To Claw or Not to Claw: Dave Ditches Google

[00:00] Dave Pengelley: Good morning, good afternoon. How are you all? **Matt:** Very good. **Marno:** Very, very good. **Dave:** Very, very good. Wow, two varies. That, that's outstanding. I mean, Matt, you are no longer by the beach as you were with us last week. [01:00] Dave Pengelley: **Matt:** No, no. Um, my filter isn't that good this [01:00] Dave Pengelley: week. Um, yeah, no, **Dave:** your lighting's much better though. We can actually see your face this week. **Matt:** Yeah, you can see the tan. **Dave:** Sure. Yeah, let's go with that. **Matt:** That's good. Adelaide was good. Good city. Far away. I think next time I'm gonna fly. **Dave:** Yeah. What, what was the drive like from your place? Because you are, you're halfway there compared to me in Sydney and much closer than, than Mano, although Kal, yeah. Still closer than Kal. **Matt:** Yeah, probably. Yeah. Uh, depending on the roads. So like, we actually went like south first. We went down to, towards Geelong, hit torque, and then did the ocean road basically up to Adelaide from there. **Dave:** Nice. **Matt:** And then on the way home, we just went straight through. So going through like, uh, like what was this towns one was called like ion and through dqu and stuff like that. But yeah. It was good. [02:00] Dave Pengelley: **Dave:** Nice, nice. We, we did a big road trip years ago and my kids were younger down all through Victoria. Uh, um, just crossed over into the South Australia border through wine country. Went to the, some caves there, um, which was awesome. And then tried to wind our way [02:00] Dave Pengelley: back along the ocean road and my kids got motion sick. We had to bail from the ocean Road cut, cut through wind land straight to Werribee. Because Tell **Matt:** what? **Dave:** And to be fair, I've **Matt:** been driving. Yeah. I was so grateful I was driving. 'cause like I would've been pretty bad. It too, **Dave:** right? Even driving it is so windy and you're up high and there's just something about it. I, I was like, okay kids, we'll do that. And I don't completely hate the idea either. Have you done, have you done the Great Ocean Road? **Marno:** Um, not necessarily. No. I, we went for the Adelaide trip from Adelaide to Kholi. Uh, we that last year, but it wasn't on the coast. We went through a bunch of. **Dave:** No. Well, the, yeah, I think the main highway through the like s and stuff is all just, **Marno:** well, it was part of the variety back, so it wasn't a straight road. It took about 10 days, stopped at random stations, um, flins range. **Dave:** Oh, nice. [03:00] Dave Pengelley: **Marno:** Um, went to the, the bridge, what do you call it, like the cliffs, whatever that's called. We went there for a little bit. Um, but it was good. It was great. It **Matt:** really imprinted on you that trip. **Marno:** It really did. Oh. The only thing that I remember is the Adelaide markets on the very first day. 'cause the food, it was great. I remember the bookstore, we got to buy chocolate for the first time and then the rest of the trip was like we tells. Bad sleep in the bloody swag. Um, first time camping as an adult, like there was so much like, okay, crap, how do we do this? Where do we go to the bathroom, wake up two o'clock in the morning, can't see anything, dunno where I am. Um, yeah, it was an experience. **Dave:** Tell you what, when I, when I went Dubai Chocolate, the first place I think of is Adelaide. **Marno:** Yeah. Hundred percent. **Dave:** Um, it is good. I don't know, I don't know if it's worth the price, like, like to buy chocolate. Have you had, have you had a mat? **Marno:** No. **Dave:** What, what do you reckon Mano **Marno:** worth it. I mean, I would not pay the $20 I paid for it in Adelaide. **Dave:** Yeah, [04:00] Dave Pengelley: **Marno:** you can now get it at bloody Woolies. There's like a, not actually the [04:00] Dave Pengelley: buy chocolate, but I could do buy Brownie. Yeah. Okay. That's like 10 bucks. That's better than the buy chocolate that's on there. And then to next level is the big Dawn's and perf the Big Dawn smokehouse. Okay. BDSM for short. They make a custom one there and that's delicious as well. **Dave:** Yeah. It's like pistachio with crispy, like a pistachio. **Marno:** Mm-hmm. **Dave:** Like paste the doors, you add these little crispy noodley things in as well. **Marno:** Yeah. **Dave:** And then it's just delicious chocolate. **Marno:** I love it. **Dave:** Yeah. It's, it's, it's pretty good. Um, there's a fruit market here that sells. You can buy the, you know, the one bar for $25, or you can buy a kit for like 50 where you can make four or five bars. You Okay. If you wanna go through the effort of melting chocolate and putting it together, like if you're a bit more kitcheny inclined than I'm, you go, okay. Well that's value for money. I'll, I'll do neither. I'm on keto at the moment. I'm trying to, trying to, **Matt:** I've got a good transition. Speaking of melting chocolate and putting it together. **Dave:** Yeah. [05:00] Dave Pengelley: **Matt:** Since I came back, like the intense backlog that I had, I had to sort [05:00] Dave Pengelley: of separate my stress from that with a little bit of like brain knowledge work. **Dave:** Mm-hmm. **Matt:** And using my, my custom agentic pathways. With my knowledge core, I've managed to completely melt down all of my content for Alex Ho Mosey. And for this podcast and for a few other people that I follow completely ingested into my core. And then through that melting pot creates some sort of intense synthesis of different really cool ideas. So I'm really, I'm actually so pre pleased and proud of that whole process that's like one little, like, I don't know, AI based win for me for the week. **Dave:** Nice. **Matt:** I'm really sure, **Marno:** I'm sure you using, using the hundred million dollar leads, was that one of the main one that you were working on, or was it **Matt:** all Yeah, yeah, yeah, yeah. So from his book Launch for Money Models mm-hmm. Um, you know, through that and got access to the, all the books and all of his other written content **Marno:** mm-hmm. [06:00] Dave Pengelley: **Matt:** Plus the like secret chapters and, and all the other things in the, yeah. Playbooks. So all of that is now fully [06:00] Dave Pengelley: ingested into my core now for just distillation use. And I even, I created an entire framework now. So that I can essentially use that as an interview platform to just discuss ideas and to break down and to decompose in from workflow stages into like full atomic level, which is fun. It's exciting. **Marno:** That's great. **Dave:** Wow. Wow. **Marno:** That's a product right there. That's a product. I, **Dave:** I just have Tropic Thunder in my head. Like you made Full Home Mosie never got full home Mossie. **Matt:** Not so good though, because like you've got in the past, Dave, with your personas. **Dave:** Yeah, **Matt:** I have now like perspectives, so like when I go just do some brain work in the core, I've got the homo perspective and then I've got the Jeffrey Huntley perspective. **Dave:** Yeah. [07:00] Dave Pengelley: **Matt:** And then a few of the other like big dev dudes that I follow. And yeah, it's cool getting like my own little flavor happening here through this melting pot. Love it. **Dave:** Nice. Nice. Um, yeah, we'll, we'll come back. I wanna talk a bit more about, you know, agentic uses and, and these sort of increasingly personalized agents and personalized personality agents and how that fits in with the future of your NEMO clause and open claws and nano claws and all the claws. Um, because I think they are tied together that generalist versus specializing type agentic stuff. Um, but, uh, yeah, my, my trip, my journey this week I subscribed to Claude finally. **Matt:** Oh my really? **Dave:** Finally. Well you guys, I appreciate you acting surprise 'cause you guys both know from the WhatsApp. Um, but for the, for the players at home that know that I've been invested heavily in antigravity up until now, uh, my, my journey as we've described, started off with chat t and chat t projects and then went to Antigravity and uh, then Google in the last week. [08:00] Dave Pengelley: They started messing around with all their pricing plans. Mm-hmm. [08:00] Dave Pengelley: Which, which freaked me out. I went from, you know, I'd waited till, like after the show Wednesday afternoon. All my quotas were made to reset. I was meant to be back to that five hour refresh window. And then immediately everything was back to like seven days. Like, then, like Thursday morning when I woke up and I'm like, are you kidding me? I barely touched this thing since I reset on Wednesday night. I wake up Thursday morning and my quotas are just trashed. Mm-hmm. What the heck? Uh, and there was a huge backlash and they reset some of 'em back to five hours, but not fully after that. And it's, it's, I think it's an evolving story, but for me it was enough to go, I have lost all confidence in this. And because Google's plan structure goes from that sort of pro plan, which is, you know, like 30 Australian dollars, 30, 35 to 400. Like, there's no middle ground. It's like you pay us $40 or you pay us $400. And I was like, I'm not paying you $400 when you've proven yourselves unreliable from what you're actually gonna gimme for my money. I, I can't trust that. I'm sorry, who else? **Matt:** They've been banning people too. [09:00] Dave Pengelley: Well, they did, and then they went, oh, we, we'll, we'll, we'll, we'll, we'll provide a path back now if you were using your, your Google credits for open core stuff, and we'll, we'll, you, you can come and ask for forgiveness and we'll let you back in. Now. **Matt:** It's crazy. **Dave:** Like, I paid for a service. Why, why do you care how I use my tokens? You've given me a token limit. I should be able to use those tokens wherever I like where we in a, some kind of gestapo AI state, um, with how like you will provide me a service, but only under very strict conditions of what I, and look, they can provide terms and conditions, but as far as you know, how you. Oh, you, you bought that Castrol GTX 10 oil. You could only put it in these sorts of vehicles, otherwise we'll take it off you and you can't use it anymore. It's like, it's ridiculous. I bought the oil, I bought the tokens, who cares which machines I put them in. **Marno:** Yeah. **Dave:** So I, I pulled the, pulled the trigger and went, I need to give this Claude thing and go now. [10:00] Dave Pengelley: Um, because I can't trust, I'm [10:00] Dave Pengelley: gonna have the credits. I need to do things and keep running my business and keep building projects in anti-gravity. **Matt:** I got a question. What, what made you go Claude, instead of just doing a, like another Codex subscription through, through chat t **Dave:** Uh, well, I, I did have Codex and I blitzed my Codex subscription completely and I've got my wife's one, but I've gotta get her to send me the two FA code and was a bit of, and I just was outta sync with that. And she was at work and I was like, oh, I'm just, I do something. Uh, and, you know, codex is good, but I also, you know, found for some of the like. Things it was doing, it was making mistakes or like getting stuff wrong. And I was like, I did like what I was seeing from some of the opus and sonnet outputs in anti-gravity. [11:00] Dave Pengelley: 'cause it gave you a taste of those models. Um, but then that was massively restricted now in the future. And I was like, ah, okay, let's, let's give entropic some money and let's finally go down that, that, that path. 