Stop Building Agents. Start Building Judgement Around Them.

[00:00] Richard Webbe: It reminds me of a- **Dave:** We, we are live now ... **Richard:** ah, good. It reminds me of a- Speaking of- ... conference call I had once where I was arguing with someone. Yeah. And we were right in the middle of this argument, and then the conference call went live, and we went very quiet. Sorry, Matt, what **Dave:** were we talking about? The, um, have... Is it just me or has Richard got massive lip sync issues? **Richard:** No? Yeah. Do I? Is it not good? **Dave:** You look like you're in a Jackie Chan film. **Richard:** I think the internet should be **Dave:** able to fix that too. Yes. Yeah, you don't, you don't need to fake it. You just talk and then, like, five seconds later, then the audio comes through. **Matt:** You know **Richard:** what it looks like? Should I, uh, turn my audio off and on? **Dave:** That's fixed. That's fixed now. **Richard:** Oh, it's, it's- No, it's fixed ... caught up. **Joanna:** Yeah. **Richard:** It's like my brain. It eventually catches up. **Matt:** The, the, the Australian internet's just had to sort of buffer a little bit. **Dave:** Ah. I, I got an invite for something the other day, and they, they said there was a w- gonna be an in-person thing, and then it flipped to being an online thing. [01:00] Richard Webbe: And the calendar invites were messed up, and they had to send me the Teams [01:00] Richard Webbe: link. And they're like, "I've sent it to you." And I'm like, "We're over time. We should be started the meeting, like, three minutes ago." And I'm like, "It hasn't arrived yet. It's stuck in the internet somewhere." And he's like, "Yeah, our emails bounce all over the place. They take a while." I was like, "Oh, my goodness. It's crazy." You just expect in 2026 you hit send and it should be bing. Yeah. This was seriously, like, a three-minute lag between him saying, "I've sent it," and me finally getting the link in my email to, to click into the meeting. I was like, "Whoa, that still **Richard:** happens?" Well, do, do you remember where that delay came from? Just for those technical people, 'cause I used to be technical once you lot, so don't laugh. In the old days, ISDN dial-up was very expensive, and so people would have their servers wait and only dial into the internet every half hour, and they'd have a half-hour server update, and then the emails would go out. [02:00] Richard Webbe: So, you didn't know how long before you... 'Cause nowadays it's all pretty much live. But, uh, and I remember when the early routers came out and they would said, "Oh, you can have it so it only uses the ISDN whenever it [02:00] Richard Webbe: needs it, but it'll automatically dial it up for you." But what they didn't realize is the dial-up fee was higher than the usage fee, so some people who had, like, $5,000 ISDN internet bills ended up getting $140,000 bills from people like Telstra, and they didn't like it. It's not like that anymore, of course. Oh, no. Hope that makes... That was the first early form of internet bill shock. **Dave:** Yeah. I, I, I used to get the, um... When I was an AE at Telstra, I'd get the mobile roaming bill shock. **instagram sketch voice:** Mm. **Richard:** Mm. ' **Dave:** Cause this was 15 years ago when data was kinda new-ish, and people would go overseas and they'd rack up, like, hundreds or thousands of dollars and things, and they'd be like, "Oh, can you, can you clear that for me? I didn't know." And I'm like, "Oh." So I'd have to go to my boss, 'cause I wanna keep my customers happy so they'll keep buying phone lines and stuff off me then, and be like, "Can we give this guy a credit? Can we do that?" And it's like, oh, my goodness. It was like all I did- ... was just k- try to give people things for free to keep them happy so they'd buy something in the future. It was- That's crazy ... wow. [03:00] Richard Webbe: **Joanna:** That, um- I think [03:00] Richard Webbe: my mom just said to me that, um, I really should have had shares in Telstra probably 30 years ago. I was, um, a very avid user of the telephone. **Richard:** Yeah. **Dave:** Uh- **Richard:** Jo, I'm guessing in your house, did you have the phone on the wall with the really long cable to the handle? **Joanna:** Oh, no. We didn't have it on the wall. It was on the desk. **Dave:** Mm. **Joanna:** And just the dial-up. We did- we never got to the- Oh ... press button one. We only went from the- Ooh ... we had the dial. The rotary. **Dave:** Yeah. **Joanna:** Yeah. **Richard:** Yeah. I took my... I should give them a plug, uh, because some good friends that we all know, but they've set up the Telecommunications Museum in Melbourne, and I took my son there the other day, and it was bizarre. [04:00] Richard Webbe: He looked up this rotary phone and said, "Dad, what's that?" I said, "That's how you call people." And 'cause I used to adjust the step-by-step exchange relays for the dialing for the decadic dialing, 'cause it's how far the, you know, the number nine, the relay jumped nine times. That's how you got your numbers and [04:00] Richard Webbe: how it found you in the exchange. And he was just blown away by it. I'll see if I can find it. I have a photo of him and I making phone calls there. But the stuff that he looked at, he went, "Dad, did they really mechanically connect phone calls?" I said, "Yes, son. That's the only way they could do it. We didn't have digital back then." He's looking at me going, "Are you insane?" **Joanna:** I, I loved reverse call. That was my favorite. **Richard:** What's that? **Joanna:** My mom and dad absolutely hated it. Oh, **Richard:** yeah. We used to- **Joanna:** 199. **Richard:** Was it 199? Yeah, 199. **Joanna:** I think so. We used to dial 199. I think it's 199, and it would say, "Pick up" at the other side, and my mom and dad would say, "Hello?" And then she goes, "It's a reverse charge call from Joanna." Like, from where, somewhere in Australia. And they'd be saying, "Yes, we received the call." And I was like, "Oh, thank goodness." **Richard:** Sorry, Dave, I think we've dragged ourselves- [05:00] Richard Webbe: **Dave:** No, no, no. I was, I was just gonna say o- on the, on top of our phone lines and things, back, um, before ADSL, when you had dial-up plans and you paid per minute for the internet, there was a mob [05:00] Richard Webbe: that came out called Dingo Blue. **Richard:** Yes. **Dave:** Over in WA, Jo, I don't know if you ever heard of Dingo Blue. That had a, um, unlimited minutes plan. Unheard of, but it was a, like this really cost-effective, like rather than pay, you know, 30 bucks for 30 minutes, it was, you know, 45 bucks for unlimited minutes- Wow ... or whatever my dad was paying. I, I don't know. But so because he had this unlimited minutes plan, he put an extra phone line on the house, so we essentially had 24/7, like, ADSL constant internet before anyone else. Like, we just had always on internet. So, like, the whole jump on a web browser and, you know, AltaVista it, as it was back in the day. Um, was- **Richard:** Now I know why. You were born of a nerd. **Dave:** Yeah. Yeah, yeah. So I, I had that always on internet from a very, very, uh, early stage of the internet. Um, [06:00] Richard Webbe: **Joanna:** yeah. I, actually, my first experience of internet and computers was I was in grade- Five, in primary [06:00] Richard Webbe: school Mm And my best friend ha- had moved from Melbourne, and her dad actually worked for the ABC, and so he was a, a research reporter for ABC, so he had a computer in the house **Dave:** Mm. **Joanna:** It was fantastic. I loved going there. I loved it, just going over, ooh, ooh, playing games. **Dave:** Right Right? It was- Mm ... it was this thing of awe when they first came out and the, the home computers were, were there and- And **Joanna:** they were so big. **Richard:** Yeah. I re- do any of you go to the old ATAG conferences? You're probably all too young for that. So it was the Australian Telecommunications User Group, and literally you'd get 100,000 people plus through the conference over a couple of days. Early Cisco, Telstra, Optus, all of those, and they were just massive, massive sort of, um, uh, meet and greets. And it was all about, um, how, um, how people got things going. [07:00] Richard Webbe: And I remember meeting kids there who, like your father, Dave, [07:00] Richard Webbe: had ordered five or six ISDN lines into their house- ... and they were becoming their own service provider for internet access, and making a lot of money- **Dave:** Mm ... ' **Richard:** cause they just knew the trick. **Dave:** But what, what was really fun is back, back in the '90s there was a computer user group, and one of the members, I w- I'll, I'll, I'll shade some of the details to, to keep, uh, anonymity for the guilty. But, uh, they worked for a government department organization, and once a month a group of computer people, like nerds, would... There was this computer user group where we would get together in the cafeteria of this government department after hours, and they would run network cables, and we'd set up hubs, and we would use the high speed, like, this was back in the dial-up days, where we'd g- get access to, you know, 100 megabit plus connections that were unheard of at home for downloading things off the internet and stuff. [08:00] Richard Webbe: And so that was, that was a treat every month, going to the computer user group. And, and we'd, we'd cart our towers and our giant CRT monitors- ... and we'd, we'd unplug it all, and we'd, we'd cart it in there, and we'd all be set [08:00] Richard Webbe: up on the, on the cafeteria tables. Like, there'd be, like, a dozen or more of us. Um, and the guy would run this, like, blue cable out from the back server room. We'd plug in our hubs, and all of a sudden, like, we're all online with this super high-speed internet. Now look at **Richard:** us. Well, well, well, well, what, with sovereign AI and some of the new announcements that you and I were discussing, and Matt, discussing earlier in the week, um, that may be a scene again where we turn up with our own sovereign AI under our arm, with a high-pitched, high-speed processor and