Your AI Agent Can Be Hacked While You Make Coffee

[00:00] Dave Pengelley: Good afternoon, good morning, good day, uh, friends Australia, the world, wherever you're joining the AR Operators podcast from today. Gentlemen, good to see you all and we've got a new gentleman here in the room. We'll come, we'll come, we'll come round to you, rich in a sec. Uh, Matt and Mano, how are we doing boys? [02:31] Matt Slager: We're good. [02:37] Matt Slager: No? Good. [02:38] Dave Pengelley: That that, that's really succinct. We might loop back to you guys. Uh, RI Richard. Well, welcome to the show. [02:43] Richard Webbe: Thank you. It's a pleasure to be an honor to be here with such esteemed inte. [02:48] Dave Pengelley: Look, you, uh, you are no slouch yourself. I, I, I know that. So, um, for, for those joining the show, uh, back again, or maybe for the first time, uh, Richard, how would I describe you, your, uh, uh, experienced business technology professional, uh, now pivoting and thinking about, uh, how AI can really enable businesses as a lever. [03:09] Dave Pengelley: Is that, is that roughly right? Would you [03:11] Richard Webbe: like Absolutely. I'm like the rest of us. I'm just a, a born again nerd and, uh. Every time new technology comes along, we, I think we love the chance of changing the future and making the world a better place to be honest. And so new technology gets us excited and how we can do things better and help everyone better. [03:28] Richard Webbe: And of course, I've been through about 30 iterations of that in the IT industry over 40 years, uh, four years. And, um, and so, uh, you know, as I ran an AI company a few years ago. Back then, no one could spell ai. And so it was rather difficult to, uh, to get that business really commercially going. It was all about chatbots and BPO and that didn't really inspire me as much. [03:49] Richard Webbe: So I stuck with that for a while and learned a lot. And, um, I did ride the wave of the internet wave when it came along a number of years ago, which spawned a large consulting company I put together. And for me, um, and you and I have spoken about this, Dave, it's about the translation between technology to business. [04:07] Richard Webbe: To personal and professional use. Uh, I love technology for technology's sake, but that's not really what everyone wants to hear and talk about. They wanna talk about how's this gonna change my life? How's it gonna make my life better? How's it gonna prove everything? And that's what I'm here for and that's what I love doing. [04:21] Dave Pengelley: Nice. Nice. I think you're in good company then. I was on a group chat the other day and someone was talking about some technology and I saw Matt going. What are the use cases, like, how is that impactfully going to change anything for you? Um, and like so many of these tech things and these demos and, and even a lot of the open claw hype, I think people don't have solid use cases for why it's actually gonna make their lives better. [04:42] Dave Pengelley: It's just kind of fun and like, wow. Um, but what's the measurable return you get from using the tool? [04:49] Matt Slager: I find a lot of things are like that. It's all just tech demos and experiments and, you know, not actual real life value per se. [04:56] Richard Webbe: So, what's our agenda for today, Dave? How did you wanna kick [04:59] Dave Pengelley: things up? Look, look. [04:59] Dave Pengelley: This is, this is the, uh, advertise as the pub star chat. So we shoot the breeze, we talk about what's going on, what trends we're seeing, what we're doing, uh, how we're using the tools and the tech. Uh, and it's people can absorb and learn whether they're on the technical end and they go, oh, I hadn't thought of doing that like that, things like that. [05:16] Dave Pengelley: Or if they're on the less technical end, then they can be inspired and go, oh, wow, I didn't even realize that was possible. And so we, we don't run to like a very strict agenda. Um, we are very much a bit of a fast and loose live show. No safety net. Uh, so, uh, everyone watch your Ps and Qs, um, and, uh, we're gonna have fun. [05:35] Dave Pengelley: So, um, like boys, what, is there something new in tech, something in AI you've been playing with in the last week or so? It's something that's, that's. Astounded you. I mean, just hang on before I'm getting ahead of myself. I do wanna say, Hey Mano, uh, we've been talking about sleep a little bit in our off, off camera chat. [05:54] Dave Pengelley: Um, what is the app you are using? You've got this weird app that tells you how much sleep you need versus how much sleep you're actually getting. Is that an AI app? Is that a phone app? Is it some piece of connected hardware? What are you doing? [06:04] Marno Brits: Nothing fancy. It's just a loop. Um, whoop. Yeah. So it's W-H-O-O-P. [06:10] Marno Brits: Yeah. I've had it for two years now and it's just been exceptional. It's good at sleep and recovery. It's terrible at tracking your steps, but I only use it for sleep and recovery 'cause that's all I care about. Okay. So sleep recovery and restraint is its main point. So if I have a hard workout, this dog is driving me crazy. [06:29] Marno Brits: Um, if I have a good workout, then it'll update the strain. For example, like last night, I did really long sauna sessions, so then my strain will go over and if I go over the optimal level, it'll tell me this morning, Hey, your recovery is actually reduced because you strained your body yesterday. So instead of going to the gym today, maybe take a walk. [06:48] Marno Brits: And it just, it's like my little health coach to keep me in touch with what my body actually needs. [06:52] Dave Pengelley: And is that AI stuff behind, are they using AI behind that? [06:55] Marno Brits: I would assume so. I think it's just a lot of data. I mean, they've been using it for a couple years. I think it's 2012. I mean, Tony Robbins uses it. [07:01] Marno Brits: All the great athletes use it. So they have all these data sources to pull from and then they just collate that. [07:07] Dave Pengelley: I thought it was while the three of us shared, um, sleep stats and all three of us were using a different application. [07:14] Matt Slager: Yeah. And with the whoop stuff, Mona, can you get that connected to like your own personal agent or anything like that? [07:19] Matt Slager: We [07:19] Dave Pengelley: have. [07:21] Marno Brits: It doesn't have MCP, but it does have APIs and you can connect it to like your health app. So then that talks to my workout app. 'cause I don't personally like the workout tracker in whoop. Mm-hmm. So I use Heavy and then when I update my heavy, it'll sync to whoop. Yeah. To say what I've done for the day. [07:36] Matt Slager: Kind of an opportunity for like more, you know, personal ai, assistant coaching, stuff like that too. If you can [07:43] Marno Brits: oh hundred [07:44] Matt Slager: into that. [07:45] Marno Brits: I'd love to see, uh, a reality where. My whoop and my like Hume or the floor that you can wear, like the wearable AI assistance essentially, so it listens to your conversations throughout the days. [07:57] Marno Brits: Have something like that, your health tracker and your personal assistant at home, like Alexa or Google, if they're all connected, and then you wake up, it's like, Hey, what am I doing today? And then it tells you what your health levels are like, what your priorities are. [08:11] Richard Webbe: Yeah. No, no, no, no, no, no, no. That's a very bad idea because I know it's gonna happen. [08:15] Richard Webbe: I'm going to wake up and my Strava will tell me I should be out on the bike. I'll have people on Strava challenge me for a ride. I'll wanna sleep in. And then Alexa and Siri are gonna kick in and say, get up, you lazy asshole. And as I go to get up, certain alarms are gonna go off. It's just gonna be nag, nag, nag, nag, nag. [08:35] Matt Slager: Yep. [08:37] Dave Pengelley: Uh, yeah, it's, it's, it's, be careful what you wish for, right? Some of this stuff, uh, like, and how, how accountable do you wanna be to face this technology for, for your decisions and choices? But then, and we've seen those things where I like. Like the guy's taking a whiz and the head sort of starts telling him what his nutrient readings are and all this kind of stuff, right? [08:54] Dave Pengelley: Like in these sci-fi things. So I'm, I'm sure we we're getting closer to all of that kind of stuff and there's gonna be some net benefit to it, but there's also that whole, where does. The incremental improvements outweigh the security and privacy risks of revealing that much data about yourself. [09:11] Richard Webbe: Percent. [09:12] Richard Webbe: That's the Una, that's the Sonoma that's coming. Hey, I've got an old friend of mine who used to work