'cause everyone's been raving about it for six months. And, [11:00] Dave Pengelley: you know, being in this AI space, I do have a bit of fomo and felt I really should, but I just didn't have enough reason and impetus to spend the money until the Google anti-Gravity. Uh, **Matt:** it's an interesting thing though, like, 'cause there there is a feel like the cord feels in a certain way. Mm-hmm. You know, even if you use their other models, like not just Opus, like, it actually genuinely feels different. It doesn't necessarily mean it's better. But you know, like with the knowledge work that I do, especially like with my core and stuff like that **Dave:** mm-hmm. **Matt:** I basically have to use Claude. Um, I haven't even attempted to use Gemini through that. And I currently don't pay open AI any money, so I haven't like tested Codex. Um, although I have heard really, really good things about 5.4 mini. **Dave:** Yeah. Well that just launched today. So in the headlines, um, open AI just launched Mini and Nano for 5.4. [12:00] Dave Pengelley: Yeah. Um, which I hope they're better than like Gemini Flash. Gemini Flash is okay for just doing bulk work and, you know, summarizing your emails and things like that. But when it came to actual strategy [12:00] Dave Pengelley: planning, even writing code. Like, and because I'm trying to do some complex projects and I'm halfway through trying to get these models like flash, even sonnet to an extent, they're just not strong enough to work out what's going on where I'm up to and make good choices. So they just, they're just hacking and slashing and go, oh, that didn't work, so let's burn that down and try something else. I'm like, no, this was mostly working. I needed you to fix this one thing. And now you're just like, because you don't have the bigger picture, you're just making guesses and just changing things willy-nilly and it's, it's killing me. So, yeah. And I think one of the issues I've been having with the, the product I'm building is I was switching between model to model to model and context. Like no one was owning the big picture between, between me, between Gemini Pro, between Codex, between, um, the little bit of Opus through the API through Antigravity. [13:00] Dave Pengelley: It was all over the shop. Uh, so I decided, you know what? Let's do that. And so on Friday afternoon after my deliberation, I, I, I grabbed some on Friday [13:00] Dave Pengelley: morning, I think, because I got a reset really quickly. Anyway, timelines on Don't Matter. Late last week I bought Claude code on just the, the pro plan, the cheap one, and was like blitzing my five hour quota, like every five hours except when I was sleeping. And then I'd wake up in the morning and I'd blitzer the five hours and have to wait for like two hours for the next thing, um, until sort of Monday when I used all my weekly quota as well. And I, yep, that took, that took, **Marno:** I can relate **Dave:** that took like three or four days. Um, but I think using Claude Code, and we can talk about this more in a little bit, but Claude Code in the terminal and just the way it manages things and the way that interface works and the way it plans. [14:00] Dave Pengelley: And I love subagent. That's what I've wanted from Gemini, from day one and Antigravity. Um. So, so yeah, so, so many, so many good things and I was like, okay, I, I need to upgrade this so I can keep working. And on five x at the moment. Um, [14:00] Dave Pengelley: I feel like I've got enough to just, just send anything I like. I'm not hitting limits yet. I'm sure you **Matt:** shouldn't, you shouldn't hit too many with, with five x. Um, and then especially with your Codex sub as well. **Dave:** Yeah. **Matt:** You know, like you are balancing between the two for me, like I'm on the 20 x, um, and I probably use about 30% of my weekly quota at the moment. **Dave:** Yeah, yeah. It's **Matt:** just above **Dave:** and, and given they're doing double the five hour during, you know, Australian daylight hours at the moment, that, that's pretty cool too. So, **Matt:** yeah. **Dave:** Um, yeah, I'm, I'm, I'm really happy. I'm, I'm running multiple terminal windows now. I haven't bolted it into A-A-I-D-E or anything like that, but, um. Code's good so far. Awesome. Mixing some, so mixing some opus, running subagent, getting it across my business os stuff. So all that kind of stuff that I was doing in antigravity, I've gotta do that. [15:00] Dave Pengelley: I've gotta do software development projects [15:00] Dave Pengelley: and other sort of things. Trying to work out, uh, how to make some side revenue to actually pay for all these AI subscriptions. Um, that's, it's like good, I've, I've gone to the five x plan, who's paying me to do that? So, uh, working out how to turn these things into revenue, which, you know, that, that's what I talk about as a business person. Like these agent and AI things are all about driving ROI. It's not toys for the sake of having toys. It's toys because they're gonna help you drive your business forward and progress things and make you money. And so I need to self apply that and go, cool, I've got this extra power now. How am I using it? And it's not just like, oh, I can drop the Google subscription. And just have this 'cause and, and, and this is where I'm caught again. I'm like, this is like buying Netflix. And then you go, oh, but I want that show. But I still want that show 'cause I'm like, oh, this still has VO and, and Nano Banana and all these other great things that are really good that I probably still need some access to. [16:00] Dave Pengelley: 'cause Philanthropic doesn't really do the image gen stuff, [16:00] Dave Pengelley: so I need something for that. But maybe Codex Open ai, I can probably do image gen through Codex. **Marno:** You could just use Open Router and then that way you get best of both worlds. Like I'm busy setting up my Ai os thanks to you David. Like giving that idea. **Dave:** Yeah. **Marno:** Now I've built it out where it's from. I'm actually using Discord instead of Telegram 'cause I just like the interface better. But from that I can check all my emails, all my calendar, my clicker, send my invoices, do everything. And then the goal is I'm now trying to learn how to do it. Where Claude Code will do, will be the workhorse and that'll spin off the subagent. But then I want to utilize Nemo tron for like normal stuff, like just checking emails. And then I want to use um, anti-Gravity, or not Antigravity, sorry, Gemini for net of banana and then open AI for Open Whisper or for for Whisper so I can actually talk to it. **Matt:** There's so many things there. You just said so many names. [17:00] Dave Pengelley: I feel like you need to have a, a, a whiteboard where you're like drawing these pieces out. **Marno:** That's what I'm getting at to create. 'cause it's, there's so much going on, but I, yeah, I don't wanna pay for 20 different subscriptions. So it's like, okay, how do I, how do I optimize this where NEMO is free, which is awesome. I a cloud code, which can do most of the heavy lifting, that'll do 80% of the work. But then, yeah, I still wanna do social media posts and then I need Nana Banana after that 'cause I like the administrator a bit better or I use Notebook L Lamb to do my presentations for me or my proposals even. Um, so bloody hell. It's been, it's been fun. But I neglected my wife and all duties for the past week because I had three days, or I had one day off last week. Tuesday, I spent that entire day building this, and then I had Wednesday all day work and then went to Perth, and then I got back Monday night. Yeah. Two nights ago, and then all of Monday night. And all of yesterday was just, again, spent building this out because I need to do a, a full rebrand. **Dave:** Mm-hmm. [18:00] Dave Pengelley: **Marno:** I'm like, if I wanna do a, like, I'm approaching it, like I'm starting a whole [18:00] Dave Pengelley: new business. If I'm wanna start a new business, how can I get it to do the heavy lifting? So, yeah, **Dave:** I hear **Marno:** you. I'll give it the name, it'll do the domain purchasing, the DNS, the Microsoft setup. Everything because holy crap, I'm not looking forward to it. Just so much boring work. So if I can get it to do it, if I can spend a week to automate a task that's gonna take me two hours, that's a win. **Dave:** Yeah, because, because as a solopreneur, I mean, uh, good morning, Joe who's joined us, Joe's fortunate to have a team. She that we had Joe on the show last week. She's, she's got people to help her with some of that sort of stuff. But, uh, if you are running a single shop, I get, I get it and I am, uh, spending, uh, a lot of my time trying to build my tool, the next generation, my AI os as a standalone tool that's not living in an IDE for those same reasons. I've got all these things I'm trying to achieve and I'm like, if I could just get the assistant ready to run and do it, then that's gonna help me move a lot faster. But building the thing is taking a bit of time to get it right. So that's. Where I'm at [19:00] Dave Pengelley: **Matt:** a hundred [19:00] Dave Pengelley: percent. Like I'm super curious Mano like, but you just settled this stuff. And I, I, I pride myself on something. Someone that can like, hear things and, and