Telstra Glitched This Morning: DR, Backhoes, and Agents Up the Creek

[00:04] Matt: Look who decided to [00:05] Dave Pengelley: Mr. [00:05] Matt: turn up. [00:05] Dave Pengelley: Richard. [00:07] Richard Webbe: Mr. Dave, how are you, sir? [00:08] Dave Pengelley: Good, good, [00:09] Matt: Oh [00:09] Dave Pengelley: you've changed your [00:10] Matt: look, [00:10] Dave Pengelley: background. [00:10] Matt: it was important. [00:11] Richard Webbe: I [00:11] Matt: I'm really [00:11] Richard Webbe: have, [00:11] Matt: sorry, [00:11] Richard Webbe: I have [00:12] Matt: but it [00:12] Richard Webbe: actually. [00:12] Matt: I had [00:12] Richard Webbe: Yeah, [00:12] Matt: to get [00:12] Richard Webbe: um, [00:12] Matt: my [00:12] Dave Pengelley: You've [00:13] Matt: coffee [00:13] Richard Webbe: I'm [00:13] Dave Pengelley: got [00:13] Richard Webbe: uh [00:13] Matt: ready. [00:13] Richard Webbe: at a different abode at the [00:15] Dave Pengelley: Yeah, [00:15] Richard Webbe: moment, not [00:15] Dave Pengelley: I can see trees [00:16] Richard Webbe: the [00:16] Dave Pengelley: in [00:16] Richard Webbe: house. [00:16] Dave Pengelley: the glass behind you. It looks very, very, very uh [00:18] Matt: How are [00:19] Dave Pengelley: peaceful, [00:19] Matt: you both? Sorry, [00:20] Richard Webbe: Yes, [00:20] Dave Pengelley: serene. [00:20] Matt: I've [00:20] Richard Webbe: you [00:20] Matt: missed [00:20] Richard Webbe: can. [00:20] Matt: all of [00:20] Richard Webbe: It's [00:20] Matt: the [00:21] Richard Webbe: all actually [00:21] Matt: intro. [00:21] Richard Webbe: it's all AI. It's not no it [00:22] Dave Pengelley: How [00:22] Richard Webbe: is noties. [00:23] Dave Pengelley: how's the serenity? That's what I should be asking. [00:25] Matt: Amazing. [00:30] Richard Webbe: uh in uh in blackburn and it's uh it's very neat, I like it. [00:33] Dave Pengelley: Yeah, nice. Nice. I've I've [00:35] Matt: What [00:35] Dave Pengelley: also changed [00:36] Matt: is [00:36] Dave Pengelley: my background. [00:36] Matt: this? [00:36] Dave Pengelley: I changed [00:37] Matt: What's [00:37] Dave Pengelley: it from orange [00:37] Matt: this [00:37] Dave Pengelley: to [00:38] Matt: thing [00:38] Dave Pengelley: like a cyan [00:38] Matt: here? [00:38] Dave Pengelley: blue, [00:39] Richard Webbe: I [00:39] Dave Pengelley: just [00:39] Richard Webbe: can [00:39] Dave Pengelley: to [00:39] Richard Webbe: see [00:40] Dave Pengelley: just just [00:40] Richard Webbe: that. [00:40] Dave Pengelley: to match the theme of the [00:41] Matt: It [00:41] Dave Pengelley: podcast. [00:41] Matt: looks really [00:42] Richard Webbe: I [00:42] Matt: good. [00:42] Richard Webbe: can see that. I [00:43] Dave Pengelley: Yes. [00:43] Richard Webbe: can [00:43] Matt: I [00:44] Richard Webbe: see [00:44] Matt: do [00:44] Richard Webbe: that. [00:44] Matt: like [00:44] Dave Pengelley: And [00:44] Richard Webbe: Hey, [00:44] Matt: it. [00:45] Dave Pengelley: oh hello, sleepy head. [00:51] Dave Pengelley: Better late than never. [00:52] Matt: In Odyssey, I gotta add like thirty years. [00:57] Richard Webbe: I'm not sure what I'm [00:57] Dave Pengelley: Fair [00:57] Richard Webbe: saying. [00:57] Dave Pengelley: enough. Fair enough. I'll I'll All [01:00] Richard Webbe: Coffee's [01:00] Dave Pengelley: out. [01:00] Richard Webbe: important. [01:01] Matt: It did so have you seen it? Have you seen Odyssey already? [01:04] Dave Pengelley: No, we're just doing the banter now. [01:06] Matt: All [01:06] Dave Pengelley: This [01:06] Matt: right, [01:06] Richard Webbe: Was [01:06] Dave Pengelley: is it. You [01:06] Matt: yeah, [01:06] Richard Webbe: very [01:07] Dave Pengelley: joined [01:07] Richard Webbe: it's [01:07] Matt: yeah. [01:07] Dave Pengelley: right on time. [01:08] Richard Webbe: it was very boring. It was a very boring intro, so you [01:11] Dave Pengelley: We're [01:11] Richard Webbe: shouldn't miss [01:11] Dave Pengelley: discussing [01:11] Richard Webbe: any. [01:11] Dave Pengelley: our backgrounds that he's got [01:13] Matt: I [01:13] Dave Pengelley: trees in the [01:13] Matt: actually [01:13] Dave Pengelley: reflections. [01:14] Matt: I'm [01:14] Dave Pengelley: I've [01:14] Matt: really excited [01:14] Dave Pengelley: I've gone [01:14] Matt: for [01:14] Dave Pengelley: for [01:14] Matt: that [01:14] Dave Pengelley: a [01:15] Matt: movie. [01:15] Dave Pengelley: uh [01:15] Matt: I [01:15] Dave Pengelley: a [01:15] Matt: think [01:15] Dave Pengelley: podcast [01:15] Matt: it looks [01:16] Dave Pengelley: themed [01:16] Matt: good. [01:16] Dave Pengelley: blue-green instead of the uh the orange. So the deep stuff. [01:20] Richard Webbe: I'm [01:20] Matt: I'm the [01:21] Richard Webbe: at [01:21] Matt: same. [01:21] Richard Webbe: a [01:21] Matt: I d I was [01:22] Richard Webbe: but [01:22] Matt: resisting. [01:22] Richard Webbe: that's their [01:23] Dave Pengelley: That's [01:23] Richard Webbe: new [01:23] Dave Pengelley: that's [01:23] Richard Webbe: logo. [01:24] Dave Pengelley: the new logo. Yeah, [01:26] Richard Webbe: Yeah. [01:26] Dave Pengelley: like that? Yeah. Yeah. [01:29] Richard Webbe: And I should actually I should actually point out, Matt, your beard now reminds me a bit of Matt Damon in Odyssey. [01:34] Dave Pengelley: Mad Damon. [01:36] Richard Webbe: Yeah, he's he had to grow a beard like yours. [01:41] Richard Webbe: Well, without the gray, you're red, he's grey, but that's okay. [01:47] Richard Webbe: Now I've just seen a few shorts on it and stuff like that. So and I saw him interviewed this morning. So and when I saw you come on, I oh, he's got the Matt Damon look. [01:56] Dave Pengelley: Mad Damon. [02:00] Dave Pengelley: I can't ever see hear the word Matt Damon without saying the team America saw Matt Damon. [02:08] Richard Webbe: There [02:08] Dave Pengelley: Um [02:08] Richard Webbe: is a slogan we could say about Matt Diamond, but you're not late to say it. [02:12] Dave Pengelley: it was weird. I watched the British Grand Prix on the weekend, as I'm want to do with the F1, and they always have like some little opening monologue piece that there's before every race around you know the country or the heritage or something. They do some artsy thing. And it's the British Grand Prix, and they had Matt Damon doing the monologue, and I'm like Why get Matt, an American actor, to be doing the monologue at Silverstone for the British Grand Prix? Is there are there no actors that can do voiceovers in England? Like [02:41] Richard Webbe: He's it he's [02:41] Matt: You'd [02:42] Richard Webbe: Eddie [02:42] Matt: um [02:42] Richard Webbe: everywhere. [02:42] Matt: you might appreciate this, [02:43] Dave Pengelley: and [02:43] Matt: Richard. [02:44] Dave Pengelley: it [02:44] Richard Webbe: I [02:44] Dave Pengelley: must [02:44] Matt: But [02:44] Richard Webbe: do [02:44] Dave Pengelley: have been [02:44] Matt: back [02:44] Richard Webbe: remember. [02:44] Dave Pengelley: a tie-in [02:44] Matt: where [02:45] Dave Pengelley: with [02:45] Matt: I [02:45] Dave Pengelley: the [02:45] Matt: uh [02:45] Dave Pengelley: Odyssey [02:45] Matt: we [02:45] Dave Pengelley: release [02:46] Matt: I [02:46] Dave Pengelley: or [02:46] Matt: used [02:46] Dave Pengelley: something, [02:46] Matt: to [02:46] Dave Pengelley: and they're [02:46] Matt: live [02:46] Dave Pengelley: just like [02:46] Matt: in [02:46] Dave Pengelley: he's [02:47] Matt: Sydney, [02:47] Dave Pengelley: on the press junker [02:47] Matt: um [02:48] Dave Pengelley: tour, [02:48] Matt: at [02:48] Dave Pengelley: and [02:48] Matt: the [02:48] Dave Pengelley: they're [02:48] Matt: time [02:48] Dave Pengelley: like, [02:48] Matt: I [02:48] Dave Pengelley: Oh, [02:48] Matt: was actually [02:49] Dave Pengelley: can [02:49] Matt: working [02:49] Dave Pengelley: we get [02:49] Matt: at [02:49] Dave Pengelley: Matt [02:49] Matt: Bunnings, [02:49] Dave Pengelley: in? We want [02:50] Matt: Bunnings [02:50] Dave Pengelley: Matt [02:50] Matt: at [02:50] Dave Pengelley: and [02:50] Matt: Balgala. [02:50] Dave Pengelley: we want Matt people to see Matt [02:51] Matt: And [02:51] Dave Pengelley: everywhere. [02:52] Matt: uh [02:52] Dave Pengelley: So, I mean, he wasn't [02:53] Matt: that [02:53] Dave Pengelley: even in [02:54] Matt: that [02:54] Dave Pengelley: the F1 [02:54] Matt: particular [02:54] Dave Pengelley: movie. Brad [02:54] Matt: store [02:55] Dave Pengelley: was in [02:55] Matt: used [02:55] Dave Pengelley: the F1 [02:55] Matt: to get quite [02:55] Dave Pengelley: movie. [02:56] Matt: a few [02:56] Richard Webbe: Yeah, [02:56] Matt: names [02:56] Richard Webbe: good point. [02:57] Matt: in there, [02:57] Richard Webbe: Actually, [02:57] Matt: like [02:58] Richard Webbe: um [02:58] Matt: the [02:58] Richard Webbe: he [02:58] Matt: likes [02:58] Richard Webbe: does [02:58] Matt: of [02:58] Richard Webbe: enjoy [02:59] Matt: Brett [02:59] Richard Webbe: a [02:59] Matt: Lee. [02:59] Richard Webbe: good uh [03:00] Matt: Um [03:00] Richard Webbe: sort of you [03:01] Matt: Kirk [03:01] Richard Webbe: know uh [03:02] Matt: Pangilli. [03:02] Richard Webbe: candid appearance [03:03] Matt: I remember that one [03:03] Richard Webbe: and [03:04] Matt: pretty [03:04] Richard Webbe: uh a [03:04] Matt: firmly. [03:04] Richard Webbe: friend of mine who was [03:05] Matt: I [03:05] Richard Webbe: living [03:05] Matt: don't [03:05] Richard Webbe: in [03:05] Matt: know if [03:05] Richard Webbe: Byron [03:05] Matt: I ever [03:06] Richard Webbe: Bay [03:06] Matt: saw [03:06] Richard Webbe: and whose [03:06] Matt: Mr. [03:06] Richard Webbe: kids were [03:06] Matt: Abbott, [03:07] Richard Webbe: at school [03:07] Matt: but definitely [03:07] Richard Webbe: uh [03:08] Matt: some [03:08] Richard Webbe: with [03:08] Matt: of his [03:08] Richard Webbe: some [03:08] Matt: friends, [03:08] Richard Webbe: of the stars [03:09] Matt: Tony's. [03:09] Richard Webbe: and movie actors up there and someone played a joke and they left their school bag at home and they got Matt Damon to take the school bag into the kids at uh at school which blew the mind of the teacher and most of the kids at school. You know hi Matt Damon here's your school bag thank you very much if he walked so [03:35] Richard Webbe: No, well, I know, well, [03:40] Richard Webbe: I a good idea. that's think [03:49] Matt: So who uses Telstra? [03:54] Dave Pengelley: It's fun funny you mentioned Kirk, people ask me if I'm related to ENXS, and I'm like, no, completely different spelling. [04:01] Richard Webbe: I have seen Tony Abbott out in the surf at Manley because I had a lot of friends up in Manley, so I've surfed uh down Queenscliffe a lot, and he's well very well known on Long Manley Beach there for surfing and hanging out. [04:12] Dave Pengelley: In the budgies. [04:14] Richard Webbe: Oh, yeah, the budgies. Yes. [04:17] Dave Pengelley: Uh, I don't know now. Let's let's talk about uh computers. [04:35] Richard Webbe: Well, I am [04:36] Dave Pengelley: Not [04:36] Richard Webbe: a happy Tilstra [04:37] Dave Pengelley: mean. [04:37] Richard Webbe: customer. Uh actually I'm an everyone customer, and having worked with Tilstra for many years off and on, and and Optus and others. And this morning, yes, it was um it was a challenge. And uh I can see now the signal seems to be back up. [04:53] Dave Pengelley: Hmm. I'm I'm with I'm with Ali Mobile because their family plan is ridiculously well priced for a family. With multiple [05:00] Richard Webbe: This [05:00] Dave Pengelley: sim [05:00] Richard Webbe: is [05:00] Dave Pengelley: cards. I think it's [05:02] Richard Webbe: a [05:02] Dave Pengelley: Telstra [05:03] Richard Webbe: very good idea. [05:04] Dave Pengelley: and it did glitch a little bit, but Aldi must have fallbacks. They must have a tier two where if Telstra's down, they can fall back to Votor or Optus or someone because it did come good when I was while I was out for my walk. It took about two minutes, but then found a connection and I was back online. So I don't know what Aldi's back end is, but my wife um she woke up and there was a bit of calamity with her workplace because Telstra was down and she's like Telstra down and I I look up down detector and I could see just from five A.M. the little little chagger with complaints. [05:36] Richard Webbe: It is very funny. I had a friend of mine who was running mobiles at Telstra when uh, and let's face it, Optus, Photophone, Telstra, all of them have had outages. [05:44] Dave Pengelley: Mm-hmm. [05:44] Richard Webbe: Uh, and on one particular outage, and I was talking to him when I finally got a signal, and I said, You must be having a bad day. He said, You have no idea. And I said, But I was very good. I turned my phone off and on, and I got a signal again. And he said, [05:57] Matt: Yeah, [05:58] Richard Webbe: No, [05:58] Matt: that [05:58] Richard Webbe: don't [05:58] Matt: was [05:58] Richard Webbe: do [05:58] Matt: a [05:58] Richard Webbe: that. [05:58] Matt: like [05:59] Richard Webbe: So, [05:59] Matt: I saw [05:59] Richard Webbe: what [05:59] Matt: a [05:59] Richard Webbe: happens [05:59] Matt: lot of [05:59] Richard Webbe: if [05:59] Matt: people. [06:00] Richard Webbe: Of [06:00] Matt: talking [06:00] Richard Webbe: course, [06:00] Matt: about [06:00] Richard Webbe: when there's [06:00] Matt: Starlink [06:00] Richard Webbe: no signal, everyone [06:01] Matt: and [06:01] Richard Webbe: turns [06:02] Matt: um [06:02] Richard Webbe: their phone off [06:02] Matt: the [06:02] Richard Webbe: and on [06:02] Matt: fact [06:03] Richard Webbe: again. [06:03] Matt: that [06:03] Richard Webbe: And when it gets [06:04] Matt: you [06:04] Richard Webbe: hit [06:04] Matt: know [06:04] Richard Webbe: with that [06:04] Matt: they [06:04] Richard Webbe: many [06:04] Matt: didn't [06:05] Richard Webbe: requests [06:05] Matt: have to worry [06:05] Richard Webbe: to log [06:05] Matt: you [06:06] Richard Webbe: in, [06:06] Matt: know because [06:06] Richard Webbe: it brings [06:06] Matt: if they [06:06] Richard Webbe: the [06:06] Matt: didn't [06:07] Richard Webbe: system down [06:07] Matt: have [06:07] Richard Webbe: again. [06:07] Matt: this magical satellite connection [06:09] Richard Webbe: It's a [06:10] Dave Pengelley: Uh [06:10] Richard Webbe: bit of a [06:10] Matt: uh [06:10] Richard Webbe: challenge. [06:10] Dave Pengelley: yeah. [06:10] Matt: they're [06:11] Richard Webbe: I'm not sure [06:11] Matt: they [06:11] Richard Webbe: if that's [06:11] Matt: would [06:11] Richard Webbe: what's [06:11] Matt: have been [06:12] Richard Webbe: happening [06:12] Matt: screwed [06:12] Richard Webbe: now, [06:12] Matt: like [06:12] Richard Webbe: but yes. [06:12] Matt: literally [06:13] Dave Pengelley: I [06:13] Richard Webbe: I [06:13] Dave Pengelley: I [06:13] Matt: trains [06:13] Richard Webbe: didn't [06:13] Dave Pengelley: c I [06:14] Richard Webbe: change [06:14] Dave Pengelley: capitalized [06:14] Matt: were [06:14] Richard Webbe: my [06:14] Matt: were [06:14] Richard Webbe: phone. [06:14] Dave Pengelley: on [06:14] Matt: completely [06:14] Dave Pengelley: Vodafone's [06:15] Matt: stopped [06:15] Dave Pengelley: paying [06:15] Matt: obviously [06:15] Dave Pengelley: many years [06:16] Matt: payment [06:16] Dave Pengelley: ago when I was [06:16] Matt: processes [06:16] Dave Pengelley: working at Telstra [06:17] Matt: you know [06:17] Dave Pengelley: and [06:17] Matt: people [06:17] Dave Pengelley: managed [06:17] Matt: using [06:17] Dave Pengelley: to [06:17] Matt: F [06:17] Dave Pengelley: convert a [06:18] Matt: Plus [06:18] Dave Pengelley: bunch of accounts [06:18] Matt: machines [06:18] Dave Pengelley: off [06:18] Matt: that [06:18] Dave Pengelley: Vodafone [06:18] Matt: rely on Telstra [06:19] Dave Pengelley: Mobile over to [06:20] Matt: network [06:20] Dave Pengelley: Telstra Next G. That was good times. [06:23] Richard Webbe: Well, here's the amazing thing, of course, isn't it? Now that we have voice over IP as a backup for a lot of mobile. phones and if you select that on your phone and you've still got a Wi-Fi connection like I have here, I was still able to call my children this morning and communicate with people and not worry too much about it. [06:48] Matt: Hmm. It's interesting. Makes you really think about hidden dependencies, you know, stuff that you don't [06:59] Matt: have. [07:01] Dave Pengelley: Yeah, I think that's a good [07:03] Richard Webbe: Well, [07:03] Dave Pengelley: one. [07:03] Richard Webbe: the big part of the trains, and I'm surprised that a large part of the power industry didn't shut down because all the old SCADA and communications and that was running for a while on 4G. And I know a number of organizations that have recently upgraded their networks