GPT 5.5s and your Claudes, your Opus and stuff, they can iterate on their previous session context a little bit quicker and easier, and it's not like every time they look at it they're like, "Oh, let me just see the, what I need to [27:00] Matt Slager: know again. Okay, well, the now I think you should do this. Oh, okay, let me just make sure I've got the full picture." And like, you get that, "Let me get the full picture," like over and over again with something like M3. So- Yeah ... it's not without drawbacks, but it's, you know, what's the, the triangle? Done well, done cheap, done fast. Pick any two. Yeah. And I'm, I'm picking, you know, cheap and a combination of fast and well. **Matt:** You know what? That, that image of like just constantly refreshing context- I kind of feel a little bit, uh, targeted 'cause that's, that's me . If I'm trying to do something... Like me trying to do a recipe, like cooking, oh man. [28:00] Matt Slager: You know, if I try and have a look, okay, I need to get that thing, that thing, that thing. All right, turn around, go to the pantry. What was it? Yeah. Like, and then I keep having to try and, you know, refresh my memory on which step is what. So yeah, I can definitely sympathize with that. That, that kind of operation in agentic operation [28:00] Matt Slager: stuff, I forget what the term is there, that, that is, um, that's what they talk about when they say, like, long-term execution. Like, you know- Yeah ... working on a job for longer time horizons. And yeah, M3, um, Minimax in general definitely suffers from that. Even with, like, a million tokens context window- Yeah ... it's more like you don't trust the million. You know, you could probably pull in a lot and get a, a nice distilled, aggregated summary question of all of that information, but then yeah, it won't do it for very long. **Dave:** Yeah. Yeah. Yeah, **Matt:** definitely. Which is one thing that they're saying that Fable does do, which is interesting. **Dave:** Which, which, I mean, we've been talking about this from day one, that, that the way context works, the way all these LLMs work is every time you give it a new prompt, it goes and loads in the entire conversation so far, plus any of its system prompts, et cetera, on top, and rereads that and then, you know, oh, cool, now I know what I'm doing. [29:00] Matt Slager: Yeah. Um, and so, you know, with something like [29:00] Matt Slager: M3, it's not as good at necessarily reinterpreting its past, and it doesn't trust... It al- almost doesn't trust its own context. **Matt:** Hmm. **Dave:** Um, and so it goes, "Oh, let me just double-check that again," and then goes out and checks the file system, does another read-through of things and other bits and pieces. And, and, um, you know, the Hermes harness isn't necessarily the simplest thing to be operating in either. There's a lot of ins and outs what Hermes can do, and so it's got to navigate around that. And a lot of what I'm doing is, you know, Hermes config choices. How do I set this agent up? I need to add this skill group. Is that a skill or is that an agent? Like, what do we do here and there? And so not even straightforward things where there's one right answer. **Matt:** Yeah. **Dave:** Which, which doesn't always help either. Um, and, and because some of my projects have been adapting from a Claude Code world into, into a Hermes world, and I'm just going through that at the moment, right? [30:00] Matt Slager: Where I had a Claude Code repo with all my Claude agents and Claude skills defined in the repo, uh, repository, for those playing at home that, uh, don't know what that is. It's just, you know, a folder full of stuff that you build something in And now with Hermes, everything [30:00] Matt Slager: kind of lives in the Hermes brain and the Hermes profile and the Hermes skills, not in the repository where you're working. Mm-hmm. It kind of lives in the Hermes, the harness layer versus the, the pa- the repository layer. And so I've got this weird blending of where I've kind of started recreating some of the, the Claude Code agent d- definitions in the Hermes world, but not all of it. And it's like, okay, are we doing handing off and telling it to pretend to be Claude Code, or are we actually gonna fully build this as a Hermes level project? **Matt:** Hmm. **Dave:** And there's no r- right or wrong. There's a, a more right or a less wrong or a more efficient or s- less efficient way of maybe doing that, but it's just- **Matt:** I've seen that. I've seen that with Hermes, that you, you can't have project level context. Which- **Dave:** Yeah ... [31:00] Matt Slager: **Matt:** s- simplest way for people who don't know what that means, or if you do, fantastic, you know, if you open up your Claude Code or your Codex agent in a particular directory, let's say it's on your computer and you open it up, your projects folder, and you've got, you know, project A, [31:00] Matt Slager: and you actually launch your agent inside that. If you have the respective guidance files, like the agents.md or the claude.md locally in that folder, those will be specific to that project. So you can jump into another project, and this information's irrelevant now. Hmm. So what Dave's saying there is all of his project specific context has to live globally in- Yeah that Hermes config. **Dave:** It's, it's like Claude Code or Codex is a battery, and you go to a different device and you plug the battery in, and then that device works. And if it's a remote control or a, or a car or a torch, whatever that battery plugs into, it makes that thing operate as its own device. Yeah. [32:00] Matt Slager: Whereas Hermes is not a battery. Hermes is kind of the person holding the batteries or something, and so you're always the person. And, uh, uh, yeah, I, I don't quite know what the Hermes analogy is versus the battery. I think the battery thing fits [32:00] Matt Slager: better. But yeah, it means that all... You always know everything. Hermes always knows