listen and learn things quite quickly. But it just sounded like noise. Can you run me through or run us through, like what do you actually have that's working right now that you're like, you're all like, this is sick. Like this is a win. I've got some stuff set up. **Dave:** Yeah. Because I'm not sure why you need Nemo chlor that set up by the sound of it **Marno:** purely just 'cause it'll be fun to play with it and it will reduce some of my token usage. Um, I also just wanna see what it's capable of because through, through the Nvidia Open Eye or further NVIDIA's open source models, there is image generation. [20:00] Dave Pengelley: There is not video generation yet, I don't believe. But I can do stuff to look at my GL location. I can help with surveillance footage. So like if I walk in front of the camera, it can respond to me talking to it like Jarvis, rather than using voice. It can look at my hand gestures and. Voice, whatever [20:00] Dave Pengelley: else. So I can interpret for a lot more things. So I wanna stop being into **Dave:** Big Brother early. **Marno:** Yeah, I wanna bring that in as soon as possible, just 'cause it seems like a lot of fun. But current MVP, what I've got working as of now, um, I actually just finished updating it this morning, but I now have, um, memory, I've got subagent set up there. And the way it works is on my phone I've got Discord, which is just the communications app similar to Slack or Telegram. Um, I've got a private channel in there and a private, um, bot that can have access to that one specific channel. So it's not gonna talk to anything else. And I'm connecting that to, um, open AI's whisper model, which means it'll understand my voice messages if I just send it a voice message instead, it'll transcribe it. [21:00] Dave Pengelley: But I can go into that channel, I can say, Hey, good morning, what's happening today? And then it will look through my calendar and my emails, what have I committed to, and making sure that if there's any emails where I said, Hey, I'll respond to you by Wednesday and [21:00] Dave Pengelley: I've not responded yet, then it'll notify me of that. Um, it can look at my task or my clickup that I used in my CRM, um, for looking at all my customers, where they're at, what the projects are going on. It can gimme updates on that and which ones I should be prioritizing. Um, and then by the end of today, it's hopefully talking to Henry, my accounting platform. So when I finish working for client, rather than me having to open up the app, click new invoice or new client, create the client and send it, it takes five minutes. But why would I wanna do that if I can just go, Hey, send an invoice to such and such for this amount, this is the line item, put my phone down and the work is done. **Matt:** Yeah. **Marno:** It's like, so it's made everything easier, [22:00] Dave Pengelley: **Matt:** the, the deterministic versus non-deterministic workflows again. Yeah. That's awesome. So good. And speaking of Joe, who's, um, once again here, thanks for reminding me Dave. I didn't have the chat open. She's re uh, recently sent us a, a video in a, in a group that we're in to do with Alex Hor talking about breaking [22:00] Dave Pengelley: down AI workflows, like the full decomposition of the workflow and that being the value, you know, the ability to see a system, not just the hype of the tool for example, but the idea of like, you know, instead of having to hire a marketing team member, you look at what that marketing team member does and each of those tasks that they do, you look at the individual single atomic steps that they do to achieve that task. **Marno:** Yeah. **Matt:** And then that's how you create your workflows and your AI agents, which I think is amazing. So thank you so much for sharing that video, Joe, that went straight into my knowledge core and I'm already using it. **Marno:** That's awesome man. I think it's such a good approach and that's something that you've been talking about for a while, Matt, is your. [23:00] Dave Pengelley: Versus output, like what inputs are you bringing in? And just taking that same methodology when you're working with a client. Not what is this person doing and what do they hate doing, but okay, what's the actual task, what's the role? And then, okay, sure, we can't automate 100% of it, but if we can [23:00] Dave Pengelley: make you a job 80% easier or 40% easier, then you can do other things that are higher leverage or higher value to the business. **Matt:** Yeah. **Marno:** Um, it's so much fun. I'm curious to see if we can, even just going off it around there, like what if we could we just build a custom GPT or custom thing where you would give it what the workflow is and then it has that, um, methodology trained into it? So for example, you say, I want to hire an editor. What, how much of this role can be done by ai? And then it just pulls from like a knowledge core of what's possible and it gives you some suggestions of what's a, like a quick win that's gonna be five minutes and what's gonna completely change the way that you see this position. **Matt:** Yeah. So that comes down to like, I forget what the term is. [24:00] Dave Pengelley: It's, it's specialist area knowledge. Mm. A good example is, is somebody, you know, you're organizing logistics and you know that at certain sites, the site manager will not deal with certain drivers. So if you [24:00] Dave Pengelley: organizing route and stuff like that, you have to make sure that those, those drivers don't go that site, for example. **Marno:** Great approach. **Matt:** Yeah. That, that kind of like specialist knowledge when you're breaking down and analyzing workflows is not something that AI can do by itself. Like that's very much where the human still is there. Mm-hmm. You know, curating that for them. But it's funny you say like, can you turn it into a custom GPT or whatever? I, I actually fully built it into a framework into my core already. **Marno:** Of course you did. Of course you did. **Matt:** And I've got, I've got two panels open, like waiting for me, like ready to interview me about like stuff that I do to then further distill everything. Mm-hmm. So. So many plates are spinning, dude, like, **Marno:** oh yes, **Matt:** I have, I have the, the concept of it being like a lead magnet where, you know, you go to a website or you, you have ads on a platform and somebody just wants to understand their, their business processes better. [25:00] Dave Pengelley: 'cause like, how many people, even like business owners, even operations managers at companies, how many people actually genuinely from like a super, super fine, you know, high [25:00] Dave Pengelley: resolution perspective can under understand and identify their entire workflow from an atomic perspective? Basically None. Right? Yeah. Like, I was on a call this morning literally with two, um, business stakeholders, like, you know, like team lead and, and mm-hmm. Um, assistant for example. And they were arguing about their own workflows being like, no, it doesn't work like that. I was like, yeah, look, me, let me screen share. I'll show you. **Dave:** Oh, you, you, you take for granted how much you do without even realizing you're doing it. And how many steps are involved in some basic tasks. Like when I was launching my chart reporter. Beta a few weeks ago, and I had to get it ready to actually put out as a beta. I had to go through the whole user experience of what does sign up look like? What does registering for the beta look like? What does that look like? What email do they get? What do they type in here? What works, what doesn't? And that last mile of going through every single little step of that process was like hours of work to actually go through each little thing. You, yeah, it's all right. [26:00] Dave Pengelley: We'll add a login page, we'll do that, or go to, but then when you actually go through the step by step and [26:00] Dave Pengelley: execute every single step, there's more in it than you realize **Marno:** always **Dave:** just off the top of your head because we just, you know, um, what, what people underestimate what they can do in the short term, but, um, overestimate what they can do in the short term, but underestimate what they can do in the long term. Yeah. So I mean, most people, some people are probably really good at their time prediction. Me. No. I always think, yeah, that'll take me five minutes and 20 minutes later I'm like, I'm coming. I'm coming. **Marno:** So either of you use tango at all, or tango or scribe. **Matt:** No, but I know you've told me about that before. I can't **Marno:** remember what it's, it's been not even a lead magnet, it's just, I yell it from the rooftop. Whenever I meet a business, I'm like, Hey, wanna make life easier for both of us? Just use tango. Hit record. Do the workflow. You don't have to think about what's going on it every time you click. It used to be a built in feature in Windows, but then they deprecated it last year. Um, and what it does is it's just an extension on your browser or on your computer, and you just hit record. [27:00] Dave Pengelley: It looks at your workflow and every time you click, it'll highlight where you're [27:00] Dave Pengelley: clicking and break down what exactly you're doing. Yeah, cool. So you've got a visual documentation following you as well as a description of what's going on. Um, and one, now it can do a lot more than that, but just that basic line is like that makes sure that you're not missing anything. 'cause if I'm creating documentation on how to create a new, per a new user in Microsoft, I'm gonna miss a few steps because I've done it 20 million times. I'm just gonna click through the buttons. But if I have tango following me, then it's not gonna miss anything because every single click is documented and recorded. And then I'll use that as like, Hey, you wanna get me to automate something, use tango, record the workflow and send that to me. That's cool. And then I've got step by step exactly which buttons need to be clicked. So there's no room there, there's no errors at all, because it's **Dave:** doing, have you heard of Nobi? **Marno:** No. [28:00] Dave Pengelley: **Dave:** So I, I've heard about Nobi 'cause Richard, who couldn't join us today, I forgot to say, up off the top of the show, uh, Richard had a, had a conflicting meeting with a client and had to, you know, do client things, which, uh, is important, uh, [28:00] Dave Pengelley: as, as business people. So he's, uh, he's doing that. Uh, but he put me on, this is an Australian product from someone that he knows who's created Nobi, which if I bring it up, sounds like it's very similar to what you're describing, um, where you record what you're doing and, and not just screen processes, but physical processes like for trade and, and stuff. **Marno:** Trade **Dave:** where, where you record what the process is and then it will. Make the SOP and then you can, you know, output it with a QR code and people can see what that is. So then this is a, an Aussie product that does that. So if, um, people are looking for that kind of record a process and get it documented, um, check out Nobi not a sponsor yet. **Marno:** That is great. That's so much smarter. 