Smart Models Do The Wrong Thing Silently. Why Your AI Agent Just Made You Dumber.

[00:00] Dave Pengelley: Next week [00:01] Richard Webbe: Next week [00:01] Dave Pengelley: Next week [00:02] Richard Webbe: We miss Mano [00:03] Dave Pengelley: Oh, there we go. The blue's much brighter than the red. The red's dark. Blue... I mean, look at that. Look at the bounce off the wall with that blue. [00:10] Richard Webbe: Mm. [00:10] Dave Pengelley: That's much better. [00:12] Richard Webbe: People are enjoying watching you, uh, do the lighting. [00:14] Dave Pengelley: Yeah, I mean, I can go, like, really crazy, like weird colors and stuff. [00:19] Dave Pengelley: Ooh, romantic. There we go. [00:21] Richard Webbe: There you [00:21] Dave Pengelley: go. I don't normally play with my lights, but I'm set up in a different area 'cause you're here. [00:25] Richard Webbe: Yeah. [00:26] Dave Pengelley: Um. [00:27] Richard Webbe: What do you- [00:28] Dave Pengelley: Let's just go the bright blue. That's... bounces. Nice. 'Cause that's one of the things I've found with playing with lights, the blue lights do s- like, as much as you go, "Oh, well, I'd like to have fancy color lights," you can't see them half the time. [01:00] Dave Pengelley: They just, they just... they're so dull if you go different colors. Greens are okay, the greens and blues. But, you know, all the reds and pinks and stuff just die. Yellow, you know? Those are normal. Like, it's nothing. What is that? Anyway. This is what we do on the show, we talk random rubbish at the [01:00] Richard Webbe: start. People watching are losing the will to live a bit [01:01] Dave Pengelley: now. [01:01] Dave Pengelley: No, they are not. Everyone's like, "Ooh, smart lights. What's he doing? How can he control the lights with his phone? That's [01:09] Matt Slager: incredible." What AI model is that? [01:11] Dave Pengelley: Uh, that is a Mirabella Genio. That is what that is. They're great. Mirabella Genio smart lights, so much cheaper than Philips Hue. I've got Hue in various places around the house. [01:21] Dave Pengelley: I think in my normal studio space that you see me in, which is, like, on the opposite side of the ducts there, um, I've got Hue strips behind bookshelves, they give me my normal color. But Hue needs a special hub, and then Hue lights are just... They've got a price premium on them because they're Philips, and they cost a lot. [01:41] Dave Pengelley: Whereas the Mirabella ones don't need a specific hub, they kind of use a Wi-Fi protocol, and they're much cheaper. [01:49] Richard Webbe: I used the Hue ones, set up my house lighting the way I wanted it to, and then every few weeks it'd lose all the memory and I'd have to reset it up, so I gave up. So I went back to just [00:02:00] normal white lights. [02:01] Dave Pengelley: We have, we, we have smart lights everywhere. Most of them are Mirabella, because Mirabella make these bigger, cheaper, like, bigger lumen ones for living rooms and, and stuff. These big UFO ones that only Kmart sells. That's the weird thing about Mirabella's Genio strategy, different retailers get to sell different products. [02:16] Dave Pengelley: There's no one shop you can go to to get the full range of Mirabella Genio lights. It doesn't work. Yeah, right. Yeah. And the Kmarts around me don't sell half the stock anymore. So we've got some lights, and a few of them break occasionally when, you know, I'm swinging the kids around and kick them in the ceiling, and I can't replace those lights ever again. [02:34] Dave Pengelley: No. I'm like, "Do not break, do not break the ceiling lights." We have to be careful. [02:38] Richard Webbe: There you go. [02:39] Dave Pengelley: Yeah. Anyway. Uh, Matt, how are you going? So this is, this is weird, Richard. This is [02:45] Richard Webbe: pretty odd. It is weird to be in the same space, but it's good. We had a good night last night, as we were explaining to Matt, with our first AI update information IRL presentation. [02:55] Dave Pengelley: It's, it's the AI Operators IRL [02:58] Richard Webbe: event. There you go. Close enough. We're [02:59] Dave Pengelley: adding too many words. Did [00:03:00] [03:01] Matt Slager: you, did you learn what IRL means, Richard? [03:03] Richard Webbe: In real life, sir. [03:05] Matt Slager: Yes. [03:06] Richard Webbe: I did think it was some technical term, and then I kind of maybe on my own worked it out. Just. [03:11] Matt Slager: No, that's very good. [03:12] Richard Webbe: Just. [03:13] Matt Slager: It reminds me of my, uh, my childhood playing many computer games, talking about being AFK and BRB and need to go- Yeah [03:20] Matt Slager: uh, see each other IRL. [03:23] Richard Webbe: Yes, very much so. But we... It was a, it was a good night last night, I thought. We had a good group of people, uh, a lot of questions. One of the, uh... He's a AI consultant, um, quite experienced, and came up and said, he said, "This is great work you guys are doing, 'cause people need this level of discussion and education." [04:00] Richard Webbe: And one of the best parts of it was, Matt, was, as you know, the interaction. So we, you know, we put up a few ideas and talk about strategy. We had a couple of users like Kelly and Sean come on and talk about how they do what they do and- Dave gave a great rendition about agentic and how agents do what they do, and, um, [04:00] Richard Webbe: everyone bounced back really positively and said, "That's great questions." [04:03] Richard Webbe: That's awesome. And yeah. So I think we have a good agenda. Well done. I think it's, uh, well needed, and it was good to have people from the audience come up and say, "Wow, that was great." [04:10] Matt Slager: Yeah. That, that's fantastic to hear. Wa- was it recorded? Like, is there gonna be clips or anything like that? [04:15] Dave Pengelley: So, um, we had a, we had a great show sponsor, uh, IKA Tech Solutions that were there- Legit [04:22] Dave Pengelley: filming and taking photos. I also took my own GoPro as a backup just 'cause I wanted my own copy as well, and this morning when I turned it on to download it, it said, "Overheating issue." I'm like, "Oh, no." Um, so my own GoPro's clapped out at 20 minutes. So I think we got you An- uh, you, Richard. [04:38] Richard Webbe: Did you? [04:38] Dave Pengelley: I probably got you and, and maybe Kelly. [04:40] Dave Pengelley: I haven't, haven't had a good look yet. Uh- That's good ... but I hopefully, hopefully IKA's recording, uh, is gonna be more complete. Uh, I did give them the brief to focus more on the snaps for, for marketing and promotion than full recordings of the, of the talks. But, uh, yeah, [04:55] Richard Webbe: hopefully we get something. If you wanna learn, come to our next one, which is gonna be in Melbourne, which [00:05:00] Matt, you will be there, Dave will be there, I'll be there, Sean will be there, and, uh, we'll have [05:06] Dave Pengelley: some- Yeah, I can, I can, I can come and give my AI agents talk again. [05:09] Richard Webbe: Yeah. [05:09] Dave Pengelley: Happy to do that. [05:10] Richard Webbe: Yeah. Cool. [05:11] Dave Pengelley: Yes. [05:12] Matt Slager: Super cool. Dave, what, that, that rabbit thing seems to be popping up a lot. What's- [05:16] Dave Pengelley: Roobot. It's [05:18] Matt Slager: Roobot. [05:18] Dave Pengelley: Yeah. It's, it's our kangaroo. It's our ro- Ah. ... robot kangaroo rabbit thing. [05:23] Matt Slager: Successfully made it [05:25] Dave Pengelley: But that, that, that's our new show mascot, the r- the Roobot. Um- [05:27] Matt Slager: Very cool. [05:29] Dave Pengelley: Yeah. Um- [05:29] Matt Slager: And that, that particular theme there, that seems very, very DC. [05:35] Dave Pengelley: Yeah, this is, um, you know, a bit of GPT-5.5 image gen helping out with the, the styling for Roobot. That's kind of the master style there. But, uh, I like it. It's nice to have, have this guy. Like, he popped up a few times through my talk in, in different poses doing different things. [06:00] Dave Pengelley: I love this one. It's so good. Um, so yeah, it's, it's good to... As, as we evolve the show, evolve the brand, we'll, we'll keep evolving things. So [06:00] Dave Pengelley: we've got Roobot now. Uh, this is... I'm using Gamma for this presentation, and branded Gamma fully up as AI Operators, which is pretty fun. [06:08] Richard Webbe: Cool. [06:09] Dave Pengelley: So yeah. Yeah, it was good to do the event. [06:11] Dave Pengelley: It was a lot of, um, a lot of those sort of first things first, like, and, and doing all the things, setting up all the... Anyone who's set up an event before, the logistics, the registration, post-event survey forms, printouts, all the promotion stuff, there's a lot to actually run even a simple event. [06:28] Richard Webbe: Mm. [06:29] Dave Pengelley: Um, so it was good. [06:30] Dave Pengelley: Good learning, good lessons, uh, good, good turnout, uh, good crew there. So- [06:34] Richard Webbe: Kelly was great ... yeah. Sean. Kelly is a, a, a user gave a very heartfelt talk about how AI had given her her family back, and then went through all the processes and things she did with her husband to speed up their lives and get themselves more time back. [06:48] Richard Webbe: I thought that was a great talk. [06:49] Dave Pengelley: Yeah. [06:50] Richard Webbe: And Sean was really good in showing what he does for the, uh, the, the, the quoting and trading and industry, and how he- Yeah ... he went through the whole