our own disc, and sitting it down and say, "Here's my AI." **Dave:** Well, seeing as you used the magic letters of AI, we should do this. Pretty sure the first person who uses the word AI on this show owes everyone else a beer. Yeah. I think that's how this works. **Richard:** I'll see you at the pub in Sydney. **Dave:** Brilliant. What date? What place? Sorry, Matt? **Richard:** What date? [09:00] Richard Webbe: **Dave:** Uh, the [09:00] Richard Webbe: 23rd. 23rd of June. **Richard:** Mm-hmm. **Dave:** Mm-hmm. Uh, we will, we will come back to that. Actually, let's, let's kick off with that. Why not? Let's, let's talk about that. Uh, we have our AI Operators IRL event, uh, which you can come to in the northwest of Sydney in the suburb of Rouse Hill. So if you are keen to come and talk AI and have a drink and catch up and hear from different people around how they're using AI, uh, there's myself and, and Richard and a few other local business people and stuff that we've got coming. Um, with Sean, who's been on the show, will be coming and joining us to share a bit of what he's doing. So we're gonna have a few different faces and voices. You can do that at, uh, go to aioperatorspod.com. That is our new website, and if I learn how to use a computer, I will bring it up and share it. I'm highly organized, as you can tell. [10:00] Richard Webbe: Uh, let's... Here we go. Here we go. Here's our new website. Here's our new website. The aioperatorspod.com website. Go to IRL, and there you can [10:00] Richard Webbe: see the details. Click on the Reserve Your Seat link, which will take you through to Luma. Um, we've got 20 going. Now, the room only hold- holds 40 or 50, so we're nearly, we're halfway through to capacity with, uh, still 20 days till the event. So if you're keen, jump in, lock in your seat. Looking forward to having a, a bunch of people there to talk AI with. It's this, but in person. **Richard:** And we're gonna run much bigger events. This is our test run, or what would I say, pilot, and, uh, it's already getting a lot of attention. Yeah. So we're doing it in Sydney, Melbourne, Brisbane, Perth- Brisbane Adelaide, Kalgoorlie, you name it. **Dave:** Well, uh, yeah. I mean, uh- **Richard:** I said that for Mano. I said that for Mano. **Dave:** Yeah. I mean, Jo- Jo's keen. She wants to be at... uh, come, come to an AI Operators IRL event. You, you, you're a bit of an events person, Jo. You run your own events, don't you? [11:00] Richard Webbe: **Joanna:** Well, I haven't had them launched yet, so [11:00] Richard Webbe: I've actually, um, just in the process of getting it up and running. I'm just sorting out partnerships and, um, yeah. So I've got a partnership in the pipework. Uh, Friday we have a meeting, and they're government backed. They were able to get partnership with government and Chamber of Commerce as well for Western Australia. So- Yeah, nice ... it's going to be, um... I'm looking forward to that conversation just to see what I can bring to the table and what their, you know, what their plan is. 'Cause their mission really is to bring AI education statewide and get it in the leadership. They've got quite a large number of members now and, um, yeah, it's really exciting. So yes. Yes, I love having an event, a good event. Um, I'd love to be over to coming, coming to your, uh, event, having a beer and talking AI with you all. Um, but- You **Richard:** know, Jo, being an, being an ex Sandgroper, um, WA government is very supportive of business over there and getting those things moved around, so well done. [12:00] Richard Webbe: **Joanna:** Yeah. Yeah. Well, I actually had no part in that partnership between that [12:00] Richard Webbe: organization, but at the same time they launched on LinkedIn was pretty much the day I was sending out all my emails to Chambers of Commerce. So as soon as I saw that, I was like, "Righto, someone's onto it. That's fantastic. Like, we need to talk." **Dave:** Brilliant. Yes. And then of course, if you're out there listening and you wanna partner and sponsor us and be part of the IRL event series or the podcast in general, uh, connecting with an interested and engaged audience around AI, uh, again, go to the website and fill in our contact form, uh, that you're interested in a partnership. We would love to hear from you. Uh, but if you're just here joining us and listening to us, we also love your partnership. Make sure you drop a comment in the chat. Make sure you hit the like button and subscribe, and do the things that let the algorithm know that more people should be here with us, uh, learning about all things happening in AI and the news and- [13:00] Richard Webbe: **ai voiceover:** Welcome to the AI Update. Let's look at what's [13:00] Richard Webbe: happening in the news **Dave:** So much is happening in the news. I'm jumping straight in 'cause we can banter forever, but there's, there's just too many things to talk about. So we're gonna kick off with one that I didn't give anyone a heads-up on, but I think is just, like, a quick and easy one to discuss, especially, m- Richard, off the back of your story that you've told, uh, on this show around the impact of intranets. You've talked about how back in the '90s, the 100 million percent improvement that they received from rolling out an intranet. **Richard:** Yep. It was astounding. And I, I used to **Dave:** love it- OpenAI. **Richard:** Go on **Dave:** OpenAI have just released a new sites functionality. So if you're using their Codex app and you're building little internal calculator tools or process management tools or anything that you're doing, like modeling tools internally for your organization, where did you host that? [14:00] Richard Webbe: Where did you put that? You had to, like, find a place or run it off your local machine or, or push it up to a public place like Vercel or somewhere like that. They have [14:00] Richard Webbe: basically built in their own SharePoint intranet service type thing- Ooh ... into OpenAI Codex. **Richard:** Oh, **Dave:** wow. So if you create a little tool, you can now share that within your organization if you're on a ChatGPT business and enterprise plan. So they are really trying to, you know, get in on that business thing, and this could be anything. It could be like, "Hey, we run the internal employee resource group, and we're gonna have a microsite that talks about our events to service this part of the population or whatever." Or like I said, a modeling calculator or a marketing brand platform or whatever businesses are building with these tools internally, they've now made it super easy, one shot to publish this internally. **Richard:** The **Matt:** question is- **Richard:** Even, um- Oh, go [15:00] Richard Webbe: **Matt:** on Matt ... visual aids. Yeah, I, I... This sort of thing's really good for visual aids and demonstrating and making sense out of data. Like, you think about a typical corporate meeting where everyone's snoozing 'cause no one really understands what's being said. Imagine actually having an interactive thing that actually looks good that you've spent [15:00] Richard Webbe: zero time preparing. **Richard:** That brings me to my, the point I was going to make. Uh, I worked on a lot of intranets with a lot of companies around the world for many years, and if the mar- uh, if I remember correctly, if the marketing te- team were in charge of the intranet, you'd get a menu that was this wide with 1,000 choices and only this deep, uh, and you were confused about where to go. But if you let the engineers and the IT people run the menu driver on the intranet, you get a menu that was 50 layers deep, and you'd have to answer 20 pull-down menus before you got to the information you wanted. And the ownership of such information sites was always very political, wasn't it? Did IT own it? Did marketing own it? Did the business own it? I see this sort of thing from someone like AI coming in, giving the power back to the business to share the right information to the right people and remove a lot of design and control issues. I think it's wonderful. Hmm. **Joanna:** Yeah, I have, um... I just, I haven't quite caught up with the news. [16:00] Richard Webbe: I'm a bit like what [16:00] Richard Webbe: Matt said. It's all, you know, uh, there's a lot going on, and it's very hard to just keep up and stay current. But, um- **Richard:** It's changed again. **Joanna:** I, I use Replit a lot. Um- Mm-hmm ... and- I've just noticed, like I did hear that so they've actually announced that they've got a business package that helps b- uh, new businesses start all the processes, and it's multi, um, agent as well. So it's literally being... So I think these organizations are really going down that avenue as quickly as they can to get everything put together as a package, aren't they? **Matt:** Yeah. There's literally a quote that I got when I was preparing for today in my research where the last week it's become really obvious across multiple different platforms in different spaces and niches that companies and even countries are no longer really seeing models as the product anymore. [17:00] Richard Webbe: It's now the control plane around it- Mm-hmm ... which we've all just called harness in the past. **Dave:** Yeah. **Matt:** So yeah, like everything that's been released, like you just said, you know, Replit realizing that this control plane around various models to provide what people want is the product. **Joanna:** Yeah. Yes. Yeah. Yeah. It's, um... I think I started with Replit probably over, I don't know, maybe 15 months ago. Mm-hmm. And the, the amount of things that now that I can, I can pretty much do absolutely everything on there. I don't only use Replit, I use a number of different programs, but, um, platforms. But yeah, it's just mind-blowing just literally in the last three months or even since, I think the Replit, um, build-a-thon, 10 year build-a-thon was, which I entered, was on around about, I don't know, early May, maybe three weeks ago. [18:00] Richard Webbe: And I think at one point they had over 25,000 users on because they did the agent free for 24 hours- Mm ... and they did the builder. And so it literally, they were, um, [18:00] Richard Webbe: on the hop at the same time troubleshooting, and I have to give them credit- Mm ... 