'cause yeah, the Tango's limitation and scribe as well. It's all just on computer [29:00] Dave Pengelley: **Dave:** desktop. Yeah, so, so this, this is, I think nobody you can run on your phone and just record the process be, and then it'll process through the thing and help generate out the SOP. So I haven't used it myself personally yet, but, uh, that's on this topic I [29:00] Dave Pengelley: thought relevant. Um, Richard was here. I'm sure you could tell us a lot more about it. **Marno:** Noby that, **Matt:** you know, what else is fun though? Having your agents actually do that sort of thing for you? I experienced this literally last night, and this is something that. One of our, um, one of our shared friends as well who, um, wink Wink might be on the show in the future, was he, he basically showed me how to reverse engineer APIs when you don't have API documentation. **Marno:** What? **Matt:** Yeah. Well that's sink in. So like if you have, if you've got a, a tool that does not provide you any sort of like, developer docs for, here's the post request for like sending and creating a record for whatever, you can literally use your agents through browser automation to scrape that, those, and find them. [30:00] Dave Pengelley: And I did it last night. Like we've got, we've got an entire like, backend process from admin level token [30:00] Dave Pengelley: and the whole public, like API. That's stuff to do with, you know, if somebody goes to their website, goes to the tool and clicks the things. All of that is like fully ingested now into this egen system that will become tools for a voice AI system. So **Dave:** that makes that, that makes sense, right? Because like if I build a site and I'm pushing up to Versal, I don't just talk straight to super base or whatever backend, I, you know, on vercel it creates its own little web API endpoint in the website, and then the forms submit into that, which then does the backend lifting. But if you can get the website to decode what's go, just going through to that vercel level endpoint, what that form is submitting to that, then you could then, if you are authorized, if you need all whatever, use that token to push it all the way through a [31:00] Dave Pengelley: **Matt:** hundred percent. And that's where, that's where a lot of security concerns come from because that's where, like when you guys are, are building your apps and building your APIs with your agents, it needs to be that level of [31:00] Dave Pengelley: protection for public APIs **Dave:** and, and, and on the versal stuff, that's why you have a secret environment variable in the backend that. The website doesn't publish out, but when you submit it, it attaches that in in the backend and then pushes that through to your final API. **Matt:** Mm. **Dave:** And so, yeah, **Matt:** it's super cool and also like, Hey Sean, like, um, you're absolutely right about the N eight N internal APIs as well. Um, you know, like something else just mind blowing that I did recently was connect through that kind of reverse engineering to the NNN internal APIs as a skill into my terminal agents. So I basically go to my agent and say, can you just check out what's going on and logs, it'll check like the whole architecture and be like, why did you do it that way? Um, yeah, it's insane. I **Dave:** mean that, that, that's why people are saying open claw is doing so much stuff for them, right? 'cause they just point open claw at a problem and it just iterates itself and, and burns tokens until it works it out. [32:00] Dave Pengelley: Um, because it just runs on loops and sets the itron drops up and, and [32:00] Dave Pengelley: kind of ralphs it to a certain extent. And that's, oh, just works stuff out. But I mean all, all, all that stuff comes back to then we were talking about heaps times is is things like context and understanding the context and what these things do with context **Marno:** mm-hmm. **Dave:** Is wild. Like I have my little workflows and things that I set up in Antigravity and I haven't really been using them heavily, but this morning I was like, I'm gonna use my Dawn workflow, which is gonna go check my inbox and calendar and gimme a summary of the news and things. 'cause we've got the podcast today. It's probably a good, good idea to run that. It's been a little while. And so I ran an incl code, which I'd never done before, and it was trying to use MCP, which was set up in anti-gravity. I was like, no, no, no. All those, all those endpoints for checking my mail. They ran out and they webhooks just work it out and, okay, cool the web hooks, bang, bang, bang, bang. [33:00] Dave Pengelley: And did it all through curl requests, which was awesome. Um, but one of the things it did is it looked through my inbox and my calendar and said, you've got the AI Operators podcast today. And by the way, I noticed that Restream has told you your LinkedIn connection is disconnected and you need to reauth your [33:00] Dave Pengelley: LinkedIn. And we've never streamed to my LinkedIn before. We've only ever gone to YouTube. But I did have LinkedIn connected in Restream when I was first seeing it up, just going, oh, I might as well connect all these services. I dunno what I'm gonna use. And so today I was like, you know what? Let's push it out to LinkedIn. So if you're joining us on LinkedIn, if you're watching this on LinkedIn for the first time, welcome, welcome to watching the show. Please drop a comment and tell your friends. But that was the context. It like goes, oh, you're doing a show and you use restream and there's an email from Restream letting you know there's an error with your service. And they highlighted, this is like a critical update. Make sure that you fix your LinkedIn or before you do the podcast. I was like, this is, this is wild. And so **Matt:** Yep. Feel the a GI. **Dave:** Yeah, it's, it's all that, that that, that stuff coming. And, um, I'm, I'm gonna drop a video up now that I, so I haven't watched this video, so I'm watching this. [34:00] Dave Pengelley: This is a blind watch, but it's from a friend of the show. Uh, one of our friends, Dominic, uh, who has just put this up on [34:00] Dave Pengelley: LinkedIn 58 minutes ago, so just before we went live, uh, and I've watched it yet, but it sounds interesting, so we're gonna watch it together. Let me share this tab. **Matt:** Feels like at school when they push the TV in? **Dave:** Yeah. **Matt:** Oh, best time ever. **Dave:** Um, we, we had, we had in, uh, in high school, the library also had these little mini TVs, these little like, uh, 12 inch, 10 inch TVs that had a builtin VHS player in the tv. Like it was an all in one unit. Oh yeah. And so you'd go to like, go to the library and get the thing and you'd, you'd carry this little TV with a little vh. Anyway. **Matt:** Awesome. **Dave:** Um, this new, so this is from our friend, Dominic friend of the show. This new cowork feature is pretty crazy. Dispatch allows you to talk to your Claude Cowork running on your computer while you're on the move. So, oh, this is that remote access thing. So this is why you are doing mine with, with Discord. [35:00] Dave Pengelley: But now Claude's built it in. I've just run a whole SEO audit create strategy. Now I'm waiting for the implementation all while on my lunchtime walk. So there's a video here. Let's watch the video. A pitcher says a thousand words. Right? **Matt:** Interesting. Unreal. So, so **Dave:** have you, have you guys played with cowork yet? **Matt:** Yeah, I have, uh, like cowork is literally just the GUI of court code. So as always, um, that, that looks like you must have a, like a tunnel session or something to your desktop. So like your computer has to be running and it's dispatching to that. Well, you've [36:00] Dave Pengelley: **Dave:** got the [36:00] Dave Pengelley: Claude, you've got the Claude app Open, which has the cowork tool built in. And if you're authenticated in on the cloud and you're authenticated in, like if you've got the mobile app, you're, you're both in the same cloud account. And so I, I see no reason why it can't go. Oh, like, like a WhatsApp, like any other app. **Matt:** Yeah. **Dave:** Central login bang bang. Um, to then fire off commands on your desktop. **Matt:** So that work for those that like have persistent running desktops, like, you know, for me when I shut my laptop at nighttime, I can't really keep that running. Um, curious though, like you can kind of see the ecosystem coming together. 'cause like that's not something new. This, **Dave:** this is then taken on open claw. That's what this is. **Marno:** I don't see the benefit like this is, might to be polarizing, but I I've yet to see the benefit of Epic. Like, I spent mostly yesterday setting up Alaw, um, followed via the NEMO law setup. Like that's what sparked the idea. **Dave:** Mm-hmm. [37:00] Dave Pengelley: **Marno:** And after setting it up, I'm like, well, how [37:00] Dave Pengelley: is this any different from Core Code, just cloud code, some skills. MCP server connected to Discord, I can talk to it. Um, it has subagent, it has all the capabilities. There's no real difference. It's just I feel a lot safer using Code Code than I do open clock for various reasons. **Matt:** You know what you're discovering, and this is something I've said since the beginning of this show, you're discovering primitives and fundamentals. **Marno:** Yeah. Yeah. **Matt:** And you're seeing comparisons and Yeah. Like it, you, you're pattern matching against the, like, there's literally only four things that exist. It's, it's models. **Marno:** Mm-hmm. **Matt:** Context prompts and tools, like that's all it is. And tools are literally just. Anything. It could be API requests, it could be just markdown files, you know, the, then, then you've got the models, which are the things that they constantly iterate on. The context is, you know, ever involving. And that's where the engineering comes in. [38:00] Dave Pengelley: And the, the prompting at the end of the day is just [38:00] Dave Pengelley: engineering of that context. But what you're seeing is exactly what everyone's been saying for a long time. Like what is the difference? But to, to kind of like segue a little bit and to use your questions that you gave us yesterday. It comes down to what you want and what you actually need. 