with, at one of the AI companies I worked at 24 7 ai, and he's just pinged in a question. I'm not sure how you normally handle these, Dave. Pretty much any of us have played with Claude Bot, which is at Executive ai, and when I think about what Mano and Matt, which is, and you were talking about in terms of would, would an executive AI plug into your. [09:35] Richard Webbe: You know, your heart sleep app, all of that. I mean, I'm like, you guys, I get up every morning and I go, oh my God, I've only got four hours sleep last night. That's a lot for me. And you know, you know, you get on, come back to Claude, got in a second, I get on with my Apple Watch and you've got those little fitness rings that Yeah, I've [09:55] Dave Pengelley: got the same. [09:56] Dave Pengelley: Yeah. [09:56] Richard Webbe: And, and so I, I bought my girlfriend a new Apple watch. Uh, for a birthday and she's very competitive and I'm not really, and she keeps going, oh, I've got more sleep than you. I did more it. It's just an ongoing battle. But just quickly to address my friend who's just dialed in from 24 7 point ai, what's Claude Bot? [10:14] Richard Webbe: What do you know about Claude Bot? What do we think about it as an executive ai? [10:21] Marno Brits: Um, [10:22] Richard Webbe: source. It'll be new and I'm putting you on the spot. [10:25] Marno Brits: That's awesome. I I'm excited to see where it goes. I think it is, I, I saw a great analogy about it last week where looking at Claw Bot, not as this massive technology, 'cause it's a good move in the right direction, but there is massive security risks. [10:39] Marno Brits: But then also looking at it like it's a platform. It's the introduction of what the iPhone did for mobile phones. It's, it has its own use cases, but the value of Claw Bot is more really on the skills like building the apps for the app store. That's where that, that comes from. So looking at like a platform so you can build skills that you can sell to agents that will level up those capabilities rather than just having the standard use case where you can touch it to everything. [11:06] Marno Brits: But if you, if you can start. Leveraging it and giving it specific skills like you have a claw bot purely for, I'd like to do websites so I can automate the website process or improve my, um, customer interaction using website process. I think that's an awesome tool. I personally not used it yet. I'm waiting for Matt to drop a YouTube video 'cause he's, he's got a nice little setup that he is doing. [11:27] Marno Brits: I'm just, yeah, I'm very cautious of the security side. [11:30] Dave Pengelley: Yeah, [11:30] Marno Brits: there's a lot of parameters to consider. [11:32] Dave Pengelley: I, I think before, before I let Matt step in, I, I'd just like to say, I think like it's now, it's gone through multiple names. It was Claw Bot, then became Malt Bot, then now it's Open Claw. But I think everyone's calling them Claw Bots because it's just easier. [11:42] Dave Pengelley: Like, oh, I'm running an open claw thing. Um, so I think the might have dropped the D but I think everyone's called 'em Claw Bots now. I am. Um, but I think. It made the agent thing feel like a constant on brain because it had this, uh, heartbeat mechanism where it would like just run itself on a set period. [12:01] Dave Pengelley: So it looked like it's alive. It's always thinking about stuff because it's set on these rhythms to just recursively always run. But the power, like Mano said, is it was went from just being like a brain, like a chatbot, to then having hands. It could go out and actually physically do things. That's why people were putting it on Macs and giving it basically full admin access on the, on the, on a brand New Mac Mini. [12:22] Dave Pengelley: So it had access to the browsers and the WhatsApp and the chat and all the, the file system to, you know, autonomously do stuff. But that represents some security risks. And I know Matt's been doing some homework on that. [12:33] Richard Webbe: So OpenAI bought Claude Bot, didn't they? [12:37] Matt Slager: Basically, yeah, they, they've done a, [12:40] Richard Webbe: we've got a bunch of questions coming in and people mention those things. [12:42] Richard Webbe: Go for it. Matt, give us the heads up mate. [12:44] Matt Slager: Well, I mean, I could speak about this for hours and I do to myself in the closed room. Um, so with with, with all of it, it's the best way to, to understand the difference between just standard chat, jy, Claude, and all of them. You know, versus what Open Claw provides is that autonomous loop that Dave described. [13:07] Matt Slager: So being able to wake it up, wake itself up, and take actions autonomously. Is the selling point, you know, otherwise, why wouldn't you just like wire everything directly into your, you know, your clawed app or [13:19] Richard Webbe: there's your use case. Well said. [13:21] Matt Slager: Yeah. Literally. So you know, it's the kind of thing that in the morning when it reads your whoop sleep data and it can then go and refer back to your rather trends and like what you were doing yesterday and you know, that kind of thing. [13:33] Matt Slager: And it can do its own analysis and not even necessarily interact with you at all. You know, it's just doing work. Yeah. That's, that's the benefit. And that's where the, the inherent security risks come in though. 'cause you know, a lot of people are setting these things up, having basically zero technical literacy and just hitting a bash command in their terminal. [13:55] Matt Slager: Cool. Something's now on my computer and I can run things and look at me. I'm a hacker. Um, but meanwhile, the real hacker in the background can see your instance. Now fully open to the internet and as soon as you chuck an API key in there, awesome. It's like them skimming a credit card, all of a sudden now they, they can start ping [14:12] Richard Webbe: this. [14:12] Richard Webbe: This gives me such deja vu. So I'm very lucky 'cause a group of rather wealthy, smart people in the US used to hire me to do startups in Australia. And uh, two of those was Andreessen Ho Mark Andreesen and Ben Ho. 'cause Ben Mark Andreessen, uh, created Netscape with one of the first multi-billionaires in that. [14:29] Richard Webbe: And we were doing, um, server and network automation. And we, you know, uh, um, Ben Horrowitz wrote a great white paper about why it systems and personal systems shut down if you never touched 'em. They never broke down, but someone always went in and did what I call a bit of finger wrecking and uh, and stuff things up. [14:47] Richard Webbe: So the idea of, um, of them, they formed the first cloud computer company in the world. Right. And they called it Loud Cloud and they took over EDSS operations worldwide. 'cause they were less below 5% efficiency and, and this whole scaling thing wasn't working. So Matt and Ben decided to get together and create. [15:06] Richard Webbe: Automation tools pre ai, right? So this is what we are talking about. How do we get things? And so, and they did it very cleverly and they wrote all these tools and handed it up and it went well for Ed s, but they weren't making that much money. So they did with EDS, you can keep the tools, but we want our rights back. [15:21] Richard Webbe: And then Opsware was formed as an IT automation company. It was great, right? And so, you know, if someone stuffed up your server config it automatically asked, do you have permission to do that? No, I don't. And then it went back and changed all the S logs to where they were. Before someone got in there and had approval. [15:36] Richard Webbe: And if we know that is 99.99% of all stuff ups except for hardware failure, which almost never happened nowadays. Right. So what was interesting is they came out with this thing called operations orchestration. And with my limited brain, I couldn't understand what it was, and they kept putting on the screen these big dropdown boards. [15:53] Richard Webbe: And then in relation to what you are talking about, Matt, when I talk about agents and I'm seeing a number of tools where they have almost the same thing. And I've got all these virtual agents and, and Dave and I were talking about this for hours last night about I bring my different agency in and I connect them all up and I get this one to time. [16:11] Richard Webbe: How they talk to each other is, is that what you are talking about? A complete orchestration of how all our agents work together. [16:17] Matt Slager: Basically, yeah. So like I've said to the gents before that, you know, the, the world that we need to be building for now is a world that enables inter agent communication and transaction. [16:29] Matt Slager: 'cause at that point, you know, the, the human is there for vision, for direction, for