'cause the amount of support that was available for anyone that was building in that 24 hours using their agent- Um, I could just email and probably 10 minutes I would have, uh, a response and I would ha- and it would be rectified because I wouldn't be the only person saying, "This is bung. This is not happening." So they were getting... You know, their AI was literally throwing all those questions and, and realizing, "Okay, there's a bunch of question that's all the same." You know, so it was... They were able to really troubleshoot so quickly. Um, but the engineers had announced that they actually put that build-a-thon and upgraded their agent one week before. I mean, everything is just moving incredibly fast. [19:00] Richard Webbe: **Dave:** Yeah. I, I... My, my first experience trying to do any kind of proper dev was Replit, and I used it, and then two weeks later they're like, "Replit new agent has come out- Yeah ... and we're gonna give you free [19:00] Richard Webbe: access to fast thing." I was like, "Oh, cool." I'm still finishing my app at the time, so I, I knocked it out and I thought Replit was amazing when I first used it, um, because I... Yeah, it was so good, um, for doing that sort of stuff. Since then, obviously I've moved on to doing local development with your Claude Codes and, and everything else- Mm. Codex and so on. But, um, for just an all in one that set up databases and could do things, and this was, you know, 12, 18 months ago, it was fantastic. Uh, um, but I think, I think back to this Codex thing with sites, organizations are seeing sprawl, they're seeing employees- Mm ... building things on Lovable and Replit, and then sharing these sort of public links that anyone could access because employees don't always think around security implications of these things. They just go, "Check this thing out, and here's a link." And it's like, "Well, is there a login?" "No, no, it's just a link." And I was like, oh my goodness, like- **Richard:** Uh-oh ... [20:00] Richard Webbe: **Dave:** that is horrific from a security point of view. Um, and so good to see Codex coming up and, and others saying, "Okay, if you were on a business plan and all your users are logged in [20:00] Richard Webbe: through, into our governed org where we can put some guardrails and so on, on it- **instagram sketch voice:** Mm **Dave:** then you can share things internally, securely between these people" just to help take all the energy and excitement from using these tools and, you know, put some kind of protection and, and security around them. Um, because- Because- **Matt:** Oh, **Dave:** there we go ... companies have a new AI problem- Yeah ... our, our logo being in the way. That's, that's... There we go. I didn't find that muter button. Put the logo of Richard Garret instead. Uh, companies have a new problem, too many agents. Uh, as it gets crea- uh, easier to create AI agents with platforms like Anthropic's Claude Cowork, some businesses are dealing with AI agent sprawl. And it's funny, I mean, this article came out a few weeks ago. [21:00] Richard Webbe: My AI just surfaced it for me this morning. I hate it when it does that. It's like, "Here's some news headlines that are gonna be useful for you." Oh, man. I thought we were prescient, that we were, uh, ahead of the game talking about this last week- [21:00] Richard Webbe: Yeah ... but this article came out the week before. Like, anyway, uh, we talked about this last week on the show, how there's gonna be this issue of too many agents, too much- **instagram sketch voice:** Mm **Dave:** tokens being burned and, and how companies are gonna need to be bringing back some of that compute onto their own processes and things, or the costs are gonna blow out. There's multiple articles I've seen this week around cost blowouts and people spending way too much on their tokens and, and all the other bits and pieces that go with that, let alone the security concerns of what are these agents doing? What are they connected to? What tools are they using? There's so many implications of this democratized- buildathon that's going on. Uh, what, what do you guys think? What are you hearing and seeing out in the field? **Richard:** I- I, I totally agree, but y- when I was... Sorry, when I was at... Oh, you go, Matt, in a sec, but I'll just say, I always say everything that's new is, it was old, and that everything that was old is new again, and the same things are happening. [22:00] Richard Webbe: And I remember, forget, with just with internet and internet access at VMware, I remember having a [22:00] Richard Webbe: conversation with several corporates and, uh, I'd say, you know, "How many virtual machines have you got now?" And they said, "Well, last year we had 200. Now we've got 4,500, and we're having to maintain all of those, put security around all of those, put some- Mm billing guardrails around all of those and find..." And these are major big banks and big companies in Australia, and I see the same, that explosion, the bill shock. The same thing happens around agents. You've now got departments in different parts of the same company building the same agents with the same resources. It's, it's exploding. Sorry, Matt, you go. **Matt:** No, my, my thing was exactly the same as what you just said, um, where y- early adopters are always gonna pay the price. That's- **Richard:** Yeah ... **Matt:** it's literally part of it. And as technology comes out, the first versions are always the most expensive. Mm. You know, and the most resource intensive, the slowest and, you know, the worst that it will be until things stabilize. [23:00] Richard Webbe: So this cost issue is just with the hardware catching up. [23:00] Richard Webbe: You know, as soon as we have the hardware to, to handle the demand, the costs will go down. You look at Cursor's model that they have, Composer 2.5, it's very capable and it's extremely cheap. When my Codex- Mm ... subscription quota capped out, um, I had to go looking at other options, and I fired up a, a Cursor subscription again, and I was messing around with Composer. I was pleasantly surprised, went super hard on it, and only spent about 10% of the weekly budget on their cheapest plan. So- Mm ... when people realize they don't have to use Opus for everything, you know- Mm ... that, that's the difference. **Dave:** W- we, we, we, we can talk about this when we get to wins of the week, uh, assuming we get there today. [24:00] Richard Webbe: But I've been using Hermes a lot in the last week since I discussed it last week, and Matt, right at the end of last week's show, you told me to check out Minimax, and I did, and they had their m2.7 model, and they've over the weekend brought out m3. And I am on their cheapest plan of $10, and I have [24:00] Richard Webbe: basically unlimited Hermes agent, and it does really great stuff in the way... Like, if I've got a meeting coming up or I wanted to ask it a question, I just flick it on Telegram or Discord, and it goes and pulls full, like, multi-response dossiers and research and pulls web information, and I'm setting up workflows- Mm ... and doing some development tasks with it for 10 bucks a month, and it's running on my local computer. Like, Hermes, the, the ha- agent harness, that, that wrapper, is living on the Windows Linux subsystem, the WSL on my, my Asus laptop, and I'm paying 10 bucks a month in tokens. It's, it's crazy. Like- Yeah ... the capability there. Yeah. Um, I'm really happy with Minimax. Uh, whereas, you know, the other models are super capable and, and there's probably a place for some of those models for, like, GPT for doing bits and pieces, but for all the day-to-day stuff, um, I'm getting more than enough out of Minimax, and it's awesome. [25:00] Richard Webbe: **Richard:** Yeah. We're gonna see an explosion of more hardware being purchased into [25:00] Richard Webbe: organizations for local processing, because it makes a lot more sense ex- put it all out onto, uh, you know, an online or a cloud or a service. And like the- Mm ... just doing things... Sorry, Joe, you were gonna say something then, but processing where it's most cost-effective **Joanna:** I think that, uh, for... It's really interesting all of this because, um, I'm sort of pivoting from the agency side of AI, and I've, uh... really pushing on the education side of it. And part of the education is just letting people know that they don't have to stick with one. **Richard:** Mm. **Joanna:** That they can run multiple, they can choose. You know? And so if they're finding that there's something like what Dave's saying and it's expensive, or that he's hitting the ceiling, then, you know, maybe choose something somewhere else. Like, and like Matt was saying, he's had to pivot because of the usage of his tokens. So yeah, I think part of the education is people become very... [26:00] Richard Webbe: You know, like we [26:00] Richard Webbe: have, you know, particular brands we all like, or particular things that we get used to and that comfort zone. But it's like, no, actually give it a try. Yeah. Give this a try. Try them out. Like, that's what I think part of the education is, just making sure people have got that mindset where they go, "Okay, I'm not just one particular platform, or LLM, or- Yeah process." **Matt:** I have a question for you, Jo, on that, because I fully agree. Like, I have zero loyalty towards any model, any harness, any lab. You know, I, I, I will constantly jump to suit what suits me and my workflows and- Mm ... the people that I serve. You know, I'm always gonna- Yeah ... grab the sharpest ax, otherwise, you know, what's the point? [27:00] Richard Webbe: But, but my question for you is for somebody that's like, "Okay, all I know about is ChatGPT. I've heard people talk about Claude." Um, Dave's spouting about Minimax. Matt's over here talking about Cursor. You've mentioned Replit. What the hell is going on? Like how, how does someone actually make sense of all of the options? And w- like how do they do it without just melting their brain? 