everything about everything you've done. **Matt:** Yeah. **Dave:** Um, which is where I've been pushing this whole profiles thing a lot, because I'm running three or four different streams of work between the podcast, between, um, client style work, between, you know, my, uh, FX trading projects and, and, you know, my, my SaaS app and other things that I'm building. Uh, they all are kind of standalone, and I don't really need them to cross over into one another. And so Hermes has this concept of profiles. It all lives within the Hermes brain, but you can actually kind of have like sub Hermeses that know specific things and can have, you know, where they work and what they do. [33:00] Matt Slager: But there's an overhead level and, and thinking about how to set all that up, but it's good. And I've got, like, now three or four different Telegram channels. It's all the one Hermies install on the one computer, but they're [33:00] Matt Slager: basically three or four different instances of the agent. **Matt:** Yeah. Can, uh... I got two sort of questions, and these, these are not, like, candid, contrived questions. These are, these are real. Um, th- for... Do you think that kind of setup for a business, 'cause, you know, the idea of you building it means you're building it for time leverage, uh, for cognitive leverage- Mm ... um, you know, you're trying to displace some things to this system, or at least building a system to run the system. You know, where could a business somehow integrate some of this, or do you think it's even ready for a business yet? Or is it, you know, is there still security issues? Um, uh, do the negatives outweigh the positives? You know, like, what, what does this look like for someone in, you know, at any business level? **Dave:** It, it all comes back to the problems you're trying to solve and what does that business need to solve, right? **Matt:** Mm. [34:00] Matt Slager: **Dave:** And that's what it comes back to. [34:00] Matt Slager: Um, for, for me, it is leveraging. It's given me a virtual team, and so, um, a- and a soundboard, and hands to build applications, and projects, and calculators, and things like that. Like, you know, we- we're planning the AI Operators IR relevant in a few weeks time, and I'm looking at what the next steps are, and, and Rich and I are looking at running some sort of actual paid workshops and things. And so looking at venue hire, and costs, and ticketing pricing, and what makes money, and trying to do all those sort of calculations Traditionally, you go build a spreadsheet, and I'll open up Excel, and you start, you know, going to websites and getting pricing from different venues and putting it in. I sent my researcher agent to go and find venues in the geographical area that I, I, I wanna run the event in, and then it went and got me, you know, this pri- this place for this price and this many hours and blah, blah, blah, blah, and this one needs catering and this one doesn't. And then I said, "Okay, cool. [35:00] Matt Slager: Now build me a web app where I can then [35:00] Matt Slager: play with pricing and working out the cost of running the event versus the cost per ticket, and maybe if people come to the IRL event, we give them a discount ticket versus if they buy it after the fact to try and, you know, encourage sales. And if I sell X amount of tickets at that price versus this price." I don't wanna build that in Excel. So I tell it to go and build me a single-page app that lets me run those calculations and do that modeling and stuff, and it, it does. So I see the AI as a tool to build tools as much as anything else. **Matt:** Mm. **Dave:** Like, at the end of the day, I don't necessarily wanna be living in an AI in a chat, but the chat is the worker that goes and creates the other things that actually drive value and do things- **Matt:** Yeah **Dave:** um, for me. [36:00] Matt Slager: **Matt:** I agree. And I, I don't wanna, I don't wanna discount, like, any individual's effort, knowledge, experience, and capabilities, but it is kind of at that point now where there's a certain [36:00] Matt Slager: tier of maybe VA or jumping onto Fiverr or Upwork, you know, maybe on that lower end where you just wanted to hire someone for cheap to get something done. **Dave:** Yeah. **Matt:** It's probably at that point now where these agents can probably replace those- **Dave:** Yep ... **Matt:** realistically. **Dave:** Yeah. As long as it's not one of those things that need, requires a lot of pointing and clicking and configuring and stuff, right? Like I'm probably w- not gonna get a, um, any of the AIs to really go set up a Shopify store because it's not kinda built for that kind of stuff or as much. But, you know, but if you want a basic logo, and, uh, design's super subjective, right? And, and in my experience trying to design logos, and yes, we went with the, uh, straight-out-of-the-box, um, AI-generated logo for our show right there. But often where I've wanted to have some, a bit more control over a logo, I'll get some mock-ups from AI, and then I'll end up over in Figma manually tracing out and adjusting the final shape so I can get a nice, tight, cut vector image- **Matt:** Yep [37:00] Matt Slager: **Dave:** that I can [37:00] Matt Slager: use for something else, right? **Matt:** Mm. **Dave:** Um, so as these tools get better and plug into more things and there's more MCPs available, so you can plug your AI straight into Figma and it's gonna move the coordinates around, you might get luckier as they get better with the vision models and, and all that kinda stuff. But I think we're also seeing a pushback against the AI slop thing. I think we're seeing more people wanting to know artists are actually able to do artistry, and that the attention to d- to detail matters. So I think it's horses for courses. I think Fiverr's a race to the bottom, and it's just hard to get on there now and try and make any money anyway. Yeah. 