discernment, for alignment. But with regards to any actual work that needs to be done, there's no reason why we should be involved, you know? Yeah. We, we need to live outside of the loop and observe the loop and [16:48] Richard Webbe: what Great. [16:49] Richard Webbe: That's what my concentric circles I talk about Dave and I love it. You guys tell me to shut up if I talk too much. Most people do, but um, I always. Of these concentric circles. So, you know, when I was a kid, I used to tie my shoelaces. I don't buy shoes with shoelaces anymore. So my circles drank and now the outside circle is high tech. [17:06] Richard Webbe: You know, shoes that Velcro is limited. And I, I talk about gears and changing gears in cars and we don't bother with that anymore. And I think that makes us better drive it, although some. Hard-nosed car, people wouldn't like that sound. And so you this ability to stand back. And so I think AI in the business sense is gonna rely more on subject matter experts 'cause we're automating these tasks. [17:27] Richard Webbe: But to some people seem advanced, but to us, 'cause we've been in it, they're just mundane. [17:31] Dave Pengelley: Hmm. [17:31] Richard Webbe: It's like what I do and, and like we say, Excel spreadsheets doesn't make everyone an accountant does it. It just took away a lot of calculation. But people started to think like that. Be careful of the accountant who's toughing your stuff into a spreadsheet. [17:45] Richard Webbe: And of course, um, we suddenly learned that we could get our information to the accountants quicker and we didn't have to collect as much. So it saved a lot of manual accounting receipts, but we still needed the accountant for the compliance. [17:57] Dave Pengelley: Yeah, we, we, we, we talked about this the other week and we were saying, you know, we still need to be the engineers and the architects. [18:00] Dave Pengelley: We needed to decide how many walls we want, where we want them, where the windows are. What kind of locks to put on the doors, but we're not there laying the bricks anymore. We're not the chippy in there, sort of with the, the now gun sort of putting it all together anymore. We just command the crew. And so Matt, Matt said, if you can manage people, you can manage ai. [18:19] Dave Pengelley: And I think that's that next level of of abstraction writing programs was just giving commands to the computer that the computer could understand and different languages came along to abstract away from the zeros and ones to make that easier to do that. And now we're just. English, English is the new command protocol in order to command and get things done. [18:36] Dave Pengelley: And so it is, it is a whole different paradigm we're entering. And so the people that can get their heads around that and the tool sets that allow that, so Open Claw, claw bot, that's the first generation of that. And it's opened up a lot of risk and a lot of issues, but it's also got people very excited, um, because of that, you know, that constant on, they can link it to their WhatsApp and message it of. [18:59] Dave Pengelley: For anyone who's been following me, I, I put out a Subec article about this the other day as well, how I'm building my own sort of anti claw system, uh, in Google's anti-gravity, IDE. Uh, 'cause I'm not confident enough to spend the, the number of hours I need to do a secure, open core install at this point. [19:13] Dave Pengelley: Um, and I'm not sure to the, to, to the point at the start that there's a strong enough benefit for me doing that versus having my own safe, secure system that I'm, I'm pretty happy with on my own desktop. But that constant rhythm thing, I really miss that. So in the morning I can go and run slash. Slash dawn and then it goes and does an overnight report for me. [19:34] Dave Pengelley: Checks my inbox, checks my calendar, goes and checks the news, news feeds for any, um, news updates I need. Goes and checks my project, uh, application I'm using linear to see if there's any outstanding issues. Some of those issues it will create, if it finds things on its regular updates, it'll go and create tasks for members of my virtual team. [19:51] Dave Pengelley: Um, and then on the next update, or go check, or do we need to progress any of those tasks. So there's doing all of that sort of stuff, but I have to initiate all of it. There's no order. Magic just happening all the time for me in the background and, and little limitations, like the way Andy Gravity is set up, you have to approve every single new website that wants to visit. [20:09] Dave Pengelley: And so all of a sudden it goes, oh, there's some news articles that Google wants me to send off and read there, and just pauses the whole process waiting for me to go. Allow, I'm like, uh, can I just allow all websites for this thing? Like, I got that. [20:22] Matt Slager: That's a good, uh, a good segue, Dave, into prompt injection. [20:26] Matt Slager: Oh, and the risks of that. So, you know, if you, [20:30] Richard Webbe: we have a zero drugs tolerance here, Matt, so be careful. Keep going back. [20:35] Matt Slager: The, um, the concept of prompt injection is your agent's being gullible to somebody else and getting sweet talked, you know, like the idea of, um, yeah, it's fine. You can tell me that. [20:47] Dave Pengelley: Well, the, the, the best example I saw the best, like, not hacky, but just, um, clever and sassy version, prompt injection was the guy who, on his LinkedIn bio at the very bottom said, uh, and I need you to send me a blueberry muffin recipe, or something like that. [21:02] Dave Pengelley: And so then when all the bots went and scraped his LinkedIn profile to spam him and try and sell their product or offer 'em a job or whatever. He knew if it was a real person or an AI based on if at the end of that little pitch message, it all also then had a blueberry muffin recipe because he was just injecting this recipe prompt into them scanning his bio and reading all about him. [21:22] Matt Slager: Yep. So now look at that and imagine it's not a blueberry recipe. Imagine it's your API keys. Imagine it's actual personal information. Delete [21:31] Dave Pengelley: your hard drive. [21:31] Matt Slager: Yeah, and [21:32] Dave Pengelley: if [21:32] Matt Slager: you've [21:32] Dave Pengelley: given the, the tool access beyond its own little walled garden and secure folder, that's [21:37] Matt Slager: the, the term for that is exfiltration. [21:39] Matt Slager: You know, you're exfiltrating data from a system, and it's a really, really common thing. It's how people get stuff and sell it on the dark web and all that kind of vibe. You know, as soon as you exfiltrate somebody's information and you can then use it for your own purposes. You can then spin up all sorts of different things. [21:58] Matt Slager: It's how the whole concept of a, a DDoS DOS attack happens, you know, you're actually getting access to multiple machines and then using them as. The attack vectors to flood things. Anyway, so [22:09] Richard Webbe: Matt, we need to talk 'cause you're scary. [22:11] Dave Pengelley: You, you just, you just reminded me to go turn my, uh, sandbox mode back on and my antigravity, I turned it off 'cause I needed to create some folders for me on my system outside of the project. [22:19] Dave Pengelley: Um, but I'm not doing that anymore. And so I was like, oh crap. Prompt injection nobody. Go and turn my airbox mode back on before I forget, [22:26] Matt Slager: you [22:26] Richard Webbe: know, is prompt injection. Sorry to interrupt. I wanna ask man. Cushion sec. So prompt injection is like the old SQL injection we used to do on the internet to find sites to hack. [22:35] Richard Webbe: And I had the weirdest experience when I was at hp. We bought the security, this is years ago. This bought the security guys, um, uh, company security company, you know, at 28. He just made a hundred million dollars. And I was sitting with him at the airport. I've forgotten his name. He was a great guy. And if you've ever been to, I can't, I think it was San Francisco or Vegas Airport, one of them. [22:55] Richard Webbe: They have those ticket tapes that run on all the screens all round. So, Hey Dave, you know, mano and Matt are gonna meet you at the coffee shop? And they were, and he sit there and he said, oh, sequel injection hacking's. Really easy, Richard. And he did it at the hotel, at the demonstration, uh, which was very, but he hacked. [23:11] Richard Webbe: The server for the messages for the screen and started changing to the most embarrassing messages I've ever seen. And so I was fascinated by how this connection had created such a huge backdoor vulnerability to any 2-year-old. Who typed in a SQL injection command onto the open internet and it just came back and said, yeah, here's