process step by step [00:07:00] of how he's linked AI together. So it was, it was... I thought it was really good. [07:05] Dave Pengelley: Yeah. Yeah, Sean reinforced during his demo that like most of the value's in the workflow, and you just use AI where you need to use AI. [07:11] Dave Pengelley: And when it comes to things like maths and quoting, don't use AI. And really broke that down and showed people- Which is good ... which is good. I mean, that was a common theme I think around- That we carried through the whole show, which we talk about on this show all the time, about deterministic versus probabilistic and where do you need agents and where, where you don't. [07:26] Dave Pengelley: And so when I started talking about agents, that was sort of a, a key aspect of my talk. So it all tied together. We all leveraged off each other. It was, it was good. Looking forward to the next one. We'll get some, uh, um, we don't expect Kelly to travel down to Melbourne. We'll get a local operator from down in Melbourne, get some, some local flavor from, from another sort of regular business person who's doing some stuff with AI. [07:46] Dave Pengelley: Tell that story and- Yeah, that's pretty cool ... Richard's on the case now to find us a pub. [07:50] Richard Webbe: I'll find a place and, uh- I'm sure there'll be [07:53] Dave Pengelley: one ... yeah. I mean- I'm sure there's a pub in Melbourne [07:55] Richard Webbe: somewhere ... I think, I think we can rustle up a few people and a bit of sponsorship and off we go. [07:59] Matt Slager: Just ask [00:08:00] Jeremy Clarkson. [08:01] Dave Pengelley: Yeah. I wish. I wish we could. Yeah, I was actually thinking about it this morning. I was like, we, um, ran the event and I was, uh... and I do think in Clarkson voice sometimes from watching way too much Top Gear over the years. I was like, "I did a thing." We, we ran the event. It all happened. I mean, that was our... [08:18] Dave Pengelley: well, I keep it on the promo, right? We're doing it in a pub. So yeah. Yeah. It was good. It was good. It was good. We'll do more. Very well. Do a bit more. Great. Great to have you. Matt, Matt, Matt, do you think you'll come down to Melbourne for the next one? Is Matt, is Melbourne a slightly closer trip for you than Sydney? [08:31] Matt Slager: Yeah, it's slightly closer. Slightly less turns on the highway. [08:35] Dave Pengelley: Yeah. Okay. [08:35] Matt Slager: Um, it's, it's about four hours to Melbourne, whereas for me to drive to, um, like, your place would be about- Eight? ... six and a half. [08:43] Dave Pengelley: Yeah. Okay. Fair enough. [08:44] Richard Webbe: You know, I was the IBM, uh, account director for Albury-Wodonga, and I used to do that trip many, many times. [08:51] Richard Webbe: You must drive slow. I used to do it in about three hours flat. From [08:55] Matt Slager: Albury to Sydney? [08:56] Richard Webbe: No, to Melbourne. [08:57] Matt Slager: Oh. Yeah, I don't know where [00:09:00] I'm measuring to with Melbourne. That just might just be Google Maps, you know, from my house to CBD sort of thing. Um- [09:05] Dave Pengelley: There's more, there's more speed traps and police than there used to be. [09:07] Richard Webbe: Yes. Oh, there's a lot more of those. [09:09] Matt Slager: True. Uh, we've got a comment from Kate. Um, I actually am really curious about what Kelly's talk actually involved and that sort of stuff as well. Um, yeah, very, very keen to see some sort of replay or, um- [09:23] Dave Pengelley: Yeah. Hopefully we get the recordings, and we will snip up if we can get them, if, like, like hopefully, um, the... [09:29] Dave Pengelley: I don't... I'm not quite sure what Seth was using for video stuff. Um, that was, like I said, this the second category of capture for us. But, uh, she talked about how building... There's two sides. She broke it into business and personal. In business, building various tools. She, she's used to run her own dance studio, and she was a full-time teacher, and now she works in her husband's business and helping streamline that 'cause it gives her more time flexibility to go to school events and be there for, for her kids, which was, was one, one of her big goals. [10:00] Dave Pengelley: And so now she's using AI to streamline processes, build custom [10:00] Dave Pengelley: apps and workflows, replace legacy software that they're not happy with. Doing lots of stuff in tools like Lovable. And, you know, she has been on the journey, as many people are, where she, uh, builds something and then realizes it doesn't actually have a proper storage back end, and all the things she did in it just vanish, and it's like, "What? [10:16] Dave Pengelley: I need a database." Stories like that, uh, which I think are pretty common in Vibe, Vibe Coding land. It is. But, uh, yeah, so she talked about how she'd built, like, various different solutions for different problems that they've found in the business, uh, to streamline some of those things that were taking a long time. [10:31] Dave Pengelley: And also just on the family side, she's been using NotebookLM to help with the kids' study and homework prep, and generating random birthday, custom birthday songs for her kids and family, and so just fun mom stuff too. Like, that was good. [10:45] Richard Webbe: And she, um, she really nailed the reduced complexity for her and her husband's business, and how the compliance part- And reducing complexity went really well. [11:00] Richard Webbe: Like, almost the processes they had, because their manual had [11:00] Richard Webbe: too many steps and too much going. Yeah. And, uh, and she showed how she reduced that down, and it gave her more time back with her family, and, uh, scheduling and all of that. So it was a, it was a good talk. Uh, a quick shout-out to Kate, by the way. [11:12] Richard Webbe: We had a long chat on the phone the other day, uh, about- Nice ... all this stuff. And Kate, while you're, uh, watching- ... go find a sponsor and we'll come to Cairns and we'll do a show up there. [11:23] Matt Slager: Definitely. [11:25] Dave Pengelley: Yeah. IRL all over the place. Uh, so it's good. It's good fun. Good to... Uh, like, like we, we said the Q&A bit, people are asking interesting questions. [11:33] Dave Pengelley: A lot of questions around cost. Everyone wants to know how much all this stuff costs. Yeah. [11:36] Richard Webbe: Yeah, right. [11:36] Dave Pengelley: How long is a piece of string? Which is where we came back to some of that, um, deterministic stuff. Not... I, I showcased the whole don't make an agent, like in the N8N thing- [11:45] Richard Webbe: Yes ... [11:45] Dave Pengelley: where everything's an agent, and let's just give tools to an agent and then watch it pass context back and forth and just context stack ad infinitum and, and rack your costs up, so. [11:54] Richard Webbe: I think you scared everyone away from using agents last night. It was- Good ... it was a good talk. Good. It was [00:12:00] a good talk. It was explaining, you know, it's the whole don't use technology for technology's sake, and, uh, avoid, uh, tokenomics and bill shock by, uh, keeping a lot of it outside the O11 till it's required. [12:11] Richard Webbe: Yeah. I think that was a really good message. I enjoyed listening to that. [12:15] Matt Slager: Yeah. I love it. I lo- that, it's so good. It's like what we were just talking about before, um, before we went live, that, you know, it's really fun seeing things catch up. You know, like we, we're uniquely positioned... Well, I know that, like I'm a super strange, like weirdo with the niches that I'm in and that I follow. [12:34] Matt Slager: Like, I always feel like I'm in this weird bubble that nobody else is aware of yet. But then those things will slowly trickle down the, the stream and it's like, "Oh, okay. People are like realizing that these things exist." And, you know, not, not just like the tools, but the mentalities- Mm ... the, the concepts. [12:52] Matt Slager: And yeah, that's exciting. [12:54] Richard Webbe: Matt, when you put yourself in the weird category, you do know who we are? We're in the same bubble. [00:13:00] [13:01] Matt Slager: Well, there's gotta be a reason why we get along. [13:02] Richard Webbe: Exactly. You find your village, as I was told from a young age. You eventually find your friends and village and, uh, certainly with us that's the case. [13:11] Dave Pengelley: Yeah. I mean, there's a reason why I really like this, this duck picture 'cause I, I just like it. I really like the, the Warhol ducks. [13:20] Richard Webbe: Okay. Is that an Andy Warhol pr- [13:22] Dave Pengelley: No, this is a Kmart. [13:23] Richard Webbe: It's a copy, [13:24] Dave Pengelley: yeah. Andy [13:24] Richard Webbe: Warhol- [13:25] Dave Pengelley: Someone had the idea ... rip-off ... they, they used that same sort of idea and theming and scheming, but put rubber ducks on it. [13:30] Dave Pengelley: I was like, "I love that." My wife does not like it. That's all right. Um, we n- Can't please them ... we normally, th- this is our, this is, you know, inside baseball. This is our dining room with just my dining table. And, uh, normally we've got a family photo there, but I was like, "For the podcast I'm bringing the ducks out." [13:43] Dave Pengelley: They were hiding in the cupboard. I was like, "Yes, the ducks." [13:47] Richard Webbe: So other than the talk last night, what are we excited about in AI this week? What has happened- [13:52] Dave Pengelley: Oh, hang on. Hang on. Oh, sorry. Oh, oh, stop. Wait. I just forgot we didn't do the thing. [13:57] Richard Webbe: Thank you.