'Cause I feel like I'm on the upper end here, and my brain is melting. **Joanna:** Yeah. I think, do you know what I would do? I would go to Claude, and I would punch in what it is you're trying to achieve and w- you know, like budget. Like put in what your requirements are and have a conversation, whether it's Claude or whether, you know, people using Perplexity, or whether they're using other LLMs. Um, you know, I think the interface is really important for people. Like the user interface, UI, is It's gotta be nice and simple, and for some people it's gotta be more tech-looking, and some people want more pretty. So it depends. Like when you go to Manus, you have a look at their user interface. You, you go through all those sorts of things. [28:00] Richard Webbe: But I would just definitely... E- every time I have a, a question or my brain's starting to blow up, I literally just talk, I talk it to the AI. I go, "Right, this is what I actually want. There's too many options. Uh, maybe give me the top five [28:00] Richard Webbe: for, uh, this process or this, uh, skill that I want." Or, you know, I just try and, um, I try and bring AI into it to have the conversation to help me choose what the heck is going on. And, and also I say, "And don't be nice to me about this, and st- and do not lie." **Richard:** You tell it, you tell it, "Don't," yeah, "Don't sugarcoat it for me," yeah. **Joanna:** Don't you sugarcoat it. **Matt:** I love, um- No ... I love what you just went through there. Like- Yeah ... you just mentioned two new names. You mentioned, um, Perplexity and Manus, which again, two more things. Um, uh- What you, what you also said about, like brain dumping to the AI and helping it make sense and giving you, "Okay, what's the top five things that... Just help me make a decision," think... Like, if you have a, a marketing ear on, think of every single company that's just like, "How do I become on that list?" **Richard:** Mm. [29:00] Richard Webbe: Oh, yeah. And that's... No, that was about to bring that up. So the new web AI search which is starting to take over, so I don't know if everyone's noticed- Mm-hmm ... but I've started to see private advertising appear in my AI answers. Has anyone [29:00] Richard Webbe: seen that yet? Yes. Now, that is, that is the new economical model that is gonna break the internet. Yeah. Uh, and I use that term euphemistically, right? Because, uh, the shortest path to a, a, a purchase or a business connection or a transaction used to be I look them up on the internet and find it. Then it was, oh, they're on my social media, so now I see a nice shirt on Facebook and I go, "Oh, I'll buy that"- Mm 'cause I love instant gratification. Now we're getting into the purchasing decisions coming inside the window of an AI when we do a brain dump of everything that's bothering us. You know, the smart marketing companies out there will be connecting- Yeah ... with the AI. They'll get their Venn diagrams out, and when, you know, Matt, Joe or Richard or Dave say, "I've got these four issues," oh, our product solves that. [30:00] Richard Webbe: Bang. And it brings us to a concern about business and commercial bias. When is it gonna come up with the AI, what I call, or what Dave and I would call the syllogism, which is a Latin [30:00] Richard Webbe: term for sort of fuzzy logic. That's what we named the company, Syllogism. When does it come up with the actual answer versus how much someone paid to give you that answer? And the best example of that was when I was working for Nokia, I'll age myself here, back in the '90, right? And I was- **Joanna:** I loved my Nokia. **Richard:** It was back in the '90s, and I was launching WAP across China and, uh, the Middle East and Scandinavia, believe it or not, which was the first location technology. And I think it was... I wasn't at the actual event, but I think it was probably, you know, some prime minister from Norway or Finland or whatever, and they're standing outside a restaurant with the phone, and they're going, "So watch, we can look for a restaurant to eat using this phone." People had never seen that before. Of course, it's natural now. [31:00] Richard Webbe: And they go, "Book." And it kept coming up with a phone about, a, a restaurant about four miles away, not the restaurant where they're outside doing the launch, right? And everyone's going, "This stuff doesn't work." It wasn't that it didn't work, it's just that the restaurant [31:00] Richard Webbe: down the road paid more money. Oh. And that's gonna happen with the confirmation bias and financial bias in AI. So it's going to be interesting how you use your prompts to get around some of those biases, and then how they've paid to get around that. Do **Joanna:** you guys remember- Yeah, a lot of it is... Oh, sorry, Matt. **Richard:** You go. I was just gonna rant **Matt:** about something confusing. **Joanna:** Uh, yeah, a lot of it is, um, also, like in regards to the marketing aspect, is to do with your branding and your authority in the space. So if you are putting out long form content, if you are being recognized in other publications, if you have got, um, newsletters going out, y- the repetitive, like whatever it is, there's certain... [32:00] Richard Webbe: You know, your keywords need to be exactly the same over everything that you do. And sometimes building that authority, it doesn't... People think that the SEO and like the GEO is not, is just three months. Uh, no, sometimes it takes up to eight, nine months for the actual authority to build. So it's really important that people actually start that, looking at [32:00] Richard Webbe: that marketing. Um, I did get a message in regards to... I'm part of a group that's in Europe, they're, uh, agency owners, and they sent out a message to say, um, they gave me a link and everything of, of how to start doing my, um, like paid ads with, uh, ChatGPT. **Richard:** And how's that going for you? **Joanna:** I, that is like, that's about down here on my list of things to do at the moment. Um, really I've, I'm excited. I, mean, I obviously need another... I need a H- Hermes as well myself- Mm ... to get a lot of this stuff done, um, 'cause I haven't integrated, uh- Yeah ... either OpenClaw or Hermes, but, um- **Dave:** That's- **Matt:** Shaman exploding **Dave:** there's so, there's so much happening that you guys... Sorry, Matt? **Matt:** That was two more things. [33:00] Richard Webbe: **Dave:** Yeah. Mm. You, you guys, you guys have touched on so many different things. I've just been like, "Ah, so many dots to try and connect and link for the audience, uh, uh, out there." [33:00] Richard Webbe: Um, and, you know, jump in the chat if you've got a tool that we, we haven't name dropped yet. Um, need a little bell for every time someone mentions a new tool. **Joanna:** Oh, yeah. **Dave:** Uh, you, you mentioned the, um, you know, the hardware costs coming down, and as the hardware costs come down and more of this stuff happens. NVIDIA just announced- Yes ... their new chip. Uh, so this is sort of the, I suppose you could almost say this is the Windows version of the Apple Silicon in some ways with the whole system on a chip. So Apple came out with their M processors, what, five years ago, where they put all the RAM, the GPU, the CPU, everything, like, in one little piece of silicon. Everything's really next to each other. So everything's very close and, and all built into one. NVIDIA's now done that with this new N1X chip, um, where they're saying you could run 120 billion parameter model on your desktop, super battery efficient, super energy efficient, as well as do all your business and your games and everything else. [34:00] Richard Webbe: So this was Jensen saying you can basically run this [34:00] Richard Webbe: AI data center in the palm of your hand. **Richard:** I think, uh, uh, Michael Dell became the, uh, one of the second richest men in the world with one of these announcements in the last few days with the price going, uh, share price going up with Dell so dramatically with the announcement. **Joanna:** Mm. Yeah. I think one of the things I did read was that, uh, NVIDIA did try to launch this in 2025, but they felt that it was underpowered. Um, and that's, you know, part of why it didn't go ahead. So clearly they've got the powering going. Um, and a bit of research I was looking at that I think, I think it might, not sure it was the Microsoft, the Dell, their battery, it's like if it's running on full hilt it's 40 minutes- Shit unpowered. **Richard:** Oh, dear. **Joanna:** So if it's... Yeah, I need to research that a little more, so no one take me to task. [35:00] Richard Webbe: **Dave:** That's like Windows 8 all over again. I remember when I was in the corporate space, I, I upgraded to a, a new laptop and it came with Windows 8, which was just not [35:00] Richard Webbe: a good operating system. It was not well optimized for being mobile and, and everything else. And I was going out to client meetings, I, I'd get less than an hour out of this thing. I'd have to take my charger for a one-hour meeting. It was ridiculous. So I pegged it back at them and said, "This is no good." Ended up, that's when I switched to Mac, 'cause I went to a MacBook, um, Pro then, and I got hours of battery life. And went, "I will learn new shortcut keys to get better battery life." uh, and moved to an Intel Mac at the time. But- **Joanna:** You- oh, it's so true. My, um, I've had a number of laptops and then, uh, this, I bought a, purchased the MSI laptop. My battery Beautiful **Richard:** Is it? Wow. **Joanna:** Mm. Yeah. **Dave:** Yeah. I, I, I've got a MacBook Air that I use when I go out and about, an old M1 MacBook Air that I got, um, and I'm using an ASUS ROG Republic of Gaming laptop, but that's like a desktop. It just lives on my desk. I don't actually take it anywhere. Um, it's just something I bought, you know, four years ago, and it was a good price and does a decent job. So a good [36:00] Richard Webbe: cheap laptop. In, in terms of power, I loved Elon Musk's announcement this week that he's gonna put data centers in space with unlimited solar panels, uh, to provide high... low-cost, high availability, AI data center and data access. The only thing I don't know how he's gonna solve is the lag time between space and terrestrial usage, but, you know, maybe there's a number of batch processing, uh, capability that could be done in space, and then- **Dave:** Lasers ... **Richard:** to, uh, yeah. The laser. You've got- **Matt:** I have a, I have a genuine question on that, Richard, 'cause you probably- Yeah ... have more infrastructure knowledge than me. Do, do you th- do you believe that the mic- microwave transmission is slower than copper? **Richard:** Um, no, no. I just think the amount of connectivity and the amount of data may mean we will feel a lag down here. [37:00] Richard Webbe: **Matt:** Mm. 