'Cause it's all, um, people sort of underselling their things, uh, to try and get the next, the, the upsell and the next sell. Um, but for me, yeah, certainly I would probably hit an AI before I hit a Fiverr for most jobs. [38:00] Matt Slager: **Matt:** Yeah. That makes sense. Like, y- you just said, like, the, you know, injecting your AI into your systems and building custom [38:00] Matt Slager: stuff. You know, last week we talked about, um, the, the time gone era of Microsoft Access and- Mm-hmm ... you know, building custom databases and that sort of thing. I Have just been trying to follow the, the current WWDC, the Apple release stuff, you know, the, the presentations and what they're talking about. And for Apple developers, so people building iOS, macOS, um, everything is having massive jumps there too right now, which, you know, that's continuing to happen this week. It's probably, um, day three is probably still going, uh, at the moment. So, you know, they, they are actually having this entire Siri upgrade now that Google's powering for them, um, which, you know, Siri is Siri, right? Everyone doesn't really like it because it's been so bad for so long, and not many people really understand what it could be. [39:00] Matt Slager: You know? Like, there's these, these apps that we're using on the computer and the, [39:00] Matt Slager: you know, ChatGPT and Claude app and all that sort of thing, but how could your phone potentially be just natively that useful? Like at the... It doesn't really make sense to most people. But what is more interesting to me is that they're exposing all of that, um, I forget what the name of it is. It used to be called SiriKit, and now it's something else because it's the, the new version. Um, they're exposing all of that so that you can build iOS apps with this native on-device AI ready to go, which is incredible. **Dave:** Developing apps on the phone itself. **Matt:** Yeah, with local LLMs ready to go. Wow. Um, so, you know, think about what kind of- Yeah non-deterministic stuff you may- [40:00] Matt Slager: **Dave:** Well, I mean, I, I saw, I saw a friend on, on, uh, LinkedIn the other day posting that after having a stressful morning with his kids and trying to get them to school and then get to the train station on time and check how many parking spots would be available and what time the next train was coming and, um, missing it by a minute or whatever, that he went and used some [40:00] Matt Slager: widget kit thing to build a little widget for his iOS home screen using some LLM as well that now shows him, you know, available bays, next train times as a widget on his home screen. And so he did that, you know, offline and then pushed it to his phone. But those little productivity life hack type things- **Matt:** Yeah ... **Dave:** um, yeah. That's, that, that's wild. I mean- The whole, people like Apple and Amazon were ahead of the game with voice interfaces. Um, they've been doing it for a long time, but obviously they weren't LLM powered. [41:00] Matt Slager: They were very instructional based. I don't know of like how, what the backend of an Alexa looks like. Um- Mm-hmm ... but they're, they're bringing out their Plus version. They've been started rolling that out in the US where you can have an LLM in every smart speaker. Um, which that's gonna change things 'cause my kids have grown up with one of those on the bench, and they ask it questions all the time, and, [41:00] Matt Slager: um, show me a picture of this, and what does that look like, and can you define this and describe that. And so putting a GPT behind that is gonna be crazy. Like, 'cause that's- Yeah ... just one of those things that's, um, ambient, ambient technology. It's just there, and you're just so used to talking to it. I mean, even I was talking about food and whimsy. Uh, and, and I was saying, uh, my daughter loves meatballs, and I never wanna cook meatballs because I'm like, "They're harder to cook. They're harder to eat. They're, they're, they're not particularly functional." And then they're not, then they're not even, they're not even particularly whimsical. Um, and I said, "If, if my food is not, not functional, then, then, then I, I at least wanted to have some whimsy." And she's like, "What's whimsy?" Uh, and I'm like, "What's whimsical food?" [42:00] Matt Slager: And I was like, "Uh, like fairy floss." And my bro- my son's like, um, "Fairy bread?" I'm like, "Yeah, pretty much any food with the word fairy in it is probably fairly whimsical." But then she, she's like, um, uh, um, I don't wanna say the name 'cause I've got one on my desk here, and she'll start talking to me, but, [42:00] Matt Slager: uh, asked the, uh, smart speaker in the kitchen what, uh, to define whimsical, and we just do that all the time, just ask for a definition of something. Like, um, yeah, we don't just use it to play the radio or, or ask for the weather and stuff. So having broader intelligence or a better knowledge source, 'cause at the moment, half the time you ask it for things, it goes, "Here's something I found on the web. Here's an Alexa Answers contributor," or whatever- Yeah uh, like has, has given us. So yeah, I'm, I, I think with Siri now bringing that in, um, with people are just gonna say, "Hey, Siri," and have access to that- [43:00] Matt Slager: I wonder how many you just triggered. Mine doesn't have it on, so I'm fine. Um, we were watching, we were watching a movie years ago, and it was a magician on the stage. And, ah, again, I'm not gonna do it now 'cause it will actually trigger, but his, uh, [43:00] Matt Slager: assistant, his magician's assistant had a name that began with A. And, um, on the movie it's like, "A, turn out the lights," and our whole house just went dark. Fwoo. I was like, ah. That's good. That is, is... Was that a deliberate, deliberate Easter egg? Um, very, very funny. Likely. Yeah. Um, **Matt:** yeah. **Dave:** Recently enough it was like one of the Adam Sandler Netflix films. I think it was one of the murder mystery ones he did. Mm. Um, yeah **Matt:** There's been a little bit of, uh, chatter in the, in the comments. R- Richard's actually joined us, um, in the comments there, you know, giving a, a little bit of encouragement. Um, and then Dustin's actually mentioned something really important that I kind of forgot about, and where Jony Ive actually did that partnership with OpenAI, and yeah, they were gonna call it IO. [44:00] Matt Slager: I think Google cracked his, the, put their foot down and, um, decided that they couldn't na- name it IO. So, [44:00] Matt Slager: I'm curious to see where it goes, 'cause it was so- gonna be some sort of wearable, you know, maybe like- Like a pendant or **Dave:** something they were talking about maybe? **Matt:** Yeah, like, like the Fieldie- Or- ... or the, the Plaud, P-A- Yeah P-L-A-U-D. **Dave:** Or the po- the Pocket. I've seen Pocket advertised a lot on YouTube. **Matt:** It makes me think of the, um, the teenage engineering rabbit and also, um... What was the other one? The, the Humane Pin. You know, like the- Yeah ... the wearable AI devices that everyone thought were hilarious and stupid. I think those were ahead of their time. **Dave:** Mm. **Matt:** Um, they, they probably were things that we're gonna see soon and everyone's gonna actually really like them, and no doubt it'll be like OpenAI's Jony Ive thing- I- ... or Apple will come out with something and everyone's like, "Wow, this is amazing. Why did nobody think of this?" You know? Well, **Dave:** that's, that's, that's weird. [45:00] Matt Slager: That's... Yeah. I mean, I, I always think of that Tom Hanks movie, The Circle, um, where everyone wears these full-time camera tracking AI things and it becomes this [45:00] Matt Slager: whole Big Brother conspiracy thing and, and it becomes pretty dark and full on. Which, look, there's always... All these AI technologies could go either way, right? They're either gonna be amazing for humanity or they're gonna be the end of us, or they're probably just gonna be eh. And I think that's, I, I think we're seeing that eh a little bit at the moment, right? We're like, "Oh, it's gonna lose all the jobs. Everything's gonna go, and, uh, and we won't need graphic designers, we won't need programmers." And then as we reach this tipping point where we go, "Yeah, it's good, but it's, it still kinda needs the humans," 'cause we're looking at what it's just doing by itself and it's, it's not amazing. Um, and like even the end of SaaS, SaaS is dead. Just build your own applications, and then people actually start trying to build them and use them for more than three days and go like, "Ah, there's reasons why companies exist to maintain and build these things out," and, and we'll swing back the other way and we'll land somewhere in the middle. But- **Matt:** Yeah ... [46:00] Matt Slager: **Dave:** I mean, even, even me trying to build my own harness, and look, it's great that PewDiePie has built his Odysseus harness and put that out. He's also probably not building that [46:00] Matt Slager: himself, but he's got all his YouTube money and is paying people to help him do that kind of stuff, right? Like, he's got technical resources beyond a, uh- small YouTuber like us, um, trying to build a harness. But I went, you know, I'm looking at what, what Noose, Noose Research are doing with Ho- Hermes and that. And when I can't compete with a team of specialist people building a thing versus just me with my AI trying to build it, um, there's a reason why these things need teams, and need specialists, and need knowledge. Because they're actually really hard and complicated to build, let alone maintain, and roadmap, and build new features, and keep them relevant. **Matt:** Yeah. **Dave:** And any SaaS, CRM, et cetera, like if you can just... Like, we go back to the Microsoft Access thing. If it's just for you and it's a bespoke tool that you wanna use for your personal needs, that's one thing. [47:00] Matt Slager: But as soon as you wanna build something that's an actual product... And we'd, we'd, we'd get this back in my Salesforce days, where you go, "Is your business building a CRM or is your business doing whatever it is your business [47:00] Matt Slager: does?" Like, why do you wanna be responsible for... A- and that was even the argument for not self-implementing and going, "Get one of our partner organizations that is a specialist in implementing to come and put it in." Because they know the rules, they know how the system works, they know the limits. Let them get you set up correctly to then use it, because your business is not being experts at implementing CRM. Your business is whatever your business was. Whatever **Matt:** it does, yeah. **Dave:** Um, a- and so I think every- everyone going, "We, we can be our own SaaS company," you probably shouldn't be, 'cause that's... Unless that's what you really wanna do and pivot your company to being a software developer, um, it's probably not. **Matt:** I've had, um, I've had my full, like, fair taste of these things. 'Cause yeah, you're right, that building something for yourself that, you know, only you or maybe, like, some of your team use as, like, an internal tool is completely different to actually building something that's production grade, hosted, secure- Yeah [48:00] Matt Slager: and has every possible edge case handled. You know, that, that [48:00] Matt Slager: kind of thing... Like, I did my best to do a hands-off build recently, you know, in the sense of, okay, I have my goal, my attention, everything's fully planned, specced out. I'm gonna delegate this off to a long te- like, a long-running agent system. Yeah. For that I was using, I was using Droid and I was using their Missions system. Yeah. So Missions were basically, like, what slash goal is. **Dave:** Yeah, **Matt:** yeah. Or it's probably, like, a mix of slash goal and Ultracode. But you know, this was organized by the factory team, and it's been around for ages, but- Basically, I ended up with more than two days straight