[00:14:00] [14:08] Richard Webbe: We can never leave out the credits [14:09] Dave Pengelley: Welcome, welcome to The AI Operators Show. [14:12] Richard Webbe: Glad you could join us. And you've been there 15 minutes. [14:15] Dave Pengelley: The show where we, we, we talk, uh, the, the, the real pub chat style, not the hyper around what's going on in the world of AI, how we're using it, how we're seeing it, how we're seeing customers and people in industry using it in the real world, not the people that are claiming that their agents are making millions of dollars for them every night on the markets. [14:34] Dave Pengelley: I wish I could make my agents do that. That'd be pretty fun. [14:37] Richard Webbe: Yeah. You know, the full spectrum of the fears, wants, needs, and questions of AI that was around, uh, you know, 12 months ago when the retail end of ChatGPT and that started to really accelerate, um, it was all there last night in the audience. [15:00] Richard Webbe: People with the same questions, people with the same concerns. Will I lose my job? How do I do [15:00] Richard Webbe: this? Like you said, how much does it cost? How do I get started? So even though we see we're being saturated with AI media, news, and education, there's a lot to be delivered to the community and the business community for them to be able to function effectively and keep going. [15:15] Dave Pengelley: Well, yeah, and e- even on agents, if they... You have a use case for agents and I'm, you know, I run Hormozi's agents. I'm still doing that. I, I... That's useful for me. But people ask me how much this stuff costs and, and where do you get started, which is the best one, and just that breaking down the stuff we talk about every week. [15:30] Dave Pengelley: And you said, Matt, like the world's catching up. But I think we are over target on this show because we are living almost on that bleeding edge constantly of testing some of these things, understanding, you know, the harness versus the model and the difference that that makes and why different models are better or different. [15:45] Dave Pengelley: I talked last night around how M3 overthinks things and over-confirms things, whereas Groq is lazy and tries to make you do all the work. Yes, [15:52] Richard Webbe: it was very funny. [15:52] Dave Pengelley: Um, ridiculous. But- It [15:55] Richard Webbe: is crazy ... [15:56] Dave Pengelley: but, but then you can plug that same model into a different harness and [00:16:00] you're gonna get a different result. Like, if you plug, you know, M3 into OpenCode, it's different than sitting in Hormozi's, for example. [16:08] Richard Webbe: I was sitting with, uh, next to, uh, one of the members of the audience, a senior telco, uh, executive that came last night, and, um, I was checking with him how we have this conversation, and he's very much n- not an AI person. And I said, "Are you understanding all of this and, and stuff like that?" He loved the Mario joke, by the way. [16:28] Dave Pengelley: Very good. [16:29] Richard Webbe: Um, and, uh, he, and he, he said, "Yes." He said, "I'm getting a real good feel for it." And he said to me, "What should I do now?" I said, "Well, keep in touch with us, and go and just get yourself and keep playing." [16:40] Dave Pengelley: Hmm. " [16:41] Richard Webbe: But don't just rewrite your resume on ChatGPT or Claude or something. Do other stuff. Keep playing, and then you'll get to see it. [16:50] Richard Webbe: Expose yourself a bit more and get your context going, so." That was an interesting conversation [16:56] Matt Slager: That's true, though. Like Richard, you, you quite often say [00:17:00] that there's the layers of AI adoption and understanding, and somebody the other day, um, brought that up with me. I think it was the... Like, uh, every week I go to a, a mental health walk around my community. [17:11] Matt Slager: It's, it's for blokes. It's from the Spoke to a Bloke commun- um, organization. Very nice. And yeah, so we, we have the blokes walk. Anyway, the, the, the guys the other day, I was talking about stuff that I do, and they're like, "Yeah, but, like, don't you just... Isn't it just Google? Like, don't you just ask it questions?" [17:27] Matt Slager: Like, what do you mean? Like- Mm ... how can you get it to do things? Like, and so I had to try and elaborate and explain stuff. But just like my bubble that I mentioned, it is so hard to integrate with, like, the rest of the world that don't have this knowledge and speak from that language. That's what I find very difficult, and that's what I think this show does a pretty good job at doing. [17:50] Matt Slager: It's the idea of, you know, how can you use these things that exist in this weird realm that nobody understands? [17:59] Richard Webbe: Anyway, and look, it's [00:18:00] the great equalizer, and we need to normalize the language so that those that we want to adopt AI... 'Cause at the end of the day, if you look at a very macro view, if everyone adopts and uses AI, ignoring the competitive advantage that we might try and get, then the world's gonna run a lot more efficiently. [18:17] Richard Webbe: We'll solve many more problems. There'll be more resources available. It'll all be a much better world to live in than it would be yesterday. And, uh, and that's one of the things we were discussing last night as well. [18:28] Dave Pengelley: Well, if you, if you look at the technology advancements since the Industrial Revolution, for better or worse, over time, one of the things that we have now is weekends, and free time- Free time [18:37] Dave Pengelley: and leisure time- [18:38] Richard Webbe: Very true ... [18:39] Dave Pengelley: and disposable income, and things like that that just weren't concepts- Yeah ... 100, 200, 300 years ago, right? [18:44] Richard Webbe: Yeah. [18:44] Dave Pengelley: I went on my daughter's camp to the gold fields in New South Wales last year, and they were saying people were working very hard six days a week, and they had their, the, the, the term when you go to church, wear, wear your Sunday best. [19:00] Dave Pengelley: That was their social time. That's when they got to see friends, when they went to church. And so they wore their [19:00] Dave Pengelley: nicest, best clothes for church, 'cause the rest of the week they were wearing their rags and work clothes, and they were all coal-smitten and dirty, and they might have one bath a week for church or whatever. [19:09] Dave Pengelley: And it was a whole different environment. People worked very hard, very long, very t- very manual labor jobs. And then they got very little leisure time- Yeah ... um, compared to now. A- and you think, you know, that whole pre-industrial revolution, subsistence farming, and everything's just the next and next and next, unless you've managed to get some level of, uh, expansion or, or inheritance. [19:32] Dave Pengelley: But now everyone, regular people, most people, the economy's changing a bit and we, we won't d- go into the, the negatives of the, of the 20th, first century, but you know, we have weekends. People work five days, 40-hour weeks on, on average, all that kind of stuff. You know, we've got time to go to the movies and do this and do that, and have all these leisure activities. [19:52] Richard Webbe: Yeah, see, we don't realize that going back 100 years, that wasn't available to anyone. [19:56] Dave Pengelley: No, I mean, it's like that survey where people said, "Would you, would you change your li- [00:20:00] trade your life with like, um, Rockefeller?" And people are, "Oh yeah, he was the richest man in America. I'd definitely do that." And they go, "You know, he had no indoor plumbing, no electricity, no this, no that, no that." [20:09] Dave Pengelley: And they're like, "No, I'll stay poor in 2026, thanks. I don't wanna go back to the 1800s and be the richest man," 'cause their life sucked. [20:16] Richard Webbe: Yeah, the Jimmy Carr, the, uh, the comedian, uh, who's quite a good philosopher as well. [20:21] Dave Pengelley: Mm. And [20:21] Richard Webbe: he goes, "Go back 50 or 60 years ago, most people in the world didn't have a hot shower." [20:27] Matt Slager: Mm-hmm. [20:28] Richard Webbe: Explain that to people today and they don't get it. So AI's gonna make those changes. Actually, just sort of changing the conversation slightly, you're gonna hit your news button soon, eh? [20:36] Dave Pengelley: I am. [20:37] Richard Webbe: Okay. Would you, [20:39] Matt Slager: would you like to know the previous question? I have a kind of, like a funny anecdote on that, Richard. [20:41] Matt Slager: Okay. You know, like if you, you go back 50, 60 years, no one had a hot shower. Now, uh, it's so funny how things switch around. People use a cold shower now as like- ... a piece of therapy or- With [20:53] Dave Pengelley: mono. [20:55] Matt Slager: Yeah. You know, it, it's funny how that sort of thing shifts and- [20:59] Richard Webbe: It is very [00:21:00] true ... [21:00] Matt Slager: yeah. Things just rotate, and like you're talking about the idea of, you know, working and having no leisure time, and maybe only showering once a week. [21:10] Matt Slager: I'm not, not gonna lie, some weeks it feels like that for me. [21:14] Richard Webbe: That is very true. [21:15] Dave Pengelley: Well, yeah. I