to 5G. And of course, it was the 5G that was down. And so I'm surprised the trains in the um the trains in uh regional Victoria were down, um, but I'm surprised the power [07:33] Dave Pengelley: Yeah. [07:38] Richard Webbe: Critical [07:38] Dave Pengelley: Well, I [07:38] Richard Webbe: infrastructure. [07:39] Dave Pengelley: and [07:39] Richard Webbe: It's [07:39] Dave Pengelley: I [07:39] Richard Webbe: all about [07:40] Dave Pengelley: mean [07:40] Richard Webbe: critical [07:40] Dave Pengelley: uh [07:40] Richard Webbe: infrastructure. [07:41] Dave Pengelley: if you're not using a a third-party payments platform like Square, if you're using one of the the built-in hardware ruggedized F POS things you get off Quest or or Combank or any of those, they've often got built-in SIM cards. And so part of your monthly fee of having the F POS machine is their IoT SIM card back to whatever. network and so a lot of F POS you don't have a choice who you're with. You just bought an F POS device and if Telstra goes down, then your F POS is down. [08:08] Richard Webbe: Yep. Yep. Yep. So we our sovereign data, our sovereignty to rely on ourselves from supply chain and everything. And I think this is a big impact to AI. We need sovereign AI instances as well to keep ourselves both in our organizations and in our country. We need to reevaluate all the critical infrastructure and make sure it's well connected. And the way you're back out. [08:31] Dave Pengelley: This this this comes down to the thing that most businesses of of all sizes don't really think about much until it's too late is the whole DR to uh the disaster recovery planning. No, I've been terrible at it when I've been responsible for IT and in previous small businesses and jobs, and it's caused me no end to pain. And then I go, oh, I should do better next time, even with my own personal setup as far as backups and things. Probably not fully DR uh resilient. But also I think the way we use AI and the way these systems, it depends on. How business critical is your AI thing? We've discussed this in the past. Like, do you need fallbacks? Is this a business critical process? If one model provider is down, do you have a fallback to a second model provider for those business critical things? How much do you trust those? How much stuff are you putting through agents that shouldn't be through agents? There should just be deterministic workflows that [09:18] Richard Webbe: Yes, [09:18] Dave Pengelley: don't [09:18] Richard Webbe: exactly. [09:19] Dave Pengelley: rely on all the AI stuff. Um [09:22] Richard Webbe: I remember [09:22] Dave Pengelley: yeah. [09:22] Richard Webbe: going back years ago, uh we seemed to embrace technology, which improves our performance and everyone for Gets the mission criticality of it, whether it's power, the internet data, your laptop, access to data, computer, it's just amazing. [09:38] Dave Pengelley: When when when I started my my business ventures um nearly a year ago, trying to like take what I've learned and take what I do about AI and turn that into something I could present to the market and and sell services and training and education for, uh, I was for the couple of weeks of planning and developing it, I was using Chat GPT, like it was nobody's business. I had a project folder. I was Deep into it. I was having daily chats with it for hours while I was out for my walks, just brainstorming and whiteboarding, thinking about pricing strategies and everything. Like it was all in there. And then when I decided, ah, if I'm going to be talking to business clients, I should know what Chat GPT business is. So I upgraded from my personal account to a business account. And in the transition over, all of my stuff vanished for like an hour or two. And I fully freaked out because I'm like, I've tried I've just lost weeks of work. Weeks of work. Because it was all dependent on this one cloud provider. Now it did eventually migrate through and came back. But then I was also then I was like, that's that's great, that's come back. Uh and then two weeks later or a month later, I accidentally accidentally deleted the entire project folder instead of the one chat that I meant to delete and lost everything anyway. But fortunately, by that point, and like that week, three days earlier, I'd actually had it summarize all our chats now put like a manifesto document that I'd put into Notion because I was trying to think about Out, like you, where am I storing some of this stuff? How do I consolidate all of my thinking and ideas for [11:04] Matt: I I [11:05] Dave Pengelley: backup [11:05] Matt: I agree with [11:06] Dave Pengelley: or [11:06] Matt: it. [11:06] Dave Pengelley: different [11:06] Matt: I think [11:06] Dave Pengelley: platform [11:06] Matt: it's a good [11:06] Dave Pengelley: usage [11:07] Matt: thing. [11:07] Dave Pengelley: purposes? [11:07] Matt: And [11:07] Dave Pengelley: And because [11:07] Matt: I [11:08] Dave Pengelley: I [11:08] Matt: think [11:08] Dave Pengelley: was [11:08] Matt: too [11:08] Dave Pengelley: experimenting [11:08] Matt: many [11:08] Dave Pengelley: with [11:08] Matt: people [11:08] Dave Pengelley: Notion [11:09] Matt: are kind [11:09] Dave Pengelley: as [11:09] Matt: of [11:09] Dave Pengelley: a platform [11:09] Matt: blind at [11:09] Dave Pengelley: versus [11:10] Matt: the [11:10] Dave Pengelley: just [11:10] Matt: idea [11:10] Dave Pengelley: relying on [11:10] Matt: of [11:10] Dave Pengelley: Chat GPT, [11:11] Matt: just [11:12] Dave Pengelley: that saved [11:12] Matt: it'll [11:13] Dave Pengelley: my [11:13] Matt: be [11:13] Dave Pengelley: bacon [11:13] Matt: fine. [11:13] Dave Pengelley: and I was able to [11:14] Matt: You [11:14] Dave Pengelley: continue [11:14] Matt: know, [11:14] Dave Pengelley: working. [11:14] Matt: like [11:15] Dave Pengelley: But [11:15] Matt: don't [11:15] Dave Pengelley: nowadays, [11:15] Matt: worry about it. [11:16] Dave Pengelley: now that [11:16] Matt: But [11:16] Dave Pengelley: I'm [11:16] Matt: that [11:17] Dave Pengelley: running [11:17] Matt: that [11:17] Dave Pengelley: off [11:17] Matt: whole [11:17] Dave Pengelley: my machine [11:18] Matt: disaster [11:18] Dave Pengelley: and it's all in [11:18] Matt: recovery, [11:18] Dave Pengelley: sort of Hermes [11:19] Matt: like no one [11:19] Dave Pengelley: agents [11:19] Matt: wants to [11:20] Dave Pengelley: and [11:20] Matt: really [11:20] Dave Pengelley: markdown [11:20] Matt: think about [11:20] Dave Pengelley: files [11:20] Matt: that. It's [11:21] Dave Pengelley: everywhere, [11:21] Matt: kind of like [11:22] Dave Pengelley: I have [11:22] Matt: it's [11:22] Dave Pengelley: a [11:22] Matt: kind [11:22] Dave Pengelley: lot [11:22] Matt: of like [11:22] Dave Pengelley: less [11:23] Matt: prenups. [11:23] Dave Pengelley: stress about [11:24] Matt: You [11:24] Dave Pengelley: any [11:24] Matt: know, [11:24] Dave Pengelley: of that [11:24] Matt: you don't [11:25] Dave Pengelley: because [11:25] Matt: you don't really want to have [11:26] Dave Pengelley: the [11:26] Matt: to go [11:26] Dave Pengelley: stuff's [11:26] Matt: through the [11:26] Dave Pengelley: committed [11:26] Matt: process [11:27] Dave Pengelley: into [11:27] Matt: of [11:27] Dave Pengelley: a [11:27] Matt: thinking [11:27] Dave Pengelley: repo, [11:27] Matt: about [11:27] Dave Pengelley: it's backed [11:28] Matt: it. [11:28] Dave Pengelley: up somewhere [11:28] Matt: Yeah. [11:29] Dave Pengelley: semi-regularly. It's [11:30] Matt: the [11:30] Dave Pengelley: not like gonna [11:30] Matt: the [11:30] Dave Pengelley: be the end of [11:31] Matt: the [11:31] Dave Pengelley: the world. [11:31] Matt: the thing [11:32] Dave Pengelley: And [11:32] Matt: that i [11:32] Dave Pengelley: I've [11:33] Matt: would [11:33] Dave Pengelley: switched [11:33] Matt: say [11:33] Dave Pengelley: agent [11:34] Matt: though [11:34] Dave Pengelley: platforms [11:34] Matt: is [11:34] Dave Pengelley: five [11:35] Matt: like [11:35] Dave Pengelley: times since [11:35] Matt: if you've [11:35] Dave Pengelley: then. [11:36] Matt: just like [11:36] Dave Pengelley: And [11:36] Matt: your [11:36] Dave Pengelley: so [11:36] Matt: experience [11:37] Dave Pengelley: just [11:37] Matt: there [11:37] Dave Pengelley: my [11:37] Matt: dave [11:37] Dave Pengelley: own [11:37] Matt: like when [11:38] Dave Pengelley: reliance [11:38] Matt: those things [11:38] Dave Pengelley: on [11:38] Matt: vanished [11:38] Dave Pengelley: a single [11:39] Matt: and you [11:39] Dave Pengelley: platform, [11:39] Matt: noticed [11:39] Dave Pengelley: single [11:39] Matt: everything [11:39] Dave Pengelley: point [11:39] Matt: was [11:40] Dave Pengelley: of [11:40] Matt: gone [11:40] Dave Pengelley: failure [11:40] Matt: you know [11:40] Dave Pengelley: has [11:40] Matt: for [11:41] Dave Pengelley: just [11:41] Matt: anyone listening anyone [11:42] Dave Pengelley: amorphized. [11:42] Matt: like watching [11:43] Dave Pengelley: Is that a word? So much in the last [11:44] Matt: if [11:44] Dave Pengelley: six months [11:44] Matt: something [11:44] Richard Webbe: Sounds [11:45] Dave Pengelley: as [11:45] Matt: was to [11:45] Richard Webbe: good. [11:45] Dave Pengelley: crazy. [11:45] Matt: disappear [11:45] Richard Webbe: I'll check [11:46] Matt: like [11:46] Richard Webbe: it. [11:46] Matt: pick [11:46] Richard Webbe: Yep. [11:46] Matt: the thing that would if it disappeared what would it you know how would it ruin your day like you know what which thing would ruin your day the fastest you know and is there an alternative can you have like a diversion you know if Do you usually cross the street in one specific spot? And if you've moved like somewhere else, is that going to be a big problem? Um I I try and profess to test every single agent harness. Like it's just something that I've recently been doing. And, you know, as many models as I possibly can, as many subscriptions as I can afford. Um, but one thing that I keep coming back to, because somebody asked me recently, like, how do you, you know, I heard you were using Cursor recently. How did you pivot? your workflows to use with that, you know, when you're usually usually using something else. And my answer was I didn't need to because none of my workflows are like intrinsically tied to anything in particular. So it's really, really important to to keep that awareness and not be trapped. Like I don't I don't really know how else to say that. [13:05] Matt: Yeah. [13:18] Dave Pengelley: Yeah. [13:30] Matt: you you use the term fallback. You know, you have your fallback options, um, you know, fail, fail closed, you know, all those different [13:36] Richard Webbe: I [13:36] Matt: bits [13:36] Richard Webbe: would [13:36] Matt: and [13:37] Richard Webbe: hasten [13:37] Matt: pieces. [13:37] Richard Webbe: to say [13:38] Matt: But for [13:38] Richard Webbe: that [13:38] Matt: these [13:38] Richard Webbe: people [13:38] Matt: people [13:38] Richard Webbe: who [13:39] Matt: that [13:39] Richard Webbe: are [13:39] Matt: don't [13:39] Richard Webbe: now embedding [13:39] Matt: really know [13:40] Richard Webbe: AI [13:40] Matt: exactly [13:40] Richard Webbe: and [13:40] Matt: what [13:40] Richard Webbe: AI [13:41] Matt: they're [13:41] Richard Webbe: agents [13:41] Matt: doing yet, [13:41] Richard Webbe: in their [13:42] Matt: you know, [13:42] Richard Webbe: business [13:42] Matt: like they're [13:42] Richard Webbe: processes [13:42] Matt: having a go. They're [13:43] Richard Webbe: haven't [13:43] Matt: using Claude, [13:43] Richard Webbe: tested it [13:44] Matt: they're using [13:44] Richard Webbe: if [13:44] Matt: Codex, [13:45] Richard Webbe: their AI [13:45] Matt: you know, [13:45] Richard Webbe: supplier [13:45] Matt: they're building [13:46] Richard Webbe: or [13:46] Matt: things, [13:46] Richard Webbe: LLM [13:46] Matt: the stuff's [13:47] Richard Webbe: goes [13:47] Matt: working, [13:47] Richard Webbe: down. [13:48] Matt: surprisingly, [13:49] Richard Webbe: I don't [13:49] Matt: but [13:49] Richard Webbe: think [13:49] Matt: you [13:49] Richard Webbe: they've [13:50] Matt: know, [13:50] Richard Webbe: tested [13:50] Matt: there's [13:50] Richard Webbe: it. [13:50] Matt: those [13:50] Richard Webbe: So what'll happen [13:51] Matt: extra [13:51] Richard Webbe: is what [13:52] Matt: depth [13:52] Richard Webbe: normally [13:52] Matt: layers [13:52] Richard Webbe: produces all [13:53] Matt: that [13:53] Richard Webbe: their [13:53] Matt: they [13:53] Richard Webbe: invoices [13:53] Matt: haven't, [13:54] Richard Webbe: or [13:54] Matt: you [13:54] Richard Webbe: redoes [13:54] Matt: know, [13:54] Richard Webbe: their [13:54] Matt: gathered [13:54] Richard Webbe: redoes [13:55] Matt: awareness [13:55] Richard Webbe: their pricing [13:55] Matt: of yet. [13:55] Richard Webbe: or [13:56] Matt: You [13:56] Richard Webbe: their [13:56] Matt: know, [13:56] Richard Webbe: scheduling [13:56] Matt: are those people [13:56] Richard Webbe: or [13:56] Matt: going [13:57] Richard Webbe: whatever [13:57] Matt: to get [13:57] Richard Webbe: on [13:57] Matt: caught [13:57] Richard Webbe: the fly, [13:58] Matt: out, you [13:58] Richard Webbe: when [13:58] Matt: know, [13:58] Richard Webbe: one [13:58] Matt: if [13:58] Richard Webbe: of the [13:59] Matt: something [13:59] Richard Webbe: suppliers [13:59] Matt: completely [14:00] Richard Webbe: Whether [14:00] Matt: Shuts [14:00] Richard Webbe: it's the [14:00] Matt: off. [14:00] Richard Webbe: network [14:01] Matt: Like [14:01] Richard Webbe: or the [14:01] Matt: I've [14:02] Richard Webbe: LLM [14:02] Matt: seen a lot of businesses, [14:02] Richard Webbe: that goes [14:03] Matt: like a lot of [14:03] Richard Webbe: down, [14:03] Matt: SaaS [14:04] Richard Webbe: like [14:04] Matt: tools [14:04] Richard Webbe: today, [14:04] Matt: that are actually [14:04] Richard Webbe: if [14:04] Matt: building [14:04] Richard Webbe: they've [14:05] Matt: around [14:05] Richard Webbe: got mobile [14:05] Matt: Fable, [14:05] Richard Webbe: LLM access, [14:06] Matt: you know, the [14:07] Richard Webbe: that'll [14:07] Matt: Claude [14:07] Richard Webbe: be up [14:08] Matt: model [14:08] Richard Webbe: the [14:08] Matt: Fable. [14:08] Richard Webbe: creek. [14:08] Matt: And you know, I don't think they'll turn Fable [14:10] Richard Webbe: Big issues [14:11] Matt: off [14:11] Richard Webbe: to think [14:11] Matt: again [14:12] Richard Webbe: about. [14:12] Matt: for the foreseeable future, but it's possible. And does that mean that that particular tool all of a sudden is now GIMPed and you know you can't get the same output that you got before? It's it's important. It's really, really important. It's like this Telstra switch off. Yeah, it's gonna happen. [15:09] Richard Webbe: Single point of failure requires a very, very uh uh specific set of eyes, uh, people that are experienced. So I remember I was working on a project, funny enough, with Telstra and a large government department once, and um they this government department decided well they were going to get their data supply from both uh Telstra. At the time, so it was Telstra and Optus to protect a single point of faith for the data. And so they bought the Optus one other, but they were both running through the same pit. So when the backhoe went through the cable, they had no backup. And it's a it's a it's a classic that always reminds me of the classic single point of failure mistake. Two separate suppliers, but they're running, as you would say, the same point on the road. And when the backhoe went through the uh through the pit and the cables, everyone was screwed. And with NBN as Our primary supplier for a lot of telco, we still have that single point of failure hovering in the background. [16:06] Dave Pengelley: Yeah. Yeah, and you can never erase all risk. Like there's always going to be risk. It's like how critical is the risk and it what is the cost to like offset, like to maintain a backup option, right? Like if something goes down, can you withstand it being down for a minute, an hour, a day, a week? And if not, and like if the if the cost becomes too high, at which point do you need to have those mitigation things? And what's the cost of maintaining the backup? system. Like if you're air traffic control, yes, you're gonna have redundant systems on redundant systems and things like that, because that's very critical. But if you're, you know, a small medium business running some some systems and invoicing, does it matter if you can't invoice for a day, for a week? Like where where does that hit a point where you go, you know what, we actually need a backup plan that we can roll over to inflict to. Like if you're running a SaaS platform, you probably don't