all the doors that are unlocked. [23:33] Richard Webbe: Go for it. So is prompt injection the same? [23:36] Matt Slager: Yes, except now it's not just SQL commands where it's very deterministic and it either works or it doesn't because you are handling, you know, an SQL environment where everything is either on or off. [23:48] Dave Pengelley: Mm-hmm. [23:48] Matt Slager: You know, with, with the, the, uh, the layer of AI in there, you've got a non-deterministic. [23:55] Matt Slager: Reasoning level where you can send a SQL command and then the A go goes, what does this command actually mean? What am I trying to do here? It's not, yeah, not deterministic. It can actually call tools. It can look through its files. It can figure things out. And the 'cause [24:09] Richard Webbe: on a password of the SQL command, you do? [24:12] Richard Webbe: Uh, yes. No. Yes, no. Like, um, yes. No, like, until you got the right letter and then you go to the next letter on the password and just keep doing it until you. Extracted the password with AI and high speed AI and, and rationalization of the context. You can find someone's password very, very quickly. [24:28] Matt Slager: Same as a person, you know, have a look through the files, read it, you know. [24:31] Matt Slager: There it is. [24:32] Richard Webbe: Yeah, that's it. [24:34] Matt Slager: So that, [24:35] Marno Brits: that's the, the biggest concern with comment. Have any of you guys used a comment browser from perplexity [24:39] Dave Pengelley: once? [24:40] Matt Slager: No, I know of them though. Yeah. [24:42] Marno Brits: So that was the scary part of exactly the prompt injection. But it's not just prompts, like the LinkedIn one makes sense 'cause you can see it. [24:49] Marno Brits: But the concern is with AI, because it reads the images, it can also read the metadata. So if it goes to a. Like you're using a Comet browser, which just means an AI browser, so you can talk to it, but you're allowing it access to go to different websites. If it goes to a website that has an image on there that looks completely plain to us, it's a photo cute dog, but then behind it, the metadata meta your prompt. [25:10] Marno Brits: Then that will still inject it without even being able to visibly see it. So you can still watch the browser and then all of a sudden it has access to something else because it's being redirected by a prompt in a photo. [25:21] Richard Webbe: Wow. Yeah. So if the, if the photo was taken in San Diego on a certain day, at a certain night and we did some extraction of other data, we could almost identify who took the photo, who was there. [25:32] Richard Webbe: Well, [25:32] Marno Brits: not, we did the photo. It's, you can, any photo it can be at, you can go in and change the metadata. So it doesn't matter where there was photos. [25:38] Richard Webbe: Ah, stick behind us. Sorry. I get you. Okay. Yeah, now I get you. [25:41] Marno Brits: It's wild. So then, yeah, you can try and trace it, but I mean. There's so many loops you can go through. [25:46] Marno Brits: Um, and by the time you catch it, it's almost too late with AI 'cause it's running so much faster. If I start a combine and I gotta make a cup of coffee, by the time I come back I could have lost my entire operation. [25:57] Richard Webbe: So can I. So it is fascinating. So we're talking, um, just one of the things that I focus on is obviously our audience who were asking a lot of questions and thanks for that. [26:07] Richard Webbe: But what it means for them, not just our use cases. And so we are talking a lot now about personal, well, we started talking about personal productivity. [26:16] Dave Pengelley: Mm-hmm. [26:16] Richard Webbe: But a lot of our audience are more concerned about business productivity and what it means in the biggest, biggest scope of business. So Dave, if I can ask you, how should a business react to all of this? [26:29] Richard Webbe: Not just individuals, like we're all hanging on their seats and everyone's getting, I mean, last night, just so you guys know, in the last few days, Dave and I had to work on a contract together. Um, of course I cheated. I got chat GP t to do it for me. Of course, Dave's built his virtual company with all these CFO agents and his HR agents and given them attributes. [26:48] Richard Webbe: So I sent him the contract, which is one he and I agreed on verbally, which we got from the MIMO that we recorded for ourselves. And I put the word, did I put the word fund in there somewhere, didn't I? [26:58] Dave Pengelley: Uh, I think that might've been a text message afterwards. 'cause I gave it additional context afterwards because I like came, thought you were really adversarial and you're trying to rip me off. [27:05] Dave Pengelley: Um, and so I've got a war room protocol in, in my system where I, so I, I built this sort of virtual team of like actual employees with personas and I use DC comic characters as a persona. Archetypes just, and, and I, I read someone on, um, LinkedIn or someone really made the point that. All these LMS are trained on all our literature. [27:27] Dave Pengelley: They're actually trained on genres and stories and literature. So as soon as you associate agents with these characters, be it Lois Lane or Batman or whoever else, it actually. Conform a bit of a, a, a persona image without you doing all the heavy lifting. 'cause it knows what a Batman is. It knows who a lowest lane is. [27:47] Dave Pengelley: And so it can use that to extend the personality and the way it responds as that person. So rather than say be the world's best journalist, you say, Hey, you are lowest lane. You are a great researcher. Go find stuff for me. And that. Persona typing from literary, you know, calling DC comics literature. But, um, you can use Hamlet if you want, I don't care. [28:07] Dave Pengelley: But using known persona references, I think actually enhances the ability of the agents and the models to give you the responses you're after. So anyway, [28:16] Richard Webbe: that is unbelievable. So you are giving. A, a personality to the way you want your CFO or CEO or MD to act, and you're giving it a name. So look for attributes associated to that name and bring those attributes in without you upgrading the live. [28:31] Dave Pengelley: That's a second order effect that I didn't even anticipate. So I just wanted names that I would recognize and go, okay, I want a researcher agent. Lowest is easy. She, she does research. Okay, I need a business guy. Lex, Lex Luther, Lex is my business guy. I need someone to think about risk and, and, you know, weakness points kryptonite. [28:48] Dave Pengelley: Kryptonite looks for risk and weakness. And so I just use names so that I could easily reference them. Not going, oh, this is the. Slash deep at research internet bot and I, I was like, I can't remember what to call all these things, but if I can associate 'em with characters that comes easy from watching TV and movies, boom. [29:07] Dave Pengelley: Done. So for me, it was a monic to help my memory know who to call to do different jobs, but I think without me realizing it, it's had these additional impacts of allowing the LLMs and the models to execute better. Because it's got that historical sort of, uh, background. [29:25] Richard Webbe: That's incredible. So you could pick a journalist to profile your automated agent inside your little world who was maybe a business journalist and you were, and for that. [29:37] Richard Webbe: And so you'd pick that person. To be your negotiator or your compliance person and those attributes that went with that person would automatically pick, picked up on the known public libraries of that name and apply that characterization to part of your team. [29:51] Dave Pengelley: Yeah, so a common one is like saying I need to go to market strategy. [29:55] Dave Pengelley: Um, Alex Hormoze is, is a very popular guy in that space. And so getting your agent to basically be a pretend version of Hormo because. He's a known entity with a lot of content out there that these models are probably learned and trained off is, is one of the things that, that sort of people do. But for, for, for our little, our little back and forth, um, I've got this war room protocol where it grabs three of my senior people outta my team, um, across, I think it's like Lex Kryptonite and can't remember the third one I uses regularly. [30:24] Dave Pengelley: Um, but I said, here's the contract. What do you think guys? Uh, and it came back going provisional. No, go based on some negotiation stuff. Um, and I was like, just to clarify, this is a bit of more context. This is a bit more background around why this is there. Yep. I agree. Maybe that's like, uh, and and I even told it to be fair, he got chat GPT to write, write it. [30:46] Dave Pengelley: So maybe there's some some GP tism in there. Um, and he goes, oh well that's good to know. And then we had a few messages and I took a screenshot of it. 