mean, that, that, that's one of the risks, right? Uh, and I said this last night in my talk. With my agents when I had this agent explosion and I had like 30-plus agents, they were staking out more work for me than I could manage. [21:26] Dave Pengelley: Like, they were giving me jobs and asking me to confirm work, and I didn't have the context of the decisions I even need to make because they were generating so many outputs so quickly, and then just waiting. So they weren't doing any more work because they were waiting on me. I wasn't getting any work done because I was cognitively just overloaded with the amount of different sort of tasks that I had to review and have opinions on, and I didn't even have opinions because I hadn't been part of any of the process. [22:00] Dave Pengelley: The robots had done four steps ahead and then gone, "Now tell us what you think, Dave." And I'm like, "W- how did we even get to here? I don't know what this is. I don't have an opinion." And so then [22:00] Dave Pengelley: I was just like- Yes ... analysis paralysis and doing nothing. And so I just hit this, this, this grind point and, uh, which is like a full circle moment for me where I, like I ended up with so much productivity, I ended up with no productivity. [22:13] Dave Pengelley: But when I was simpler and getting started, and I sort of maybe just got into 3D printing at the same time, I felt like I always had to have my a- my, my, my Claude Code or my Antigravity or whatever it is doing the next job. I need to be using it. It needs to be working all the time, writing the next line of code, building the next part of my application. [22:28] Dave Pengelley: I don't wanna waste any time, so I gotta make sure I get it doing something, and then I can go for a walk. Get it doing something, then I can go to bed. I can't leave my computer doing nothing. I have to have it always on, which again, to your point, Matt, I was always working more than ever because I always wanted to keep the robot working. [22:43] Richard Webbe: You were trying to make your computer ADHD. [22:46] Dave Pengelley: I just wanted it to build stuff for me Yes And I didn't wa- I didn't wanna waste time. Yes. And so, yeah, I mean, and that, that's been one of the arguments around this whole work-life balance, work from home, all that kind of stuff. The fact that with laptops and, um, and mobile phones and [00:23:00] things, back in the days, you did go to the office, you did your job at the office, and then when you went home, you would... [23:05] Dave Pengelley: That was it. You didn't work from home. All your stuff was at the office. Exactly. And now we've got these atomized lives where we're always working. We always have access to our email. We can always be accessible, always contactable. We can always bring up a spreadsheet. We can always do these things all the time. [23:17] Dave Pengelley: And so, yes, the benefit of that is in the middle of the day when it's the only time you can go to the bank, you can actually step away from work and go to the bank now. But also, you'll then be distracted from your personal life and your friends and family in the evenings, potentially, because you can always still be on working. [23:34] Dave Pengelley: And so it is this weird thing as humans. We went, we're at the risk, uh... That's, that's, that's weird. [23:40] Richard Webbe: Where does it get easier? [23:41] Dave Pengelley: I don't know, but let's, let's, let's talk about some news. [23:43] Richard Webbe: Go. [23:48] Matt Slager: Welcome to the AI Update. Let's look at what's happening in the news. [23:55] Dave Pengelley: What is happening in the news, Richard? You seem very keen for me to hit the news button. [23:57] Richard Webbe: Well, I heard... I mentioned this last night, but [00:24:00] I heard a security, cybersecurity AI news article where someone was scammed out of, uh, a large fortune, and they're on a video call, on a virtual one, and they didn't realize, but they were the only real human on the call And this group of people masquerading as other people, uh, made them sign their money away. [24:22] Dave Pengelley: Wow. [24:23] Richard Webbe: That's quite scary, isn't it? And I do remember, I, you know, I ran an AI company a few years ago, and I was working with a New Zealand company that had produced really clever humanoid avatars that would appear in a meeting or a podcast like this, and you would not know that they weren't human. I'd- I actually didn't at the meeting. [24:40] Richard Webbe: And so it's quite believable that someone could be invited to a, a video podcast and you convince them, "Yeah, no, we're all real. He's the CFO." A- and apparently at this meeting, none of them were real it was only the person who was getting scammed. [24:54] Matt Slager: So- That's crazy. [24:55] Richard Webbe: It is crazy. It is crazy. [24:57] Matt Slager: So- There's probably ways, like little, little weird tricks that you [00:25:00] can do. [25:00] Matt Slager: Like I've seen, um, people, like yeah, you put your hand in front of your face and what they can't do, like specific amounts of fingers. [25:07] Richard Webbe: You know, it's funny, they can't do three fingers in front of the face at the moment. There's something... And I, but I don't know if that's true or not because I, you know, I would- Not a [25:15] Matt Slager: robot. [25:16] Matt Slager: Yeah. And it will probably get, like that'll get patched in like- It looked like when a [25:19] Richard Webbe: demon in, in The Exorcist dancing. [25:24] Matt Slager: But yeah, like how, like when that is, when that's solved, when, you know, you can have your AI do that finger thing and, you know, how do you tell? And like, you know, a lot of people, uh, on their voice AI systems, like they'll- Yeah [25:36] Matt Slager: have disclosures at the beginning, you know, like, "This is a r- uh, an automated voice call. You can press whatever button to speak to a human." Um, is that gonna be a requirement? And if you don't disclose it, is that gonna be like a legal requirement and- Well, [25:51] Dave Pengelley: I think in the US you have to disclose if you're using an AI voice. [26:00] Dave Pengelley: I think in Australia it's a bit grayer and you can do outbound calling with AI without disclosing that [26:00] Dave Pengelley: upfront. I think, I think that certainly on inbound you wouldn't have to. Um, but- Oh, there- But there are, there are people doing the whole like scam voices. I dunno where they're getting the voices, but getting daughters ringing elderly parents saying, "I'm in trouble, my car's broken down. [26:15] Dave Pengelley: Send me money." Mm. There's people doing those sorts of voice notes- Oh yeah, [26:18] Richard Webbe: that's happening right now ... [26:19] Dave Pengelley: scams already. I mean, it takes you back to Terminator 2. It's like, "What, what is your dog's name?" "Max." "How's Wolfie?" "Wolfie's just fine." Your- Your dogs are dead. [26:32] Richard Webbe: I never spoke about this. There's nothing better than getting your 11 and 14-year-old children when you've got a scam call and you know, 'cause we used to play in the car, um, scam call, and I'd pretend- [26:41] Richard Webbe: to be the scam caller. And the kids would, the kids would pretend to answer the call and handle it the way I taught them to handle scam calls. So when I get a real scam caller in the car, I just hand them the phone and, and off they go. Wow. [26:55] Dave Pengelley: Amazing. [26:56] Richard Webbe: I hear this Indian voice going, "You are just drag- stretching this out [00:27:00] and you're just confusing me and you're wasting my time." [27:02] Dave Pengelley: I've, I've never heard of playing that as a car game. Scam call. [27:06] Richard Webbe: And my daughter goes, "Yes, that's what I'm here to do, waste your time, 'cause you're a scammer." [27:11] Matt Slager: Amazing [27:12] Richard Webbe: It's very funny. [27:13] Matt Slager: Anyway- [27:14] Richard Webbe: So that's the kind of like- What else is in the news? [27:15] Matt Slager: That's the kind of literacy stuff that we always talk about, you know? [27:17] Matt Slager: Like the awareness and the ability to like arm yourself, you know, against this world that we're moving into, that's already here. Mm. [27:28] Dave Pengelley: Yep, yep. Um- You're big on that education awareness piece, Matt. Like, I think, uh, that knowing the risks is, is half the battle. [27:38] Richard Webbe: Yeah, yeah. Um- [27:38] Dave Pengelley: I think it's [27:39] Richard Webbe: really important ... [27:39] Dave Pengelley: but, but people have such little, even, even just before this AI stuff, just the number of phishing attacks and email and, and other random stuff that goes on. [28:00] Dave Pengelley: People's general digital literacy, it, it's shocking 'cause computers and the internet have been out there, what, 30 years old now, give or take, when people started Windows [28:00] Dave Pengelley: 95 and Internet Explorer and, and Netscape and all that was popping up. That would've been, yeah, 30... Oh my goodness, I'm old. [28:06] Matt Slager: You probably would've been about that, yeah. [28:08] Dave Pengelley: Yeah. And when, so people have been had access to start getting on the internet, and especially the, the, my generation and younger have grown up with a lot of this sort of stuff for, for, for more often. But people still just have this thing with the computers where they're like, "I don't know about computers. [28:23] Dave