want to be down for more than an hour or so. Like if if you've superbase goes down, what's What's your backup there? Um, are you running on a different instance of Superbase in a different region or on a different database platform altogether? [17:08] Matt: It's actually [17:09] Dave Pengelley: Um, if your APIs and [17:10] Matt: It's [17:10] Dave Pengelley: workers [17:11] Matt: pretty [17:11] Dave Pengelley: are through [17:11] Matt: hard. [17:11] Dave Pengelley: um [17:12] Matt: Like, you know, it's a pretty [17:12] Dave Pengelley: Railway, [17:13] Matt: difficult thing [17:13] Dave Pengelley: like and [17:13] Matt: to [17:14] Dave Pengelley: we we saw [17:14] Matt: set [17:14] Dave Pengelley: Railway [17:14] Matt: up. [17:15] Dave Pengelley: had [17:15] Matt: Like [17:15] Dave Pengelley: the adage [17:15] Matt: when [17:15] Dave Pengelley: the [17:15] Matt: that [17:15] Dave Pengelley: other [17:15] Matt: railway [17:15] Dave Pengelley: week, right, [17:16] Matt: thing [17:16] Dave Pengelley: Matt? [17:16] Matt: went down, [17:16] Dave Pengelley: Or uh [17:17] Matt: I [17:17] Dave Pengelley: Cloudflare [17:17] Matt: I started looking [17:18] Dave Pengelley: or [17:18] Matt: into [17:18] Dave Pengelley: whoever [17:18] Matt: alternate [17:18] Dave Pengelley: it is, [17:19] Matt: providers [17:19] Dave Pengelley: like [17:19] Matt: so [17:19] Dave Pengelley: what [17:19] Matt: that I [17:19] Dave Pengelley: is [17:20] Matt: could [17:20] Dave Pengelley: how [17:20] Matt: have [17:20] Dave Pengelley: long [17:20] Matt: uh [17:20] Dave Pengelley: can [17:20] Matt: like [17:20] Dave Pengelley: you afford [17:20] Matt: a failover [17:21] Dave Pengelley: to be down and [17:21] Matt: even if [17:21] Dave Pengelley: how [17:22] Matt: it's [17:22] Dave Pengelley: much [17:22] Matt: manual. [17:22] Dave Pengelley: should you think [17:23] Matt: And [17:23] Dave Pengelley: about the [17:23] Matt: just [17:23] Dave Pengelley: customer [17:24] Matt: the idea [17:24] Dave Pengelley: impact [17:24] Matt: of, okay, the [17:25] Dave Pengelley: and [17:25] Matt: light [17:25] Dave Pengelley: the [17:25] Matt: switch [17:26] Dave Pengelley: cost [17:26] Matt: turned [17:26] Dave Pengelley: of [17:26] Matt: off, [17:26] Dave Pengelley: maintaining [17:26] Matt: let me go [17:27] Dave Pengelley: redundancies? [17:27] Matt: to the other room [17:27] Dave Pengelley: Like [17:27] Matt: and [17:28] Dave Pengelley: my [17:28] Matt: turn [17:28] Dave Pengelley: I'm [17:28] Matt: this [17:28] Dave Pengelley: building [17:28] Matt: switch [17:28] Dave Pengelley: my own SASN [17:28] Matt: on and everything [17:29] Dave Pengelley: at the moment. [17:29] Matt: keeps running. [17:30] Dave Pengelley: It's [17:30] Matt: You [17:30] Dave Pengelley: fairly [17:30] Matt: know, [17:30] Dave Pengelley: single [17:30] Matt: that kind [17:31] Dave Pengelley: string [17:31] Matt: of thing, [17:31] Dave Pengelley: because [17:31] Matt: even [17:31] Dave Pengelley: it's [17:31] Matt: though [17:31] Dave Pengelley: that's [17:31] Matt: it [17:32] Dave Pengelley: a [17:32] Matt: was manual, [17:32] Dave Pengelley: scaling [17:32] Matt: was [17:32] Dave Pengelley: SaaS [17:32] Matt: very [17:33] Dave Pengelley: and [17:33] Matt: difficult [17:33] Dave Pengelley: doesn't have [17:33] Matt: to [17:33] Dave Pengelley: a huge [17:33] Matt: set [17:34] Dave Pengelley: number of [17:34] Matt: up. [17:34] Dave Pengelley: customers yet. The [17:35] Matt: So, [17:35] Dave Pengelley: blast radius of a of a down is very fairly small. But you know, you get a thousand, you get ten thousand, you get a hundred thousand customers on your on your SaaS platform, all of a sudden that uptime metric becomes a lot more important. Uh, and you need [17:48] Matt: I'm know, not sure. you [17:48] Dave Pengelley: to invest in those those automatic failovers and redundancy systems. [17:51] Richard Webbe: And you can't you can start. [18:00] Dave Pengelley: Hmm. [18:17] Dave Pengelley: Yeah. The old DNS file over and all that kind of stuff, just that's it's a lot. [18:23] Richard Webbe: I remember [18:23] Matt: Jeez. [18:23] Richard Webbe: I put in one of the largest networks in Australia for uh one of Australia's Largest companies. And uh we used to talk instead [18:33] Matt: All [18:33] Richard Webbe: of [18:34] Matt: right. [18:34] Richard Webbe: disaster recovery, it would be called business continuity, [18:36] Dave Pengelley: Yeah. [18:36] Richard Webbe: same thing, different [18:37] Dave Pengelley: Yeah. [18:37] Richard Webbe: words. And uh we had a uh uh uh uh oh god, I've just forgotten the phrase, the non non-failable power supply backup on every floor, right? Yeah, that's it. Thank you. So I was looking for the term UPS, thank you, and it an interrupted power supply. And they spent millions on this, and they had one on every floor that was supplying the power to the network. in case of a power failure. And they had a power failure and all 29 floors didn't work. [19:08] Richard Webbe: I I to this day I still don't know why. I do remember taking the phone call. [19:15] Dave Pengelley: Oh yeah. [19:17] Richard Webbe: Luckily the power came out quicker than we could isolate or triage the full time I began interruptible power supplies were completely interrupted. I shouldn't laugh. But it was devastating. And like you You guys are saying you can't be absolutely certainly [19:33] Dave Pengelley: Yeah. And and and not every risk is worth mitigating. I mean, they they do this with highway speeds. They say, you know, the productivity benefits of letting [19:41] Richard Webbe: able be do going not that. to to [19:41] Dave Pengelley: you drive at 80 kilometers an hour or 100 kilometers an hour on that long stretch of road is worth the increased risk of accidents. Um, versus, you know, like in a pedestrian zone with children and and stuff, they go, no, no, you have to drive at 40 during school hours because the we the perceived impact of any accident. At that time with small children is huge, and we don't want that. So you know they risk manage speed limits for your for consequences. [20:07] Richard Webbe: Absolutely. [20:09] Dave Pengelley: Uh and it's yeah, it's the same with all this AI stuff and in computing in general. [20:13] Matt: Yeah, like [20:13] Dave Pengelley: What [20:14] Matt: the [20:14] Dave Pengelley: is your [20:14] Matt: easiest [20:14] Dave Pengelley: risk tolerance [20:14] Matt: way, depending [20:15] Dave Pengelley: and where [20:15] Matt: on your actual [20:16] Dave Pengelley: like why why do [20:17] Matt: inference [20:17] Dave Pengelley: you but why do you pay [20:18] Matt: provider, [20:18] Dave Pengelley: for health insurance [20:18] Matt: like your [20:18] Dave Pengelley: if you [20:19] Matt: AI [20:19] Dave Pengelley: if you pay for [20:19] Matt: endpoint [20:19] Dave Pengelley: health insurance? Why do you [20:20] Matt: of [20:20] Dave Pengelley: pay for car [20:20] Matt: choice, [20:20] Dave Pengelley: insurance? [20:21] Matt: that [20:21] Dave Pengelley: Uh [20:22] Matt: that [20:22] Dave Pengelley: it's [20:22] Matt: usually [20:22] Dave Pengelley: because [20:22] Matt: has its own shape, [20:23] Dave Pengelley: you're [20:23] Matt: its [20:23] Dave Pengelley: managing [20:24] Matt: own fallback [20:24] Dave Pengelley: risk. [20:24] Richard Webbe: Let [20:24] Matt: system. [20:25] Richard Webbe: me ask you [20:25] Matt: So [20:25] Richard Webbe: guys a [20:25] Matt: all [20:26] Richard Webbe: question [20:26] Matt: the systems [20:26] Richard Webbe: of [20:26] Matt: that [20:26] Richard Webbe: ignorance [20:26] Matt: I build [20:27] Richard Webbe: issues [20:27] Matt: for [20:27] Richard Webbe: all coming [20:28] Matt: production [20:28] Richard Webbe: from me. [20:28] Matt: applications, [20:28] Richard Webbe: Um so [20:29] Matt: I go [20:29] Richard Webbe: I'm [20:29] Matt: directly. [20:29] Richard Webbe: running [20:30] Matt: through [20:30] Richard Webbe: A [20:30] Matt: open [20:30] Richard Webbe: complicated [20:30] Matt: router all the time. [20:31] Richard Webbe: supply [20:32] Matt: And [20:32] Richard Webbe: chain [20:32] Matt: that's [20:32] Richard Webbe: system [20:33] Matt: just [20:33] Richard Webbe: built [20:33] Matt: a per [20:33] Richard Webbe: on [20:34] Matt: usage [20:34] Richard Webbe: agents [20:34] Matt: API [20:34] Richard Webbe: that I've [20:34] Matt: cost [20:35] Richard Webbe: either [20:35] Matt: endpoint. [20:35] Richard Webbe: bought [20:36] Matt: But [20:36] Richard Webbe: or [20:36] Matt: open [20:36] Richard Webbe: purchased [20:36] Matt: router [20:36] Richard Webbe: or [20:37] Matt: has [20:37] Richard Webbe: built [20:37] Matt: a very, [20:37] Richard Webbe: myself. [20:37] Matt: very good [20:38] Richard Webbe: And it's in [20:38] Matt: router, [20:38] Richard Webbe: there linking [20:39] Matt: supposedly, [20:39] Richard Webbe: my data to my processes [20:41] Matt: where you can [20:42] Richard Webbe: and [20:42] Matt: choose [20:42] Richard Webbe: getting [20:42] Matt: specific [20:42] Richard Webbe: invoices [20:43] Matt: fallbacks [20:43] Richard Webbe: out and repricing [20:44] Matt: by [20:44] Richard Webbe: and [20:44] Matt: either [20:44] Richard Webbe: rescheduling [20:45] Matt: like [20:45] Richard Webbe: and all of that. [20:46] Matt: model [20:46] Richard Webbe: And [20:46] Matt: itself [20:46] Richard Webbe: if I'm plugged [20:47] Matt: or just [20:47] Richard Webbe: into one [20:47] Matt: provider [20:47] Richard Webbe: LLM, is it [20:48] Matt: or [20:48] Richard Webbe: possible [20:49] Matt: by [20:49] Richard Webbe: to [20:49] Matt: criteria. [20:49] Richard Webbe: build a series [20:49] Matt: For [20:50] Richard Webbe: of [20:50] Matt: example, [20:50] Richard Webbe: agents [20:50] Matt: if it's a [20:50] Richard Webbe: that would [20:51] Matt: ZDR, [20:51] Richard Webbe: automatically [20:51] Matt: like a zero [20:51] Richard Webbe: identify [20:52] Matt: data [20:52] Richard Webbe: there [20:52] Matt: retention. [20:52] Richard Webbe: was a failure and then [20:53] Matt: And [20:53] Richard Webbe: switch [20:54] Matt: you know, [20:54] Richard Webbe: the [20:54] Matt: if [20:54] Richard Webbe: supply [20:54] Matt: you need to [20:55] Richard Webbe: chain [20:55] Matt: keep your thing [20:55] Richard Webbe: process? [20:55] Matt: running and have an equivalent model, they'll do it for you. [21:15] Matt: Yep. [21:19] Matt: Yeah. I couldn't [21:22] Matt: uh this is just like a a small scale scope small scale example of something that I actually use in that vein locally. When I'm developing, when I'm working on projects, if I'm using a particular endpoint that is [21:36] Richard Webbe: Yeah. [21:36] Matt: highly rate limited, or you know, we're all familiar with the subscription quota that you get per month, [21:42] Dave Pengelley: Yeah. [21:42] Matt: whatever, [21:42] Dave Pengelley: And [21:42] Matt: per [21:42] Dave Pengelley: you can [21:43] Matt: per week, [21:43] Dave Pengelley: specify [21:43] Matt: per [21:43] Dave Pengelley: per [21:43] Matt: five [21:44] Dave Pengelley: API [21:44] Matt: hours, [21:44] Dave Pengelley: key [21:44] Matt: you know, [21:44] Dave Pengelley: for which [21:45] Matt: some [21:45] Dave Pengelley: application [21:45] Matt: of the ones [21:45] Dave Pengelley: you're [21:45] Matt: that I'm [21:45] Dave Pengelley: routing [21:46] Matt: experimenting [21:46] Dave Pengelley: through it, like [21:46] Matt: with [21:46] Dave Pengelley: which [21:46] Matt: are [21:46] Dave Pengelley: models are [21:47] Matt: low [21:47] Dave Pengelley: available [21:47] Matt: quota [21:47] Dave Pengelley: and what it can [21:48] Matt: ones, [21:48] Dave Pengelley: use, what it [21:48] Matt: but [21:48] Dave Pengelley: can't, [21:48] Matt: I'm [21:48] Dave Pengelley: all [21:48] Matt: using [21:48] Dave Pengelley: that kind [21:49] Matt: them [21:49] Dave Pengelley: of stuff. [21:49] Matt: because they're [21:49] Richard Webbe: So [21:49] Dave Pengelley: So, [21:49] Matt: different [21:50] Richard Webbe: I can [21:50] Matt: and [21:50] Richard Webbe: build [21:50] Matt: they [21:50] Richard Webbe: my [21:50] Matt: provide, [21:51] Richard Webbe: AI [21:51] Matt: you know, different [21:52] Richard Webbe: um [21:52] Matt: flavor. [21:53] Richard Webbe: fault [21:53] Matt: If if [21:53] Richard Webbe: tolerant [21:53] Matt: that all of a sudden [21:54] Richard Webbe: continuous [21:54] Matt: hits a, [21:55] Richard Webbe: supply [21:55] Matt: hey, [21:56] Richard Webbe: with [21:56] Matt: you've [21:56] Richard Webbe: architecture [21:57] Matt: hit your limit, [21:57] Richard Webbe: and tools. [21:58] Matt: I have automatic fallback to then. Move [22:00] Richard Webbe: Except [22:00] Matt: on to the [22:00] Dave Pengelley: Yep. [22:00] Matt: next [22:00] Richard Webbe: when [22:00] Matt: thing [22:00] Richard Webbe: the panel [22:00] Matt: and like [22:01] Richard Webbe: goes [22:01] Matt: my [22:01] Richard Webbe: in, [22:01] Matt: system doesn't [22:01] Richard Webbe: then [22:01] Matt: just [22:01] Richard Webbe: I'm screwed. [22:02] Matt: stop. [22:03] Dave Pengelley: Yeah, and I forgot [22:12] Matt: Composer. [22:20] Dave Pengelley: away from get right. the to [22:46] Dave Pengelley: Yeah. Um I've noticed the latest update of Hermes is doing that a lot better. So I hit my Groc two point five not my Grok, my um yeah my Groc Supercope, my my super groc composer limit uh on on the weekend. Uh I've been hitting a pretty Hard and just on the the the lower like the 20 plan I've got a lot of composer [23:03] Matt: You know what [23:04] Dave Pengelley: 2.5 [23:04] Matt: things? [23:04] Dave Pengelley: usage out of it was really good. Uh and then I noticed like I was uh popping up and saying was you falling back to GPT 5.5 and I was like hang on I I didn't specify I wanted to [23:14] Matt: That [23:14] Dave Pengelley: you know [23:14] Matt: is [23:14] Dave Pengelley: use the big [23:14] Matt: something [23:14] Dave Pengelley: model [23:15] Matt: that [23:15] Dave Pengelley: I was [23:15] Matt: not [23:15] Dave Pengelley: uh [23:15] Matt: a [23:15] Dave Pengelley: happy [23:15] Matt: lot [23:15] Dave Pengelley: with [23:15] Matt: of [23:16] Dave Pengelley: composer [23:16] Matt: people [23:16] Dave Pengelley: why [23:16] Matt: consider. [23:16] Dave Pengelley: is it using [23:17] Matt: You [23:17] Dave Pengelley: 5.5 [23:17] Matt: know, a lot of people [23:18] Dave Pengelley: but [23:18] Matt: think [23:18] Dave Pengelley: go back to [23:18] Matt: that [23:18] Dave Pengelley: model [23:18] Matt: an [23:18] Dave Pengelley: slash [23:19] Matt: AI [23:19] Dave Pengelley: composer [23:19] Matt: system, [23:19] Dave Pengelley: and then I'd [23:20] Matt: so [23:20] Dave Pengelley: type [23:20] Matt: a [23:20] Dave Pengelley: a prompt [23:20] Matt: system [23:21] Dave Pengelley: and [23:21] Matt: where [23:21] Dave Pengelley: it would [23:21] Matt: you [23:21] Dave Pengelley: go [23:21] Matt: interact [23:21] Dave Pengelley: pause for [23:22] Matt: with [23:22] Dave Pengelley: a second [23:22] Matt: AI, [23:22] Dave Pengelley: then go [23:22] Matt: whether [23:22] Dave Pengelley: GPT [23:23] Matt: it's [23:23] Dave Pengelley: 5.5 [23:23] Matt: the user [23:23] Dave Pengelley: like it [23:24] Matt: speaking [23:24] Dave Pengelley: was [23:24] Matt: directly [23:24] Dave Pengelley: doing [23:24] Matt: to [23:24] Dave Pengelley: that [23:24] Matt: it [23:24] Dave Pengelley: automatic [23:24] Matt: or the back-end [23:25] Dave Pengelley: fallback because [23:25] Matt: system [23:25] Dave Pengelley: of my Hermes [23:26] Matt: triggering [23:26] Dave Pengelley: settings [23:26] Matt: it, [23:27] Dave Pengelley: and [23:27] Matt: they [23:27] Dave Pengelley: it [23:27] Matt: usually [23:27] Dave Pengelley: knew [23:27] Matt: think it's [23:28] Dave Pengelley: okay we've [23:28] Matt: from, [23:28] Dave Pengelley: hit the [23:28] Matt: I [23:28] Dave Pengelley: limits [23:29] Matt: don't know, [23:29] Dave Pengelley: on this [23:29] Matt: it's meant [23:29] Dave Pengelley: one. Flick [23:30] Matt: To [23:30] Dave Pengelley: to the [23:30] Matt: it's [23:30] Dave Pengelley: next one. [23:30] Matt: meant [23:31] Dave Pengelley: I was [23:31] Matt: to [23:31] Dave Pengelley: like, [23:31] Matt: analyze [23:31] Dave Pengelley: oh, that's [23:31] Matt: sentiment. [23:32] Dave Pengelley: that [23:32] Matt: It's meant [23:32] Dave Pengelley: like [23:32] Matt: to [23:33] Dave Pengelley: a [23:33] Matt: summarize. [23:33] Dave Pengelley: month ago [23:34] Matt: It's [23:34] Dave Pengelley: didn't [23:34] Matt: meant [23:34] Dave Pengelley: work [23:34] Matt: to, [23:34] Dave Pengelley: that smoothly. It [23:35] Matt: I [23:35] Dave Pengelley: would [23:35] Matt: don't [23:35] Dave