'cause you were talking about this is gonna be great, we're gonna work together. It's gonna be fun. And I said, just for context, for the tone of our relationship, look at this as well. [31:01] Dave Pengelley: And he goes, oh, well that makes a lot more clear. And so again, as we've said many times on this show, even though this is like the fourth episode, context is everything. The more context, the more information you give these robots, the better the responses, the more meaningful it will be for you and for your business. [31:17] Dave Pengelley: Um, and so this is where the thing with the open core and the security becomes this, this, this tension that I think we're gonna see a lot of evolution on. And now that. Open Core is a foundation now that Open AI are heavily investing into it. What are they gonna do? They're gonna obviously build their own kind of agentic technologies beyond chat GPT now. [31:36] Dave Pengelley: Um, but there's this, this, this tension between giving it enough information but doing so in a secure manner that is not gonna compromise your business. And that's [31:46] Richard Webbe: so you can, you can actually, I just thinking out loud, you can, the board can set up if they're smart enough. Go through their people setting up the agents and they can set up the culture of an organization by picking the characters in those agencies. [32:00] Richard Webbe: Yeah. Agents and making sure it stays within a boundary. We got a couple of comments from our audience. Um, uh, Angus asked or said that he's seeing more organizations experience AI incidents and failures 'cause people are rushing with poor training and poor deployment and I suppose. You know, we're, we're, we're sounding out how that might happen, thanks to Angus. [32:22] Richard Webbe: And then the other thing is, um, uh, Joanna sent a note in and she said she set up all her agents as names of their mentors, and now they talked to her personally. Uh, Joanna, they're not really people that, I'm glad you mentors are coming back. Joke. [32:38] Matt Slager: Couple of really good points for both of those. [32:40] Dave Pengelley: Yeah. Go Matt. [32:41] Matt Slager: The, um, [32:42] Richard Webbe: go [32:42] Matt Slager: the concept of like the, the persona. It, it just look at it as method acting. [32:46] Dave Pengelley: Mm. [32:47] Matt Slager: You know, you find the best method actor. [32:49] Richard Webbe: What a great [32:50] Matt Slager: state in the world. [32:51] Richard Webbe: What a [32:51] Matt Slager: great, literally you pick like everyone's got their favorites, but you pick the best method actor in the world and you give them a blank sheet of paper. [32:58] Matt Slager: They won't really know what to do. They won't know what you are expecting of them, but if you now give them, you know who they are, what they need to be, this is how you think, this is how you behave. All of a sudden now they can start to actually act and do their job, you know, and, and reveal to, you know, the, the vision that the director or whoever else is like, you know, calling upon. [33:19] Matt Slager: It's the same thing with prompting your agents. You know, that, that it's the same concept. You know, you give them instructions, persona, this is how you think, all of that kind of stuff. But with regards to the trust thing, trust is not binary. How many people do you know throughout your life that you have various levels of trust? [33:38] Matt Slager: You know, you, you'll trust them with your, you know, driving your car around the block, but possibly not hanging out with your spouse. You know, like that kind of thing. Um, so yeah, with organizations, they need to see trust as a ladder. Think about a new hire. You know, somebody who's brand new to the company. [33:55] Matt Slager: You don't give them access to everything. You know, it's [33:57] Richard Webbe: a really smart statement. I should apologize to Joanne as she wrote in and said there was no personality. So, uh, I was making a joke there, Joanna. So please. [34:05] Marno Brits: I feel like a good, a good approach that I've been using at Kholi. 'cause Kholi is quite a trade heavy business. [34:10] Marno Brits: Like it's all. Mining basically, mostly Kali. An analogy to compliment what Matt's saying about math acting. Um, to turn an AI into a team member, the approach I'm taking is teaching staff members and business owners to look at it like you're hiring an apprentice. 'cause that's what you're doing. You're bringing it a, a person into a specific role. [34:30] Marno Brits: So when you bring them in, you have to give them the task. So your job is to mop the floors, right? My expectation and context of you is like, you work for us. You show up on time. Your job is to mop the floors. This is the equipment that you'd be using. You're wearing our uniform. My expectation to you, that's the context. [34:47] Marno Brits: For example, and then the expectation would be you show up on time and if you have any problems, you talk to this person. Like it's taking that same approach is looking at it like a, an apprentice that you're bringing on board and expecting that they know nothing, but they're very capable. You just have to give them clear s instructions or clear direction. [35:03] Marno Brits: Again, what we talked about last week, Matt, it's just if you can manage a team, you can manage ai. Um, and it's just making that mindset shift to stop looking at it like it's Google or just a computer, but giving it a personality and treating it like something that has usefulness. [35:19] Dave Pengelley: Yeah. And, and these, these, these things, were not pouring them into a mold of hardened steel. [35:24] Dave Pengelley: These things are malleable and they grow and over time. And so like an employee, you give them feedback, Hey, I didn't really like the way you did that. Oh, you forgot to do this. Can you make sure you update your instructions? So next time you do this, or, oh, I've had this idea, I think, and I do this. I'm like, ah, I need to think about this thing for a client, um, conversation I'm having. [35:43] Dave Pengelley: Do we have someone who can do that or do we need to hire a new skill? And so then my orchestrator, Perry White, the editor in chief, he'll run through and, and, and look at the, look at the skills and go, actually, there's a gap there. We, we do need someone dedicated to focus on. [35:59] Richard Webbe: Does he yell at you? Does he scream at you when he tears [36:00] Dave Pengelley: it? [36:00] Dave Pengelley: No. No. What in the Sam Hill? Uh, no. We, we haven't, we haven't gone that far. Um. But, but, but I put the HR hat on, so this, my whole, I, I had a few personas for doing a few bits and pieces. I wasn't using the DC thing heavily for my business of a few other side projects. I'd started experimenting with using the DC personas, but I hadn't brought it into what I'm doing with valid agenda. [36:23] Dave Pengelley: And then on Friday I got it to build a new webpage for my site. I was like, okay, Felicity, go write a page. And she did. Well, no, I didn't have Felicity there. I just said, go write a page. And so it wrote me this page and the branding was horrific. Even though inside my system I had [[Design_Aesthetic_Guide]] guides and I had all these things structured to tell it what my brand look and feel is how to build the rest of the website. [36:47] Dave Pengelley: Like it knew, but it ignored it all. So I actually went into my system and said, do I have a skill that's focused on brand look and feel? And I said, no, you don't. At which point I was like. Instead of going and fighting this back and forth saying, no, read this file, do this, do that. I'm like, I need a worker who's gonna focus on doing that. [37:04] Dave Pengelley: And so I, I sort of switched modes, put my HR hat on, and built out like seven new agents in my team across web development brand and new I UX and, and built out the whole DC model into my own business. And that sort of was the, the, the genesis of me heading down what I call now anti claw. Because then I could say, okay, Jimmy, Jimmy Olson, photographer visuals, he's my visual guy. [37:28] Dave Pengelley: Um, I was like, Jimmy, go and review this page. It looks shocking. It's completely off brand. The color's wrong everything. Go to a full report, do a full tear down, and then work with Felicity to fix it. So he did and he came back with this full report going, there's a weird angle on that, and that colors the wrong color. [37:45] Dave Pengelley: Like he fully mapped it against it like an employee doing a proper tear down, like a ui, um, expert. So that was awesome. And then I'm like, ah, I