'cause you know, if you think about what do you need as an individual, it doesn't matter like which tool you use to do it. **Marno:** Yeah. **Matt:** But one might make it easier than the other. So like I can give you like a, an instant apples to oranges comparison between Claude Code and something like Open Claw or this NEMO code is the persistence. Like the, the availability and the, the on time and the, you know, ability to trigger it. So with dispatch, you know, if you have, if you don't have it running on your computer, it's not running. **Marno:** Yeah. **Matt:** Whereas if you have OpenCL running on like a, a cloud server like that, you host through VPS or whatever, it's always there. [39:00] Dave Pengelley: And [39:00] Dave Pengelley: then that's where you have your persistent workflows. Like when people use the term heartbeat or chron, plus being able to trigger it remotely. You know, like my cord code on my computer, I can't trigger it remotely. **Dave:** It, that's, it's a, it comes back, it's a, it's another harness. It's another way of interacting with the technology. It's all, it's almost, this is a, this abstraction doesn't quite fit, but it's like, what's the difference between a V eight and a Monro? Right. Um. So one is the backend engine. They both have v eights in them. Mm-hmm. But one of 'em is actually pre-configured and accessible with and usable outta the box with wheels and steering and, and chassis and so on. Yeah. Um, so Claude code is more just the straight V eight and you've gotta do all the configuration. You've gotta build your own chassis, you gotta put your own steering wheels on it. You've gotta attach all the things. You get a registered roadworthy, whereas you buy them in the V eight Monro and it comes outta the box ready to drive. [40:00] Dave Pengelley: And I think open claw, what it did is it attached a whole bunch of stuff onto whichever [40:00] Dave Pengelley: model you happen to use. Mm-hmm. And initially I think it was using the, you know, the old clawed bot. It was, it was leveraging off, off clawed more than anything else because of the plans and the tokens and the usage and the intelligence. Uh, yeah. So that's the, that, I think that's, that's the big difference. And so yes, you can build very similar functionality in anything. My, my process this morning, your process that you're talking about, we, it'll go check your inbox and do this and do that and do that. That's why I'm saying why do you need to add NEMO claw to that mix? It feels like you're already doing all the things that a, a claw bot would need to do. And I, I argue with my build team all the time, my build team being code around what I'm building as far as my own, um, multi-agent hierarchical harness going. Should I just be using open claws? Like, like am I over engineering? [41:00] Dave Pengelley: Like you're doing all this sort of stuff anyway with your pedantic agents that you're, you're, you're building, it does this, it has access to that, it's doing that, but you've also built these other structures and things which are unique and special to you. And I'm like, okay, thank you. I just, [41:00] Dave Pengelley: you just gotta work through the FOMO sometimes and go, why do I need that particular thing Now? We've said the word nemo claw way too many times for not explaining what it is. This is news that came out of the Nvidia conference yesterday. This is hot off the presses. **Marno:** Yep. **Dave:** Um, there, there he is. Mr. Shiny jacket man in a 3D rendered version of his shiny jacket for you, Matt? Um, I watched the 20 minute keynote from Jensen Wang. Is that his name? Yesterday? **Marno:** Yeah. **Dave:** And he's saying Open Claw is revolutionary. It's like the new operating system for working with agents. And every business needs to not just have like, what's their AI strategy, but what's their claw strategy? What are they doing as far as enabling people with these intelligent tools with hands? [42:00] Dave Pengelley: That's what we've been saying from day one, right? The claws are the hands, uh, what is that? And the problem with enterprise is the security layer, which we've also been talking about is a problem with, with open claw. And so now they've built this new layer called Open Shell, which [42:00] Dave Pengelley: is a, you know, a protective layer, a shell around, uh, open claw. And, uh, a big part of it is then, you know, trying to sell Nvidia hardware and NVIDIA models to offer that other thing we talk about a lot, which is sovereign models. And so rather than sending everything to these third party companies, buy some Nvidia hardware, run your own models, give all your Staff NEMO clause, uh, running safe, secure, sovereign data through your own data centers, and giving everyone these claw tools with your government policy and f your governance and policy frameworks, which is what the Open Shell is managing and, and attaching those guardrails onto everything that goes in route of open claw. **Marno:** And would you say that without extra steps would just be the perplexity computer where you don't have to buy the hardware, you just give up the data, so, and just use Perplexity computer instead? **Dave:** I, I haven't looked into Perplexity computer enough. You seem like the expert, so, um **Marno:** Oh, not the expert. Yeah, I just saw Joe asked about it. [43:00] Dave Pengelley: I've looked into it, but after the effort I've put into this broad [43:00] Dave Pengelley: code workflow, it looks to be very much the same approach. Except, um, I'll just be using anti-gravity. **Dave:** I, I think there are people that are like hosting these, these 24 7 semi-autonomous robot things, right? So Perplexity is getting in on it. Um, Kimmi, we're running Kimmi Claw where I think you Kimmi claw license or on Kimmi in the cloud. **Matt:** Yep. Minimax has one. **Dave:** So clause's become the generic term now for a memory based CR job. Ai **Matt:** yeah. **Dave:** Assistant that has access to third party tools. **Matt:** It's funny, like just to use Manos line before, like I look at NEMO law and I go, this is nothing different to what I've been building in railway. Like, it's literally the same security abstraction, you know, making it simple for people. [44:00] Dave Pengelley: **Dave:** Yeah. And. I think Docker, Docker partnering with Nano Claw and they're, they've, they've done a thing to make it easy for you to spin up nano clauses in [44:00] Dave Pengelley: their own secure little docker shells, which again, my PI agents are, I'm running like a hypervisor type thing where they, they run up in their own specific little shell for, um, sort of a bit of protection security. So everyone's approaching this in similar, different manners. Uh, and it's gonna be interesting to see which ones get to market, who their ICP is, how they're different from one another who wins The problem with a lot of these agent shell things like NVIDIA's got a monetization strategy. They wanna sell hardware to run models. If I release my own harness, kinda like open claw, how do I make money off that unless I'm selling implementation services to help people implement it because it's such an amazing piece of technology, which it is. Uh, then, but we don't sell software licenses. You don't walk into Harvey Norman and buy a box of software off the shelf anymore like you used to. That's how you used to sell apps. Now. You either have a SaaS model or you give them away, seems to be the two main options. [45:00] Dave Pengelley: **Matt:** I got a, um, I got a good way of, of sort of approaching this. [45:00] Dave Pengelley: So, um, I alluded to minor's questions that he asked us yesterday that, um, I was gonna bring up on this episode, which to say them really quickly, like as if they're small things, was who do you serve? What problem do you own? And what have you learned or unlearnt about yourself in that process? Um, with what you're saying, Dave, like how do you monetize the thing? What is the thing that you're monetizing it? Is it the software or is it the problem that you're solving? And you know, like there's a really good line that I've got here in my notes, um, which was, oh, where is it? It was so good. It hit really hard. It was something like, you know, when you, um, when you first. Yeah, this is it. So when I was looking through the, the a hundred million dollars stuff, you know, the first question isn't what should I build? It's who is this for? **Dave:** Yeah. [46:00] Dave Pengelley: **Matt:** So, and that comes down to like what [46:00] Dave Pengelley: their issues are. So yeah, like the, the point I guess that I'm trying to make here is there's so many options and tools and stuff, like, there's gonna be constant, it's literally gonna be like hockey stick exponential with the amount of claws that are coming out and environments and harnesses and all the stuff, but the problems don't necessarily change. Yeah. So **Dave:** yeah, **Matt:** how we're solving them **Dave:** and, and the whole nicheing issue, right? The, I said to my Claude code after sort of finishing off some functionality today, I forked off another chat and said, um, just help me double check who is my ICP? And it just went through, what did you have the requirements? And did its own little research and went, okay, based on this. And this versus these, these competitors in the market and what they do versus what you do versus this and that. Here's your ICP. It's, you know, small businesses, five to 50 people that understand hierarchies and structures and this and that. Mm-hmm. Okay, cool. Um, but all these things, it comes down to knowing who you're serving and what the customer is. [47:00] Dave Pengelley: Right. Like I mentioned monetization and I'm looking up doing some, um, [47:00] Dave Pengelley: some quick web flipping, web design flipping, and it was niching down and