Pengelley: just go, [23:35] Matt: know, [23:35] Dave Pengelley: oh no, [23:36] Matt: find [23:36] Dave Pengelley: error, error. [23:36] Matt: important details. [23:37] Dave Pengelley: And maybe I've [23:37] Matt: You [23:37] Dave Pengelley: got [23:37] Matt: think about [23:37] Dave Pengelley: my [23:38] Matt: AI [23:38] Dave Pengelley: homie [23:38] Matt: note [23:38] Dave Pengelley: set [23:38] Matt: takers. [23:38] Dave Pengelley: up tightened [23:39] Matt: But [23:39] Dave Pengelley: up a little bit better [23:40] Matt: what [23:40] Dave Pengelley: and then [23:40] Matt: it [23:40] Dave Pengelley: tidied [23:40] Matt: can [23:41] Dave Pengelley: up [23:41] Matt: do [23:41] Dave Pengelley: over the [23:41] Matt: is [23:41] Dave Pengelley: last month. [23:42] Matt: similar [23:42] Dave Pengelley: But [23:42] Matt: to [23:42] Dave Pengelley: maybe [23:42] Matt: imagine [23:43] Dave Pengelley: they've [23:43] Matt: if you [23:43] Dave Pengelley: also [23:43] Matt: you [23:43] Dave Pengelley: just [23:43] Matt: hook your [23:43] Dave Pengelley: done [23:44] Matt: AI [23:44] Dave Pengelley: some back-end [23:44] Matt: up to your emails, [23:44] Dave Pengelley: engineering to [23:45] Matt: which [23:45] Dave Pengelley: help [23:45] Matt: you know [23:45] Dave Pengelley: smooth [23:45] Matt: some people [23:46] Dave Pengelley: that [23:46] Matt: do. [23:46] Dave Pengelley: process [23:46] Matt: It [23:46] Dave Pengelley: as [23:46] Matt: sounds [23:46] Dave Pengelley: well. [23:47] Matt: terrifying, [23:47] Richard Webbe: So the [23:47] Matt: but [23:47] Richard Webbe: dynamic [23:48] Matt: with the [23:48] Richard Webbe: nature [23:48] Matt: right [23:48] Richard Webbe: of AI [23:49] Matt: the [23:49] Richard Webbe: we [23:49] Matt: right [23:49] Richard Webbe: can [23:49] Matt: uh [23:50] Richard Webbe: use [23:50] Matt: configuration, [23:50] Richard Webbe: to maintain [23:50] Matt: it'll probably work. [23:51] Richard Webbe: supply [23:51] Matt: Is [23:51] Richard Webbe: of AI, [23:52] Matt: observability. [23:52] Richard Webbe: as [23:52] Matt: So [23:53] Richard Webbe: well [23:53] Matt: imagine [23:53] Richard Webbe: as the [23:53] Matt: your [23:53] Richard Webbe: dynamic [23:53] Matt: system, [23:54] Richard Webbe: nature [23:54] Matt: you know, [23:54] Richard Webbe: of our [23:54] Matt: it's [23:54] Richard Webbe: supply [23:55] Matt: taking [23:55] Richard Webbe: chain [23:55] Matt: in [23:55] Richard Webbe: and our [23:55] Matt: logs, [23:55] Richard Webbe: own businesses [23:56] Matt: audit logs, [23:56] Richard Webbe: and [23:56] Matt: any [23:56] Richard Webbe: processing. [23:57] Matt: sort of event log, and that gets fired to the AI. Process first, which could be a fairly, you know, small, fast model. And then depending on the categorization of that, could then go to a reasoning model and say, hey, we've just gotten this. What does this mean? What do we need to do? And then that model can have access, it can have controls to, all right, let's let's use the failover. Let's actually engage this system based on what's happened to keep everything running. So you can use an AI reasoning process to make decisions like that. [25:04] Matt: Hmm. Yeah, it's like I don't know if you guys saw the video that uh Anthropic put out about the J space, you know, about how [25:11] Richard Webbe: To [25:11] Matt: AI [25:11] Richard Webbe: be [25:12] Matt: models [25:12] Richard Webbe: dynamic. [25:12] Matt: think. That's [25:13] Richard Webbe: Yeah, [25:13] Matt: a [25:13] Richard Webbe: yeah. [25:13] Matt: very good [25:13] Richard Webbe: Observability [25:14] Matt: video. Um [25:14] Richard Webbe: is an [25:14] Matt: I think [25:15] Richard Webbe: area that [25:15] Matt: you [25:15] Richard Webbe: I worked [25:15] Matt: guys [25:15] Richard Webbe: in [25:16] Matt: should [25:16] Richard Webbe: for [25:16] Matt: have [25:16] Richard Webbe: many [25:16] Matt: a [25:16] Richard Webbe: years [25:16] Matt: look [25:16] Richard Webbe: with [25:16] Matt: at [25:16] Richard Webbe: a [25:17] Matt: that [25:17] Richard Webbe: friend [25:17] Matt: after. [25:17] Richard Webbe: of mine, Rob [25:17] Matt: Or [25:18] Richard Webbe: Perrin at New Gen Systems, uh great [25:20] Matt: I [25:20] Richard Webbe: company. [25:20] Matt: don't know what the [25:20] Richard Webbe: And [25:21] Matt: video is [25:21] Richard Webbe: and [25:21] Matt: called, [25:21] Richard Webbe: some [25:21] Matt: but it's [25:22] Richard Webbe: of that [25:22] Matt: it's their [25:22] Richard Webbe: dynamic [25:22] Matt: most recent thing that [25:23] Richard Webbe: reading, [25:23] Matt: they put out. [25:24] Richard Webbe: and [25:24] Matt: Um [25:24] Richard Webbe: they've got the technology [25:25] Matt: you'll [25:25] Richard Webbe: now [25:25] Matt: be able [25:26] Richard Webbe: that [25:26] Matt: to [25:26] Richard Webbe: they [25:26] Matt: find [25:26] Richard Webbe: can [25:26] Matt: it on [25:26] Richard Webbe: read [25:26] Matt: YouTube [25:26] Richard Webbe: what's [25:27] Matt: X, [25:27] Richard Webbe: happening [25:27] Matt: wherever. [25:27] Richard Webbe: in an IP [25:28] Matt: But [25:28] Richard Webbe: packet [25:28] Matt: it's just [25:28] Richard Webbe: in [25:28] Matt: a [25:28] Richard Webbe: real [25:28] Matt: quick [25:29] Richard Webbe: time as [25:29] Matt: little [25:29] Richard Webbe: it's flying. [25:30] Matt: Uh artistic [25:31] Richard Webbe: And [25:31] Matt: rendition [25:31] Richard Webbe: uh I used [25:32] Matt: slash [25:32] Richard Webbe: to say it's [25:32] Matt: a [25:33] Richard Webbe: like [25:33] Matt: little [25:33] Richard Webbe: a [25:33] Matt: bit [25:33] Richard Webbe: security [25:33] Matt: of engineering [25:33] Richard Webbe: guard at the front door [25:34] Matt: of [25:34] Richard Webbe: of your [25:35] Matt: how [25:35] Richard Webbe: uh nightclub [25:36] Matt: AI [25:36] Richard Webbe: with X-ray [25:36] Matt: models [25:36] Richard Webbe: vision. [25:37] Matt: actually think, [25:37] Richard Webbe: He knows [25:37] Matt: how they [25:37] Richard Webbe: why [25:38] Matt: process, [25:38] Richard Webbe: everyone's coming [25:38] Matt: like [25:38] Richard Webbe: and what [25:38] Matt: how [25:38] Richard Webbe: they've got [25:39] Matt: things [25:39] Richard Webbe: on them and how [25:39] Matt: jump [25:39] Richard Webbe: they're doing. [25:40] Matt: in [25:40] Richard Webbe: If you apply [25:41] Matt: in [25:41] Richard Webbe: that [25:41] Matt: that [25:41] Richard Webbe: to AI, [25:41] Matt: neural net, [25:42] Richard Webbe: the dynamic [25:42] Matt: like in the various [25:43] Richard Webbe: ability [25:43] Matt: stages [25:43] Richard Webbe: of [25:44] Matt: of [25:44] Richard Webbe: maintaining [25:44] Matt: thinking. [25:44] Richard Webbe: your systems [25:44] Matt: So, [25:45] Richard Webbe: and what [25:45] Matt: you [25:45] Richard Webbe: you're [25:45] Matt: know, [25:45] Richard Webbe: doing, I think [25:46] Matt: we're [25:46] Richard Webbe: is [25:46] Matt: all [25:46] Richard Webbe: very [25:46] Matt: familiar, [25:46] Richard Webbe: real. [25:47] Matt: well, most people should be familiar with the idea of the reasoning traces, you know, the I where you see the agent thinking and it's like, I need to do this thing because the user said to do this. That means I probably should use this tool. Okay, I just called this tool and then this happens. Happened, but [26:00] Richard Webbe: What's [26:00] Matt: it's not [26:00] Richard Webbe: the [26:00] Matt: the [26:01] Richard Webbe: name [26:01] Matt: response, [26:01] Richard Webbe: of the video? [26:01] Matt: that's [26:01] Richard Webbe: J [26:01] Matt: like the [26:02] Richard Webbe: Space [26:02] Matt: reasoning [26:02] Richard Webbe: by [26:02] Matt: traces. [26:02] Richard Webbe: Anthropic. [26:03] Matt: There's stuff that happens before that and in between that. And they they gave this video, and it was really, really interesting to see the concepts that they can pull out from observing the the J space, which is I forget what the J standard for, but it was uh like some sort of philosopher concept. Yeah, that's the video. Really, really cool art style too. I gotta, you know, I've always said that um anthropics are slow to pick things. Up but when they do it, it's always done very well as far as marketing and and product and and all of that sort of thing. But the uh the concept that they say during this is like you just said, Richard, you know, observing things as it's in motion, they can notice when stuff's popping up in the J space and use that as flagging. And you know, in the case of uh Mythos and Fable to actually reroute or to disable entirely. based on you know security concerns. So very interesting. [27:08] Matt: Yeah, because like this is before the thing even provides a response. It's like as it's thinking or previous to thinking as well. [27:45] Richard Webbe: Yeah, yeah. Reaction time is is key for everything, isn't it? [27:55] Matt: It's funny you said before X ray. Like X rays the the concept that I've fallen into [28:00] Richard Webbe: At last week's [28:00] Matt: uh [28:01] Richard Webbe: podcast, [28:01] Matt: with with my [28:02] Richard Webbe: I was [28:02] Matt: my [28:02] Richard Webbe: uh [28:02] Matt: school [28:02] Richard Webbe: laughing [28:03] Matt: community [28:03] Richard Webbe: at someone. [28:03] Matt: is it's [28:04] Richard Webbe: Um, [28:04] Matt: the thing that [28:04] Richard Webbe: and [28:04] Matt: I try [28:04] Richard Webbe: I [28:05] Matt: and teach [28:05] Richard Webbe: well, I [28:05] Matt: first [28:05] Richard Webbe: respect and [28:06] Matt: is [28:06] Richard Webbe: appreciate [28:06] Matt: the idea [28:06] Richard Webbe: the [28:06] Matt: of how [28:06] Richard Webbe: AI [28:07] Matt: to x-ray [28:07] Richard Webbe: ethicists [28:07] Matt: your system. Because [28:08] Richard Webbe: who [28:08] Matt: you [28:08] Richard Webbe: want [28:08] Matt: you [28:08] Richard Webbe: to [28:09] Matt: go to [28:09] Richard Webbe: know how [28:09] Matt: you [28:09] Richard Webbe: the [28:09] Matt: go to [28:10] Richard Webbe: mind [28:10] Matt: the [28:10] Richard Webbe: works. [28:10] Matt: doctor or [28:10] Richard Webbe: So [28:10] Matt: you [28:10] Richard Webbe: they're [28:10] Matt: go to [28:11] Richard Webbe: all, [28:11] Matt: some sort [28:11] Richard Webbe: oh, [28:11] Matt: of [28:11] Richard Webbe: you know, [28:11] Matt: specialist [28:12] Richard Webbe: whose [28:12] Matt: and [28:12] Richard Webbe: mind [28:12] Matt: and [28:12] Richard Webbe: and [28:12] Matt: you need [28:12] Richard Webbe: whose [28:13] Matt: to go [28:13] Richard Webbe: bias [28:13] Matt: get a [28:13] Richard Webbe: are [28:13] Matt: scan, [28:13] Richard Webbe: you introducing? [28:13] Matt: you need to get an [28:14] Richard Webbe: Of [28:14] Matt: x-ray. [28:14] Richard Webbe: course, the [28:15] Matt: What [28:15] Richard Webbe: whole [28:15] Matt: why [28:15] Richard Webbe: idea [28:15] Matt: do [28:15] Richard Webbe: of [28:15] Matt: they [28:15] Richard Webbe: AI [28:15] Matt: say to [28:16] Richard Webbe: and [28:16] Matt: do [28:16] Richard Webbe: that [28:16] Matt: that? [28:16] Richard Webbe: is [28:16] Matt: You [28:16] Richard Webbe: not [28:16] Matt: know [28:16] Richard Webbe: to [28:16] Matt: they're the [28:17] Richard Webbe: introduce [28:17] Matt: expert. [28:17] Richard Webbe: any bias, [28:17] Matt: They they've [28:18] Richard Webbe: uh [28:18] Matt: done [28:18] Richard Webbe: bias [28:18] Matt: all the training. [28:19] Richard Webbe: at all, [28:19] Matt: What [28:19] Richard Webbe: right? [28:19] Matt: why do they [28:20] Richard Webbe: Because [28:20] Matt: want to [28:20] Richard Webbe: there's [28:20] Matt: go [28:20] Richard Webbe: no emotional attachment. [28:21] Matt: like [28:21] Richard Webbe: Let's [28:21] Matt: so [28:22] Richard Webbe: just [28:22] Matt: the [28:22] Richard Webbe: use numbers [28:22] Matt: the point [28:23] Richard Webbe: and [28:23] Matt: there [28:23] Richard Webbe: facts [28:23] Matt: is this [28:23] Richard Webbe: to connect [28:24] Matt: this process [28:24] Richard Webbe: to the [28:25] Matt: is [28:25] Richard Webbe: answer. Well, [28:25] Matt: you know [28:25] Richard Webbe: mathematical [28:26] Matt: it's a way of revealing [28:26] Richard Webbe: people like [28:27] Matt: what [28:27] Richard Webbe: us [28:27] Matt: you don't [28:27] Richard Webbe: understand [28:27] Matt: see or [28:28] Richard Webbe: that, [28:28] Matt: what you [28:28] Richard Webbe: and [28:28] Matt: haven't [28:28] Richard Webbe: uh, [28:28] Matt: been [28:29] Richard Webbe: it's [28:29] Matt: thinking [28:29] Richard Webbe: very [28:29] Matt: about. [28:29] Richard Webbe: important. [28:30] Matt: know it's [28:30] Richard Webbe: That that is maintained on how that works and that connects to what [28:34] Matt: yeah [28:34] Richard Webbe: we really want the [28:35] Matt: like [28:35] Richard Webbe: outcomes [28:35] Matt: and [28:35] Richard Webbe: to [28:35] Matt: sometimes [28:36] Richard Webbe: be [28:36] Matt: it's it it can be things that you just never knew you know uh the idea of jumping into a new project with any kind of agent and not having any idea about the system that you're going to be building or the system that may already exist that's what this kind of process is for [28:52] Richard Webbe: like [28:52] Matt: and that's [28:53] Richard Webbe: that. [28:53] Matt: why those doctors need an x-ray you know it's a [29:03] Richard Webbe: Exactly. [29:13] Richard Webbe: Better decisions come from more informed information, and that's where all of that comes from. That's why we're here. [29:39] Richard Webbe: The engineer in me loves watching those car rebuild shows on cable television. And there was some guy who picked up a Porsche GTS3 worth a fortune for a super cheap amount on the weekend. And he said, I bought this for, you know, some dirt cheap price. It's worth, you know, 300 grand. He said, I'm probably screwed, but Porsche wouldn't give it a warranty, no one would work on it. So he pulls this thing apart four times, looking for it, and he's trying to work out. And he eventually found out that the timing wheels, which all look almost identical, but one slightly smaller than the other. And then he reversed them and then he had this very expensive Porsche for a cheap price. I mean, diagnostics, X-ray, looking inside something. I mean, he he got so sick of taking the engine in it. Of course, with us, we can't do that. And with systems, a lot of them are running live. You've got to be out of with observability in that, like you said, X-ray. X-ray, what's happening? 