Pengelley: I'm, I'm computer illiterate. I don't know that stuff." Which I got in the '90s when computers and personal computers were still pretty new, and you'd get people sort of like they go, "Ah." But come on, this is, like, standard stuff in life now. [28:36] Matt Slager: Mm. [28:37] Richard Webbe: Yes. Yeah, ignorance is dangerous. It's very much like that. And, uh, the way I kept my kids on, on their toes during the car trips when I played scam caller, and I learnt this from a psychologist, is I'd have a bag of snakes there, and if they answered correctly, they got a snake. [28:52] Richard Webbe: If they answered incorrectly, I threw one out the window. That really got them up for, for- [28:58] Matt Slager: I've got- Is that where I got all those [00:29:00] snakes from? J- jelly snakes, right? Not real snakes. Jelly snakes, not real snakes. Thanks for clarifying. [29:09] Matt Slager: Very good. [29:10] Richard Webbe: So what other news do we have? [29:12] Dave Pengelley: All right. Uh, there's one I saw that I wanna bring up. Uh, let me share it with the... No, not that one. I nearly jumped on a stinger. Hang on. There we go. Uh, Perplexity. We don't talk about Perplexity a lot. I know, um... Have you used Perplexity much, Matt? [29:28] Matt Slager: Yeah. Mano is actually a big Perplexity, um, promoter. [29:32] Matt Slager: He is. Uh, he teaches it a lot. And yeah, I used to use it a lot. It was one of the, one of the good alternative search indexes. [29:41] Dave Pengelley: Yeah. [29:42] Matt Slager: Um, but yeah, they've expanded their product line in a lot of different ways. I, I'm honestly confused how they still exist, you know, as strongly as they seem to, 'cause I don't even think about them anymore. [29:53] Dave Pengelley: Well, no, so, and they've, they've gone down the Perplexity computer route where the, the independent agents with [00:30:00] their own in- independent workspace to do things for you off your computer in a secure space thing. I think they've, they've pushed that way. But it's interesting that you say, "Well, I don't think about them." [30:08] Dave Pengelley: Who- They're not really a model. [30:10] Richard Webbe: Who, who's the guy we're looking at? [30:11] Dave Pengelley: He's the Perplexity CEO. We're gonna, we're gonna look at him [30:13] Richard Webbe: again. He, he needs to shave three times a day, doesn't he? [30:18] Dave Pengelley: He, he shaved and by the time they set the camera up and got the lighting ready, he was back like that. [30:22] Richard Webbe: Uh-oh. We, we- [30:23] Dave Pengelley: Some people are gifted like that [30:24] Richard Webbe: we're shave shaming someone. We better move on. Keep going. [30:27] Dave Pengelley: No, it's fine. That's, it's, it's a glorious beard that the man has. Um, this is, uh, Aravind Srinivas. Mm-hmm. Uh, so breaking the CEO... So thank you to Guillermo Flor for this one. Breaking, the CEO of a $20 billion company just said the model is no longer the product Wow, that sounds like he's been watching our podcast. [30:50] Matt Slager: Yeah. [30:50] Dave Pengelley: Um, Aravind Srinivas went on 20VC for 95 minutes and made a case that most of the AI industry has the wrong mental model, uh, and he says that the [00:31:00] model is a commodity now. Wow. That's pretty much what we've been saying for weeks, if not months, right? Talking about the versus, the harness versus the model. [31:09] Richard Webbe: Mm. Yeah. And it, it, it signals- 'Cause it- ... market acceptance. [31:14] Matt Slager: There's a really, really interesting thing there, and this might be, like, like a step up in complexity, so bear with me, but, like, as the models increase in their power- Mm ... in their intelligence, in the... Y- as the models get better, whatever better means, I found something really interesting this morning as I was thinking this through, and you would think that as these models get better, like even though Aravind says that it's not the moat, you know, like the- Yeah [32:00] Matt Slager: the actual harness, the product is, it's not the model. Y- you would think that as the models get better, that we have to care less about what they're doing 'cause they're actually smarter and they, they know what to do, and we can teach them what to do and all that kinda thing. But here's the difference, and I hope that if people are listening to this and watching it back, they hear this and it actually [32:00] Matt Slager: clicks. [32:00] Matt Slager: I believe, truly, as the models get better and better, the human, and by extension the human's judgment, needs to be razor sharp. [32:12] Richard Webbe: Mm. [32:12] Matt Slager: Yes. Because a dumb model is super obvious, and it's very obvious when it does the wrong thing. It's not so obvious when a smart model does the wrong thing. [32:22] Richard Webbe: Yeah. Yeah. That's... [32:23] Richard Webbe: And, and, yeah, I go back to some research I did years ago on AI. When a, uh, a particular model comes back with an answer, I think there was four categories that they could put the answer in, and the scary one was, "We don't know how it got that answer, so we can't verify how it goes." But of course we use AI to check AI, a little bit like you were talking about the guardrail unhooking, uh, uh, last week when we were talking about using one AI to check another AI. [33:00] Richard Webbe: But yeah. And, and the thing that we spoke about a bit last night as well is, is relying on AI answers as be- the, the golden truth. Yeah. That's such a [33:00] Richard Webbe: dangerous thing 'cause you don't know, right? And I, I said to... You know, it's, whether it's cyber scamming or answers or empirical or, uh, uh, other answers, if you... [33:11] Richard Webbe: And the common person will do this, won't they? Mm. They'll just assume it's all the right answer, and they'll go on that basis, which is a great place for scamming, but it's a great place for making a massive mistake, and technology's been planned for massive, massive mistakes. Uh, I remember, I won't say the bank's name, but I believe a wrong number in a spreadsheet where one of the executives at an, an Australian bank put the wrong number in a spreadsheet. [33:36] Richard Webbe: And of course the whole company's business process at one stage was relying on this part of this spreadsheet, which is a single point of failure, deathly danger. And I think the bank lost 40 or $140 million just on that one bit of data being put in the square wrong. Now, if we make AI a critical single point of failure, we're all in trouble. [34:00] Richard Webbe: So we need humans still to be riding shotgun. [34:00] Matt Slager: Yeah. Yeah, there's absolutely that, that element of just judgment and discernment of synthesis. You know, you, you can, you can spin up as many agents as you want. You know, get them thinking, get them exploring, get them prototyping. Yeah, get Dave to build you 32 of them. [34:16] Matt Slager: And, you know, those, those results when, when you go to try and, like, as, you know, sift through that, of course it's gonna be nauseating and overwhelming. So that's, that's literally the, the, the thing I'm trying to identify here is it's your judgment needs to be razor sharp. [34:31] Dave Pengelley: Yeah. [34:32] Matt Slager: Yeah. [34:32] Dave Pengelley: Yeah. You, you gotta keep the human in the loop. [34:34] Dave Pengelley: And, and this is what I was saying with my 32 agents. There was too many things, too many autonomous decisions that I didn't have a say in, and so I felt completely overwhelmed and out of control. And, and I made the point, like, if you've ever bu- hired anyone into a team, try hiring 32 people with no managers all at once and trying to onboard them and get them to be effective and doing the right things, or put some managers in, but the managers are brand new as well, and they're all actually just super int- intelligent toddlers. [35:00] Dave Pengelley: Like, that's [35:00] Dave Pengelley: just- [35:00] Richard Webbe: That was... I did enjoy your talk last night describing agents as super intelligent toddlers. It was a good metaphor and it, and, and got over very well. Yeah. And it's [35:09] Dave Pengelley: true. Yeah. And, and, and even, even little things, whatever your... Even if you've trained them, and you've given them all the rules. [35:14] Dave Pengelley: And, and you guys know that, uh, this week I was working on some new content formats for us and, and looking at the whole, the comic style format. There... Oh, there's Ribot. Um, but yeah, and then sort of taking the four of us and building out these comic versions of us. Trying to get consistent look and feel on these was ridiculous. [35:32] Dave Pengelley: And, like, all of a sudden I'd have your beard, Matt, and, and, and, you know, like, you were really, like, fat in one of the pictures out of nowhere. I don't know why, but all of a sudden it gave me this fat Matt. And, like, it was just doing random stuff. And so trying to even, even with strict guardrails and models to what to look at and so on, just trying to get consistency. [36:00] Dave Pengelley: And so if I just fully automate this process to spit these out, I'm not gonna get a good result. I actually do need to review the result, [36:00] Dave Pengelley: double check it. No, no, why, why is Matt 200 kilos heavy? Fix that. He goes, "Oh, okay. No worries." Bang, trims you right down again. Um, [36:10] Richard Webbe: but- It is those, it is those random