of this mission running, and I just kind of kept feeding it. You know? I just... If it needed questions answered, I would answer it. Came out the other end, and it was magical, and nothing worked. Like, it was hilarious. Like, I had to go through and, you know, use my, uh, limited discernment to figure out what it had done, completely defragment the whole thing- **Dave:** Yeah ... [49:00] Matt Slager: **Matt:** before anything even started to [49:00] Matt Slager: work. And that was something that needed to be run in production live for a client. So I, I, I just tested it, you know? I was like, "Let's just see how this goes," and, yeah, fill in the space. So that kind of full hands-off is, is not possible yet, you know? Especially for a production grade system. **Dave:** I see with debugging stuff, w- whether it's debugging an app that it's built or whether it's debugging my, my own agent setup where something's not working properly and, and I'm trying to use... Homie's has this sort of Kanban project or orchestration system and- You, um, say, "Oh, that's not working. Can you check that or should we do it like this?" And as soon as you watch it start looking at different files and how it's actually going through the debugging process, you go, "Why, why are you doing that? Stop. Stop. Just go and do X, Y, and Zed. Like, I, I thought you were clever enough to work it out and I didn't have to be super specific and I thought you'd might know better. Clearly you don't know better. Clearly I need to be a bit more directive on exactly where I think you need to look for certain log files or whatnot." [50:00] Matt Slager: Because, and maybe that's [50:00] Matt Slager: an M3 limitation. Maybe if I was using, you know, Fable I'd have a lot less of this sort of rework issue. Uh, and maybe that is the consequence- Yeah ... um, of a cheap model in that I do need to be more directive, but that's not necessarily a bad thing either, right? It keeps my hand- Mm on the, on the steering wheel, um, **Matt:** a little bit more firmly. It's like it's that fatigue that, um, that atrophy that we've talked about in the past, you know, where people are saying that they're, they're, they're delegating all of their thinking and their decision-making and their discernment, like the reason why humans are the way that we are, **Dave:** Mm. **Matt:** You know, they're delegating and deferring all of that off and then all of a sudden realizing that they feel really disconnected from the, from the system or from the project- Yeah ... and they have no idea what's going on. So, you know, if, even if they did wanna jump in manually, let's say that they're a, a, a super experienced coder and they wanted to go in and check stuff, they would almost not know where to start because of- **Dave:** It- [51:00] Matt Slager: that system ... it's true. Like, like I've, I've got [51:00] Matt Slager: pipelines to try and build trading strategies and stuff, uh, you know, 'cause I'm still, you know, dabbling on, on... It's a strategy, it's not gambling. Um, and it'll go and, and it'll come back and go, "Yeah, we've tried like this, this, this, this," and ba, ba, may, a list of like six different variants with acronyms that I don't fully understand, and I'm like, "This is different from when I had an idea around how I thought price action moved," and said, "Go investigate. If it bounces here and goes there and does that, let's run a research project around that strategy." Now it's just building a dozen things. I'm like, "I don't even know what those are." Like, like, you can't... And it says, um, "We, we need your input. Do you wanna go in this direction or this direction with this one?" I'm like I don't know anything about any of these. You freaking pick. You're, you seem to be the expert that knows everything, so you pick a decision and just work out how to make money. That is your goal. Go make money. Find profitable strategies. Stop coming and asking me for input that is irrelevant from my point of view. I don't, I don't care. I don't need to know, so. [52:00] Matt Slager: **Matt:** Yeah. It makes me actually really, really conscious of [52:00] Matt Slager: there's, there's a dissonance right now between how the human should operate and orchestrate and navigate, and then how the agent should, and, and what those two separate SOPs actually look like and how they should communicate to each other. B- uh, picture it like your, your corporate experience. You know, if you, if you're coming to your, your team lead or your, your department head or whatever that chain of command is, they don't necessarily want to listen to your huge story or all of the random context that you feel like is absolutely essential. They just wanna hear what's the problem, what's been your attempted solution, and is there a means for me to provide advice or a decision to make. You know, that kind of thing exists with this AI and human building loop where we only wanna know what decision do you need me to make right now. Okay. You know, what do you need me to actually look through and think about? [53:00] Matt Slager: Because if you [53:00] Matt Slager: try and tell me all those random acronyms and all that random technical jargon i- in the strange, confusing way that they do sometimes... Like, I'm a technical person, and I try and read responses sometimes and I just end up yelling at them saying, "Can you just say that again in English? 