going, yeah. If they've got a blog, not my target audience, if they've got a WordPress, not my target audience, because WordPress sites, um, as you would know Mano, um, from from your work are not simple things. Those sorts of people have advanced plugins, they've got lots of stuff on. They, like the CMS, they go in, they tweak their copy every second day. **Marno:** Mm-hmm. **Dave:** Um, if it's, if it's one of those sites, I don't want to touch it. If it's a Wix site that looks like it hasn't been updated in three years. That feels like something that you could sell them a quick and, uh, quick upgrade onto a versal next JS site with no monthly hosting fees. [48:00] Dave Pengelley: **Marno:** Exactly. Yeah. I mean, it, it's so, so easy. The, the website flipping is really, really fun. I've got, I'm actually, there's, last week has been pivotal, so, um, a quick recap was realize that I would get paid significantly less if I get to teach people rather than me sitting in an office and doing nothing. Um, [48:00] Dave Pengelley: by nothing, I mean like literally just sitting in front of the computer typing away the whole time. Mm-hmm. Although I enjoy it, I feel a lot more fulfillment, um, and joy when I'm teaching for people. **Dave:** Mm-hmm. **Marno:** And I actually chose to do a job that paid me 500 bucks for the day as opposed to a job that paid me $1,500 for the exact same day. **Matt:** Mm. **Marno:** Um, and I wasn't even sad about the money. I was like, I'm just happy I get to do this and this is awesome. **Dave:** Nice. **Marno:** So realizing that that's where I shine and that's what I'm naturally drawn to. Completely pivoting the business and going into just education. So I'm gonna find my niche by working backwards. I'm gonna look at associations that I care about and work for. Um, I wanna look at which businesses have the most economical impact and then how do I serve them to then improve the economical impact and output of Australia, which would be awesome 'cause that's me building my intrinsic why, my intrinsic motivation. [49:00] Dave Pengelley: But as a result of that is I found a bloke who's really, really good, um, [49:00] Dave Pengelley: in networking. He loves it, loves ai, she and web design. And he's just gotten lost in it. Like the last week. He's keeps running out of tokens. He's just loving it. So I'm gonna help him. We're gonna launch a separate business called LY Web Design, which is just gonna do web design. Um, and then help him in supporting that. And then I'm going fully into education. And after making that decision, a competitor in town, or not really a competitor, but the local IT company **Dave:** mm-hmm. **Marno:** Showed me that people are getting confused between our business or my business and his business because the naming is very much the same. Um, I, in hindsight, maybe I was a bit arrogant thinking that people won't see the difference or people would see the difference because for me it's two separate disciplines, AI and it completely difficult. Ah, but I did not consider the end user. And they'll probably see it as like, oh, it's computers, which means you can fix my printer and my con and my phone. [50:00] Dave Pengelley: So, um, with that, I had a conversation with him [50:00] Dave Pengelley: yesterday and the result of that meeting was I need to do a rebrand completely. So completely move away from Kaku ai. **Dave:** Oh wow. **Marno:** Find a new business, buy a new domain, build a new website, set up Microsoft licensing. Um, it's gonna suck, but. With all that. I'm taking this as an opportunity to go hard into the education side and actually step away from the web design completely and just run that as a separate little business. Nice. That's why I asked you guys those questions. 'cause I had that meeting and I was like, crap. Like this is an opportunity for me to essentially start a new business. So what is it gonna stand for? Why am I doing it, and what am I gonna say no to? And I'm just gonna not do anything that isn't my education. Any jobs that I get. That's automation. Hey Matt, I'll pay you a pretty penny. Can you do this? Or Dave, can you do this? Um, I just don't wanna be touching it. I'll orchestrate it or do the, what do you call it? The triaging of it. **Dave:** Yeah. **Marno:** But I don't wanna be building anymore. Yep. I'm done. [51:00] Dave Pengelley: **Dave:** Yep. I, I, as someone who's recently going, or still [51:00] Dave Pengelley: in, like on the tail end of a rebrand from Valid Agenda to [[Syllogism|Syllogism]], now that I'm working with Richard. Mm-hmm. I completely understand. So I, I'd be keen to chat offline around what that looks like and feels like. Um, sure. Because, uh, we won't, we won't bore the, uh, the, the, the punters out there, uh, with sort of a bit of a inside baseball on business branding and stuff. 'cause that's not what this show is. This is not the business branding show. This is AI operators. But, uh, and how using AI in order to execute a new business like that, that is interesting. And it sounds like you're building all your automate the whole thing. Touch, yeah. Hundred percent. Working out where, but where do you need a human in the loop? I don't think it's safe to fully hands off everything at this point. Um, there's always gonna be little bits and pieces where you need a bit of human in the loop. And I think where those touch points are and how, how comfortable you are giving, uh, your AI credit card details and the ability to make payments. **Marno:** I think it's just making the decisions that that's my, at least my approach to it is AI can do all the thinking. [52:00] Dave Pengelley: I have [52:00] Dave Pengelley: to make the decisions because I want it to do the work. I'm the one that has the final say off to say, is this correct? From something as small as anti-gravity, let's explore this directory. Do I allow it outside of that specific folder structure that I gave it to? Yes or no? Or do you send this email to this client stipulating this? Like, I wanna still have the final say off on all of it, but it just allows me to not have to do the actual manual workload because I can just talk to it, give it my idea, it can formulate it for me and make it look pretty and I can accept. So I'm still the originator, but then it executes it and I, I just implement it. **Dave:** You just gotta work out over time, which of the jobs do you really need to do that? Because if you become the bottleneck, then the whole thing breaks down. And I dunno if you guys ever put in, um, smart doorbells and get the, the, the camera alerts on your phone. So the first week or so, you're like, Ooh, it detected motion. [53:00] Dave Pengelley: Ooh, it detected motion by the, by, by by the second week. You're like, ignore, ignore, [53:00] Dave Pengelley: ignore, ignore, ignore. Because this is like, ah, a leaf blew in front of the camera and it becomes spammed to the point that you stop paying attention to it. **Marno:** Mm-hmm. **Dave:** And I find the approval process with these AI and things, if there's easy tasks that should auto approve, you should build that in pretty quickly so that way you don't start ignoring the big decisions and just auto approving the big ones as well. Fair **Marno:** point. **Dave:** Because you get sick of it and you go, I just approve everything. Swipe, swipe. Yeah. Um, so yeah, I mean, that, that'd be interesting to see how, how you evolve that, because I'm, I'm a shocker for auto approve stuff with my ais. **Marno:** Well, how's the capabilities? I mean, Matt was a big proponent of this, um, a couple weeks ago, and I, it still repeats in my brain to today. You asked him, how long would it take you to write this amount of code? And he thought about it for a few seconds and he said, I just wouldn't, why wouldn't, and I was like, shit like that, that shift in. I have this, these voices in my head, especially this one friend of mine that's incredibly intelligent. [54:00] Dave Pengelley: Like yeah, he is just up there and [54:00] Dave Pengelley: his approach to anything that's developing is incredibly just meticulous. And he has to see every little syntax, everything. Um, and I always think like, no, no. What would, what would this guy say? What would Nick say basically? And then now it's, oh wait, I, it doesn't really matter as long. If I, if it's secure, the output's there and um, I have control over it, then I'm pretty happy. Like I feel more comfortable to have the autonomous there rather than overseeing and understanding all of it, because I don't benefit from understanding every line of code anymore. **Matt:** No, it comes kinda, it's a good point to bring up validation again and like, you know, order validation and just everything is a loop. So, you know, with that, that concept of you, you just said like, it doesn't matter as long as it's secure and it does what it needs to. How do you know what secure is and how do you know what it needs to do? I'll **Dave:** figure that. **Marno:** Yeah, [55:00] Dave Pengelley: **Matt:** yeah. Like the, that, that sort of self validation is so important, and that's the reason why I said that now, [55:00] Dave Pengelley: because it's not so much about just mindlessly trusting the model that it knows what it's doing. It's me knowing that, you know, it's a slot car and I've set up the track, it can only follow that track. **Dave:** Yeah. **Matt:** You know, I can't go anywhere else, **Dave:** but if you're writing random code that does random things not, you know what I mean? Uh, then and you're not reading all the code 'cause there's thousands of lines of it and you don't really understand the language Anyway. That's where I think having the secondary models, like you talked about the fact I've got Codex, I probably would want to keep that because after Opus builds something or sonnet, then I can hand that out to the Codex CLI stay in the Claude code environment, but say, can you just palm that out and just get a third point party point of view. [56:00] Dave Pengelley: Joanne, Joanna was talking Oh yeah. I think Joe was talking about doing that last week. Or maybe Amber. Amber was talking about doing that last week with her agents, how she. Outsources to a third party agent to review things that she's doing with her little agent workforce. And so I found that when I was in Gemini with Antigravity, it was building all this stuff [56:00] Dave Pengelley: for me. And then I started that process where I integrated a skill