mistakes, 'cause there's so many complex processes going on. [36:14] Richard Webbe: You know what I found enjoyable about yesterday? And just for people to understand, we had a long chat last night about personal productivity versus corporate competitive versus business streamlining processes, and they are different things in the AI world, but most people feel the personal productivity thing. [36:30] Richard Webbe: But if I can make a point on that, we were finishing our slides on the way into town to do the presentation, and we were finishing them on our phones and our computers in a very quick- Timeframe on a train while moving on the way to the presentation. I couldn't have imagined me doing that m- me, you know, a year ago, right? [36:49] Richard Webbe: But it was really productive. It was really good It... [36:52] Dave Pengelley: And, and, and again, like I got Gamma, I used Gamma and I used my agents and they helped me frame out some stuff and then pushed out some content [00:37:00] up to Am- Gamma via the API, built a prompt, ran that through Gamma. I had to rework 70% of it to make what, what I wanted because... [37:08] Dave Pengelley: But ag- and, and I've written about this on my, on my own substack. One of the advantages you get with AI is it breaks the blank page. If you're like me where, where you struggle with the inertia and when you've got nothing sitting in front of you, it's like- [37:18] Richard Webbe: Getting started, yeah ... w- well [37:18] Dave Pengelley: I don't, what do I write? [37:21] Dave Pengelley: What's the first word? Do I get the first right word right or wrong? But having something just get you started and break the blank page can be a huge productivity driver. And then, but making sure that you then review the blank page, you fix it, you edit, you modify, you finalize, and don't just rely on it being the all and end all. [37:38] Dave Pengelley: Um, I'll bring up a few links on, on, on that whole business productivity versus, uh, product, uh, other things- [37:44] Richard Webbe: Personal [37:45] Dave Pengelley: productivity, yeah ... but, uh, and then what we're talking about right now, probably relevant is this one that you shared with me, Matt, from Anthropic Uh, agentic coding and persistent returns to expertise. [38:00] Dave Pengelley: And persistent returns. Get the [38:00] Dave Pengelley: emphasis right. Uh, agentic coding and persistent returns. No, I, I, I don't even know how to pronounce that properly. What does that even say? But what it's saying is that if you want cons- persistent, consistent, good returns from your agentic coding, that you need to have good expertise in. [38:15] Dave Pengelley: And I think the crux of this article is talking around the main knowledge, uh, matters. [38:21] Richard Webbe: Yeah. Right. The- Actually, Kelly demonstrated that, uh, really well last night. And, uh, and actually... And, and, and also did, uh, Sean with their, uh... And I asked Sean a question while I was presenting, and he said that, you know... [38:34] Richard Webbe: And it's what I emphasize, it's the critical part, is know your business better than you know yourself, and know your business process flow, and know where, as you do that, know where you could improve things. [38:45] Dave Pengelley: Yeah, there we go. The greater domain expertise a person brings to a session, the more work Claude does per instruction. [38:50] Dave Pengelley: Yeah. [38:51] Richard Webbe: Yeah. [38:52] Matt Slager: It... Yeah, this, this article, uh, I forget if that one includes the diagrams or not that I saw earlier, but it, it basically was showing that in [00:39:00] a, in a scenario where you have domain expertise, where you do, and you're the one steering Claude, you'll achieve such higher results from somebody who tries to do the same job without that domain expertise. [39:14] Matt Slager: Like, it sounds obvious, but a lot of people keep thinking that, oh, if I just use the smartest model, the smartest agent, I can just do the thing that these other people do without having that domain expertise. But this... Not that it needed to be proved, but this, this article proves that to be completely false. [39:29] Richard Webbe: Exactly. And what's, what's hilarious is that people, um, incorrectly sit inside their organization or their personal productivity bubble, and they're using AI to do something, and they think they're the only ones doing it. And while they're doing it, everyone else is. Mm. And so it's not really competitive advantage, it's just keeping up. [39:49] Richard Webbe: Yeah. Everyone's doing the same thing. [39:53] Dave Pengelley: Yeah. Yeah, it's the, the SaaS that I built, Chart Report, I could only build because I'd had [00:40:00] experience as a trader. If I didn't bring trading knowledge and experience to it, I couldn't have built the same app, right? No. Like, like anything, if you wanna build a solution... [40:07] Dave Pengelley: A, a friend of mine was telling me about SEO stuff and, and I thought, oh, maybe I could build a SEO thing that generates pages with backlinks and, you know, like pages per suburb and uses AI to go and work out what are the keywords and how to build those sites. And I started building it and, and I was playing, this was before I was Antigravity. [40:23] Dave Pengelley: This was, like early, like this was, you know, six months ago. Like early, early AI. [40:27] Richard Webbe: Way back then. [40:28] Dave Pengelley: Yeah, way back then. Uh, and I was playing with Lovable a little bit, and realized really quickly, and I was just using the free, free tokens per day, per week, whatever, on Lovable just to, to ration this thing out. [41:00] Dave Pengelley: It was like a Christmas project before I discovered the idea for Chart Reporter. And I realized I didn't know enough about SEO to actually build an effective SEO tool. I couldn't build an AI tool that would generate SEO pages 'cause I didn't know enough about SEO. I didn't know enough about the problem that, well, I was trying to solve, so I couldn't actually build a good solution [41:00] Dave Pengelley: 'cause I didn't understand the problem. [41:01] Richard Webbe: Touché. [41:02] Dave Pengelley: Um, and then, you know, days later I was like banging my head against the wall trying to do some trading robot stuff and went, oh, hang on. Hang on. This is a problem I understand. [41:12] Richard Webbe: And off you go. Yeah, yeah. And that is true. Yeah, yeah. Yeah. None of it, no, no tool is any good unless the trades person behind the tool knows what they're trying to do, and they have the expertise in the environment they're doing. [41:22] Richard Webbe: And AI is no different. [41:24] Matt Slager: Yeah, exactly. Like, it's similar to the concept of bringing agents into your company and doing stuff. Like, you know, Dave, you mentioned like a hiring example. It's the same thing with agents. Like you gotta treat them like you're hiring someone important or something important, so you have to understand what they're going into. [41:41] Matt Slager: Like, you would never hire a, you know, a brand new junior and say, "Okay, reorganize my entire organization." [41:48] Dave Pengelley: Yeah. [41:49] Matt Slager: But people were doing that with their agents . [41:52] Dave Pengelley: And, [41:52] Matt Slager: and, [41:52] Dave Pengelley: and I made the point like, like, uh, and we discussed it afterwards as well, around my number of agents. And, [00:42:00] and we know on this show I've talked about how I, I give them personalities and names and, and sort of the- have the Harry as the boss of the different agents and orchestrate stuff down. [42:08] Dave Pengelley: And so when I was building my sergeant harness, I was trying to build that containerized hierarchy thing. What I've realized since over-profiling in Hermes, I think that was actually a flawed con- concept in some ways because over-compartmentalizing, over-containerizing your agents, they actually do need some level of shared context. [42:28] Dave Pengelley: If you over-compartmentalize and rely on it, just giving them that little snippet they need... And I talk about ambient context on the show all the time, that like, yeah, over-containerization, over-compartmentalization and isolation of agents has negative impacts as well. What I'm much better off doing, which is what I've returned to, which is more like the repo Claude Code IDE Codex kind of model, is having, you know, agents per project rather than a project with 50 agents And the agent in that project has little hats it can wear. [43:00] Dave Pengelley: And I... Ma- the Mario thing I [43:00] Dave Pengelley: mentioned, Mario puts a squirrel suit on. He's still Mario, but now he's got squirrel powers. And I do that with my agents now, where, like, it's the one agent, but they've got different hats they can put on to get different powers or different insights in order to do things. They do? [43:12] Dave Pengelley: But then they, then they remember who they are across that, and they have the shared context of what each of the hat-wearing people did, and they can work more effectively in that manner. [43:21] Richard Webbe: Someone sitting next to me had never played Super Mario. They were concerned about Mario and squirrel. But I, I explained it, so- Okay [43:28] Richard Webbe: you're [43:29] Matt Slager: all good. Well, that's good. Lucky. That, um, that kind of compartmentalization thing, m- people may not be able to connect