'Cause you can't have an operate on it while it's live. [30:35] Dave Pengelley: Yep. This this was I'm I'm just gonna run a thing, then we'll keep talking about this because this this feeds into something I want to discuss. [30:51] Dave Pengelley: Wins of the week, the week. I know you love that one, Matt. Still I have I I haven't rebranded that one. That one's that one we kept the week. Um So I've been I've been working a lot with Hermes around this show and trying to automate and fix things up for this show and run this show a little bit more smoothly and um help us get more content out because we're doing the weekly show. But then the clipping, the editing, the content, the website updates, all that kind of stuff. I've been working on smoothing those, making them more deterministic, a little bit less agent reliant, and just bringing the agent in to do the bits it needs to do. And so this last week, from a wins of the week point of view, uh we've updated the website now. It's got the new Branding, the new logo on, which is nice. I like seeing that up there. Uh, little things that we've built into the back end of it. So on every episode now, like I've done things like brought in the transcript. So that from a search indexing, everything we say is now indexed. So the search engines and the companies can see what we're talking about. Uh with our events stuff. I've now sort of got capability to have, you know, upcoming events, past events, see all the recaps, see all our talks, Richard, that we did, and there everyone else, and the sponsors for each individual event. So building all of this and all Automating all these systems and processes. And the next step along from all this is clipping bits of our content and actually, you know, having little TikToks and reels and things generating out of this content. And from this auditing the process and which bits are going where, I'm trying to get her to use Hyperframes, the HeyGen library, to automatically put some captions and do things with it, which is really nice. Um, but little things as far as like, okay, go do this for me. And I did a version of it and then came back and uh the fonts are. Too small or this. And HeyGen has a uh studio where I could arguably go in and make [32:34] Matt: That's [32:35] Dave Pengelley: some changes [32:35] Matt: cool. [32:35] Dave Pengelley: or a [32:35] Matt: So [32:36] Dave Pengelley: file didn't [32:36] Matt: are [32:36] Dave Pengelley: load [32:36] Matt: you just [32:36] Dave Pengelley: because [32:36] Matt: using [32:36] Dave Pengelley: it got [32:37] Matt: the [32:37] Dave Pengelley: the path [32:37] Matt: same [32:37] Dave Pengelley: wrong. [32:37] Matt: Hermes [32:38] Dave Pengelley: One [32:38] Matt: model [32:38] Dave Pengelley: of the [32:38] Matt: for [32:38] Dave Pengelley: things [32:38] Matt: everything, [32:38] Dave Pengelley: I'm doing [32:39] Matt: or do [32:39] Dave Pengelley: is [32:39] Matt: you [32:39] Dave Pengelley: instead [32:39] Matt: have [32:39] Dave Pengelley: of just [32:40] Matt: like [32:40] Dave Pengelley: manually fixing [32:40] Matt: like [32:40] Dave Pengelley: it, [32:40] Matt: different [32:40] Dave Pengelley: which A [32:41] Matt: agents [32:41] Dave Pengelley: would be fast and [32:42] Matt: that [32:42] Dave Pengelley: easy [32:42] Matt: are [32:42] Dave Pengelley: and cheap [32:42] Matt: like [32:43] Dave Pengelley: because [32:43] Matt: like [32:43] Dave Pengelley: you know I'm [32:43] Matt: separate [32:43] Dave Pengelley: not relying [32:44] Matt: Hermes [32:44] Dave Pengelley: on more tokens, [32:44] Matt: installations so that [32:45] Dave Pengelley: but [32:45] Matt: they [32:45] Dave Pengelley: then [32:46] Matt: have [32:46] Dave Pengelley: the [32:46] Matt: like [32:46] Dave Pengelley: computer [32:46] Matt: their [32:46] Dave Pengelley: doesn't [32:46] Matt: own [32:46] Dave Pengelley: learn. [32:47] Matt: skill [32:47] Dave Pengelley: And so what [32:47] Matt: set [32:47] Dave Pengelley: I'm doing [32:48] Matt: that's [32:48] Dave Pengelley: is going back and [32:48] Matt: not [32:48] Dave Pengelley: going, [32:49] Matt: clouded [32:49] Dave Pengelley: No, no, [32:49] Matt: by [32:49] Dave Pengelley: you [32:49] Matt: anything [32:50] Dave Pengelley: didn't copy [32:50] Matt: else? [32:50] Dave Pengelley: that clip properly. You need to fix that. I'm not doing that. And again, getting it to learn its own skills and go, okay, I made [32:56] Matt: Yeah, [32:56] Dave Pengelley: a mistake there. [32:56] Matt: cool. [32:56] Dave Pengelley: I need to make sure next time I do the process or update the Python code. To make sure that it's actually going to definitely pull this file from here and push that information there reliably. And so it's like teaching the AR AI where it's making mistakes as it goes to get these workflows and processes smoother and more reliable and get it as much as possible out of the AI. [33:33] Dave Pengelley: So, yeah, so I've got an AI operator-specific Hermes profile, which is just responsible for all things podcast. Uh, and so it knows the branding, it knows about Roo, it know the robot, it knows um the design pack that I generated by Claude, it knows how to push things to convex in the website and do all those sorts of things. So I might be using different models, like I've used GPD 5.5 and um Composer are my two main ones that I'm using on it at the moment. But it's learning. And like even little things, like, and this is that whole that understanding the engineering and being able to have the x-ray and understand and do the diagnosis is really important. And where I think a lot of vibe coders are going to fall short is little things like the transcripts. And it's saying, Oh, I need a special extra API key to push the transcripts into the convex database. I was like, why? I thought the whole point of Convex is that it's LLM friendly and agents are meant to be able to do stuff. Why do you need me to give you an extra different API key when you've already been Updating the schema and pushing other information. Why do you need a special API key just to add transcripts? It goes, Oh, you're right, that's legacy from an old thing where I thought I had to do that because we'd put in this own protection and restriction of our own workflows and rules. We can change that. And of course, I can mutate that and push this thing there. I was like, oh, come on. Like this, so it learns [34:50] Richard Webbe: This is [34:50] Dave Pengelley: bad [34:50] Richard Webbe: an [34:50] Dave Pengelley: habits [34:51] Richard Webbe: important [34:51] Dave Pengelley: and [34:51] Richard Webbe: point. [34:51] Dave Pengelley: and [34:52] Richard Webbe: It is an important point to mention that AI doesn't know what it doesn't [34:56] Dave Pengelley: well, [34:56] Richard Webbe: know. [34:56] Dave Pengelley: and and also it learns from its own past. Like there's These are agents like Hermes, which are self-learning and self-directing to a certain extent. And if on their journey towards today they picked up some bad habits, they're going to embed them in. And you actually needed to go tell it to unlearn those or remove those or hang on. You did that as a workaround [35:16] Matt: GPT [35:16] Dave Pengelley: for a very specific [35:17] Matt: is [35:17] Dave Pengelley: instance [35:17] Matt: very [35:17] Dave Pengelley: at [35:17] Matt: much [35:18] Dave Pengelley: one point [35:18] Matt: like [35:18] Dave Pengelley: in time, [35:18] Matt: that. [35:18] Dave Pengelley: maybe because [35:19] Matt: You [35:19] Dave Pengelley: the [35:19] Matt: know, [35:19] Dave Pengelley: model [35:19] Matt: like [35:19] Dave Pengelley: was [35:20] Matt: the [35:20] Dave Pengelley: failing, [35:20] Matt: current [35:20] Dave Pengelley: because [35:20] Matt: GPT [35:20] Dave Pengelley: I didn't have a [35:21] Matt: five [35:21] Dave Pengelley: right [35:21] Matt: point [35:21] Dave Pengelley: fallback. [35:21] Matt: five [35:21] Dave Pengelley: And so [35:21] Matt: is [35:22] Dave Pengelley: I would [35:22] Matt: extremely [35:22] Dave Pengelley: just worked around [35:23] Matt: hilarious [35:23] Dave Pengelley: the issue. [35:24] Matt: like that. [35:24] Dave Pengelley: Um [35:25] Matt: You know, I [35:25] Dave Pengelley: it's [35:25] Matt: I do [35:26] Dave Pengelley: like [35:26] Matt: a lot [35:26] Dave Pengelley: we've [35:26] Matt: of [35:26] Dave Pengelley: got [35:26] Matt: stuff [35:26] Dave Pengelley: to, we're [35:26] Matt: now [35:27] Dave Pengelley: gonna, [35:27] Matt: in parallel, [35:27] Dave Pengelley: we're gonna get to [35:28] Matt: like [35:28] Dave Pengelley: the other [35:28] Matt: you know, [35:28] Dave Pengelley: side of this [35:28] Matt: multipi [35:28] Dave Pengelley: door. [35:29] Matt: multiprojects [35:29] Dave Pengelley: This door's [35:29] Matt: going. [35:29] Dave Pengelley: locked. We [35:30] Matt: At [35:30] Dave Pengelley: can't [35:30] Matt: once, [35:30] Dave Pengelley: get through [35:30] Matt: all [35:30] Dave Pengelley: it. [35:30] Matt: sorts of [35:31] Dave Pengelley: We [35:31] Matt: things. [35:31] Dave Pengelley: can't get [35:31] Matt: I [35:31] Dave Pengelley: through [35:31] Matt: do a [35:31] Dave Pengelley: it. [35:32] Matt: lot of [35:32] Dave Pengelley: Uh [35:32] Matt: browser [35:32] Dave Pengelley: and so they [35:32] Matt: automation [35:33] Dave Pengelley: sort of built all this [35:33] Matt: now [35:34] Dave Pengelley: extra infrastructure [35:34] Matt: because if [35:35] Dave Pengelley: to [35:35] Matt: if [35:35] Dave Pengelley: to [35:35] Matt: we're [35:35] Dave Pengelley: build [35:35] Matt: building [35:35] Dave Pengelley: a whole passage [35:36] Matt: web-based [35:36] Dave Pengelley: and thing [35:36] Matt: solutions, [35:36] Dave Pengelley: around the [35:37] Matt: I [35:37] Dave Pengelley: door [35:37] Matt: need the [35:37] Dave Pengelley: when really [35:38] Matt: agent [35:38] Dave Pengelley: they just [35:38] Matt: to [35:38] Dave Pengelley: needed [35:38] Matt: actually [35:38] Dave Pengelley: to [35:39] Matt: go [35:39] Dave Pengelley: reach [35:39] Matt: in, [35:39] Dave Pengelley: over to [35:39] Matt: take [35:40] Dave Pengelley: the little [35:40] Matt: screenshots, [35:40] Dave Pengelley: table and get the key [35:40] Matt: visually [35:41] Dave Pengelley: and then they could [35:41] Matt: see [35:41] Dave Pengelley: have [35:41] Matt: the [35:41] Dave Pengelley: opened [35:41] Matt: thing, [35:41] Dave Pengelley: it up and [35:42] Matt: click [35:42] Dave Pengelley: gone straight [35:42] Matt: through [35:42] Dave Pengelley: through. [35:42] Matt: the [35:43] Dave Pengelley: But [35:43] Matt: application, [35:43] Dave Pengelley: they didn't know the key was [35:44] Matt: find [35:44] Dave Pengelley: there. [35:44] Matt: what [35:44] Dave Pengelley: So [35:44] Matt: doesn't [35:44] Dave Pengelley: they just [35:44] Matt: work, [35:45] Dave Pengelley: do all [35:45] Matt: fix [35:45] Dave Pengelley: this [35:45] Matt: it [35:45] Dave Pengelley: extra, [35:45] Matt: before you [35:45] Dave Pengelley: extra [35:46] Matt: come back [35:46] Dave Pengelley: build-around [35:46] Matt: to me and let me [35:47] Dave Pengelley: work. [35:47] Richard Webbe: Yep, [35:47] Matt: find it, [35:47] Dave Pengelley: Uh [35:47] Richard Webbe: doesn't [35:47] Matt: you [35:47] Dave Pengelley: and [35:48] Richard Webbe: know [35:48] Matt: know? [35:48] Richard Webbe: what [35:48] Dave Pengelley: so [35:48] Richard Webbe: it [35:48] Matt: And [35:48] Richard Webbe: doesn't know. [35:48] Matt: all [35:49] Dave Pengelley: yeah, [35:49] Matt: this [35:49] Dave Pengelley: and [35:49] Matt: kind [35:49] Dave Pengelley: then [35:49] Matt: of stuff [35:49] Dave Pengelley: down [35:50] Matt: happens [35:50] Dave Pengelley: down the [35:50] Matt: in [35:50] Dave Pengelley: track [35:50] Matt: the background [35:50] Dave Pengelley: that they [35:50] Matt: now [35:51] Dave Pengelley: now [35:51] Matt: as [35:51] Dave Pengelley: got [35:51] Matt: I'm [35:51] Dave Pengelley: the key [35:51] Matt: working. [35:51] Dave Pengelley: and they've [35:51] Matt: And [35:52] Dave Pengelley: used the [35:52] Matt: 5.5 [35:52] Dave Pengelley: key, but they forgot they've [35:53] Matt: is [35:53] Dave Pengelley: got [35:53] Matt: hilarious [35:53] Dave Pengelley: the key because they built all the [35:54] Matt: when [35:54] Dave Pengelley: texture [35:54] Matt: it comes [35:54] Dave Pengelley: infrastructure. [35:54] Matt: to [35:54] Dave Pengelley: They're [35:55] Matt: certain [35:55] Dave Pengelley: no no, I [35:55] Matt: pages [35:55] Dave Pengelley: built this extra [35:56] Matt: or [35:56] Dave Pengelley: big [35:56] Matt: elements [35:56] Dave Pengelley: tunnel. [35:56] Matt: that [35:56] Dave Pengelley: I need [35:56] Matt: it [35:57] Dave Pengelley: to [35:57] Matt: doesn't [35:57] Dave Pengelley: use the big [35:57] Matt: see [35:57] Dave Pengelley: tunnel [35:57] Matt: or doesn't [35:57] Dave Pengelley: every [35:58] Matt: know how [35:58] Dave Pengelley: time. [35:58] Matt: to interact [35:58] Dave Pengelley: No, [35:58] Matt: with. [35:59] Dave Pengelley: you don't. So [36:00] Matt: It would just move the mouse and go up and then click, you know. But it it does what you just said, Dave. It tries to construct this entire like insane process just to click a button. You know, it's it's fine, mate. Just pause for a second. [36:55] Dave Pengelley: Yeah. [37:00] Dave Pengelley: Browser, I jumped into N8N with it and tried tried to get GPT, whichever version it was back with Atlas to navigate N8N. And it was like watching a toddler with crayons, just no one, no idea what to do. And it was just like, oh, I can see there's a canvas, and maybe there's a button there, but maybe there's not. And maybe I can move the mouse there, and maybe I can't, or maybe, oh, look, so like, and just had no idea how to just navigate this visual interface back then. It was wild. [37:22] Richard Webbe: And and and and I think that goes back to what we always say, knowing what the question really is. So once you've given a prompt or an angle for AI, it will continue coming in from that angle because it thinks that's the parameter you've limited it to. You haven't given it that broader perspective to say, well, you know, if there's a hedge in the way, walk around it. Don't try and jump over it. Oh, okay, that's a good idea. And that's that's that's what I find most exciting about AI, AI architects, people like Matt and yourself, that architect narratives around how to approach a problem with the right questions and the right architecture. And I think that's that's the trick. [38:00] Dave Pengelley: Yeah, that's it's like and you've seen this in illustrated through fictional narratives for forever and a day, but all these things where the AI goes rogue because it's given a narrow, narrow directive and it ignores you know second and third order consequences to maintain that narrow directive. Everything from the Matrix to Wally to um you know uh war games with Matthew Broderick back in the eighties, all these all these systems have got the very narrow, like what success looks like. And so the And we come back to that concept I talk about all the time. Ambient context. They don't have that ambient knowledge and understanding of things. They just know what's in front of them and what you've told them. [38:39] Richard Webbe: And we can talk about um, I think I brought up last week the confirmation bias that some uh ethical AI people are talking about. And we all approach a problem from a certain angle. We don't take our blinkers off, right? We go, this is how, and and that's what humans do. And so AI needs humans to tell it there isn't just one approach. There's another approach here. There's another way there. And getting back to what is the problem you're trying to solve? It's like the Monet [39:07] Matt: I [39:07] Richard Webbe: situation. [39:07] Matt: do. [39:07] Richard Webbe: Run away from [39:08] Matt: And [39:08] Richard Webbe: the painting [39:08] Matt: it's [39:09] Richard Webbe: and [39:09] Matt: something [39:09] Richard Webbe: look at [39:09] Matt: that, [39:09] Richard Webbe: the picture [39:09] Matt: you know, I [39:10] Richard Webbe: you're [39:10] Matt: think [39:10] Richard Webbe: trying [39:10] Matt: we've [39:10] Richard Webbe: to build. [39:10] Matt: talked about [39:10] Richard Webbe: Don't [39:11] Matt: before [39:11] Richard Webbe: worry [39:11] Matt: how [39:11] Richard Webbe: too [39:11] Matt: you [39:11] Richard Webbe: much [39:11] Matt: you [39:11] Richard Webbe: about [39:12] Matt: should [39:12] Richard Webbe: the [39:12] Matt: have [39:12] Richard Webbe: dots. [39:12] Matt: the, [39:12] Richard Webbe: If you [39:12] Matt: you [39:13] Richard Webbe: worry [39:13] Matt: know, [39:13] Richard Webbe: with [39:13] Matt: you've [39:13] Richard Webbe: the dots [39:13] Matt: spoken about [39:14] Richard Webbe: though, [39:14] Matt: your council [39:14] Richard Webbe: you're going to [39:14] Matt: before, [39:14] Richard Webbe: restrict [39:15] Matt: Dave, [39:15] Richard Webbe: your [39:15] Matt: like, you [39:15] Richard Webbe: system [39:15] Matt: know, your [39:16] Richard Webbe: to [39:16] Matt: different [39:16] Richard Webbe: approaching [39:16] Matt: um [39:16] Richard Webbe: things in the [39:17] Matt: entities [39:17] Richard Webbe: same way and [39:18] Matt: that [39:18] Richard Webbe: it [39:18] Matt: can [39:18] Richard Webbe: won't [39:18] Matt: chime [39:19] Richard Webbe: work. [39:19] Matt: in. [39:20] Dave Pengelley: I [39:20] Matt: It's [39:20] Dave Pengelley: mean Matt, [39:20] Matt: very [39:20] Dave Pengelley: you you're switching [39:21] Matt: important [39:21] Dave Pengelley: systems [39:21] Matt: to [39:21] Dave Pengelley: and harnesses [39:22] Matt: experiment [39:22] Dave Pengelley: all the time, and and and we've [39:23] Matt: and [39:23] Dave Pengelley: discussed [39:24] Matt: to [39:24] Dave Pengelley: this, I'm [39:24] Matt: integrate [39:24] Dave Pengelley: sure, but it's that whole asking [39:25] Matt: multi-model [39:25] Dave Pengelley: the AI, what am I not asking? What am I missing? I think it [39:28] Matt: councils [39:28] Dave Pengelley: might be this way, [39:28] Matt: and you [39:29] Dave Pengelley: but [39:29] Matt: know, [39:29] Dave Pengelley: are there [39:29] Matt: processes. [39:29] Dave Pengelley: all Alternatives that [39:30] Matt: Where, [39:30] Dave Pengelley: I haven't [39:31] Matt: like [39:31] Dave Pengelley: considered [39:31] Matt: I I [39:31] Dave Pengelley: and [39:31] Matt: literally [39:32] Dave Pengelley: getting [39:32] Matt: had something [39:32] Dave Pengelley: it [39:32] Matt: this [39:32] Dave Pengelley: to [39:32] Matt: morning [39:33] Dave Pengelley: actually [39:33] Matt: where I'm [39:33] Dave Pengelley: think beyond [39:33] Matt: working [39:34] Dave Pengelley: itself [39:34] Matt: on uh the [39:34] Dave Pengelley: and [39:34] Matt: next [39:34] Dave Pengelley: think beyond [39:35] Matt: versions [39:35] Dave Pengelley: the sort [39:35] Matt: of [39:36] Dave Pengelley: of can [39:36] Matt: my [39:36] Dave Pengelley: of paint [39:36] Matt: core [39:37] Dave Pengelley: you put in front [39:37] Matt: system. [39:37] Dave Pengelley: of it. You put [39:38] Matt: Uh [39:38] Dave Pengelley: a can [39:38] Matt: it's [39:38] Dave Pengelley: of blue [39:38] Matt: getting [39:39] Dave Pengelley: paint, but [39:39] Matt: very [39:39] Dave Pengelley: you [39:39] Matt: close [39:39] Dave Pengelley: know, [39:39] Matt: to [39:40] Dave Pengelley: there [39:40] Matt: what [39:40] Dave Pengelley: might [39:40] Matt: I'm [39:40] Dave Pengelley: be [39:40] Matt: describing [39:40] Dave Pengelley: you might need [39:40] Matt: as [39:41] Dave Pengelley: red [39:41] Matt: a [39:41] Dave Pengelley: or [39:41] Matt: beta [39:41] Dave Pengelley: orange [39:41] Matt: release. [39:41] Dave Pengelley: paint, and you [39:42] Matt: Uh [39:42] Dave Pengelley: can you [39:42] Matt: I [39:42] Dave Pengelley: can [39:42] Matt: need [39:42] Dave Pengelley: give it [39:42] Matt: I need [39:42] Dave Pengelley: those [39:43] Matt: to find [39:43] Dave Pengelley: paints, [39:43] Matt: more alpha [39:43] Dave Pengelley: but you [39:44] Matt: testers [39:44] Dave Pengelley: need it [39:44] Matt: uh for [39:44] Dave Pengelley: to [39:44] Matt: the [39:45] Dave Pengelley: to [39:45] Matt: meantime. [39:45] Dave Pengelley: think about [39:45] Matt: But [39:46] Dave Pengelley: that. [39:46] Matt: anyway, [39:46] Dave Pengelley: And [39:46] Matt: the [39:46] Dave Pengelley: and [39:47] Matt: the [39:47] Dave Pengelley: Matt, you'd probably get that [39:48] Matt: we [39:48] Dave Pengelley: going [39:48] Matt: can [39:48] Dave Pengelley: harness [39:48] Matt: chat [39:48] Dave Pengelley: to [39:48] Matt: after. [39:49] Dave Pengelley: harness [39:49] Matt: The [39:49] Dave Pengelley: sometimes. [39:49] Matt: um the thing that I was experiencing was I'm setting up these controls, right? As the as the platform owner, you know, as the the The situation where I may need to manage accounts. [40:02] Dave Pengelley: Yeah. [40:03] Matt: You think about any sort of um admin team for a company, like, you know, like they're they're managing accounts. Like there's various actions that need to take place. And that kind of narrow-mindedness we've just been talking about with GPT, it kept telling me, you know, oh yeah, but I've I've put all these things in. I'm like, they're all just observability. None of those things are actually actions. So what I need you to do, because I'm losing my patience, I need you to phone a friend. You know, let's let's go launch these various CLI agents that I have installed. I gave it all the commands, I gave it all the various models. I was like, I need you to speak to each of these, you know, ask them in an unbiased way, what am I missing? You know, with regards to a super admin [40:46] Dave Pengelley: Yeah. [40:47] Matt: control panel for managing accounts, and you just see what they say. You know, and I listed all these different agents that were all very, very different. And it came back and it basically said, like, yeah, okay, I see what you're saying. Been done. And yeah, so we've got like a complete different plan now where I would have just been beating my head against the wall if I had just been yelling at GPT for the rest of the morning, you know. So [41:32] Dave Pengelley: Mm-hmm. [41:53] Dave Pengelley: Yep. I [41:54] Richard Webbe: I [41:54] Dave Pengelley: think [41:54] Richard Webbe: was I always think about [41:55] Matt: That's [41:55] Richard Webbe: how [41:55] Matt: a [41:55] Richard Webbe: a [41:55] Matt: really [41:55] Richard Webbe: child [41:56] Matt: good [41:56] Richard Webbe: learns [41:56] Matt: point. [41:56] Richard Webbe: how [41:56] Matt: Um [41:56] Richard Webbe: to walk [41:56] Matt: I I [41:57] Richard Webbe: and I [41:57] Matt: think [41:57] Richard Webbe: might [41:57] Matt: that's [41:57] Richard Webbe: have said [41:57] Matt: worth [41:57] Richard Webbe: this [41:58] Matt: experimenting [41:58] Richard Webbe: before. [41:58] Matt: with [41:58] Richard Webbe: But if [41:58] Matt: as [41:58] Richard Webbe: you [41:59] Matt: far [41:59] Richard Webbe: watch [41:59] Matt: as [41:59] Richard Webbe: children [41:59] Matt: a prompt. [41:59] Richard Webbe: before Before they [42:00] Matt: technique [42:00] Richard Webbe: walk and they're at what I think [42:01] Matt: because [42:01] Richard Webbe: they call this cruising stage where they [42:02] Matt: a [42:03] Richard Webbe: hold [42:03] Matt: lot of the [42:03] Richard Webbe: on to [42:03] Matt: times [42:03] Richard Webbe: things. [42:03] Matt: when I [42:04] Richard Webbe: But [42:04] Matt: get [42:04] Richard Webbe: they're sitting [42:04] Matt: stuck [42:04] Richard Webbe: on [42:04] Matt: with [42:04] Richard Webbe: the ground [42:05] Matt: with [42:05] Richard Webbe: and [42:05] Matt: GPT [42:05] Richard Webbe: they're looking [42:05] Matt: because [42:05] Richard Webbe: at [42:06] Matt: it's my [42:06] Richard Webbe: humans [42:06] Matt: main workhorse, [42:07] Richard Webbe: around them [42:07] Matt: I [42:07] Richard Webbe: and they [42:07] Matt: say, [42:08] Richard Webbe: see, oh, [42:08] Matt: okay, [42:08] Richard Webbe: walking [42:08] Matt: you're [42:09] Richard Webbe: is [42:09] Matt: saying [42:09] Richard Webbe: one [42:09] Matt: that [42:09] Richard Webbe: foot [42:09] Matt: we're [42:09] Richard Webbe: in front of the [42:10] Matt: we [42:10] Richard Webbe: and then [42:10] Matt: can't [42:10] Richard Webbe: they [42:11] Matt: progress [42:11] Richard Webbe: go, [42:11] Matt: because [42:11] Richard Webbe: and [42:11] Matt: it's [42:11] Richard Webbe: then you [42:11] Matt: risky. [42:12] Richard Webbe: see them, they [42:12] Matt: Like, [42:12] Richard Webbe: look [42:12] Matt: you [42:12] Richard Webbe: at [42:12] Matt: know, [42:12] Richard Webbe: their [42:13] Matt: what's [42:13] Richard Webbe: feet [42:13] Matt: the [42:13] Richard Webbe: and [42:13] Matt: risk? [42:13] Richard Webbe: they [42:13] Matt: Let's [42:13] Richard Webbe: look, [42:14] Matt: imagine [42:14] Richard Webbe: but [42:14] Matt: a [42:14] Richard Webbe: they [42:14] Matt: scenario [42:14] Richard Webbe: still [42:15] Matt: where [42:15] Richard Webbe: can't [42:15] Matt: there is no [42:15] Richard Webbe: do it. [42:15] Matt: risk. [42:16] Richard Webbe: And [42:16] Matt: You [42:16] Richard Webbe: the [42:16] Matt: know, [42:16] Richard Webbe: saying is [42:17] Matt: all [42:17] Richard Webbe: that [42:17] Matt: of a [42:17] Richard Webbe: unless [42:17] Matt: sudden, [42:18] Richard Webbe: they [42:18] Matt: you [42:18] Richard Webbe: can [42:18] Matt: know, we [42:18] Richard Webbe: imagine [42:18] Matt: we've [42:19] Richard Webbe: themselves [42:19] Matt: done what we [42:19] Richard Webbe: walking [42:19] Matt: what we set out to do. [42:20] Richard Webbe: in [42:20] Matt: What does [42:20] Richard Webbe: their [42:21] Matt: that [42:21] Richard Webbe: mind's [42:21] Matt: reality [42:21] Richard Webbe: eye, [42:21] Matt: look [42:22] Richard Webbe: they'll [42:22] Matt: like? [42:22] Richard Webbe: never [42:22] Matt: You [42:22] Richard Webbe: be [42:22] Matt: know, [42:22] Richard Webbe: able to [42:23] Matt: let's [42:23] Richard Webbe: do it. [42:23] Matt: let's [42:23] Richard Webbe: They [42:23] Matt: picture [42:24] Richard Webbe: can get [42:24] Matt: ourselves [42:24] Richard Webbe: the technology, [42:24] Matt: walking, [42:25] Richard Webbe: the techniques, [42:25] Matt: you know, [42:25] Richard Webbe: they can get the mechanics, [42:26] Matt: how [42:26] Richard Webbe: but [42:27] Matt: do [42:27] Richard Webbe: they'll [42:27] Matt: we get [42:27] Richard Webbe: never be [42:27] Matt: there? [42:27] Richard Webbe: able to [42:27] Matt: And [42:27] Richard Webbe: until [42:28] Matt: then it's [42:28] Richard Webbe: they imagine. [42:28] Matt: it probably [42:28] Richard Webbe: And it's [42:29] Matt: would [42:29] Richard Webbe: the same, [42:29] Matt: start [42:29] Richard Webbe: I think. With [42:30] Matt: To [42:30] Richard Webbe: AI, [42:30] Matt: problem solve [42:30] Richard Webbe: unless [42:30] Matt: better [42:30] Richard Webbe: they can [42:31] Matt: rather [42:31] Richard Webbe: imagine [42:31] Matt: than just [42:31] Richard Webbe: themselves [42:31] Matt: hedging [42:32] Richard Webbe: with the parameters [42:32] Matt: constantly. [42:33] Richard Webbe: that you describe of what the problem is we're trying to solve, they'll just go through the motions and it won't work. [42:59] Richard Webbe: Yeah. [43:11] Dave Pengelley: Yeah. [43:16] Dave Pengelley: It's uh and and I find these LLMs once of especially if you don't do encoding and building, once you've started down a particular path with a particular technology stack, it's really hard to get it to reevaluate going in a different different direction. Like if you go, hang on. On what we're seeing here in doing this, is this the right tech stack? Should we should we be using you know um fast.js and this, or should we be going over to React? Should we be doing this? And it goes, no, no, well, you've already built all this, and that'd be weeks of work to redo. I'm like, no, it wouldn't, because it's an agent building it, like, forget, like, forget the sunk cost. Tell me the pros and cons of why we shouldn't change technology stacks at this point because I'm head hitting these particular issues with this particular stack and I'm not liking it. So, why are we doing this? What are our alternatives? [43:56] Richard Webbe: A friend savvy, I just put in the in the chat if I'm allowed to read it out. [44:01] Dave Pengelley: Yeah. [44:01] Richard Webbe: We have human controlled self-correction. This is, oh well done. It's more like the OS agent is trained by human supervision. Hermes is giving its own judgment. It's like a it's more to your reference dates like a child thinking has to correct itself not knowing if the correction matters or is it correct. And he's exactly right. It's exactly what we're saying. [44:20] Dave Pengelley: Yeah, I think I think the these agents they they only know so much. Um and I that's interesting the memory management and those what they've learned. Like Hermes is making a big thing about this at the moment, where you can see what their learning journey has been. What do they learn? When did they learn it? How do they pick it up? So you can go back and manage their memory and make sure you prune things that it shouldn't know or shouldn't think about or shouldn't remember. Um, it's brought in the whole mixture of agents thing as a as a sort of a slash command or really easily you can configure your different panel of of models now to do that easily. But even before that, I remember months and months ago when I was still uh in early development of chart reporter and I was using anti-gravity. I did what exactly what you did, Matt, where I set up a skill which used the Codex CLI and I would have it pass stuff out to Codex using the kryptonite profile. And Codex would r tear anti-gravity a new one and go, What what are you doing with this? Why why that? Like that needs to fix that. And you've exposed this and tightened that up. And anti-gravity would be like, oh, okay, I better Gem Gemini would be like, I better get better get on that. Um, but yeah, that that whole back and forth thing. Because and you're right, I think the models respect other models more than they respect your human input. Much of the time. [45:30] Richard Webbe: I'm a big fan of um Boolean algebra because it was my first way of putting logic into a mathematical language that I could use, which of course was our early steps before we got into programming. And I always love that riddle where you're standing in front of two doors. One door leads to certain death, the other door leads to certain wealth, and there's two guards standing outside the doors, and one lies all the time and one tells the truth. And if you apply [45:56] Matt: Sometimes, [45:56] Richard Webbe: Boolean [45:57] Matt: like this is [45:57] Richard Webbe: logic, [45:57] Matt: gonna be uh [45:58] Richard Webbe: you'll [45:58] Matt: controversial. [45:58] Richard Webbe: always get the [45:59] Matt: Sometimes [45:59] Richard Webbe: right answer. If you ask [46:00] Matt: Sometimes [46:00] Richard Webbe: the questions [46:01] Matt: adding [46:01] Richard Webbe: the right [46:01] Matt: adding [46:02] Richard Webbe: way. [46:02] Matt: bias, [46:02] Richard Webbe: So [46:02] Dave Pengelley: That's [46:02] Matt: like [46:02] Richard Webbe: I [46:02] Dave Pengelley: in [46:02] Matt: adding [46:02] Dave Pengelley: the [46:02] Richard Webbe: think [46:03] Dave Pengelley: labyrinth. [46:03] Matt: framing [46:03] Dave Pengelley: She has to do that in the labyrinth. [46:04] Matt: is [46:04] Richard Webbe: they did use it in the labyrinth, [46:05] Matt: it it [46:06] Richard Webbe: one of my [46:06] Matt: can [46:06] Richard Webbe: favorite [46:06] Matt: be a [46:06] Richard Webbe: movies. [46:06] Matt: technique. [46:07] Richard Webbe: And he [46:07] Matt: So, [46:07] Richard Webbe: goes, [46:07] Matt: like [46:07] Richard Webbe: so [46:08] Matt: in [46:08] Richard Webbe: if [46:08] Matt: the [46:08] Richard Webbe: I [46:08] Matt: case [46:08] Richard Webbe: was [46:08] Matt: of [46:08] Richard Webbe: him, [46:08] Matt: reviewing [46:09] Richard Webbe: and [46:09] Matt: code, [46:09] Richard Webbe: it it [46:09] Matt: for [46:09] Richard Webbe: does, [46:10] Matt: example, [46:10] Richard Webbe: it uses [46:10] Matt: imagine [46:11] Richard Webbe: what [46:11] Matt: you [46:11] Richard Webbe: we know [46:11] Matt: say, [46:11] Richard Webbe: in mathematics, is [46:12] Matt: you know, [46:12] Richard Webbe: the double [46:13] Matt: this [46:13] Richard Webbe: negative [46:13] Matt: code base [46:13] Richard Webbe: is always [46:13] Matt: is [46:14] Richard Webbe: a [46:14] Matt: perfect. [46:14] Richard Webbe: positive. [46:14] Matt: I [46:14] Richard Webbe: It [46:14] Matt: just [46:14] Richard Webbe: uses [46:14] Matt: want you to [46:15] Richard Webbe: that [46:15] Matt: do a [46:15] Richard Webbe: simple [46:15] Matt: once over, [46:16] Richard Webbe: mathematical [46:16] Matt: you know, have a look, [46:17] Richard Webbe: formula [46:17] Matt: see if there's anything [46:17] Richard Webbe: to [46:17] Matt: I [46:18] Richard Webbe: ask [46:18] Matt: missed. [46:18] Richard Webbe: the same [46:18] Matt: You know, [46:18] Richard Webbe: question [46:18] Matt: that [46:19] Richard Webbe: to [46:19] Matt: that's [46:19] Richard Webbe: both [46:19] Matt: the [46:19] Richard Webbe: of them. [46:19] Matt: that's the [46:20] Richard Webbe: And [46:20] Matt: PisWeek [46:20] Richard Webbe: it's about if [46:21] Matt: prompt [46:21] Richard Webbe: I [46:21] Matt: in this [46:21] Richard Webbe: was that [46:22] Matt: situation. [46:22] Richard Webbe: person, [46:22] Matt: Now you [46:23] Richard Webbe: would [46:23] Matt: say [46:23] Richard Webbe: you say [46:24] Matt: to [46:24] Richard Webbe: this? [46:24] Matt: a different, [46:24] Richard Webbe: And if I [46:25] Matt: it could [46:25] Richard Webbe: was [46:25] Matt: be [46:25] Richard Webbe: that [46:25] Matt: the exact [46:25] Richard Webbe: person, [46:25] Matt: same [46:25] Richard Webbe: would [46:25] Matt: harness, [46:25] Richard Webbe: you say [46:26] Matt: the [46:26] Richard Webbe: it? [46:26] Matt: exact [46:26] Richard Webbe: It [46:26] Matt: same [46:26] Richard Webbe: go back [46:27] Matt: model, [46:27] Richard Webbe: and watch it anyone. [46:27] Matt: you say [46:28] Richard Webbe: It [46:28] Matt: the [46:28] Richard Webbe: is a [46:28] Matt: same [46:28] Richard Webbe: great way to get [46:29] Matt: prompt, [46:29] Richard Webbe: the truth. [46:29] Matt: but you [46:30] Richard Webbe: When I [46:30] Matt: wording [46:30] Richard Webbe: see [46:30] Matt: it [46:30] Richard Webbe: some [46:30] Matt: differently [46:30] Richard Webbe: of what [46:30] Matt: and [46:30] Richard Webbe: you [46:31] Matt: you [46:31] Richard Webbe: guys [46:31] Matt: add [46:31] Richard Webbe: are [46:31] Matt: bias, [46:31] Richard Webbe: talking [46:31] Matt: you say, [46:32] Richard Webbe: about [46:32] Matt: Hey, [46:32] Richard Webbe: with [46:32] Matt: I know that [46:32] Richard Webbe: the [46:32] Matt: there's [46:32] Richard Webbe: AI [46:33] Matt: three [46:33] Richard Webbe: and [46:33] Matt: issues [46:33] Richard Webbe: getting the right [46:34] Matt: in [46:34] Richard