'Cause I have no idea what you just said." Like the, the logic and the, the, the cadence- Yeah ... and whatever throughout the, the sentence, so. **Dave:** Now, now I feel targeted, Matt. I, I have a story where in my corporate days I had to go to, like, a QBR, quarterly business review, and report to, you know, someone who was three levels above me about something. And I was explaining this story of success that we'd had. And I had, you know, it started with this person in this room, and then we talked about this thing, and that triggered this idea, and then we did this and this and this. And my boss tore it apart and said, "That is way too much detail." I'm like, "But it's important though." "Just, just get to the point." I'm like, "But it's a good story." "Nope. Doesn't matter. They don't care." Like, "Oh, but it matters." "No, it doesn't." I'm like, "Freaking [54:00] Matt Slager: **Matt:** hell" Don't worry. I'm exactly the same. Like, it, it's been How am I, what am I now? It's been 34 years for me to try and, [54:00] Matt Slager: like, actually communicate properly without telling a story like Sheldon Cooper from The Big Bang Theory. **Dave:** Mm. Um, I think, I think with all this AI stuff, and, and, and I've talked about this before, I use the term ambient context that humans have. **Matt:** Mm. **Dave:** We have just, we know stuff because we're around or we're online all the time, and we've absorbed so much over our, our 34, 40, 50, 60 years, whatever, however, 20, what- wherever you're at on the spectrum, right, of age, the age spectrum. **Matt:** And the other **Dave:** ones. Um, and the other spectrums. All the spectrums. Um, there's, there's spectrums. Everyone's on there somewhere. Uh, and wherever you, you land. But, um, we absorb and know so many things, and our brains are these complex neural nets that just connect memories and connect patterns, identify things, that we've been doing our entire lives. [55:00] Matt Slager: And so to put a few paragraphs of text into a prompt and expect the robot [55:00] Matt Slager: To reason and knowledge and think as well as you can. And yes, it's got a big training parameter set and it's been trained on millions of ideas and memories and things, but it's not the same- **Matt:** No ... **Dave:** for your situation, for the problem you're trying to solve, for the thing you're trying to do. And so when I'm asking it to do stuff, in my head I've already connected half the dots and go, "I'm pretty sure it's gonna do it like this and this and that, so... And it's pretty smart. If I just give her this, then..." And this, this comes back to as we've seen advances in these things, as I've now got, you know, agents with sort of reasoning personalities and soul MD files, et cetera, I become lazier in my prompting 'cause I expect more from the robot. **Matt:** Mm. **Dave:** And I think it comes back to, uh... 'Cause I used to, with the ChatGPT days, I'd always do prompt inception where I'd tell it all my details and then tell it to write the prompt. But now I just, just expect it to know things. I'm not meta-prompting, I'm not inser- prompt inceptioning as much, and maybe I need to go back to that. [56:00] Matt Slager: **Matt:** I think, um, I [56:00] Matt Slager: think there's a lot of logic, and I'd be curious to know your opinion and anyone who's still here live with us, or even if after the fact, um, if you agree or disagree with this, and i- this will help a lot if you've had sales experience or some sort of, um, personal kind of coaching in the past, or if you are a coach. The... I reckon what we describe as, like, prompting- Mm ... you know, speaking to AI systems and agents, you know, when they have all of that, that parameter training stuff you talked about, but they don't activate it the same way that we do naturally. **Dave:** Yeah, **Matt:** yeah. I think it requires a certain type of framing and, you know, like pre-suasion. [57:00] Matt Slager: I've literally got Cialdini's book here. The, you know- Right ... the idea of y- actually steering a conversation with, like, conceptual bias, so you start to lead towards a reality that you're heading towards without [57:00] Matt Slager: using specific terms. So a really good example of this is somebody this morning, I was reading, they were screwing around with Fable, and they were saying, you know, instead of saying, um, "Don't over-engineer this. Keep it simple," you know, that's such a weird thing to say. They were like, "Hey, this is probably gonna be deleted in a few weeks. We just wanna experiment here." It's like- Yeah ... they're the same thing, but they're different because there's all that extra framing involved. **Dave:** Yeah, I, I was having a chat with one of my, my team today and saying to it, uh, 'cause, and this is the, the story where it's adapted the Claude repo and I'm going to Hermes, and I'm like, "Okay, what do we do next? [58:00] Matt Slager: Do we try and continue to live this more like a Claude thing, and we try and delegate and outsource those, those definitions, or are we bringing this fully into Hermes?" And I use terms like, um, has, is, was one of the issues that the, it's tried to absorb all of the different agents that we used to have into a single skill like a Borg, uh, uh- [58:00] Matt Slager: Mm and, uh, rather than have individual people, or should we actually have the individual personalities? So, and then, like, 'cause I'm not sure which... And you sort of frame and go, "What, what are the pros and cons? Like, which, which way do we go? I'm thinking this, but maybe this. What am I missing?" And I, I say, "What am I missing or what don't I get?" all the time, because I want to bias it into where I think we should go, because I have that human context and that human intelligence, but I also know that there's a lot of gaps in my knowledge of the technicalities of how some of these systems work, and it knows itself in some ways better than I do. **Matt:** Yeah. **Dave:** And so I want its point of view as well, um, so that way I don't just, you know, force a square peg in a round hole because- Yeah ... I only know about square pegs. **Matt:** It k- it makes me think of Sycophancy and, um, you know, the idea of the agent sort of feeding your own ego and dopamine throughout the thread, whereas you kinda need to do it- Opposite **Dave:** Yeah, I, I get none of that. [59:00] Matt Slager: I get no satisfaction from my agent telling me how [59:00] Matt Slager: clever I am. I'm just like, "I'm clever when this fricking does what it needs to do, and until it does what