'Cause I- I've been running Starlink for a few months now, which is- Mm ... obviously [37:00] Richard Webbe: wireless to satellite and, you know, **Richard:** onwards. **Matt:** Mm-hmm. And my latency, so, like, the actual milliseconds latency, is faster than when I was on NBN, um, which is- Yeah. No ... probably things to think about our Australian infrastructure. **Richard:** Yeah, yeah. Yeah. So electrons and the speed of light, they're very close to each other, right? So, um, yeah, look, I, I remember I was involved with RMIT when they were setting up their, uh, connections with microwave links across the park for Melbourne Uni and stuff like that. Yeah, I just think about the size of the data and how much they have, you know, the, the bottleneck of getting- Mm through down here. Um, they'll probably solve that with some technology. I certainly think terrestrial power in the planet is a major issue, and countries that don't take steps forward to building large amounts of power availability will be left behind. Mm. I believe power is the new, sort of, uh, the new thing, uh, the new oil, right? [38:00] Richard Webbe: **Dave:** Like our- I, I saw [38:00] Richard Webbe: a, I saw a, an, a video- We- ... where they were looking at this stuff, um, in, in the US, and there are towns where their power costs are going up because they're putting big data centers in without the new power infrastructure. **Richard:** Mm. **Dave:** Um, and so there is this debate over the benefit of the data centers versus the cost versus the noise and, and some of this is just, you know, timing. There's an upfront cost while they build their own power infrastructure, or is there a regulatory issue where they're not being allowed to build their own power infrastructure, or they're doing it in parallel? So- In which case they should probably be subsidizing the local community to an extent while they eat all their power as they set up until they come online, and then everything will drop down lower. It's all complicated and, and no one wants to pay more power bills than they have to. And in Australia we are Can definitely empathize with power bills- Yeah ... going up for no apparent reason. Um- **Joanna:** Oh, it's... And even on the weekend we had a massive low front come through, and it wiped out... I was 24 hours without power. Now- **Matt:** Wow ... [39:00] Richard Webbe: **Joanna:** I, there was a project that was north of the city, sort of out towards the airport, sort of [39:00] Richard Webbe: northeast, I guess. Yeah. And that was earmarked for a data center. Um, but they can't actually... I mean, at the moment, trying to get coal out of the ground, yes, we have a lot of people that are on solar and batteries and things like that, but we are still experiencing extended power outages just from, um, I guess this area, the power is still above ground. In the areas where it's below ground, not so many blackouts and things like that. But we are still experiencing that. So to put a data center in the northeast of Perth area, um, is the actual infrastructure, like you said, Dave, is the actual infrastructure for the city and for the surrounding suburbs, areas, is it actually up to scratch? Before they go, you know, the- **Dave:** Yeah ... **Joanna:** suck of that, uh, electricity. **Dave:** Yeah. Yeah, like, uh, the, it's a multi-pronged factor and, you know- Mm ... people talk about, oh, they use so much water, but the, the new generation data centers are using all, like, closed system cooling. So they, they, they recycle their own water and stuff- Mm [40:00] Richard Webbe: so you're not getting- Exactly ... those water usage losses and so on. So it's complicated and, and [40:00] Richard Webbe: people love to dress these up as black-and-white issues- Yeah ... but obviously there, there's a lot more nuance to it. But it, it's definitely an evolving space where we're seeing a lot of stuff. And in Australia, I don't know how much investment, um, we're getting. I, I know, like, um, we see OpenAI, et cetera, saying they're, they're putting more stuff down here, and they're working with, what's the name of the company? Uh, the- I can't remember ... I can't remember the name. NextGen, NextGen Systems. **Matt:** Oh. **Dave:** Mm. NextGen run, do data centers, Australian company. Yes. So it is happening here in, in, in bits and bobs and so on, but, um, we, we need, like, there's people... This is gonna be this interesting thing. I think we've, we've, we're sort of seeing this big rush of people getting o- onto the AI train and using AI, and it will plateau and will drop off a little bit as people realize, I'm not seeing the gains, I'm not seeing the results, the token costs are too high- Mm. [41:00] Richard Webbe: It's too darn high. Uh, and things will drop down uh, until we find that, that, that medium. And I suppose the, the hope is that [41:00] Richard Webbe: it doesn't just become something that the big corporates can afford and the small-medium businesses can't. Like, how do we find those models? And I think the, the difference between the big, you know, the OpenAIs and the, the Anthropic Claudes and stuff, they're gonna charge a premium, versus what is happening with the Chinese models, the Minimaxes and the Kimis and the DeepSeeks- Mm and so on, and can you get enough benefit out of those? And what is the wrapper and the experience that you get from using those, whether it's coming directly from the model provider, or whether it's something like Hermes or OpenClaw that you're accessing via Telegram. Uh, Hermes have just today announced they've got their new desktop app. So everyone's got a desktop app now. Uh, so for people that... A- and Hermes even has a native Windows installer. So for your general punter who doesn't wanna set up Linux on their Windows machine, who just has a Windows computer and wants to have one of these agents, and they don't really care if it's 24/7, they just want it to be there when their computer's on, or they leave their, their tower or their, their desktop on all the time anyway, whatever. [42:00] Richard Webbe: They can now point and [42:00] Richard Webbe: click, run an executable, install the Hermes agent. They can have a little desktop app where they can do their chats and set up their Telegram or their Discord, and they can then Telegram it from their phone and connect whichever model they like, so they're model agnostic. If they wanna pay ChatGPT plans, they can put, put ChatGPT, and if they wanna pay Minimax, they can put Minimax in. If they wanna pay OpenRouter and just, you know, pay per token, they can do that. Like- **Joanna:** It sounds fantastic for a non-dev. It sounds beautiful for me. **Dave:** Yeah. So I mean, Herm- There's like another layer ... Hermes are really going after that easy to install consumer computer market. **Joanna:** Yeah. What was that, Matt? **Matt:** There's a really, really interesting layer to that, Jo. Like- **Joanna:** Yeah ... [43:00] Richard Webbe: **Matt:** what the Hermes agent can do, which many different agent harnesses can do this now too, w- which is the idea of computer use, which for people who don't know what that is, it's literally the agent gets their own cursor and it moves around the screen and clicks on things. So imagine a scenario, I don't know, maybe you are disabled in some way, um, but you still need [43:00] Richard Webbe: to interact. Maybe you are just, you know, brain busted and you can't remember how to do something on the computer. I know for my dad, for example, this would be great for him to be like, "I just need to do this. I forget where to go. What do I click?" And then imagine the agent then just ta-talking you through it and showing you where to go and pulling that file out that you knew existed somewhere that you couldn't find. Brilliant. You know? That kind of thing is crazy. Like, um- It is ... and I, I use that now. I, I l- for my development work, I literally will ask the agent to just run it and go click around like a user and figure out what's broken. And, you know, it's this tight feedback loop of development. But just for the average user to be able to say, "Hey, I've got a new computer. What do I do?" You know, I think that's great. [44:00] Richard Webbe: **Richard:** I, I, um, I laugh, um, about, uh, things from the old days 'cause I'm old. But, um, you know- ... I remember the Victorian government bringing out a advice for the early computer purchasers, [44:00] Richard Webbe: and it was a headline in the paper, I wish I'd cut it out, and it said- Pick your software before you pick your hardware. Now, this was official advice to the smartest brains in the States, right? Now, today, that's just natural for us, right? Right, why would you go buy a piece of hardware and hunt around for a bit of software that would run on it? Because the software's more closely connected to the business problem. And now, you know, I sent you those pictures earlier, uh, Dave, and I sent them to everyone. The real thing is, unless you're looking at a business problem inside your business that you wanna solve, amongst all this technology we're discussing that you can use, you can save tokens here, that's a relevant conversation. Mm-hmm. But only when you're starting at the business end of the issue. And, and there's this whole argument of people now are so excited about AI, they're doing AI for AI's sake, not AI for the business' sake, and they forget to step back and go, "Well, how much is this costing me?" [45:00] Richard Webbe: Mm. "And what was the problem I was gonna solve in the first place? Was it