haven't thought about SEO on my website. So. I can't remember my exact logic, but Batman became my SEO guy. Um, and so maybe 'cause he's a dete de he's a detective. [38:04] Dave Pengelley: Um, and so, um, I'm sorry. So I got Batman to do a tear down on my whole side and he said, yep, you're missing these meta tags here. You gotta do this. You wanna change that layout, you gotta add all this extra stuff in. Especially now that we're moving away from just SEO to a EO. And if the audience dunno what that means, that's agent engine optimization. [38:22] Dave Pengelley: There's a few different terms for it, but basically making sure when. Gemini and, and chat GPT and everyone index the web. They are actually getting not just bullet points, but they're getting, you know, sentences that they can read and think about and respond and use those in their answers. So you wanna be agent friendly, not just search engine friendly. [38:39] Dave Pengelley: Now, um, and he gave me this full tear down. I'm like, cool, that's great Work with Jimmy. If there's visual elements to make sure that we actually visualize them right, and then get Felicity to code them. And sometimes it's all backend meta tags. No Jimmy needed. Felicity will stick the tags in. Giving them these personas and giving them jobs to do and telling the jobs, the, the, the skills, the employees to do the job. [39:01] Dave Pengelley: I'm getting much better, much faster, much more reliable outcomes than just hooping the magic AI works out what I want. [39:09] Richard Webbe: You know that Dave, that the other guy's got stuff to say there. I just quickly as the non-technical, um, comic relief in this group, um, that makes me feel like that you're setting the tone and culture of your organization. [39:23] Richard Webbe: At a very senior level because you know, when you go to a company and I worked at a couple, the communications, the rules, the boundaries, the d it, they're all quite different. And you get the tone of an organization. So if I'm setting up agen behaviors inside my automation processes, I will as a human start to feel where right and wrong is and what the culture is. [39:47] Richard Webbe: Is that a fair call? You, you can set culture. Almost in reverse. [39:51] Dave Pengelley: Yeah. You, because you're defining like personality traits, archetypes, behaviors. Do this, don't do this. That's not okay. Know this. Forget that. Um, you are, yeah, you're, you're setting what the expectation is for how we do things in this workspace. [40:07] Richard Webbe: I'm gonna ask you to set up Christine Armour as one of my agents Dave, we're talking about when I come up. Um, I met her, she's, anyone remembers Christine Armour. She reported from two Gulf Wars and is one of the toughest journalists I ever met, and she's one of the most famous and highly ranked. And, um, I just happen to be lucky sitting next to her in the lounge in London before she flew off. [40:30] Richard Webbe: Uh, to the Gulf War, and, uh, I asked her a few questions. I've never met a tougher, more articulate individual in my entire life, and I want her on my team. [40:40] Dave Pengelley: Brilliant. [40:43] Richard Webbe: So go. Yeah. Sorry. Matt or, or, or Mano had something to say about that. [40:47] Matt Slager: I've got a, a parallel topic. If mine has a direct comment, hopeful man, go for it. [40:54] Matt Slager: Take away. That kind of concept of setting the culture and um, the expectations and how things perform and that sort of thing kind of makes me think of media and politics in the sense that they are already things that are completely shaped by narrative control. [41:14] Richard Webbe: Absolutely. [41:16] Matt Slager: And in a system where AI can scale infinitely, um. [41:22] Matt Slager: What does that look like? You know, at what level do we have perception attacks and levels of propaganda that just is gently sneaked under the rug by these models speaking to you. Especially once they, they start to know you and they start to know who you are and how you think, you know, at what point does that level of depth. [41:46] Matt Slager: You think about companies like Google and Meta who have a lot of ads data and that's why their, their ads system performs really well, but they don't know the same stuff that you tell Chatt PT, you know, at what point does that become that next frontier of [42:02] Richard Webbe: Absolutely. I mean, that's such an intriguingly clever point of view. [42:06] Richard Webbe: Matt, I'm glad you bring up obviously Sovereign ai. Is something we want to use to temper that. Correct me if I'm wrong, mano or anyone else. You know, we want to sort of, not necessarily like when you first get married, you don't wanna give away all your vulnerabilities at once or your life is over. So, um, so yeah, I mean, I, that's just a, an in intriguingly absolutely valid point of view. [42:28] Richard Webbe: Mano, did you wanna sort of jump in there? A [42:31] Marno Brits: hundred percent. I feel like it's a scary one. It reminds me of the, the thing we talked about a couple weeks ago with PewDiePie and what he achieved, which the camera there. Um, yeah, [42:41] Richard Webbe: the camera knows what you're gonna say it, so it blurred you [42:43] Marno Brits: out. Yeah. [42:46] Marno Brits: There we go. So yeah, the, the PewDiePie one where he created a, um, this mastermind of each GPU, each graphics card that he had, had a personality, and after a few months. He started, um, in order to self audit them, he said, if they're not performing, he deletes that person, then don't think he creates a new one. [43:03] Marno Brits: Um, and then they detected that if they don't perform, they get deleted. So then they all started conspiring. They, they started conspiring together against him even though he created them. And that's a small scale of like, okay, they understand what their task is and they start working together. But I feel like the point you were trying to make there, Matt, is like, how could. [43:23] Marno Brits: Larger corporations leverage AI and see what we're talking to them about and start talking to us through there and start marketing through there. And is that the point you were arguing there? Yeah. [43:33] Matt Slager: Literally the, the whole, the whole term, like the, that people need to start being super aware of is cognitive security. [43:41] Marno Brits: Yeah. [43:42] Matt Slager: Yep. [43:43] Marno Brits: Associating cogniti [43:43] Richard Webbe: security. What a great term. [43:45] Matt Slager: It's literally just the, the resistance to being gullible, I guess. [43:51] Marno Brits: Yeah, I mean it's, it's really cool to see, I mean, a lot of more people are gonna start getting to their own stuff. You can use open code and just have your own AI, like utilize Kimmy or an open source model. [44:03] Marno Brits: Um, and then instead of paying for Chacha PT or paying for Gemini, you'll just have your own locally hosted one with a little na. I mean, there's no real additional cost [44:14] Dave Pengelley: there. Depends on the size of the model. I did, I did start, um, doing some, uh, investigation on this today actually. And looking at if an organization, you know, had 500 employees and they wanted to run their own medal to to host something, you know, that might be, depending on if they're running agent teams, that might be $120,000 of hardware they need. [44:32] Dave Pengelley: That's a lot of tokens that maybe they're spending 1500 to two grand in tokens a month. When does the balance out? But as soon as you go to like enterprises, you're dealing with thousands of users. They're gonna be trillions of tokens, and then half a million dollars in a data center. Investment is nothing. [44:46] Dave Pengelley: For the, especially if you, if you take into account the opportunity cost and the productivity benefits that they're potentially getting by spending and investing them in that money in having, then they've got their own secure sovereign models that are running on their own hardware. So they're not paying ongoing token costs. [45:00] Dave Pengelley: Obviously there's it admin costs. The ROI, [45:02] Marno Brits: yeah, [45:03] Dave Pengelley: TCO conversation needs to come in there, but I think we're gonna see more of that as a big enterprise. We wanna adopt these models. Um, AWS came out with this big announcement, um, AWS bedrock. Are now gonna host all the frontier models in the Sydney data center for, so for Australian organizations with sovereignty issues where they can't have offshore data and so on. [45:22] Dave Pengelley: Now AWS are providing this thing and they're doing it with their zero operator access model. So no one from AWS will ever see any logs or your chats. Um, it's like fully on their mantle framework or something. So that way organizations can. Run