it needs to do, and until you do what I tell you to, I'm not clever." This is just me, like, smacking my head against the wall. Yeah. So just get it done. Like, I'll feel clever- One of my- ... when you do your job. **Matt:** One of my favorite, uh, loops, you know, you probably see in the tech community, people are talking about, like, what's your loop? You know, that's gonna be, that's gonna be just the bigger thing that's gonna keep coming around. But it's, it's what, it's what we've been talking about for ages. Since we started this podcast, like, I've been preaching the words of Geoffrey Huntley, you know, Mr. Ralph Loop. Um- Yeah ... and one of the things I love to do is force it to get a third-party opinion, you know? Mm. Literally I'll say, like, use one of the, the headless executable commands, like whether it's through Codex or whichever harness you have, um, and go ask. [01:00:00] Matt Slager: Go, go delegate a third party, you know, review with zero context of all of this thread that we've gone through so far. And, [01:00:00] Matt Slager: or I'll do it manually, where I'll have, you know, one agent working and then the other one I'll be like, "Can you just check what's happening right now? You know, do you have any opinions?" And I'll take that opinion and feed it back to this one and say, "Hey, I peer reviewed this. What do you think? Can you find the false positives?" And then it will go through and, you know, one by one- Yeah ... check it and be like, "Actually, these things are good." So, you know, that's the kind of stuff that you'll miss. It's like you, you're saying, like, what are you missing? You know, what am I missing? Um, there's so many ways to frame things psychologically when you're speaking to agents such that they aren't just missing things in plain sight. **Dave:** Yeah. And may- maybe I need to do more mix and model stuff, 'cause I'm just leaning just on M3, and, and the way I've gotten myself stuff set up now is I kind of got the Hermes default agent, who's kind of my cis admin across all the different Hermes things. [01:01:00] Matt Slager: But then I've got the, you know, my, my Perry White personas, who are my orchestrators across my different projects. Um, and they're... At the moment, everyone's on M3. So if I wanna, you know... That [01:01:00] Matt Slager: says the architecture of my orchestrators, I go to the Hermes default level, but as soon as I wanna work within any specific project context, I go to those Perrys now. Yeah. Whether that's, you know, through the desktop UI or through Telegram or Discord. I haven't, haven't really got them set up in Discord yet. But Telegram's been really good. Like, last night My son, he's, uh, wanted to play some Minecraft and, and my computer's the best one for doing that. So found out and, you know, I can, you know, switch users and leave my user logged in in the background, which means all my Homie's gateways stay running. So he can sit here playing Minecraft, and I can go sit on the couch with Telegram and still be working away doing, get my agents to progress projects and, and ideate with me and, and build things- **Matt:** That's **Dave:** good ... in the background. So, um- Yeah, **Matt:** very nice. **Dave:** Yeah. But I think that cognitive separation for me, I know I'm talking to this, this, this channel, this robot about this particular project, and it keeps that in context. [01:02:00] Matt Slager: I go to this one for this. People when they f- were first getting their OpenClaus and they, "I've got OpenClaus in Telegram, and I just ask it all these things, [01:02:00] Matt Slager: and I ask it all these things, and I ask it all these things." If they're doing all of that on one single Telegram thread- Yeah ... that's a nightmare. How are they getting anything done? Yeah. **Matt:** Yeah. **Dave:** Now even then I've realized now I've gotta, like, 'cause the Telegram session resets at 4:00 AM, so if I wanna... If it did something when I went to bed, then I wanna refer to it, I've actually gotta, like, reply to the previous message to make sure it pulls in that context, 'cause otherwise- Mm it's a fresh session. Or if I've been doing stuff and I wanna go, "Actually, a new topic now," I need to go /new in Telegram like any other terminal interface to clear the context and start with a fresh conversation. So all those little things that the general populace who are buying into these things and just going nuts with it, I'm like, "No wonder people have mixed reactions to these and don't think they're as good as they could be." **Matt:** Mm. **Dave:** Um, or they just find it all way too hard, and now they just stick with ChatGPT 'cause they go new chat and talk to it, and it gives me a recipe. [01:03:00] Matt Slager: **Matt:** Yeah. That's it. And- Yeah ... it's a good, um, good way to end it, uh, [01:03:00] Matt Slager: being that, you know, right now Claude's back in the spotlight with Fable. **Dave:** Yeah. **Matt:** When 5.6 comes out, that next GPT model, they believe that it's Mythos level, so. **Dave:** Well, w- do they jump to GPT-6 at that point then? **Matt:** Yeah. Well, people were saying that 5.5 should've been 6, but- Mm ... they didn't do that because it would've overestimated what 5.5 was actually releasing as. So yeah, I don't know. Do they jump to 6 now, or do they do 5.6 and it's almost like a, a, a flex to be like, "We haven't even gotten to 6 yet." **Dave:** I don't know, man. Yeah. I don't know. **Matt:** It's interesting. I don't know. **Dave:** There's, there's so many ways to solve every problem. There's so many platforms and solutions that even for people that are out there feeling overwhelmed, that's fair enough, right? Um, don't be afraid to go get a second opinion. People are like, "Oh, I, I don't... [01:04:00] Matt Slager: I'm so far behind," or, "It'll look stupid if I ask questions." But you've gotta have the conversations. I mean, I value this podcast where we get to chat and talk out ideas, and I, and I hear things