shortening my [45:00] Richard Webbe: invoice cycle? Was it looking after my staff, making my sta-" I mean, there's some new great careers in AI coming up, whereby that ex- experience, that local experience, it gives you competitive advantage. Get the girl off, and the guy off working on the call center with the invoices, and have them over here on the AI system looking at its usability, and its CX, and its UX, and what it does, and how it connects, and where the security is, and what it means. It's a really important point. I am going to take leave from this fabulous conversation, 'cause I do have a double-up, which I mentioned to Dave in this call. But, uh, always a engaging discussion. So I will say goodbye, and it's great to see everyone again. And, uh, keep pumping away. Uh, and, um, yeah, I'm out of here for now. **Joanna:** See you, Richard. **Dave:** Hey, Richard. Good **Richard:** day, everyone. Thank you. Bye. [46:00] Richard Webbe: **Matt:** On that note, I, um, I've got another couple of lines that I, I put together for when I was preparing for today. There's, [46:00] Richard Webbe: you know, like what you, what Richard just said, like this is- Mm ... this is what I wrote. So the winners will not just have better models, they'll have better judgment wrapped around those models. **Richard:** Mm. **Matt:** It's like you don't need, you don't need to put AI in. Like what's it doing? You know, what's, what's the judgment there? Mm. Like what is the purpose? What are you trying to solve? And AI could be that machine, or maybe it doesn't need to be. You know, what Dave said before about people realizing that they didn't need to just go and shove AI into everything and spend all these tokens, um, 'cause it's not actually getting the returns that they wanted, you know? **Richard:** Mm. **Matt:** The demos, you know, the, the shiny tech demos, like the period for that is kind of over. You know, controls- Yeah ... and that control plane, like I mentioned before, really is that product now. Like what's it doing? [47:00] Richard Webbe: **Joanna:** Oh yeah. I think that, um, my thoughts with all these, uh, s- you know, people's jobs being moved around and, and all that sort of change and, you know, the f- the, uh, fear-mongering that's going on at the moment [47:00] Richard Webbe: with people losing, you know, that... And, and they are cutting lots and lots of jobs. But the beauty of these, um, setups where like Hermes, things like that, people, that people can actually leave their job or be laid off and they can actually start their own like entrepreneurial journey with their own business. And I think it's sort of, it's interesting that, you know, the AI's coming up and the- The jobs situation is having to reshape, and yet, uh, and also the technology is reshaping on the other hand as a benefit for people to be able to be more creative and actually use their gifts, talents, abilities, and it may be in something completely different to what they've been doing for the last 29 years. Mm. Like myself, you know? Totally. Um, I don't think of any of my colleagues actually. Um, I think they'd be like, "I can't, I can't believe Jo's in the AI space." [48:00] Richard Webbe: **Matt:** Mm. E- even kids- Yeah ... like y- 16, 17-year-olds, like, in high school- Mm-hmm ... figuring out that somebody had [48:00] Richard Webbe: some need or, you know, people are looking for a thing, and it could just be novelty, but if people are paying for it, all of a sudden they're making money. And then it could be a real problem. They find out that their parent or grandparent or someone in their family has some issue and they're like, "I can fix that for you." Mm. Next minute, that thing starts exploding. Yeah. All it takes is an idea, a little bit of human ingenuity, and then applying it- Mm ... to these systems, and all of a sudden you're off. **Dave:** I w- **Joanna:** I was talking to- Yeah, you need to take... Oh, sorry, Dave. **Dave:** I was gonna say, I, I was talking just before this call with, uh, one of the speakers who's coming on to the IRL event, and she's, uh, her and her husband run their own business, and she's been building tools in, in Lovable and other bits and pieces, uh, for running internally, and she's using AI to build tools. She's not building AI. She's not really deep in using agents. There's no Homies or anything. So she's not doing that, but she's using the AI to build tools to solve problems and- Mm ... and fix workflows in the business. And so there's those two sides of the AI coin. **Matt:** Mm. [49:00] Richard Webbe: **Dave:** Is it an agentic solution where you want the AI [49:00] Richard Webbe: actively involved doing things for you, or are you using those agent tools to build things that then run independently of AI? Like, just use them as a, as an outsourced engineer and develop something for me, and then bring it back. Uh, she also has a teenage son who is, uh, trying to start his thing and working out what he's doing entrepreneur- preneurially. And so he's, you know, experimenting with this i- AI idea, and he's doing a bit of safe production there. He's doing a bit of this, and he's, he's mixing it up, and he's using this as one of the things he's got access to in his arsenal- Mm ... in order to work out what is he gonna do with his life. That's, that's crazy. **Joanna:** That's just some... Uh, this is why I'm, I'm very passionate about people that are starting at the very beginning and not knowing anything about it. [50:00] Richard Webbe: The best way for them to learn about what's possible is by building an app. Like, okay, so all the things that you use every day on your phone or things that you've got an issue with or there's a problem or that you need support, like it might be a self-care- Mm ... or something that you want to be organized in, or it might be meal [50:00] Richard Webbe: preparation or whatever like that, um, it's just the easiest access to get in and do some vibe coding and just build up your confidence with, um, working within, you know, certain platforms. So I, I find, like I built so many. I think I've built over 25 just from mucking around and just for my own, to build my own confidence- Yeah ... over the last couple of years, and so it's made me realize that- When I meet people and they're talking to me about something, I go, "Ah, ah, you could do something with that." And they're like, "Oh, no, I couldn't." **Dave:** Mini, **Joanna:** it u- And I go, "Yes, you can." **Dave:** It used to be y- you, "I, I better go put some of this stuff in Excel and, and try and work out the maths or the do a thing." And now it's like, "No, I'll just build an interactive web page that does all the calculations for me." Mm-hmm. Like, like, just little things like that, which where you'd spend a day just trying to work out some formulas in Excel, and it wouldn't look half as good and wouldn't be tw- like, nearly as interactive. And now you go, "No, no, just build me a little micro app where I can have sliders- Yeah ... and change values and show dynamic charts and things." Like- **Joanna:** Yeah ... [51:00] Richard Webbe: **Dave:** all these sorts of things are [51:00] Richard Webbe: just so just there to be done. **Joanna:** And the wonderful thing is, with when you do your build, you can actually pull the, uh, code and put it in your GitHub if you want, um, if you, you can get to that point. But you, you can actually just keep republishing. So if there's an update or you get some feedback from someone and they say, "Well, this would be better if it was here or this," you can just literally jump back in and then republish. Mm. And the updates are done. And then you can just pop that code that, whatever they've used to put together, uh, just, just save it in your GitHub. **Matt:** Yeah, I love... You make the, the w- the land of developers and, and DevOps so, so simple. It is kind of just that though, realistically, when you boil it down. It's like the Midwip meme. Just do it. **Joanna:** Yeah. **Dave:** Um, what, what, what, what, what was the thing that I sent you, Matt, where it, it's just code? That meme? I, I don't have it **Matt:** easy **Dave:** to pull up. [52:00] Richard Webbe: **Matt:** Yeah, I, it, the, the meme template's mid-wip, mid-wip. Um, [52:00] Richard Webbe: but yeah, no, it's true. And some of these things, like, no one really knows where to start. So if I was to give any advice these days, like, I, I keep jumping around between different things, like I said before. Zero loyalty- Mm-hmm ... on any tool. I'll just use what works and whatever is the sharpest ax at the time. **Joanna:** Yeah. **Matt:** For me right now, my recommendation is the Codex desktop app. So ChatGPT subscription. You don't have to use ChatGPT. Yeah, that's the one. Um, y- with that- ... you get everything that we've just talked about during this episode. You get all those functionalities, all that computer stuff, uh, everything that you could possibly need, everything that Jo you mentioned that Replit does for you, you do through Codex. [53:00] Richard Webbe: Um, I've been a terminal head forever, you know, living in terminal interfaces with those agents. Mm. And Codex still exists in there as a CLI, but I've literally been using their desktop app now 'cause it just seems to work better for my, for my mind and for the work that I do. So yeah, that would be my recommendation for [53:00] Richard Webbe: people these days if they **Dave:** want- Th- th- this is what I was saying to you a few weeks ago, Matt, like graphic user interfaces were created for a reason because sometimes it just makes things more accessible than, than, than through text on a, on a terminal. As m- as... And then your Tmux setup was amazing, and the way you had everything linked, like you'd almost built a GUI out of terminals, was incredible. Um, but- Yeah ... yeah, look, it's... A- and that's interesting how they all are having it now. So the Claude app came out and then the, the... There was a ChatGPT app that was short-lived and has now been replaced by the Codex app. Hermes have brought out a desktop app which, um, it's just they, they know that people want that app experience. They wanna reach everyone. They