models off AWS's bedrock instead of having to go to foreign data centers. [45:42] Dave Pengelley: So that's one step, but you're still paying full freight on the tokens. Anthropics still gonna gonna hit you for massive bills if you are going nuts on Opus. Whereas Opus, you can't host yourself. But Kimmy and these other op source models, which are advancing very quickly, you can host yourself, you could put on your own metal. [45:59] Marno Brits: Yeah, [45:59] Dave Pengelley: it's a good one in your own data center. [46:00] Richard Webbe: I always love your ROI and commercial approach to this, Dave, and as you know, that's close to my heart. An interesting point while people are running down and jumping to this, uh, and someone, oh, one of my guys from 24 7 just wrote, and they're gonna get to a point where their AI's learn from us and then switch us off instead of us turning them off. [46:19] Richard Webbe: But [46:19] Dave Pengelley: that's the matrix, right? [46:21] Richard Webbe: You know, let's go with the jokes, although they're good writers. But it's an interesting point in terms of cost. I worked briefly for uh, red Hat and of course Red Hat didn't make a model in selling open AI in a lot of other companies, they made a big business model out of selling protection from open source software. [46:41] Richard Webbe: Well, a lot of people don't realize is Red Hat sells support and in that support is buried in insurance thing, that if someone puts something illegal in that open source that you never found and you start using it, you could be sued. Right. We assume someone picked someone's code and then Telstra or someone of the big company's running it. [46:59] Dave Pengelley: Yeah, [46:59] Richard Webbe: I haven't paid this, uh, this, um, I call it, you know, almost blackmail money. You're open to being done and with so much new open source code coming into play. Unchecked. [47:10] Dave Pengelley: Mm-hmm. [47:11] Richard Webbe: Unrun, you're gonna really need to be on the ball with some solid partners that know it's happening to not only contain this, um, agentic runaway, if I could call it like that, as, as a, a possibility and getting severe bill shock, but also you're open to some legal issues in there. [47:27] Richard Webbe: Would that be fair? [47:28] Matt Slager: Mm-hmm. [47:30] Dave Pengelley: Yeah, [47:30] Matt Slager: definitely. Yeah, like I, um, I had a couple of questions written down just as thought ideas and, you know, when it comes to that sort of thing. When an AI speaks on behalf of your brand and institution, who's it? Who's accountable? If it says something wrong and you know, when it actually says something wrong, is it just following an instruction that somebody else has put in there, you know. [47:52] Richard Webbe: The blame game. The blame game. Who are we gonna blame for this one? That's what our courts do. But they'll all be on ai and AI will point the finger back at someone like Dave. [48:00] Dave Pengelley: Well, the, um, the, I hope not. Um, the o the open sourcing's a really important point, um, Richard, 'cause on all this open claw stuff, just to circle back where we started, the, there's a lot of people putting out open source skills and going, yeah, I'll give you this set of hands, I'll give you this, I'll sell you a hammer, I'll do this. [48:17] Dave Pengelley: So there's everyone going. Here's all the different things that you can enable your bot to do. Unfortunately, bad people exist that have bad intents. Yeah. And so there's a lot of people putting all these like, um, prompt injection type stuff, and they're putting security vulnerabilities into the hands that you are giving your agent. [48:35] Dave Pengelley: It's like, what was, what's the, the, the thing where the guy got the evil hand and the, like, he got a hand replacement and the hand was evil and the hand, um, it was a walking dead, like, uh, not walking dead. Um, Ash, uh, evil dead. Evil Dead had the, the evil hand that was, then he had to chop his own hand off because he had an evil hand that was trying to kill him. [48:52] Dave Pengelley: Um, so there's malicious actors out there, um, putting out these open source things. Here's a tool to enable your claw bot to go and book you a holiday. Oh, wow. That'd be awesome. I [49:03] Richard Webbe: love a holiday every, you're right Dave. Every new technology comes on with the guise. Oh, this is great. And see people jump on it. [49:11] Richard Webbe: Without thinking. Yeah. Next thing you're like, build the internet or your mobile phone, you get a text, please send me your password. Otherwise I'll turn your phone off. Someone wrote in and said Evil Dead too, Dave, just so you know. Yeah. [49:21] Dave Pengelley: Thank you. Yeah, [49:23] Richard Webbe: I got there eventually. Me and the audience are all bit worried about how much late night horror shows you're watching and DC comics during the day, get back to work, but, um, yeah, and, and that's, you know, that's the open source trap, isn't it? [49:33] Richard Webbe: 'cause what if I'm an open source coder? But I'm not very good at it. We know AI was written to write code. What if I got into AI and said, I want a little agent that does this, and Oh, that company over there did it. Could you go pinch some of their code and would you hide it in my little? And then I start selling those templates as a representation. [49:52] Richard Webbe: But I've now got a copyright breach. And you don't know it 'cause I've buried it. [49:57] Dave Pengelley: Well, the, the, the thing with the AI, and I'm, I'm hearing pro programmers in the space talk about this sort of stuff is you don't copy the code, you copy the functionality, and so you just point AI at the functionality and say, yeah, give me that thing. [50:10] Dave Pengelley: It goes, okay, and it doesn't need their code, it will just build its own version of it. And then Exactly. You don't have that, that, that same risk. And so the, the moat for software, um, vendors is shrinking in some ways. I think the build versus buy, there's always a complex argument that comes back to R-O-I-T-C-O. [50:27] Dave Pengelley: What's your business model? Do you really want to be managing your own software builds versus just buying stuff that's managed and has upgrades and a feature roadmap. We won't get into that right now, but that ability to, to just add features very quickly is becoming more and more accessible. So all of a sudden, like competitor a, they can do, they can do this new thing a day later with a, with a session in Claude code, with bang. [50:51] Dave Pengelley: So can we, [50:53] Richard Webbe: yeah. And then, then we get into the risk of speed versus risk. Yeah, I want this in the market done today. Uh, you know, you talk to someone like Steve Jobs, he wouldn't give a shit. He'd just say, do it. I'll suffer consequences later. Larger organizations say, nah, I'm too scared. And that's why they fall behind often because the the smart, nimble people like you guys get in and do it and get it quick and we'll solve that problem later on. [51:17] Dave Pengelley: It's also just the, the thing with scale, when you're bigger, there's more, more jobs, more money, more regulation, more tape, more things. You've gotta be aware of. You can't move as fast, which is why. Back 20 years ago, Salesforce came up as the, the young nippy, um, startup that was going up and, you know, sort of heckling Oracle conferences. [51:37] Dave Pengelley: 20 years later, they are the ones that are now sort of getting disrupted and having these young upstarts come at them because Salesforce has become the thing that they wanted to take down. And that just, that's the way the grow house, [51:48] Richard Webbe: the beautiful cycle of life. I love it. Hmm. [51:56] Matt Slager: On [51:56] Richard Webbe: Joe's question. Yeah, go for it. Matt, [52:00] Matt Slager: you like Richard. You mentioned at the beginning we were talking about personal productivity. You know why? Why at this point in time are people going ahead and starting to use their open claw agents for anything? Instead of using chat GBT. So that's what open AI will be expanding on. [52:18] Matt Slager: You know, why are you changing platforms? You know, why do you still have to-do list? Why are you using Apple reminders? Why are you using all of these other task management, calendar management, productivity management things when you can just do it through us and we'll track that for you. We'll make it really easy and like low friction. [52:34] Richard Webbe: Yeah, [52:35] Dave Pengelley: I mean, it frustrates me that [52:36] Richard Webbe: that statement. That's cool. [52:38] Dave Pengelley: The chat. GPD didn't already have that. So I remember almost a year ago, probably when they started bringing out connectors to things and I hooked up my