as well, right? [01:04:00] Matt Slager: Because there are so many ways to skin the cat, and there's so many different tools, and are you doing things in this model or that model? Are you doing it, like, on your desktop? Are you building tools? Are you using Convex or Airtable or Supabase or which, which back-end system do you need to use? Do you need to use any back-end system? Should I be using Hermes or OpenClaw or Cowork or Codex or- It's all relative, right? **Matt:** Yeah, exactly. You know, someone in our circle literally the other day was saying, "What are you doing? I wanna learn." And my response was that of, "What do you need?" Like, you can't... Unless you actually wanna go play, and, and that's fine, and that, you know, I totally endorse that. If you don't have a need for it right now, and you can't see a need for it right now, you probably don't need it. [01:05:00] Matt Slager: **Dave:** But yeah, I mean, they even wanted to play. I, I told the story a few weeks ago how my son said, "I want an agent." And I was like, "Cool." And he had Hermes before I [01:05:00] Matt Slager: did. And I set, I set Hermes up on a VPS for him, and he didn't touch it. And like, look, yeah, I gave him, like, rubbish free cheap quota open router plans, and they were really slow. But I've since moved him onto my Minimax OAuth because I've got enough room, and he barely uses it, that now he has access to M3 as well. He still barely touches it. I actually went and set up a cron job for him that sends him a 3D printing idea every day just so he'd start seeing it and using it and maybe getting, getting creative and getting some ideas out of it. But he doesn't really have a use for it, and so it's like I walk through Bunnings, and I see so many amazing toys, and I'm like, "Ooh, I'd like some of those toys." "What do you do with it?" "Um, don't really have a project. I'd just really like the toy." **Matt:** But you would have it. **Dave:** I would have. If a, if a, if money wasn't an object, I would have a... [01:06:00] Matt Slager: And I had space in the garage, I would have more toys 'cause I might need it one day. That'd be handy. I'd, I'd love, I'd like to have one of those. But you don't, you don't need it. And I think so many of these things, they are [01:06:00] Matt Slager: tools that are just toys if you don't have a purpose for it, and it's just gonna sit there and cost you money and collect dust. **Matt:** Yeah. **Dave:** And I think businesses are seeing that. They tried rolling out agents. They didn't really know what they needed agents for. They didn't really drive value for them. Um, they don't know how to build tools for themselves. They, they might think they wanna build a tool, and I know some small businesses out there where they are doing that, and they're using Lovable and Claude and other bits and pieces, and they're building their own little internal apps or websites. That's awesome. If they are technologically wide enough to do that themselves, that's sick. Many aren't, and they should just be building their businesses and talking to people like us if they want guidance on how to use the AI to build tools, if they want- uh, optimize their processes, 'cause so much of this is just process and workflow optimization, and the AI is just a tool to help with that. Yeah. **Matt:** Agreed. Yeah. [01:07:00] Matt Slager: **Dave:** Well, just as we're, uh, we're spiking on viewers, we're, uh, we're gonna wrap up. We've hit the hour. We try to keep this about an hour show, podcast. [01:07:00] Matt Slager: Uh, thank you for all that have interacted in the chat. Uh, good to have everyone with us. If you've got questions, don't be afraid to ask. Drop a comment. Make sure you like, subscribe, hit the bell icon. That's how you know when we go live every Wednesday, midnight, min- midday, min- mid- midday. **Matt:** Midday as well. **Dave:** Midday, yeah. Well, min- Yeah ... no, we're not on Mondays. Mid- midday on Wednesdays, uh, Eastern, Eastern Time. Uh, when I say Eastern, I mean Sydney and Melbourne, 'cause Queensland changes its time half the year, which is weird. **Matt:** They're the... Are they the smarter ones, though? They probably are. **Dave:** Yeah, actually, we're the ones that change our time. They stay the same. Yeah. I, I know in WA they, they all feel very intelligent for never changing their clocks- **Matt:** Hmm ... **Dave:** uh, being, being from there. So- **Matt:** On that note of, um, comments and chats though, yeah, thanks very much everyone today for sending in. [01:08:00] Matt Slager: And if you're watching this after the fact, not live, absolutely put in a comment because, you know, the... [01:08:00] Matt Slager: We, we wanna go back and actually read those comments, and it gives us another opportunity to build a system to check for new comments and that sort of thing too, to interact with you guys. Um, yeah, once again, thanks for listening, and thanks Dave for- And- for having me. **Dave:** And, and if you're in the northwest of Sydney, make sure you go to aioperatorspod.com, and you can hit the IRL link at the top and get the details for our upcoming event in two weeks time. Uh, click on the link and reserve your seat. We're, uh, half the seats are gone, so we are filling up fast, which is good to see. So if you're in the northwest of Sydney and wanna come and join me and Richard and others, um, for a night talking about AI, hearing some stories around how small businesses are using AI, trends in AI, and just rub shoulders and chat with other people that are pursuing and, and understanding this, just, uh, come on. Free event. So look forward to see you there. **Matt:** Yeah. I'd love to be there. [01:09:00] Matt Slager: **Dave:** One day. [01:09:00] Matt Slager: We'll, uh, we'll, we'll run one in your neck of the woods. **Matt:** One day soon. **Dave:** All right, I'm gonna wrap it up there. We'll play the stinger. And thank you all, thank you Australia, thank you the world. We'll see you all next time.