the dots of what that actually means. So, like, I, I picture a, a typical organization, you know, like a, uh, like a knowledge e-com high-ticket, like a coaching program type thing. So imagine you've got your sales team, and then you've got your fulfillment team that handles the clients after the sale. [44:00] Matt Slager: If the sales team is promising all these things, and then the fulfill- fulfillment team's like, "No, we can't do that." "What? Oh, but the [44:00] Matt Slager: salesperson said you could." Yeah. That's an example of, like, separated, you know, compartmentalized knowledge. There's no source of truth there that, that people are actually referring to. [44:08] Matt Slager: Yeah. So yeah, the agent scenario is the exact same thing. [44:11] Dave Pengelley: I, I re- I remember back in my CRM days, one of the w- first ideas I had around IoT integrating for business purposes came up. It was, like, a, a company that made steel cabling and wiring, and so they had all these big, old, 50-year-old steel extrusion machines and stuff, like old factory machinery that does, like, big, heavy work. [44:31] Dave Pengelley: But your exact scenario map, the sales guys was, like, selling rolls of cable that the factory, they had a, a machine that was offline. They couldn't actually produce that much cable in that, that amount of time to do it. So the sales team are overselling compared to what production was capable of producing. [45:00] Dave Pengelley: And I was like, imagine- Even though those are really old machines and they're not gonna have APIs, but surely electronically you can wire on some kind of sensor, even if it's like jury-rigging something onto the, the red flashing [45:00] Dave Pengelley: warning light that detects that that machine has a red warning light, it is offline, that would then send a signal up and then post that back into the, into the CRM and let the sales team know that there's a, a product gap, so they shouldn't be selling that product right now. [45:13] Richard Webbe: Yeah. [45:13] Dave Pengelley: Wouldn't that be great? Exactly. Like automating those processes a- and giving them that, that sort of awareness. [45:18] Richard Webbe: Who is the famous mathematical philosopher? I'm gonna embarrass myself 'cause I can't remember it, but they said- [45:25] Dave Pengelley: Pythagoras ... [45:25] Richard Webbe: given it, that'll do, close enough. Given enough monkeys tapping on a typewriter, if you get enough monkeys, you'll produce the world's greatest novel. [45:34] Richard Webbe: So in other words, random tapping will eventually come out, but I'm not sure anyone actually mathematically calculated how long that would take. And it could be a trillion years to get the perfect novel if you've got a bunch of illiterate monkeys tapping on a typewriter. [45:46] Matt Slager: Yeah. You know what? On that note, here's an interesting historical thing, which is not actually that old. [45:51] Matt Slager: Um, the way that large language models actually came to be. D- do you guys know that story? It's a bit of an anecdotal story. [45:59] Richard Webbe: [00:46:00] Please explain [46:01] Matt Slager: So a lot of people call them fancy autocorrect, right? [46:04] Richard Webbe: Yeah. [46:04] Matt Slager: So that's literally how they came to be. They had an autocorrect model. Have you, you guys ever done the thing where you, you open up your message stream and on your keyboard it gives you, like, three predicted words? [46:15] Dave Pengelley: You just press the next word? [46:16] Matt Slager: Yeah. You just keep pressing that middle word or whatever- Yeah ... and just see what comes out. Yes. That's what they did. That's what they did. And they ran that thing for, like, a long time, and you would think that eventually it would just, just devolve into chaos, right? But it actually ended up turning into this sort of, you know, like a neural net of, like, what actually is realistically being predicted next. [46:39] Matt Slager: And then they were like, "Wait a minute." So- Which is [46:42] Richard Webbe: mathematical statistics with a lot of firepower thrown at it so that eventually the right answer comes out. [46:48] Matt Slager: Yep. [46:49] Richard Webbe: Yeah. No, that is fascinating when you think about it. It, [46:51] Dave Pengelley: it was Emile Borel, the French mathematician. [46:55] Richard Webbe: Yes. [46:56] Dave Pengelley: Uh, the concept of... Uh, it's, it's called the infinite monkey theorem. [46:58] Richard Webbe: That's it. [46:59] Dave Pengelley: Uh, it [00:47:00] doesn't have a single inventor. It evolved from, uh, Aristotle. [47:03] Richard Webbe: It was Aristotle. There you go, one of my favorites. [47:05] Dave Pengelley: Back in the day. Um, his Physics and, and Cicero's De Natura Deorum, which discussed whether the universe could be formatted by random collisions of atoms. And the probabilities are so infinite you require as much faith to believe in no creator as one. [47:20] Richard Webbe: Entropy takes over. Mm. [47:22] Dave Pengelley: Yes. To believe in nothing is a big belief to have in itself. [47:29] Dave Pengelley: Um, that's as, uh, deep as we're gonna g- get, uh, theologically and philosophically, I think, right now. But on, uh, just, just conscious of time, um, Rich's got a hard stop 'cause he's gotta actually get to a, a plane, uh, so we will be wrapping on one. But there's a few other things going on in the world. We talk about, you know, a lot of this agent stuff is personal productivity, but how does that translate across an organization? [47:53] Dave Pengelley: Last week we had, um, Savio on talking about one of- Yeah ... his solution that he's trying to build- [47:58] Richard Webbe: Yep ... [47:58] Dave Pengelley: with, uh, [00:48:00] Theos or The OS to achieve that, and, and I, I touched on my Airtables with hyperagents. But everyone is catching up to us because, you know, we are the oracles of AI. We should have called ourselves AI oracles, not AI operators perhaps. [48:14] Dave Pengelley: Um, but we've got this one, achieving success with AI from Microsoft, and where they're... In this, it's a lot of stuff in here, but they're talking about, you know, the control plane. Here we go. It's Agent 365, the control plane. So they're getting beyond the model. Microsoft haven't built their own specific model. [48:34] Dave Pengelley: They, they invested in OpenAI, they're partnered with Claude, they're partnering a- around the world. They did, they did announce, well, we did discuss on the, the channel a few- sessions back, they did release some models, but it hasn't been their focus. Uh, and you, you talk to lots of people who go, "Yeah, all I've got access to is Copilot." [49:00] Dave Pengelley: Um, and I think they say that, but Copilot as a harness has access to lots of things and lots of integrations into Microsoft Power [49:00] Dave Pengelley: Automates and Flows and SharePoint and lots of data knowledge and stuff. People write it off, but it, it is giving them that enterprise level, uh, harness accessibility if they, they set it up right. [49:14] Matt Slager: Yeah, [49:15] Dave Pengelley: 100%. Mm. [49:16] Matt Slager: It's actually a really good ecosystem to, to learn, you know? Not a lot of people really realize what it exists and what it can do [49:25] Dave Pengelley: Yeah. Um, yeah, I think that's right. Most people don't. And, and it's because the UI is complicated, and it's hard to administer, and because Azure is just such a beast of different things plugged in together, and anyone who's tried to set up, like, SharePoint sites and stuff will tell you, like, it's, it's not easy to get it working. [49:43] Dave Pengelley: And they have so many different tools that kinda do the same thing that are overlapping with each other. And so, uh, do I do tasks or do I do, like, planning central, whatever it is, or do I do this or do I create that, or should that be a SharePoint table or a list? And there's, like, 800 different ways to solve the problem. [50:00] Dave Pengelley: So some people just go, "Ah, [50:00] Dave Pengelley: no, doesn't make sense." But I know people that are using Power Automate, and they're setting Power Automate up, and they're using Copilot agents built in as part, kinda like an end-to-end workflow, where this step does this, and then get that data and pass that into the agent, and then process that output there, and then go save it over there and And then they can standardize those at an enterprise level, and they can share those sort of workflows and those power automate flows out across users so everyone's getting the same thing, and they're standardizing those processes. [50:25] Dave Pengelley: And it goes beyond that individual productivity to then becoming a, a business productivity tool. [50:30] Richard Webbe: Well, it's... And it's interesting now, having spoken to some of the hardware manufacturers and, uh, network suppliers now, that the sales of hardware, uh, which were declining because of cloud computing and off- offsite computing, are now increasing at a great rate of knots 'cause people are building their own RAGs, their own, uh, you know, their own data management strategies, uh, and applying that power inside the walls before it touches the LLM unit- Yeah, the, [50:55] Dave Pengelley: the, the sovereignty stuff we talked [50:56] Richard Webbe: about [51:00] Richard Webbe: 'cause it's a lot cheaper to buy a big, hard, expensive computer [51:00] Richard Webbe: and