that had Codex CLI, and it could pa pa it out, and Codex came back and went, no. And it's like, Ooh, ooh, codex. Codex really ripped into that. And, and so having those, those third party points of view, uh, should help close those gaps. **Matt:** That's still non-deterministic though, like **Dave:** that's It is, it is. But unless you've, you've got a pro software engineered dev that will read through everything. Uh, and, and to your point, Matt, that's when you want the deterministic of evals and tests, right? Where you got your test functions and, and you give it the inputs and then you see what the output comes out, and you do a whole bunch of like pen testing and automated pen testing stuff. [57:00] Dave Pengelley: **Matt:** Yeah. If you could do that from scratch, that'd be lovely. My, my approach is. Why do you say that? Why is it that, you know, when you ship it out to Codex and they come back saying, no, this is no good. Why is it no good? Show me. Like, let's prove it. Yeah. And you know that that kind of questioning loop is a little, a huge part of my workflow. **Dave:** Yeah. Nice **Marno:** approach that I've had where, when I'm giving it a task, for example, this anti-gravity one is I would spend time on complexity and or Google Gemini Pro, um, even now 5.4 thinking. But I'll choose a model and I'll talk to it to say, Hey, this is what I'm trying to achieve. Spend about a good hour explaining exactly what the end goal is, what frameworks to use, what technologies to use, um, and what I wanted to do and not do. After I've done that and I understand and agree of what it's trying to tell me, then I'm saying, great. Now help me explain this to my security team, or help this, explain this to me, to my dev department. Then it writes an email that's to my dev department. But taking in that context, it makes sure that it has all the technical terminology in there. [58:00] Dave Pengelley: It explains it clearly so they understand the instructions, the end goal, um, and how to build it. And the, the phases to go through creates all of that for me. And then in that little persona, I just copy paste that. I was like, Hey, this is your job. This is the [58:00] Dave Pengelley: skill. Now build this. Um, but then that avoids the, the, I guess the weakness of me trying to explain it because I'm definitely going to miss something because I'm assuming that it knows, especially because it confirmed it to me in a previous chat. But that email format or that, Hey, explain this to my dev team or explain this to my security team has been so, so beneficial. **Matt:** Yeah. That's cool. I, my, my radar went off when you said copy paste, that, that, that is a, another opportunity for, you know, automated handoff like **Marno:** Yep. **Matt:** It's so good. Love **Dave:** it. Baby steps. **Matt:** Yeah, **Marno:** baby steps. **Matt:** Oh, it's all fundamentals and principles at the end of the day. Yeah. **Dave:** Yeah. Lately I, and, and you know, I've been building these sort of systems and, and you run it and then you go, oh, how do I automate this? Can you bash this sort of stuff up? And what does that look like? And okay, brilliant. [59:00] Dave Pengelley: I, okay, now I've gotta do that step. I, I wanna post, deploy this site on versal versa. All's got ACL I, [59:00] Dave Pengelley: right? Can we just make a deploy script so you can push it there? So I'm not going into the web UI to be this, to do that. How do I, how do I push this to a GitHub repo? How do we spin up a new reper repo automatically and push this there and then link that straight to cel and mm-hmm. And so now I've got a couple of bash scripts and, and it just does that for me without having to go in and click things because I asked the questions and it worked it out. **Marno:** That's awesome. That's so good. **Dave:** Yeah. Yeah. So, um, but, but a lot of people out there like, like with like all this stuff's gone way over their heads. Um, and it's great that you're gonna focus on education minor. There's a huge gap out there and people, I dunno what you're finding in your areas, in your conversations, but I'm just finding from just trying to put out some free help on LinkedIn. People are scared. Um, I've been putting out this like really open offer. **Marno:** Mm-hmm. [01:00:00] Dave Pengelley: **Dave:** Open calendar, no shame, no judgment, no sales pitch, no strings attached, just half hour chats. If people just wanna talk through and ask some questions and I would've thought my calendar would just be getting smashed with people going, oh, [01:00:00] Dave Pengelley: brilliant. One of the AI operators, I can get 30 minutes at his calendar and just ask questions and not feel stupid. Or even if I feel stupid, he is told me it's gonna be fine. And. People just are not like, yeah, it's just not the, the posts aren't getting traction on LinkedIn, which is weird. Um, and people are not taking up the offer. So there is both. I, I don't believe that all the people that are seeing these posts are fully, like gungho a hundred percent ready to run on ai. **Marno:** I think they just don't care, man. Like that's my experience so far is there's a lot of business owners that are massively, like, they, they can definitely benefit from using AI and even just introducing g BT into their daily life to just have conversations with would make a benefit. But it's that intent of buying, like, are you talking to a starving crowd? [01:01:00] Dave Pengelley: Because, and, and that's why, um, or one of the few excuses that I have for the social media stuff is I don't wanna just make content for the sake of content. I wanna make content that [01:01:00] Dave Pengelley: gives me and talks to the people that actually care about the product. Um, and that's where I'm not sure what that was gonna be yet until last week, where I know it's just gonna be education. So now all the content I'll be producing will be surrounding education because I know the people that want to learn will be engaging with it. Where if I get 200 likes, but it's just from random people that don't really care about education, they're just interested in the new technology. And that doesn't serve me at all. Um, what I'm trying to say there is, I don't think it's a lack of the post not performing. I think it's just there's so many people that are too busy in their own lives that they're like, I can't understand this technology. I don't understand it, and I don't see why learning it would make my life better yet. Um, and they're just lazy. Like we have paste, like avocado paste. That's my best example. We have an avocado next to it and then you'll pay an extra, what, $2 for avocado paste, because you don't wanna cut it open. People are lazy. They want you to just do it for them. Don't ask them any questions, just get it done and **Dave:** move on. Yep. [01:02:00] Dave Pengelley: **Marno:** Which is a shame. **Dave:** Joe, Joe reckons, let's just, just make a, an office hours only too hours each week. It's, it's too many. Too many options. Too much availability, scarcity. We'll see. I'll, I'll try. I'll try, Joe. I'm just trying to, trying to help as many people as I can just get over the hump. Um, **Marno:** yep. **Dave:** Because, you know, it's good for all of us, all of us that are working in this space. The more people that understand it, the more people that can sort of start getting their head around it. Mm-hmm. The, the, you know, rising tide lifts all boats. I'm just trying to be part of the rising tide. So, yeah. **Matt:** As a, as a good, like a good thing to remember, you know, like, yeah, we're the AI operators, you know, promoting ai, but it's not always about just ai, you know, it's, it's your systems and your processes **Dave:** a hundred percent. **Matt:** Like, you know, the businesses that win won't just have better ai. You know, they'll be the ones that have better processes that AI plugs into. [01:03:00] Dave Pengelley: **Dave:** AI is the marketing term. It's like this TV now has ai, everything has to have AI built in. My fridge [01:03:00] Dave Pengelley: has ai. **Matt:** Yeah. **Dave:** But, but I mean, that's the, the marketing term. Uh, and yes, we are all using and leveraging AI and we can help businesses do that. But I, I sent through a report for business this week on how to help them get our, like 20 hours back a week of time, potentially. There's no AI in it. It's just centralizing data and installing some better systems and, and setting up those, those workflows and those pipelines to talk to each other, uh, to save the time. Not actually ai, just better systems integration. **Matt:** A hundred percent. Like even like standard companies that exist now, like a lot of Australian companies would, would be like this. Mm-hmm. You know. Stop just trying to tell your people to try harder. Like, you know, you can, you can actually build systems where the right thing happens, you know, **Dave:** and, and free, free that human capital up to do other things. This isn't about firing people and, and sort of direct cost cutting and by, by cutting heads. **Marno:** Yeah. [01:04:00] Dave Pengelley: **Dave:** This is saying, Hey, your people are burning out doing these manual copy paste tasks. How about you automate and [01:04:00] Dave Pengelley: free up those people to interact with customers, to grow your business, to think about new product lines, to all the different things that the humans should and could be doing creatively and relationally, that the AI is less equipped and automated and good at doing and grow your business through that. **Marno:** I'd probably even argue that it's, it's a disadvantage to implement AI if you don't have their foundations because AI amplifies AI scale. So if you have a crappy foundation, then you're scaling crap. **Dave:** Yeah, I, I just had a mate round for coffee before the, before the, the podcast this morning, and he was saying his business partner has started playing with Claude. And he is like, oh, so amazed we can do this. And, and the next day he's like, oh, actually no, we, we, we shouldn't do that. Oh, we do this. And I was like, I was like, that's exactly why I exist, mate, because that, that whole ROI thing, not just like these sort of like little toys and tech for the sake of tech, but working out where in your business is it gonna actually drive value? [01:05:00] Dave Pengelley: But unfortunately, I think, you know, in the friend zone, sometimes friend zone and business doesn't, doesn't cross. So