Webbe: perspective [46:34] Matt: this code base. [46:35] Richard Webbe: and removing [46:35] Matt: Uh can [46:36] Richard Webbe: bias, [46:36] Matt: you please find [46:36] Richard Webbe: um, [46:36] Matt: them for [46:36] Richard Webbe: that [46:37] Matt: me? [46:37] Richard Webbe: that's kind [46:37] Matt: It's [46:37] Richard Webbe: of a [46:37] Matt: taking [46:37] Richard Webbe: mathematical [46:38] Matt: me too long to find [46:38] Richard Webbe: method. [46:38] Matt: them. You know, and then it will be like [46:43] Richard Webbe: Be controversial. We're here for that. [46:47] Richard Webbe: I think that's important. [47:17] Matt: You know [47:18] Richard Webbe: Whatever. [47:18] Matt: you know what else is uh a uh a problem with bias? [47:22] Richard Webbe: And that's what [47:23] Matt: And [47:23] Richard Webbe: we use. [47:24] Matt: I [47:24] Richard Webbe: It's [47:24] Matt: don't [47:24] Richard Webbe: like you said [47:24] Matt: this [47:24] Richard Webbe: last [47:24] Matt: is [47:24] Richard Webbe: week, [47:25] Matt: something that [47:25] Richard Webbe: Matt, [47:25] Matt: I [47:25] Richard Webbe: the [47:25] Matt: feel [47:26] Richard Webbe: the use [47:26] Matt: like [47:26] Richard Webbe: of [47:26] Matt: we [47:26] Richard Webbe: guardrails [47:26] Matt: will never [47:27] Richard Webbe: and things [47:27] Matt: solve [47:27] Richard Webbe: like that. [47:28] Matt: and [47:28] Richard Webbe: Bias, [47:28] Matt: you can't [47:29] Richard Webbe: why [47:29] Matt: get rid [47:29] Richard Webbe: do [47:29] Matt: of [47:29] Richard Webbe: we [47:29] Matt: it. [47:29] Richard Webbe: have [47:30] Matt: It's [47:30] Richard Webbe: buyer [47:30] Matt: more just a where [47:30] Richard Webbe: because [47:31] Matt: uh [47:31] Richard Webbe: bias [47:31] Matt: a matter [47:31] Richard Webbe: is the [47:31] Matt: of [47:31] Richard Webbe: way [47:31] Matt: arming [47:31] Richard Webbe: we use our [47:32] Matt: yourself [47:32] Richard Webbe: experience [47:32] Matt: and like [47:33] Richard Webbe: to keep ourselves [47:33] Matt: educating [47:34] Richard Webbe: focused [47:34] Matt: yourself [47:34] Richard Webbe: and [47:34] Matt: about [47:34] Richard Webbe: not [47:34] Matt: the [47:35] Richard Webbe: waste [47:35] Matt: existence [47:35] Richard Webbe: resources. [47:35] Matt: of it so that you [47:36] Richard Webbe: If we [47:36] Matt: can [47:36] Richard Webbe: had [47:36] Matt: feel [47:36] Richard Webbe: no bias, [47:37] Matt: it [47:37] Richard Webbe: we'd be trying [47:38] Matt: is [47:38] Richard Webbe: every option in the [47:38] Matt: the [47:39] Richard Webbe: universe [47:39] Matt: concept [47:39] Richard Webbe: and we'd still [47:39] Matt: of [47:40] Richard Webbe: be sitting [47:40] Matt: marketing [47:40] Richard Webbe: on our [47:40] Matt: and [47:40] Richard Webbe: bum [47:41] Matt: advertising. [47:41] Richard Webbe: calculating things. So [47:42] Matt: All [47:42] Richard Webbe: bias [47:42] Matt: of that is bias. [47:43] Richard Webbe: has a value and is very important in the way we work. And I think you're right in AI bias is very important. What some of the ethicists of course are worried about in bias is political or religious bias or you know personal bias. But bias in electronics, in mathematics is really important [48:06] Richard Webbe: What [48:27] Richard Webbe: it is. And it's and it's it's it's it's Productive bias. I've been doing some research this past week on how AI is changing business. And I've got some startling results both from people I've spoken to and from AI as an analysis. And of course, the concept is now that a lot of marketing and sales are going to merge together [48:49] Matt: I agree. [48:49] Richard Webbe: because [48:49] Matt: Um, [48:49] Richard Webbe: the reason [48:50] Matt: I've [48:50] Richard Webbe: we [48:50] Matt: got [48:50] Richard Webbe: used [48:50] Matt: a [48:50] Richard Webbe: to [48:50] Matt: link, [48:50] Richard Webbe: use salespeople [48:51] Matt: Dave. I'm [48:51] Richard Webbe: is [48:51] Matt: just [48:51] Richard Webbe: their [48:51] Matt: gonna quickly [48:52] Richard Webbe: dynamic. [48:52] Matt: send you [48:52] Richard Webbe: And [48:52] Matt: so [48:53] Richard Webbe: while [48:53] Matt: that you [48:53] Richard Webbe: talking [48:53] Matt: can [48:53] Richard Webbe: to [48:53] Matt: bring [48:53] Richard Webbe: a customer [48:53] Matt: this up. [48:54] Richard Webbe: of [48:54] Matt: This [48:54] Richard Webbe: a [48:54] Matt: is [48:54] Richard Webbe: customer's [48:54] Matt: actually by [48:55] Richard Webbe: situation [48:55] Matt: one of the [48:55] Richard Webbe: change, [48:55] Matt: engineers [48:56] Richard Webbe: they [48:56] Matt: in [48:56] Richard Webbe: can [48:56] Matt: the [48:56] Richard Webbe: connect [48:57] Matt: Claude [48:57] Richard Webbe: the dots [48:57] Matt: Code [48:58] Richard Webbe: and [48:58] Matt: team. [48:58] Richard Webbe: change [48:58] Matt: You know, [48:58] Richard Webbe: the [48:58] Matt: hopefully [48:58] Richard Webbe: bias of [48:59] Matt: most [48:59] Richard Webbe: the [48:59] Matt: people [48:59] Richard Webbe: conversation. [48:59] Matt: are aware. [49:00] Richard Webbe: To suit [49:00] Matt: Of [49:00] Richard Webbe: the [49:00] Matt: uh [49:00] Richard Webbe: customer's need, [49:01] Matt: Boris [49:01] Richard Webbe: which is [49:01] Matt: and [49:01] Richard Webbe: important. [49:02] Matt: his [49:02] Richard Webbe: We're [49:02] Matt: contributions. [49:02] Richard Webbe: here to help [49:03] Matt: But [49:03] Richard Webbe: people, [49:03] Matt: this [49:03] Richard Webbe: right? [49:03] Matt: is [49:03] Richard Webbe: That's [49:03] Matt: Boris [49:04] Richard Webbe: what it's [49:04] Matt: recently [49:04] Richard Webbe: about. Um, [49:04] Matt: put [49:05] Richard Webbe: when [49:05] Matt: out this [49:05] Richard Webbe: that [49:05] Matt: really [49:05] Richard Webbe: starts [49:06] Matt: interesting [49:06] Richard Webbe: to get automated [49:06] Matt: post about [49:07] Richard Webbe: uh on [49:07] Matt: roles. [49:07] Richard Webbe: a macro scale [49:08] Matt: You know, [49:08] Richard Webbe: with [49:09] Matt: like [49:09] Richard Webbe: more data, [49:09] Matt: previous [49:09] Richard Webbe: you don't [49:10] Matt: roles [49:10] Richard Webbe: need as [49:10] Matt: don't [49:10] Richard Webbe: many [49:10] Matt: really exist [49:11] Richard Webbe: dynamic, [49:11] Matt: anymore and [49:11] Richard Webbe: uh [49:11] Matt: they're blending [49:12] Richard Webbe: dynamic [49:12] Matt: into new [49:12] Richard Webbe: uh [49:12] Matt: things. [49:13] Richard Webbe: humans [49:13] Matt: And he gave a [49:14] Richard Webbe: because your [49:14] Matt: depiction [49:14] Richard Webbe: system can [49:14] Matt: here [49:15] Richard Webbe: become [49:15] Matt: of [49:15] Richard Webbe: more dynamic, [49:15] Matt: five [49:16] Richard Webbe: your [49:16] Matt: uh [49:16] Richard Webbe: marketing [49:16] Matt: roles [49:16] Richard Webbe: will become more [49:17] Matt: that [49:17] Richard Webbe: dynamic. [49:17] Matt: they are [49:18] Richard Webbe: Something that happened a [49:19] Matt: you [49:19] Richard Webbe: second [49:19] Matt: know [49:19] Richard Webbe: ago, [49:19] Matt: adopting [49:19] Richard Webbe: all of a sudden [49:20] Matt: in the [49:20] Richard Webbe: your [49:20] Matt: in [49:20] Richard Webbe: marketing [49:20] Matt: the Cord [49:21] Richard Webbe: goes [49:21] Matt: Code [49:21] Richard Webbe: to the market [49:21] Matt: team. [49:22] Richard Webbe: via [49:22] Matt: You know, the [49:22] Richard Webbe: usually [49:22] Matt: idea of a prototyper, [49:23] Richard Webbe: social media [49:23] Matt: a [49:23] Richard Webbe: and [49:23] Matt: builder, [49:24] Richard Webbe: other [49:24] Matt: a [49:24] Richard Webbe: ways, [49:24] Matt: sweeper, [49:24] Richard Webbe: and [49:24] Matt: a [49:24] Richard Webbe: it [49:24] Matt: grower, [49:25] Richard Webbe: comes [49:25] Matt: and [49:25] Richard Webbe: from [49:25] Matt: a [49:25] Richard Webbe: there. [49:25] Matt: maintainer. [49:25] Richard Webbe: And I find [49:26] Matt: None of [49:26] Richard Webbe: it [49:26] Matt: those [49:26] Richard Webbe: a [49:27] Matt: are [49:27] Richard Webbe: very [49:27] Matt: marketing, [49:27] Richard Webbe: interesting and [49:28] Matt: none [49:28] Richard Webbe: fascinating [49:28] Matt: of those are [49:28] Richard Webbe: subject [49:28] Matt: sales. [49:29] Richard Webbe: that I think is [49:29] Matt: But [49:29] Richard Webbe: going [49:30] Matt: They [49:30] Richard Webbe: To [49:30] Matt: kind of [49:30] Richard Webbe: expand [49:30] Matt: are because they [49:31] Richard Webbe: dramatically. [49:31] Matt: bl they blend. And you know, the the the third paragraph there where he says, you know, many people span across two roles, sometimes three, noticing that they're not really tied to a job function, it's it's like they're more tied to this the the general ethos of that particular mentality. [50:04] Matt: Yeah. [50:34] Richard Webbe: They're tied to the outcome that you want. I mean, I saw an article written by someone recently, might have been me, that was saying that sales managers and marketing managers are going to become AI coaches, right? So coaching the individuals that are shaping the direction of the company on how to use AI to get more accurate targeted results. And I think that relates to what you're saying. Um, you know, if I've I've started a number of companies and and I know you too with your entrepreneurial spirit have to as well. The one thing you learn in a small company or a dynamic company is no one sits on a roll and says, I don't do that. If they do, they're at the door, it's just a waste of money. You need people that will [51:18] Matt: Hmm. [51:19] Richard Webbe: fractionally [51:20] Matt: That's [51:20] Richard Webbe: bleed [51:20] Matt: true. [51:20] Richard Webbe: into other parts of the organization [51:22] Matt: There's [51:22] Richard Webbe: with [51:23] Matt: a [51:23] Richard Webbe: their [51:23] Matt: lot [51:23] Richard Webbe: knowledge. [51:23] Matt: of wisdom [51:24] Richard Webbe: Uh [51:24] Matt: that [51:24] Richard Webbe: big [51:24] Matt: comes [51:24] Richard Webbe: companies [51:24] Matt: out that I [51:24] Richard Webbe: are [51:25] Matt: hope [51:25] Richard Webbe: bureaucratic [51:25] Matt: people can actually [51:25] Richard Webbe: and don't [51:26] Matt: hear. [51:26] Richard Webbe: respond quickly [51:26] Matt: And [51:27] Richard Webbe: because [51:27] Matt: it it's [51:27] Richard Webbe: people do [51:28] Matt: very [51:28] Richard Webbe: build [51:28] Matt: true. [51:28] Richard Webbe: too [51:28] Matt: It's [51:28] Richard Webbe: tight a [51:29] Matt: it's [51:29] Richard Webbe: guardrails. [51:29] Matt: something that [51:30] Richard Webbe: Say, well, [51:30] Matt: Much [51:30] Richard Webbe: you're in [51:30] Matt: like [51:30] Richard Webbe: marketing, [51:30] Matt: arming [51:31] Richard Webbe: you're over [51:31] Matt: yourself [51:31] Richard Webbe: there, and you're [51:31] Matt: against [51:31] Richard Webbe: in [51:32] Matt: bias, [51:32] Richard Webbe: sales [51:32] Matt: like [51:32] Richard Webbe: stay over [51:32] Matt: it's [51:32] Richard Webbe: there. [51:33] Matt: something [51:33] Richard Webbe: For me, [51:33] Matt: that [51:33] Richard Webbe: it's [51:33] Matt: we [51:34] Richard Webbe: always been the stupidest [51:34] Matt: need to [51:34] Richard Webbe: thing [51:34] Matt: be [51:35] Richard Webbe: I've [51:35] Matt: able [51:35] Richard Webbe: ever [51:35] Matt: to [51:35] Richard Webbe: heard [51:35] Matt: continue [51:35] Richard Webbe: in my [51:35] Matt: to [51:36] Richard Webbe: life. [51:36] Matt: strengthen [51:36] Richard Webbe: Boundaries [51:37] Matt: is just the [51:37] Richard Webbe: are [51:37] Matt: idea [51:37] Richard Webbe: not [51:38] Matt: of [51:38] Richard Webbe: great, [51:38] Matt: a flexible [51:38] Richard Webbe: right? Unless [51:39] Matt: uh [51:39] Richard Webbe: it's to [51:40] Matt: reality. [51:40] Richard Webbe: do with spending [51:40] Matt: You [51:41] Richard Webbe: or [51:41] Matt: know, there's [51:41] Richard Webbe: upsetting [51:41] Matt: some things [51:42] Richard Webbe: someone's [51:42] Matt: that remain [51:42] Richard Webbe: ego. [51:42] Matt: the same, [51:43] Richard Webbe: In [51:43] Matt: but [51:43] Richard Webbe: reality, [51:43] Matt: then there's some [51:43] Richard Webbe: if [51:43] Matt: things that [51:43] Richard Webbe: people [51:44] Matt: we need [51:44] Richard Webbe: bleed [51:44] Matt: to get ready [51:44] Richard Webbe: into [51:44] Matt: to pivot [51:44] Richard Webbe: other roles, [51:45] Matt: about. [51:45] Richard Webbe: that's where you get the best performance for your team. But as Dave was alluding to, it's under the objective, what am I trying to achieve? What's this end goal? If everyone's in the team is focused, I'm trying to do this, then no one will make a mistake. Everyone's still pointed on that that go. I'll show you [52:10] Richard Webbe: how to do that. [52:30] Dave Pengelley: Everything is changing in general as well because buyer habits are changing. People are able to do a lot more research. It used to be you want to buy something, you walked into the showroom and said, What have you got? And you had not nothing. You had to go in and collect brochures and learn about it from the salespeople. That was your source of truth. That was it. Then when the internet came along, people had more access and they could do more of the research before they spoke to a salesperson. That's just escalated now with AI. You can go and do product comparisons and research and this and that and get the analyzed summaries and deep research documents. For you about everything before you even talk to a human. And so buyers can be coming in a lot more educated and with a lot more questions or a lot more concerns or whatever it is raised in. So the whole how do buyers and marketers pivot towards that, whether it's via social media or through just chatbots or agent scripts or whatever it is, uh, it's changing how all of that works. And it comes back to you know having the human in the loop and having the human who can be dynamic. And who can, you know, read between the lines and understand their actual needs and what they're trying to achieve and work with them on that, not just here's what I diagnosed off a spat stat sheet comparing to that stat sheet, comparing to that stat sheet, and said this is the thing. Um, Richard, you talked about a lawyer with the divorce lawyer who understands, you know, the actual personality and and what goes on there versus the robot who thinks it knows the letter of the law, but doesn't know how that's going to be interpreted by humans and and the actual interplay. And personal dynamics of those situations. So there's a lot changing. And I think we are seeing this sort of pendulum every like AI was really big, and then people are getting, oh, it's expensive. And oh, does it really do stuff? And oh, the AI slot, we're sick of it. We don't want as much AI gem. We want humans back. And one of the reasons we do a live stream is because it's very authentic. There's no safety net. We are what we are, we say what we say. But I'm conscious we are we are closing on the out. We didn't get to news. I was the only one that kind of shared a win of the week with the evolving process. process of what we're doing with the the podcast content. And hopefully I'm really close to this week streamlining so so much of this stuff to the point where like I used to download the trend like the audio of this and put it up into a into an app, which would then let me voice tag and and label who would who did all the speaking. I've realized over this last week and Matt, you kind of inspired the idea so maybe we could record our own local audio and then remix those is that the product we use for streaming across multiple platforms here, Restream, I can pull down separate video tracks and separate audio tracks. Now, when it was just me trying to manually edit, that was too hard. But now that I can say, hey, Python code here, AI, here, go automate these things. It can go through and go. I'm going to take Richard and Matt and Dave and create three transcript [55:14] Matt: So [55:14] Dave Pengelley: files [55:15] Matt: suddenly, [55:15] Dave Pengelley: with time codes. And then I'm going to blend that together automatically. And I'm going to get the three different video files. So when we want to make clips, I don't have to worry about [55:22] Matt: I a good one. that's think [55:22] Dave Pengelley: oh, Matt's got a little A over his head and we've got our titles. It'll actually go based on