run your algorithms prior to hitting the LLM to create your RAG or your AI answers. So the whole architecture is starting to change, and like with client server, I always tend to... [51:11] Richard Webbe: The argument with client server and cloud interesting. When it was expensive to buy a computer, it was cheaper to put it on a cloud or a, or a server somewhere else and just have a client. Then as soon as that and the network got too expensive, everyone brought it back in. It was a lot cheaper, and it tends to wane back and forth on where the most cost effective. [51:29] Richard Webbe: So not only are companies and technologies competing with each other, but I think the architecture of how we process computing is competing with itself and changing. And as we go forward, it's going to look very different to what it is today. The old star hub, star hub architecture is not gonna work, and it's gonna have to be very different. [51:49] Richard Webbe: And of course all on that, we have in the background quantum computing coming along saying it's gonna change things. [51:54] Matt Slager: Hmm. Yeah, I actually... Like, I know that's a really high level topic, but we should touch on that maybe next [00:52:00] week or something, 'cause Australia is actually gonna be one of the leaders in quantum computing. [52:04] Dave Pengelley: Hmm. [52:04] Matt Slager: They're- Yeah ... they're actually putting together some big lab thing now. [52:08] Dave Pengelley: I'm getting promoted posts from, like, the Australian Signals Directorate around quantum computing readiness and stuff on LinkedIn. Like, they're, they're pushing that out, um, from- [52:16] Matt Slager: I think, like, it'd be important for us to take it, like, bring our take to that as in- Yeah [52:21] Matt Slager: what does it even mean? Like- Yeah ... is, is it just the next Intel chip? Like, you know, like, what, what actually is quantum, quantum computing- Yeah ... and why does it matter, and why is it better? You know? [52:31] Dave Pengelley: Yeah, and it's gonna be interesting to see the, the sort of, you know, the cloud. I remember trying to sell cloud, but no one understood what it was and, and every... [53:00] Dave Pengelley: Like, most small businesses in Australia had a server, uh, like, running probably Microsoft SBS, Small Business Server 2003, and, and they hadn't had an upgrade in, in eight years since they'd put it in there. But that was running their mail. That was their Exchange server, or their email was coming into their office through their own ADSL connection to run their email for all their staff And, you know, but for better or worse, like [53:00] Dave Pengelley: Microsoft, they might have had a m- maintenance contract or something with their annual licensing or so on, but they might not have. [53:04] Dave Pengelley: They might have just done that as a one-off thing, and then that was their thing, and they just had that in a corner of RAM. We've been convinced to have this sort of SaaS subscription economy for the last 15 years. Um, and we're now seeing with the AI, the LLMs, they're actually going, "Actually, we can't afford to run some things as a subscription. [53:21] Dave Pengelley: It's gonna be per usage," which is gonna push people to go, "Well, maybe I can just run it internally. Maybe now I buy the hardware and I run it myself, and that way there's no subscription. I just do a one-off capital expense, and then, then it's mine." So it's gonna be interesting how this pendulum swings back and forth. [53:36] Richard Webbe: E- every time someone starts talking quantum computing, my brain slides into quantum physics, and then my brain slides into the multiverse, and I just lose myself down some rabbit hole about trying to understand. It would be nice. Maybe we'll bring a little explanation of quantum computing to next week, and how that underlies and supports AI. [54:00] Richard Webbe: But of course, it, it relies on the state, correct me if I get this wrong, Matt, [54:00] Richard Webbe: that we used to work with memory and physics and things where things had one state or the other. Now they've got three states. The same state, a different state, or the same state as another particle on the other side of the universe. [54:10] Richard Webbe: That's a- [54:11] Matt Slager: Superposition. [54:12] Dave Pengelley: Yeah, superposition. Yeah. Yeah, some- something can be zero, it can be one, or it can be both. Yeah. [54:16] Richard Webbe: Yeah. And that's just very confusing. [54:18] Dave Pengelley: Yeah, so the whole binary bit stuff. Reminds me of- Um, last... Sorry. Sorry, Matt. [54:23] Matt Slager: It just reminds me of four unit maths from high school, where we used to talk about imaginary numbers. [54:27] Dave Pengelley: Oh. Yes. Yeah. Com- complex- That, oh ... complex numbers, imaginary numbers, and frac- Oh ... fractal generation. I love fractals. Fractals are so cool. [54:34] Matt Slager: Yeah, yeah. [54:35] Dave Pengelley: Um- Look at here. Yeah. [54:35] Matt Slager: Life, [54:35] Dave Pengelley: life, life is fractal. Uh, the last thing I just want to touch on, uh, on this whole enterprise transition adoption thing, uh- [55:00] Dave Pengelley: is back on Anthropic, our friends over at Worldland. They've introduced Claude Tag. This is hot off the presses. Uh, this is them getting in on the whole Slack thing, right? [55:00] Dave Pengelley: And introducing Claude as a shared context coworker in your corporate Slack channels, which they're obviously trying to take on Slack's owner, Salesforce, with their Agentforce stuff and say, "No, no, no, we're the AI specialists. [55:15] Dave Pengelley: We want to own that market." So. [55:17] Matt Slager: What it translates to, it, it, like it's only for teams and enterprise plans, and it's only API usage only, so it's all fully cost. There's no subsidized subscriptions. Yeah. But it translates to your business's knowledge plane and control plane in one. So it's basically what Copi- Copilot's doing. [55:35] Dave Pengelley: Yeah. [55:35] Matt Slager: It's Anthropic's version of the, the Copilot situation. [55:38] Dave Pengelley: Or, or, or, you know, Slack's bought, owned by Salesforce, and they've got their own Agentforce that they're saying you should be using- Yeah ... for all that kind of stuff 'cause it's plugged into your CRM and, and everything else. So. [55:48] Matt Slager: Yeah. [55:48] Dave Pengelley: Um, I mean, but Slack has a broad audience beyond Salesforce customers, so Claude's going after that. [55:53] Matt Slager: Yeah, exactly. [55:55] Dave Pengelley: Yeah. All right, well we are on the hour. Uh, we are gonna wrap it up, 'cause like I said, we've gotta get this [00:56:00] man to a plane so he can be back in front of his own camera next week, and I can... If you wanna see what, what it took me to do this, follow us on our new Instagram, Facebook, or X channels. [56:13] Dave Pengelley: We are AIOperatorsPod everywhere. Uh, and we're starting to post some more content across those different places there. So I did post a little cheeky behind the scenes picture of what it took me, what this looks like in my dining room from the other side of the camera. Uh, and so if you wanna see more of that as we post up those comic strips and things and other bits of content around, we're gonna try and get m- I've got my agents been setting up my AI operators agent, Rubot, affectionately as I call him, uh, that we see featured in the comics. [56:43] Dave Pengelley: Rubot's gonna help me generate some content and help us get out there and share some more ideas more often, uh, as well as helping manage the [56:50] Richard Webbe: channel. Yeah. Absolutely. [56:52] Matt Slager: Very cool. [56:53] Dave Pengelley: Very good. And I think we might have Mano back next week. I think he said he's back on the 1st. Oh, wow. Uh, or whatever that is. So I think that's next weekend. [56:59] Dave Pengelley: Yeah. We'll have... Think next, next Monday we'll have Mano back. That'd be good. [57:02] Matt Slager: Amazing. All right, looking forward to it. [57:04] Dave Pengelley: Brilliant. Thanks, Matt. Thank you, Richard. Been a pleasure having you. Thank you. It's been a pleasure [57:08] Richard Webbe: being here. Thank you. [57:09] Dave Pengelley: It's a bit weird. I don't know what the dynamic's been for you, like for you, Matt, having us both on the same camera. [57:13] Matt Slager: Uh, it's been fine. Yeah, it's good. Yeah. [57:16] Richard Webbe: There were no punches thrown. It's all good. [57:18] Dave Pengelley: Yep. Yes. So anyway, back same time, same channel. We'll be back next week. Uh, thank you, uh, Kate, for your chats. If you're watching this after the po- after the fact, drop a comment. Make sure you like and subscribe. We know most of you who watch this do not have a... [57:34] Dave Pengelley: You haven't hit the sub button. So please hit the sub button. Yes. You don't have to hit the bell icon to get notifications. We'd love it if you did. Correct. But at least sub so that way it appears in your feed more often. Hit the like button so it appears in more people's feeds, and drop a comment, ask us a question. [57:46] Dave Pengelley: What do you want us to talk about? Uh, we wanna make this show useful and, uh, required watching for you. [57:53] Richard Webbe: Productive, yeah. [57:54] Dave Pengelley: So let us know in the comments, and we'll see y'all next [00:58:00] time. Thanks, Matt. [58:01] Richard Webbe: Thanks, Richard. Thanks.