wanna reach business users. Most people aren't gonna open up a terminal and look at ASCII. They, they're not interested in that. Mm-hmm. They want an app. We, we're living in the smartphone generation where everything's an app. There's an app for that. Um- **Joanna:** Well, I tell you what, how interesting. With this blackout, when we had the blackout on the weekend, so I realized, like, how dependent I was on power. [54:00] Richard Webbe: **Dave:** Mm. **Joanna:** Um- ... and I was just like, we were charging the phones on the compu- by the computer, and then the computer goes flat, and then we've got the phones. But I literally was still working from the phone. And so when I looked at it I thought, "My laptop could actually go." You know, please don't go. Please don't go anywhere. But my laptop was flat. I mean, and I could use my phone to work with everything. I have a... Sorry, another topic. I have a genuine question for you both. So with the, uh, news that you were talking about in regards to being... that oversaturation of agents and- Mm-hmm ... you know, like having to put guidelines in and really try and keep this secure nature of everything, um, I'm sort of feeling like there's not really any need to have these multi-tiered agent- Am I missing something? **Matt:** What's your question there? [55:00] Richard Webbe: **Joanna:** So the question is, like, do you actually need to have all these... Like, so with businesses where they've, it was very popular to, [55:00] Richard Webbe: over the last, you know, three to five months, to be building multiple agents within your- Mm ... um, business system. So is that actually... It's actually saying that it's saturated within it, they're getting a little bit out of control because there's just so many different agents. Is it really necessary to be building all these agents when you can use, like you said, Hermes and like- **Dave:** Well, I mean, I mean, as y- as you say that, I will, I will counter by saying within my Hermes I've got about 15-plus different agent profiles- **Joanna:** Right ... **Dave:** set up within Hermes. So Hermes is- Okay ... the master- Yeah handler. **Joanna:** The hub, yep. **Dave:** But within that I've, I've set up a bunch of profiles with their own little soul files, they call them soul.soul, um, which is just their system prompt- Mm ... which defines their thinking mode. Okay. Um, and so one of the things I've done is I've evolved my system, just to nerd out a little bit as we wrap up the show on what I've been doing with Hermes- **Joanna:** Please **Dave:** do [56:00] Richard Webbe: is I, I started, I started my journey, you know, like everyone, ChatGPT. And I was using ChatGPT projects, and I put system prompts in the project definition, which means I [56:00] Richard Webbe: could be really lazy with my prompts inside ChatGPT because I'd sort of defined these guardrails through the project parameters, and it kinda did things and knew who I was and knew what I was doing in that space. Did that for ages, but then I was, you know, hitting limits and, and just getting frustrated with it a little bit. And I discovered in my research and going, "What else can do projects?" Like, can anything else do this? And no one else had that concept. Gemini didn't really have it, their gems were n- and no one did it. And then I discovered, you know, IDE development, and I discovered Google's Antigravity, and went, "Wow, if I do this, then I can have these skill files that define basically my project things," and I actually- I had this whole hierarchy of my projects as sort of folders with skill files, and that's how I was doing things. **instagram sketch voice:** Mm. [57:00] Richard Webbe: **Dave:** And then that evolved, you know, as Antigravity and agents and, and agent definitions, and the whole ecosystem evolved, how it defines agents versus skills, et cetera. That evolved in Antigravity until, you know... And one of the things I said I loved about doing that, and I was [57:00] Richard Webbe: p- writing all this on my Substack at the time, was before this pod- podcast existed, is it gave me vendor agnosticism. I was now not tied to- Mm ... ChatGPT with all my histories in that web browser. Now all my files were on my desktop, and I was in the Antigravity tool, but all the files and all the definitions were now living on my computer. And if Gemini went rogue, I could take them and put them somewhere else, which is exactly what I did. Yeah. Okay. They changed all their licensing and their usage limits, and I moved to Claude. Um, and I've been with Claude and pushing Claude Cowork really hard and, and between Claude Code and Claude Cowork, doing heaps for months, token maxing, hitting my 100% on my 5X plan week after week, pushing really hard until I've gone, you know what? [58:00] Richard Webbe: I'm s- now stuck just in a different ecosystem, and I don't want to be paying this much money for some of these things. It's too much. Mm. So then, you know, as Hermes evolved, and, you know, talking to Matt and Mano on their show and then hearing stuff like, Matt, you're talking about where do people go, how do I keep track of all these things? I would say listening to people like us talk about this stuff, not in hype bro mode, but just discussing the realities of it. I said to you at the end of last week's show, Matt, like, there's so many models, and Joe, I, I, I went and said to Claude, I was like, "Oh, which model can I use for this agent?" And like, there's so many things, and I wanna reduce my spend, and which plans? And it was still gave me too much info, and I'm like, and who's vetted this info? How can I trust the LLM to pick which model? I wanted a real person. So when Matt said Minimax, I was like, "Done. That's it." Yeah. I trust Matt. I'm gonna give Minimax a go. So, um, I've moved... So I was with Cowork and I explo- I started exploring with Hermes and Kind of told Hormozi, just take my co-work instance and just bring it in. And after using that for a week, I realized this is not optimized because it was kind of built for a different thing, and it's not really working properly. And even back to my Antigravity days, I'd kind of confused definitions of what's an agent and what's a skill, and some of the skill stuff was mixed in with the agent definition, and it was just- Yeah [59:00] Richard Webbe: there was a lot going on. So over the [59:00] Richard Webbe: weekend, I got it to refactor my whole thing and go, okay, the profiles, the agents, that is the mindset, that's the thinking, that's the reasoning. So d- I've had, you know, an agent for every task. A lot of those tasks were just doing things. So I've actually got it to say, no, no, when you wanna push something up to YouTube, that's not an agent who manages YouTube, that's a skill that can push updates to YouTube. You got the YouTube update skill that my CMO marketing agent thinks about the messaging, thinks about the titling, and then she, she can then use the YouTube skill to push that up versus, you know, here's a specialist YouTube person that reports to the marketing person. So I think it's in... Like, and I... When you talk about do you need multiple layers of skills, I think it is useful to still have these, uh, this idea of profiles or agents that have a reasoning context, and this is their reason for living. [01:00:00] Richard Webbe: Mm. This is how they frame context. This is what they're positioning, they're putting around it And then what tools, what hands have you [01:00:00] Richard Webbe: given them, what skills in order to execute on any of that? And one of those skills might be hand off to another agent who's more specialized. But I think, yeah, that, that, that's a process journey I've been on, disaggregating the thinking bit from the doing bit, and working out those combinations and trying to refactor and re-optimize for the modern world- Mm that is different from where I was six months ago. **Matt:** Did that answer your question, Jo? Yes. **Dave:** There was a lot there, I'm sorry. **Matt:** I just wanna take a moment to say, like, congratulations. Like, you've gone through so many different phases, and- Yeah ... g- it, it almost feels like I was in a, um, like a support group session then, you know, in a circle, and it's your turn to share, and you've just, you've let it out, and it, it, it must feel amazing. So, um, for anyone listening that just freaked out, like, ki- that's kind of what you have to go through. You have to just- Yeah ... sort of figure out, and figure out which bits and pieces work for you. Yeah. Um. [01:01:00] Richard Webbe: **Dave:** And it, it's, it's getting there. It's not finished a- and I haven't fully replaced my co-work processes that were giving me my daily updates to the same [01:01:00] Richard Webbe: level of fidelity. Um- Yeah ... I've still, still got a bit of work to go on sort of doing some of those things. I'd, I had kinda mashed up and just sorta hammered into place over the last three months. Uh, it takes more than a day and two prompts to fix that. But we're getting there. Um, and I'm starting to get updates pushed out to my Discord channels with different news and events about different things, and the headlines that we looked at today came out of my daily report this morning, and so on. So yeah, it's **Joanna:** getting there. That, that's amazing, that journey, because now you have that experience, so somebody that's five steps behind you that's trying to integrate what you were doing at the very beginning, you can actually go, "Well, this is the path, this is the path, this is the path. Okay, this is, this is where I'm at now." Um- "And I've been on, in your shoes, and I can show you the pitfalls." **Dave:** And I can confidently say if someone wants a semi-autonomous agent that can do things, and they can have on their phone remotely and stuff to their desktop computer at home- **Joanna:** Mm ... [01:02:00] Richard Webbe: **Dave:** I can say, like, I'm, I'm not using the, the new [01:02:00] Richard Webbe: default built into Windows. I'm still using the WSL version. But install Homies, hook it up to Minimax- Yeah ... for 10 bucks a month, and you've got a huge headstart on everyone. Absolutely