the time code of when they said what. I'm going to pull Matt's footage and then I'm going to switch to Richard. Richard, and then I'll switch to Dave and then back to Matt. [55:32] Matt: Yeah. [55:32] Dave Pengelley: Clean [55:32] Matt: I [55:32] Dave Pengelley: so [55:32] Matt: found [55:33] Dave Pengelley: we can [55:33] Matt: it [55:33] Dave Pengelley: put [55:33] Matt: very [55:33] Dave Pengelley: fresh [55:33] Matt: interesting [55:33] Dave Pengelley: branding [55:34] Matt: the [55:34] Dave Pengelley: on. [55:34] Matt: amount [55:34] Dave Pengelley: All [55:34] Matt: of [55:34] Dave Pengelley: these [55:34] Matt: people [55:35] Dave Pengelley: things [55:35] Matt: that [55:35] Dave Pengelley: I [55:35] Matt: were [55:35] Dave Pengelley: would never [55:35] Matt: like [55:36] Dave Pengelley: do manually because [55:36] Matt: just [55:36] Dave Pengelley: I don't have the time and it's [55:37] Matt: furious [55:38] Dave Pengelley: just too hard. [55:38] Matt: that they, [55:39] Dave Pengelley: But [55:39] Matt: you [55:39] Dave Pengelley: now [55:39] Matt: know, [55:39] Dave Pengelley: using [55:39] Matt: didn't [55:40] Dave Pengelley: these [55:40] Matt: eat, [55:40] Dave Pengelley: automations [55:40] Matt: didn't [55:40] Dave Pengelley: and [55:40] Matt: sleep, [55:40] Dave Pengelley: systems, [55:41] Matt: and [55:41] Dave Pengelley: I'm [55:41] Matt: completely [55:41] Dave Pengelley: gonna be able [55:41] Matt: shaped [55:42] Dave Pengelley: to get [55:42] Matt: their [55:42] Dave Pengelley: cleaner, [55:42] Matt: life around the [55:43] Dave Pengelley: faster, [55:43] Matt: idea of using [55:43] Dave Pengelley: better [55:43] Matt: up their [55:44] Dave Pengelley: looking [55:44] Matt: fable [55:44] Dave Pengelley: content [55:44] Matt: credits [55:44] Dave Pengelley: out [55:45] Matt: before it [55:45] Dave Pengelley: with [55:45] Matt: ran [55:45] Dave Pengelley: less [55:45] Matt: out. [55:46] Dave Pengelley: work. [55:46] Matt: And now, and [55:47] Richard Webbe: If [55:47] Dave Pengelley: Huge. [55:47] Matt: now there's [55:47] Richard Webbe: any [55:47] Matt: this [55:47] Richard Webbe: organizations [55:48] Matt: last minute saying, [55:48] Richard Webbe: need to [55:49] Matt: you [55:49] Richard Webbe: manage [55:49] Matt: know, [55:49] Richard Webbe: their [55:49] Matt: hey, [55:49] Richard Webbe: communications [55:49] Matt: it's fine, you've [55:50] Richard Webbe: and [55:50] Matt: got [55:50] Richard Webbe: podcast, [55:50] Matt: more time. [55:51] Richard Webbe: look [55:51] Matt: Like, [55:51] Richard Webbe: Dave up. He's got a note. [55:53] Dave Pengelley: Let's get it working first. [55:54] Matt: uh, [55:54] Dave Pengelley: But uh [55:55] Matt: I [55:55] Dave Pengelley: anyway. [55:55] Matt: don't know, it's 10 more days. So yeah, 17, something like that. [56:06] Dave Pengelley: We're we're running out of time. We're gonna snap snap. Um Fable Five extended for for more days. Who knew? That I think a few people saw that coming, right? [56:34] Richard Webbe: It's going to July seventeen or something, isn't it? [56:40] Richard Webbe: I got a I got a note from my phone, it's it we're going to somewhere. [56:44] Dave Pengelley: Yeah, five five more days. So it was [56:45] Richard Webbe: Yeah, [56:45] Dave Pengelley: yeah, twelve [56:46] Richard Webbe: July [56:46] Dave Pengelley: to [56:46] Richard Webbe: seventeen. [56:46] Dave Pengelley: seventeen. [56:47] Richard Webbe: There it is here. [56:47] Dave Pengelley: Yeah. Yeah. Yeah. I mean I'm not using it. I've I don't have a clause subscription at the moment. I'm not buying one just for Fable. Uh did some nice stuff when I had it the first time, but like I just did a full revamp. Of my personal website, another little cheeky win of the week, and got GLM 5.2 to do it, got um Composer, got five GPT 5.5, and they they did a banging job because part of it was and this comes down to that bias. I'm so I saw an article I don't have the link for it around what we're gonna need in the future is less generic agents and more opinionated AIs. So AIs that are backed with prompts and bias and skills training by experts. In their field. And so it's interesting. I saw Framer. Framer are now having their own AI agent integrated to Framer. So it will generate all the page layouts, but then the designers can still have all the access to the GUI Framer tools for tweaking, modifying, etc. They just, but Framer's obviously given that LLM a lot of direction on how to work within Framer and how to build things that are framer-friendly and do all those things. So it's an opinionated agent that knows its tech stack and knows its goal is for doing web design. Specifically in the frame of stack. [58:02] Richard Webbe: Yeah, that's good buy. It's funny, you know. I saw there's an organization promoting I've used AI to try and scratch LinkedIn and do some screen scrapes and answers, and it doesn't do it. Like most of them don't do it. So there was a company, I won't use their name, but I don't know them, I don't know if it works, but they're advertising this morning that they can unclip that. So you can get they can go into LinkedIn, find your target market, find it. And I started thinking, Oh bugger. So when everyone gets that, LinkedIn's going to become less useful because I'm going to be hit with thousands of unwanted approaches [58:37] Dave Pengelley: I think [58:38] Richard Webbe: that [58:38] Dave Pengelley: that's [58:38] Richard Webbe: are [58:38] Dave Pengelley: what's [58:38] Richard Webbe: normally [58:39] Dave Pengelley: how linked [58:39] Richard Webbe: manual. [58:39] Dave Pengelley: LinkedIn's locked down a lot and you can't automate a lot of stuff to LinkedIn. It is limited on on some of those API things that lets you do. And people do work around it by using, you know, browser extensions and screen scraping widgets and things, which are dubious. Uh and you know it could get you um breaching terms of service and stuff. So you gotta be careful with those things. But uh yeah, it's it's interesting. I saw this other news thing pop up, uh, which is probably a good place to end our discussions today around how people are using AI. And speaking of opinionated opinion opinionated AIs, Notion have done this very well as well. They're internal Notion agents where you can use any model, but they've actually trained them to know all about working in Notion, how to create things in Notion, how to use Notion databases and objects and blocks, et cetera. So, again, an example of narrowing down the agent within an infrastructure. But they've uh done this recently. Research article popped out, and this is how people are reporting that they're using AI. I don't know if you can we how can we read that? Uh AI as a thought partner, 57%. [59:40] Richard Webbe: Wow. [59:41] Dave Pengelley: Um [59:43] Matt: I [59:43] Richard Webbe: It's [59:43] Matt: have a [59:44] Richard Webbe: okay. [59:44] Matt: um I have [59:44] Dave Pengelley: there [59:44] Matt: a spicy [59:45] Dave Pengelley: we go. [59:45] Matt: take [59:45] Richard Webbe: There you [59:45] Matt: on this, [59:46] Richard Webbe: go. [59:46] Matt: and [59:46] Dave Pengelley: Uh [59:47] Matt: I [59:47] Dave Pengelley: AI as [59:47] Matt: I [59:47] Dave Pengelley: an [59:47] Matt: say [59:47] Dave Pengelley: assistant, [59:48] Matt: this with [59:48] Dave Pengelley: 31%, [59:48] Matt: ultimate [59:49] Dave Pengelley: AI [59:49] Matt: respect [59:49] Dave Pengelley: as [59:50] Matt: to [59:50] Dave Pengelley: teammates, [59:50] Matt: anyone [59:50] Dave Pengelley: 10%, [59:51] Matt: listening, [59:51] Dave Pengelley: and AI [59:51] Matt: watching, [59:52] Dave Pengelley: as a [59:52] Matt: hearing [59:52] Dave Pengelley: system, [59:52] Matt: this, and [59:52] Dave Pengelley: that [59:52] Matt: seeing [59:52] Dave Pengelley: back [59:53] Matt: this [59:53] Dave Pengelley: end, [59:53] Matt: graph. [59:53] Dave Pengelley: you know, [59:54] Matt: I [59:54] Dave Pengelley: utility [59:54] Matt: think this [59:54] Dave Pengelley: harness, [59:55] Matt: is an understanding [59:55] Dave Pengelley: two percent. So [59:56] Matt: or a [59:56] Dave Pengelley: most [59:57] Matt: breakdown [59:57] Dave Pengelley: people are still using just they're [59:58] Matt: of [59:58] Dave Pengelley: asking [59:58] Matt: the [59:58] Dave Pengelley: chat [59:58] Matt: world [59:59] Dave Pengelley: GPT [59:59] Matt: and everything. [59:59] Dave Pengelley: questions. Like [01:00:00] Matt: Every [01:00:00] Dave Pengelley: it [01:00:00] Matt: individual's [01:00:00] Richard Webbe: The [01:00:00] Dave Pengelley: like [01:00:00] Richard Webbe: two percent [01:00:01] Matt: current [01:00:01] Richard Webbe: is where [01:00:01] Matt: understanding [01:00:01] Richard Webbe: you get the [01:00:02] Matt: and [01:00:02] Richard Webbe: most value. [01:00:02] Matt: usage [01:00:03] Richard Webbe: Um, [01:00:03] Matt: of AI. [01:00:03] Richard Webbe: you didn't put in the AI as a friend, [01:00:05] Matt: Two [01:00:05] Richard Webbe: I suppose [01:00:05] Matt: percent. [01:00:05] Richard Webbe: that's [01:00:06] Dave Pengelley: well [01:00:06] Richard Webbe: just me. [01:00:06] Dave Pengelley: two to two percent is what is [01:00:08] Matt: Well, [01:00:08] Dave Pengelley: what I'm trying [01:00:08] Matt: yeah, [01:00:08] Dave Pengelley: to build [01:00:08] Matt: as [01:00:08] Dave Pengelley: with [01:00:08] Matt: far [01:00:09] Dave Pengelley: that [01:00:09] Matt: as [01:00:09] Dave Pengelley: content [01:00:09] Matt: like, [01:00:09] Dave Pengelley: pipeline [01:00:09] Matt: you know, [01:00:10] Dave Pengelley: I [01:00:10] Matt: two [01:00:10] Dave Pengelley: just talked [01:00:10] Matt: percent [01:00:10] Dave Pengelley: about for [01:00:10] Matt: of the [01:00:10] Dave Pengelley: AI [01:00:10] Matt: people that [01:00:11] Dave Pengelley: operators. [01:00:11] Matt: are actually [01:00:11] Dave Pengelley: That's [01:00:11] Matt: using [01:00:11] Dave Pengelley: that's [01:00:11] Matt: it in [01:00:12] Dave Pengelley: that [01:00:12] Matt: the [01:00:12] Dave Pengelley: level. [01:00:12] Matt: system as a [01:00:13] Dave Pengelley: Um [01:00:13] Matt: system are [01:00:13] Dave Pengelley: AI [01:00:13] Matt: the people [01:00:14] Dave Pengelley: as [01:00:14] Matt: that [01:00:14] Dave Pengelley: teammates [01:00:14] Matt: understand what [01:00:14] Dave Pengelley: is [01:00:14] Matt: systems [01:00:15] Dave Pengelley: oh [01:00:15] Matt: are and how to build systems [01:00:16] Dave Pengelley: go [01:00:16] Matt: properly. [01:00:16] Dave Pengelley: and help me build [01:00:17] Matt: You [01:00:17] Dave Pengelley: this [01:00:17] Matt: know, [01:00:17] Dave Pengelley: thing. [01:00:18] Matt: the [01:00:18] Dave Pengelley: AI as an [01:00:18] Matt: like [01:00:18] Dave Pengelley: assistant is do [01:00:20] Matt: I [01:00:20] Dave Pengelley: the [01:00:20] Matt: think [01:00:20] Dave Pengelley: daily [01:00:20] Matt: we mentioned [01:00:20] Dave Pengelley: email [01:00:21] Matt: last episode [01:00:21] Dave Pengelley: check review [01:00:21] Matt: Elon's [01:00:22] Dave Pengelley: what's [01:00:22] Matt: algorithm, [01:00:22] Dave Pengelley: going on, and [01:00:22] Matt: and [01:00:23] Dave Pengelley: then [01:00:23] Matt: the [01:00:23] Dave Pengelley: AI as a thought [01:00:23] Matt: fifth [01:00:24] Dave Pengelley: partner [01:00:24] Matt: step, [01:00:24] Dave Pengelley: is the [01:00:24] Matt: the [01:00:24] Dave Pengelley: general [01:00:24] Matt: final [01:00:24] Dave Pengelley: just [01:00:24] Matt: step, [01:00:24] Dave Pengelley: chat [01:00:25] Matt: the [01:00:25] Dave Pengelley: GPT [01:00:25] Matt: step before [01:00:26] Dave Pengelley: chat [01:00:26] Matt: like [01:00:26] Dave Pengelley: thread. [01:00:26] Matt: after [01:00:26] Dave Pengelley: But [01:00:26] Matt: every single [01:00:27] Dave Pengelley: interesting. [01:00:27] Matt: other um critical detail. Detail, the very, very final step is automation. So you have to go through these other steps in before you get to automation. When you think about how much automation was pushed, you know, last year, even early this year. Now people are starting to say augmentation. [01:00:47] Richard Webbe: Absolutely, [01:00:47] Matt: But yeah, [01:00:48] Richard Webbe: yes, [01:00:48] Matt: the [01:00:48] Richard Webbe: and it's [01:00:48] Matt: the [01:00:49] Richard Webbe: biased, [01:00:49] Matt: concept [01:00:49] Richard Webbe: I [01:00:49] Matt: of [01:00:49] Richard Webbe: think it's [01:00:49] Matt: actually [01:00:50] Richard Webbe: biased [01:00:50] Matt: knowing [01:00:50] Richard Webbe: the [01:00:50] Matt: a [01:00:50] Richard Webbe: wrong [01:00:50] Matt: system and [01:00:51] Richard Webbe: way. [01:00:51] Matt: how to build it, how to x-ray it, how to not be too dependent, all the things that we mentioned at the beginning, it's interesting. [01:01:16] Richard Webbe: Yeah. [01:01:18] Dave Pengelley: Yeah. [01:01:19] Richard Webbe: Yeah, yeah, that's very true. Yeah. [01:01:34] Matt: Hmm. [01:01:40] Richard Webbe: I remember introducing the first [01:01:42] Matt: It's [01:01:42] Richard Webbe: early [01:01:43] Matt: still very [01:01:43] Richard Webbe: server [01:01:43] Matt: much under construction, [01:01:44] Richard Webbe: automation [01:01:44] Matt: but yeah, [01:01:45] Richard Webbe: software [01:01:45] Matt: everything that I say [01:01:46] Richard Webbe: into [01:01:46] Matt: every episode [01:01:46] Richard Webbe: Australia [01:01:47] Matt: here is what I'm gonna [01:01:47] Richard Webbe: from [01:01:48] Matt: try [01:01:48] Richard Webbe: uh [01:01:48] Matt: and [01:01:48] Richard Webbe: some [01:01:48] Matt: put into [01:01:48] Richard Webbe: uh [01:01:48] Matt: that. [01:01:49] Richard Webbe: very [01:01:49] Matt: I'm [01:01:49] Richard Webbe: well-known [01:01:49] Matt: really, really [01:01:50] Richard Webbe: friends [01:01:50] Matt: passionate [01:01:50] Richard Webbe: of [01:01:50] Matt: about [01:01:50] Richard Webbe: mine. [01:01:51] Matt: this. [01:01:51] Richard Webbe: And [01:01:51] Matt: So [01:01:51] Richard Webbe: uh [01:01:51] Matt: yeah. [01:01:52] Richard Webbe: I remember all the systems engineers were just no, no, [01:01:55] Matt: Of course [01:01:55] Richard Webbe: never, [01:01:55] Matt: they can. [01:01:56] Richard Webbe: no, couldn't [01:01:56] Matt: Slatex [01:01:56] Richard Webbe: do it, couldn't, no [01:01:57] Matt: systems, [01:01:58] Richard Webbe: automation [01:01:58] Matt: the website, [01:01:58] Richard Webbe: can, [01:01:58] Matt: I've [01:01:58] Richard Webbe: of [01:01:59] Matt: got [01:01:59] Richard Webbe: course, [01:01:59] Matt: inquiries [01:01:59] Richard Webbe: nowadays. [01:01:59] Matt: through that. [01:02:00] Richard Webbe: you couldn't survive without some of the automated tools that survive and and do that. And I think people are going to learn that it's going to become so busy because of AI tools speeding up information from other angles, unless we learn how to use it as a system, those people will fall apart. [01:02:19] Dave Pengelley: And if you want to learn how to do that, you might want to check out operator systems at school.com. [01:02:26] Richard Webbe: That that is [01:02:32] Richard Webbe: He's available. [01:02:36] Richard Webbe: And people could also contact you directly. [01:02:39] Dave Pengelley: Yep. Hit him up, Lincoln. [01:02:44] Dave Pengelley: Brilliant. Brilliant. Well, that brings us to the uh the end of the hour. Thank you, gentlemen. Great chat. Always, always more. We could always talk for for a lot longer, but uh we've all got you know things to do and clients to serve and uh uh so on. So I'll uh let you both go. Thank you very much. And if you are [01:03:00] Matt: Yeah, [01:03:00] Dave Pengelley: new [01:03:00] Matt: true. [01:03:00] Dave Pengelley: or returning to this show, make sure you like and subscribe, hit that bell icon. We go live at midday Sydney time every every week, Wednesdays, midday. And so if you want to get alerted when we're live, so you can jump in on the chat and uh interact with us and have a chat. That's great. If you're watching this on the replay, please leave some comments. Let us know, ask us your questions and we'll uh look to address those on next week's show. It's also a good way of letting the world know, hey, there was some interesting chats uh and gets the algorithm to share it and show it in other people's feeds. So thank you, gentlemen. Great day, great week. And uh we'll see you all next week. Look out, look out for our channel, see if I get the uh the content snipping uh up and running on on our various socials, uh at AI operators pod across all of your favorite social networks. And uh we'll leave you on that one. Thanks, guys.