I, I don't think I'll be helping him and his business partner out with, uh, any strategy. But [01:05:00] Dave Pengelley: that's out there. There's people that are like, oh, there's called code thing, this co-work thing. I'm gonna use this. They're gonna do that. Uh, I, I barely started using Cowork, cowork. I had a bunch of issues with, um, hard drive space on my desktop. So I tried getting cowork to look at a few folders to see whether it could help me clear up some, some things. So people were like, yeah, cowork moves all the files around, or indexed all my photos and did all this sort of stuff. But I'm like, I did, I did actually. Claude code at some of my development folders and say, can you clean up all the old build logs and files and things? I take it up gigabytes of space on my machine. So it did that and I was like, brilliant. Okay. That's, that's space back. But **Matt:** that scares me so much. **Dave:** RO minus RF from an agent. **Marno:** We've got, we've got a question from Shania. You guys reckon that business owners and managers you speak to actually see what their staff can do to save time? Like do they have tasks waiting for them? [01:06:00] Dave Pengelley: **Matt:** My answer to that is no. Yeah, they don't, and you know, this, this relates [01:06:00] Dave Pengelley: exactly to, like, if you can't specify what good looks like in specific terms, you're not managing, you're just hoping. **Marno:** Yeah. **Matt:** Like, yeah. If they go, yeah, I'm gonna save all this money, like, it's great. And then we'll do, uh, that thing, you know, like there's, I haven't had anyone give me a good answer on that yet. **Marno:** It's a good reminder that most people are just winging it. We're all just trying our best, just making it up as we go along. And some people would just hit it. Lucky. Um, yeah, it is common **Dave:** as we discussed. Um, like, uh, who knows when we've been doing this show forever now it's great. Uh, inertia is real. Uh, and people are used to doing things the way they're used to doing them, and they get really efficient at doing things in on inefficient interfaces and mm-hmm. [01:07:00] Dave Pengelley: They just think that's the norm. That's just, this is what normal is. And to change from normal is really, really tricky. Like a, a [01:07:00] Dave Pengelley: customer I was talking to about potentially helping them, they're doing everything by hand, but like, oh, it's not so not that bad. I, I, I, and it's like, well. It is. And then the, the, they show me the next step in their process and I'm like, oh, oh my goodness. Like, you're doing what now? Yeah. You're using that how, and for them it's just normal and they're just like the, and, you know, changing, uh, even if there's a potential better outcome, the potential outcome is unknown and hard to see and hard to visualize. It's why in my previous career in presales, our whole job was understanding the need and then working out how to tell the story and present it with live demos and things. And I think that's something that, you know, a lot of us, um, smaller operators that are getting started here, we don't have the demo infrastructure to really give customers the clear vision of what good looks like. **Marno:** Mm-hmm. **Dave:** Before they, they sign the check and so they're scared of signing the check. And it's that whole mosey thing, right? [01:08:00] Dave Pengelley: Like how do you de-risk the process? Yeah. In presales, in big tech, we de-risk it by [01:08:00] Dave Pengelley: showing them a live, working demo of what the future could look like for them so they can sort of taste and see a little bit more. I think that's something that as, as sort of operators, we need to think about how do we do the taste and see, and obviously us building our reputations on social media and talking about these things. **Marno:** Yeah. **Dave:** Helps with that subject matter expertise and that trust, but also physical demos and how we do that. So as you do content and your educational content, content, how do you make that real for people? I think it's something to really consider. I, **Marno:** there's a good bloke, there's two, two points to that. One, there's a bloke that I know of in the UK and he doesn't have any technical experience. Um, he's just like, for lack of better terms, it's the focused fool. Like he doesn't understand technical side, but he knows exactly what he wants to do. He knows he can help people and he wants to help people. And he has a small repertoire of tools. Yeah. So his workflow is who goes in, talks to the business. [01:09:00] Dave Pengelley: Says I could build you something. Build him a lovable demo, sends over a lovable demo, [01:09:00] Dave Pengelley: not connected to anything. It's all like made up. Yeah. None of this makes any sense. It's not even potentially even possible to build what he wants to build, but it gives him a lovable demo. They're immediately sold. That's one approach for the the like physically showing you something. But then the other side, um, that I wanted to comment on there is we're talking, we talked about people not knowing and then showing them something physical. I can't remember my second point now. **Dave:** Demo. Making a reel. Taste and see de-risking or mosy. **Marno:** Oh yeah. De-risking. Yeah. So I was fortunate enough that some of my clients in helping them building automations was, it was at the point of someone moving into a different position or leaving the organization. So like they're being promoted from, they're an admin lady, they're not gonna go into helping with the mechanical side or the logistics. [01:10:00] Dave Pengelley: Yep. That that position is going to become vacant within like a month. So we have to act on this and they know exactly what that person does. So we automated that entire [01:10:00] Dave Pengelley: person's position, which was such a win for the company because if we didn't do that, they'd have to spend months hiring someone. Hopefully they work and in the meantime, three other office ladies would have to share that workload, which sucks. Mm-hmm. And then the other one is businesses where they're busy scaling, so they don't have anyone in that seat yet. They know they need someone. And then building that someone for them. 'cause that, that way it's an easier sell rather than a, Hey, I wanna replace Janet down the road. It's gonna be, Hey, let's build Janet before she even exists. And then when you bring her in, then they can work congruently and then the workload is less. And that person is enjoying that. They like life a bit more and their work. So they're still a bit longer. Rather than leaving at six months for a high job at the most. [01:11:00] Dave Pengelley: **Dave:** Yeah. We, we in, in our CRM, uh, in my, my old role, there was a field where the sales rep had to put in, like, why, why is the customer gonna make a decision? Like, well, what, what's, what's their why? And you'd want something time bound generally. Like, oh, they've got a new product going to market and they need to [01:11:00] Dave Pengelley: have the system in place for that, et cetera. And you get one. And we were selling CRMI was at Salesforce. If people didn't know that and you'd get CRM is clunky existing, it's like, that's not a decision making criteria. They're not gonna spend money in investment project because the current system is clunky. It's like, uh, so even, even salespeople at one of the best sales organizations in the world didn't always understand that whole decision making criteria. And, and you. I'd, I'd coach other pre-sales people and they'd come and say, oh, the customer's doing this and this, and this and this. And they'd explain all the different workflows. I'm like, that's cool. We could probably build stuff and automate half of that, but which one's gonna drive value for them? Yeah. Which one do they actually care about? Because you've, you've identified a whole bunch of different options. I don't know which one they, that, that they actually even care to invest in. Oh yeah. Okay. We'll go back and ask some more questions [01:12:00] Dave Pengelley: **Matt:** like a direct, direct reference to, to the homo thing that I captured recently, which like, I'm just gonna read this out [01:12:00] Dave Pengelley: verbatim. This is a quote. It's getting granular is the thinking most people won't do, which is why they will lose. **Marno:** Yep. **Dave:** But it's, I, I find I'm, I'm a much better coach than I'm an executor on some of this sort of sales stuff too, so. **Matt:** Yeah. True. **Dave:** Um, actually having those conversations and establishing stuff with the clients. Is a whole skill set in itself of communication skills and confidence and trust building and so on. And I've always, you know, part of my solution ideation skillset is I'm really good at breaking down, analyzing other people's problems much better than my own. So that's why you should hire me 'cause I'm really good at assessing your problems. [01:13:00] Dave Pengelley: Um, alright boys. Well, we are well over time. Quarter past one. What is going on? Uh, thank you Australia. Thank you. The world. Thank you. Wherever you watched this from, at whatever time you've watched this on, uh, live or the replay, if you caught this on [01:13:00] Dave Pengelley: LinkedIn, please give it a thumbs up on YouTube, especially thumbs up. Comments, comments, drive everything. Uh, we're gonna be trying something new on this episode. You might be watching the replay, which is not the live anymore. We're gonna repost the replay up as its own thing to see if YouTube loves it more. Um, we're gonna be doing a few experiments with the channel, so make sure you, uh, subscribe to the channel. Make sure you hit that bell icon to get updates. 'cause we're looking at putting a few different types of content and doing a little bit more interesting things on the AI operators channel in the coming days, weeks, and months. So keep an eye out for that everyone. Any last thoughts before we wrap up? **Matt:** Yeah, thank you so much to Sean and Joe for all of their questions and comments as well. As always like, **Dave:** yes, legends, love the live chat. Well done. **Marno:** A hundred percent. That's all I wanted to say as well. Thank you, Joe. Thank you, Sean. **Dave:** Brilliant. Well, to all of those out there, experimenting, operating, clawing, we wish you well and we'll see you on the next one. [01:14:00] Dave Pengelley: Thanks [01:14:00] Dave Pengelley: boys.