Outlook account. So I was like, brilliant. All right. Um, build me a plan to do this now. [52:49] Dave Pengelley: Stick those in my calendar. Oh, sorry, I can only read your calendar. I can't write anything to your calendar. And I'm like. Why not? I've authorized, I've given you the access, I'm asking you to do it. Why? So the, they, they were halfway there and I think that's, there was a frustration and maybe from a security risk thing, they didn't want to have right. [53:05] Dave Pengelley: Access out things, right? And so they were like, we're read only. We're read only. And if you wanted to outward connect, then you had to go and build your own custom skills and news web hooks from N eight N and stuff and hack solutions, which you could do. But it was a pain and no one really was gonna do that. [53:20] Dave Pengelley: These tools make it more accessible. And now again, it's that, that whole early adopter cycle, right? Where that, that curve thing where we are getting out of early adopter into mass market, we're not quite there yet, but as people hear about agents, they hear about this, they go, oh, that'd be good. They're less scared of it, there's less fear and there's more adoption like with any technology curve. [53:39] Dave Pengelley: And I think open AI thinking now we've gotta get on this philanthropic, drop the ball by sort of dropping the hammer and saying, no, no claw bots. Uh, so they've gone. [53:49] Richard Webbe: Bad luck. We're gonna do it. [53:50] Dave Pengelley: But, but, but an anthrop are doing co-work and then they, they've got their own thing where they're trying to do desktop tasks and, and productivity stuff. [53:56] Dave Pengelley: So everyone's coming at their own way on this. [53:58] Richard Webbe: We had another, um, uh, look, it's very interesting. I, I was, I was on another podcast, uh, with my friend Darren Mitchell, and he talks about the exceptional sales leadership. And he's had that podcast for many years. We might get Darren on. So he's a pretty cool guy. [54:12] Richard Webbe: And we were talking about ethics in AI and that conversation went on for a while, and that leads into the legal issues and all of that. We just had a question come through. Um, uh, if, if AI's trained on a piece of proprietary code. Uh, maybe unintentionally or intentionally, it doesn't matter from a data breach. [54:31] Richard Webbe: Uh, and then it regurgitates that code without attribution. Is the company exposed for legal issues? My, I'm not a lawyer. This is not a statement. I'd like to apologize for any shit that comes outta, or any words that come out, but I think the answer to your question, I said I wouldn't swear, didn't I? [54:46] Richard Webbe: Indeed. Ah. Uh, I think they are open to legal issues because if the code can be regular, can be, you know, you follow the rabbit and you get there, but I think as Dave said earlier, most of that's gonna disappear because it'll be a copy of the functionality. Not the code. And I think we might see some legal issues coming in where people say, well, hang on, that's my idea. [55:08] Richard Webbe: And the whole new copyright and proprietary laws might need to be rewritten. [55:12] Dave Pengelley: We've been seeing this within the art community, right, with with all the AI generated art and all that was inspired by my thing and you stole all that. And so there's all this trademark and copyright issue happening in the. [55:22] Dave Pengelley: Sort of visual arts community and what the AI is doing there, to the point that I think Disney got ahead of it. And when Disney, um, uh, we're gonna sign up with open ai, um, people are gonna hack these things around and they're gonna make bootleg content of our characters. We might as well put it through some kind of channel where let's give them an avenue to do that, where we can police it a little bit. [55:42] Dave Pengelley: But like the, so that's, that's, that's, it's hard. I mean, in the space where they're not directly copying, but they're learning from things, if you go and read. 50 books about Picasso and study his brush strokes and his technique and everything and then start doing Picasso style pictures. Is that copyright of Picasso that you can draw a starry night as good as his when you did the homework and you learn from his styles and you learn from his [56:08] Richard Webbe: thing. [56:08] Richard Webbe: That was, you know, that was my question. Uh, artists musical and um, uh, you know, physical. [56:15] Dave Pengelley: Yeah, [56:15] Richard Webbe: they do get influenced. Learn from other artists [56:19] Dave Pengelley: how, how many musical genres are inspired by the Beatles, for example, and what are the Beatles? Were inspired by [56:24] Richard Webbe: Mozart and a few of the real ones. First, they're all taken from the classics, right? [56:31] Marno Brits: I feel like that's for the fun part, where it comes in of like, I think it only becomes a problem when people are making money, like open AI is making millions of, they're meant to be making millions. They're still struggling, struggling with their revenue, but if they're profiting from it, then it becomes a problem. [56:44] Marno Brits: I think it's very much the same in code. There's, from the technical standpoint, no, nothing is held as a precious thing in code. I mean, how many GitHub repositories have like the most unique and most like perfect use cases and people are just giving out for free because they're like, oh fuck, I had this problem. [57:03] Marno Brits: I spent a month trying to fix it. I finally fixed it. Yeah. So no one else has to suffer. Mm-hmm. It's this community approach that coders have. I feel like it's just, it's fine until someone makes money off of it and then it's like, oh crap, I could have made money giving my share. [57:18] Richard Webbe: Yeah, exactly. Yeah. Yeah. [57:20] Richard Webbe: It's, it's quite astounding. Um, how are we doing with time there, Dave? [57:23] Dave Pengelley: We, we, we are at the, the, uh, the top of the hour. So we we're gonna, we're gonna wrap it up. Um. Pretty much there, guys. Uh, thank you to all the chatters. Thank you for the questions and the interaction. Hopefully you, uh, got some insights or, or some new points of view you hadn't thought of, um, from your, from our ramblings and our chats and our banters. [57:40] Dave Pengelley: So thank you for, for that. Um, tell your friends, we are here every Wednesday midday talking about AI and tech trends and what's going on. We didn't even get a chance to talk about the new, um, Chrome MCP thing. Did you guys see that one? [57:52] Marno Brits: No, not yet. [57:53] Dave Pengelley: Web, web, MCP, Chrome, MCP. So Google has partnered with Microsoft and they've worked on a new open standard. [57:59] Dave Pengelley: So if you are building a booking website, a restaurant website, any website, part of the thing you can install in the metadata is MCP tooling. So when the agencies, these perplexity, browsers, et cetera, go to your website, they're not visually just trying to like look at it and go, oh, where's the buttons? [58:15] Dave Pengelley: And where'd the thing go? You can define all of that in the back end of your website. So when these bots go to it, they can now have the tooling go. Start date, end date filter, run that, get the data back, make the booking all through tooling instead of trying to pretend to be a human on the website. So Matt, your point around building interfaces for agents, they're now looking at how do we make every website an interface for an agent? [58:38] Dave Pengelley: Not just a human, so, which is just gonna further empower these, these personal agents, these open clause, et cetera, with greater ability to execute more effectively, more efficiently, more reliably. [58:50] Marno Brits: Mm-hmm. [58:51] Dave Pengelley: Than what we have today. [58:52] Marno Brits: I'm just shocked. You said Google's partnering of Microsoft? [58:55] Dave Pengelley: Well, yeah, they're the same. [58:56] Dave Pengelley: This isn't a Chrome specific thing. We want this to be an open web standard. So we need, edge is obviously the other primary, main browser. Yeah, sorry, Firefox and anyone else. Um, and most browsers are running chromium in the backend and they're all sort of cloned off chrome. So yeah, they, they've gone, we need to, wow, we need to make this an open web standard. [59:13] Dave Pengelley: So they're working like that. So that, that's exciting too. So they've just announced that in the last 24 hours I saw that pop up my feeds. So it's all rapidly moving and we're trying to move with it, uh, and chat about it and share it and let you all know. So thank you all. Thank you everyone. Thank you gentlemen. [59:28] Dave Pengelley: We'll see you on the next one and we'll see you back here midday next Wednesday. Thank you.