everyone. Um- **Joanna:** Yeah ... **Dave:** crazy. **Joanna:** Wow. Yeah, that's a huge journey. Um- Thank you for sharing that, 'cause it t- it's so, uh, beneficial. Like Matt said, it's really beneficial for us to, you know, have these open discussions about where we're actually at with our planning and what we're doing, actioning, and, uh, where things are hitting a wall and, you know, where we have to pivot. Mm. And also, I think Matt said, just don't be, you know, like, not don't be, but try and be, um, with your products, you know- Yeah ... anything that's on. And- Like, just be able to flow and move and change, and try not to be too loyal- **Dave:** Yeah. And then like I, I changed because there was a need. Like, I was actually- **Joanna:** Yeah [01:03:00] Richard Webbe: **Dave:** like from a productivity point of view, Cowork and stuff was doing really well, but it was expensive and I didn't wanna pay that [01:03:00] Richard Webbe: much money anymore. And so- Mm ... that's why I moved to the next thing. I moved to Cowork again, 'cause Gemini was blowing out and wasn't giving me what I needed. So for me, it's always been a cost-benefit ratio discussion, and I'm not moving for the sake of moving. I'm moving because the- I see a benefit to doing something different. And when I go to the, the next platform, I go, okay, now what are the new toys I've got access to? Like, how can I maximize this and, and use it differently? And Hermes has this Kanban thing, which is still hit-and-miss for me. It's kinda working sometimes. It's not always working. I'm having some issues getting the, the agents to hand over properly. I had a huge issue on the weekend because I was using Minimax, and the way Hermes breaks up tasks, it didn't know how to use my Minimax login for doing half the things that Hermes needs to do. So I had to spend hours getting my, my Hermes to debug itself to write new updates into its own code to let it use Minimax properly, um, because they hadn't patched it through the official things yet. [01:04:00] Richard Webbe: So, you know, they- these things are still new. They're still experimental. You- we're getting them for free. We're not paying for [01:04:00] Richard Webbe: these things when it comes to things like Hermes. Um, and so you gotta roll with that a little bit. Um, but- Yeah ... uh, yeah, it's, it's crazy. And then again, it comes back for me making sure I'm actually getting benefit from these things. And I wasn't feeling I was, you know, getting the ROI from having all those Claude tokens and just throwing everything at Claude all the time and, um- **instagram sketch voice:** Mm ... **Dave:** trying to be more efficient and effective, um, A, for myself, and B, so yeah, when I'm having these conversations, when I'm talking to potential customers or so on, um, I'm speaking from authority of been there, done that. Here's- things to consider. **Matt:** Yeah. It's very much a case of, um, you can get kind of Stockholm syndrome, um, when you, when you're stuck in one thing. Like, if you're just a Claude head, you know, you'll be like, "No, nothing else is as good. You know- Yeah ... this is just where I need to be." Um, but then, yeah, like if you start to increase your usage, then all of a sudden you realize you're spending so much. [01:05:00] Richard Webbe: Mm. You know, like all the big enterprise companies that have shut off their teams' usage to Claude 'cause they realized they were just [01:05:00] Richard Webbe: using too much. Like, you know, there's a, there's a level where when you actually go outside of that strange psychosis and you go, "All right, there, there is other things here," and you can still respect it. Like, I have massive respect for Claude as a thinking partner. Like, the way that they've structured that model- **Richard:** Mm ... **Matt:** it just works really well as like I have an idea. But the reason why it works well is 'cause it has these layers of sycophancy still in there. Like, it feels good to use it because it makes you feel good. It's like dopamine. So anyway, um, I don't know if that, if your question was fully, fully answered there, Jo. Like, did, did you have a follow-up or is that, is that all good? **Joanna:** I think it's just trying to work out, like with all these, like with the, um, the news and things like that where the people have got organizations with these massive amounts of agents, I think it's more, uh, sort of like what Dave said, you know. [01:06:00] Richard Webbe: Like yes, but not... You, you make sure that you have y- lots of [01:06:00] Richard Webbe: skills that you're giving them, and then maybe you don't actually need to have, you know, 30- Yeah ... 40 agents. You can actually run it with 15 **Dave:** It all comes back to context management and what are the- Mm. What, what, what's the context? What, what- Yeah what guidelines and, and rules are you given them to operate within? Yes. And are they doing some of the drives value? So in the big, bigger corporate context, that's why I think it's interesting that OpenAI is going with the get people on a business plan where they're in a, a environment where we can share these micro apps with one another in a secure way. Mm. And people are running on a subscription thing, which isn't... Yeah, that's interesting seeing where these things are going. But, um, the world is definitely still gung ho with lots of AI stuff happening- Mm ... and plenty of jokes and memes to go along with that. Let's wrap up with a bit of humor. Saw this one a few weeks ago, um, adopting Claude Speaking in my regular life. [01:07:00] Richard Webbe: "Did you do the dishes tonight?" "Yes, they're done." "Why are they still dirty?" "You're right to push back." I didn't actually do them. Um, but I saw, I saw this one. **instagram sketch voice:** Runs on business, but your business' business is stuck in the past, until now. Introducing Business AI for Business. Business AI for Business leverages the power of business AI for your business, transforming your business into AI business. With Business AI for Business, Business AI does your business in seconds, leaving your team to focus on the important business of your business, business. Now your business is free to run your business at scale, at speed, at AI. The future of business isn't coming, it's here. Business AI for Business, Business AI, Business Business AI, AI, AI. You're- **Richard:** I love it. [01:08:00] Richard Webbe: **Joanna:** Yeah. I, I think the other thing is that we just need to also remember that there's fashions. You know, like, fashions through clothing, uh, there's things that are... that we always stay with certain particular products that we, you know, that are required [01:08:00] Richard Webbe: and in context in our life, and then there's fashions. And so you have to sort of try and work out what's the fashion, you know? What's the spring fashion? What's the autumn fashion? What's, you know- Yeah ... and see whether it's a, a bit of hype or whether it's actually something that's gonna stick that is relevant for your, um, you know, in context for yourself. **Dave:** Yeah. I, I, so I saw, I saw another one, um, I, I lost the link, I didn't save it, where it was like someone coming in and it's like the CEO realizes tokens cost money, and it's like, "What does this token cost?" It's like, "We told you when we use tokens. Um, when you put me on a token improvement plan, the tokens use money." It's like, "But why do they cost more money than you?" Like, this whole thing at which point of the token spend is now becoming bigger than, you know, just hiring a human- I know ... and are you getting the same benefit? And it is, it is so much of it is that, that cognitive thinking, the engineering around it- Mm [01:09:00] Richard Webbe: and planning and coaching the AI and actually steering what you actually want because even, uh, Farm With Hermes, in Discord- I'm in a thread where it's given me some [01:09:00] Richard Webbe: feedback or something, and I reply directly below that message and say, "Actually, can you pull this file and do this for me?" And it goes, "Ah, Dave's asking me about this thing. I wonder what he means. It could mean this." I'm like, "Dude, the thread context is there. Why are you not reading from the thread context? It- here's the message again." It's like, "Oh, I see. Yes, you're right. I should have done that. I should have read that message and got that context. Of course you want that specific thing." I'm like, "Oh my **Richard:** goodness." Like, **Dave:** yeah, they are still dumb robots in many ways that, that have the appearance of intelligence. They're still in primary **Joanna:** school. You gotta, you, you can't just let them go loose. **Dave:** That's, that's like people say intelligence is a misnomer. They're not artificially intelligent, they're artificially good at regurgitating facts. Mm. **Joanna:** Mm. Yeah. [01:10:00] Richard Webbe: **Dave:** Yeah. All right. Well, thank you, Kym. We are well over the hour. We've run a long one. I blame you, Jo. Thank you very much. Good to have you. Thank you for, for filling in and representing the west in Mano's stead, as Mano's off doing cha- [01:10:00] Richard Webbe: business chamber, um, stuff in Kalgoorlie. I think that's what he's doing, and the business chamber has sponsored some free business coaching for people for AI stuff. So- **instagram sketch voice:** Yeah ... **Dave:** he's down at the library or something doing that, so good on him. **instagram sketch voice:** Yeah. **Dave:** Um, yeah, been a pleasure, everyone. Good chats as always. It's, uh, so much to get... catch up on and think about, and it's always good to hearing different perspectives. I learn every week, and, uh, it helps me, so. **Joanna:** Oh, thank you for having me on the AI Operators again. Uh, it's been great. I learn so much as well, and good to see you, Matt. Good to see you, Dave. **Richard:** Brilliant. Thanks **Dave:** for having me. All right. Thanks everyone. We'll, uh, play the bumper and see you all next week. Make sure to like and subscribe, and go to the, the website and register for the event, and do all the things. You know what to do, audience, and we'll see you next time.