AI Operators
Contact
All episodes

OpenAI Dev Day: Dots, Pricing and the Everything App

livestream1:02:191 July 2026

Dave, Matt and Richard assess OpenAI's Dev Day through an operator lens, focusing on Dots, pricing and the emerging everything-app strategy. They weigh voice agents, privacy and practical workflows against the need to start with a real business problem.

DP
Dave Pengelley
RW
Richard Webbe
MS
Matt Slager
MB
Marno Brits
YouTube
Show notes

Episode notes

OpenAI Dev Day prompted a wide-ranging operator discussion about Dots, subscription pricing, fast modes, Spaces, app integrations and plug-in extensions.

Dave Pengelley, Matt and Richard Webbe examine whether ChatGPT is moving towards an everything-app model, how voice agents could change day-to-day work, and the trade-offs between convenience, privacy and control.

The practical takeaway: start with the business problem and friction, then choose the right tool rather than forcing a tool into the work.

Topics discussed

  • Dev Day demos and the challenge of explaining business value
  • Dots, quotas, cached tokens and subscription choices
  • Spaces, apps, plug-in extensions and the everything-app question
  • Voice agents, personal data and the privacy-convenience trade-off
  • Image-led design workflows and AI-assisted building
Transcript
[00:00] Dave Pengelley: I'll [00:00] Matt: Have little [00:00] Dave Pengelley: have a [00:00] Matt: sneaking [00:00] Dave Pengelley: nap then. [00:01] Matt: apples. [00:02] Richard Webbe: Deep eight. [00:06] Dave Pengelley: Uh I'm back. I'm back twice in one day. This is wild. [00:11] Matt: Oh, your other one was basically yesterday, let's be real. [00:14] Dave Pengelley: No, no, I haven't slept. That means it's today. [00:18] Matt: Oh, that's [00:18] Dave Pengelley: Like [00:19] Matt: terrible. [00:19] Dave Pengelley: I got I got up at 2 a.m. went to bed at like 11 with an alarm set for 2 30, 240, and then you know, kept woke up at midnight, woke up at 12 30, woke up at 1, just you know that the body panic. That I'm gonna miss my alarm. And so by about two, I couldn't get back to sleep. So I finally got up at about sort of quarter past two or whatever, and came and started getting ready to do a 3 a.m. live stream. [00:43] Richard Webbe: I noticed you were on at three AM live, yes, yes. [00:47] Dave Pengelley: All [00:47] Richard Webbe: Very [00:47] Dave Pengelley: by myself. [00:47] Richard Webbe: exciting. [00:48] Dave Pengelley: I put an offer out there on our on our uh on our group to see if there was any uh international people in the group that wanted to jump in with me. Nah. So I had a watch watch party just for one. I did have a few people jump in live, I saw a few people. People on the on the viewer numbers jump in and out, but um no comments, no interaction. [01:04] Richard Webbe: We [01:05] Dave Pengelley: It was just me [01:05] Richard Webbe: Dave, [01:05] Dave Pengelley: heckling from [01:06] Richard Webbe: we [01:06] Dave Pengelley: the [01:06] Richard Webbe: re [01:06] Dave Pengelley: cheap seats. [01:07] Richard Webbe: we rely on you to be the filter of knowledge and information to digest all of that and then present the answers to us. [01:14] Dave Pengelley: Well, yeah, I mean I did. I I I uh straight straight afterwards. We'll we'll get into Dev Day um in a in a minute, but um LinkedIn's driving me insane because I I I did a sole summary. Here's all my thoughts about it. This is like the top story of the day in tech. I opened our dev day announcements, I thought this is Good. I'm bringing a different opinion. It's got all the right keywords. I've used all the right things around OpenAI. I'm fresh. I'm like in the first hour of release. Um, I just did a live stream of it. And then when I look at the impression stats of it like an hour ago, my live stream got three impressions on LinkedIn, three impressions, and all my thoughts and uh and commentary got like 27 impressions on LinkedIn in [01:53] Richard Webbe: Yeah, [01:54] Dave Pengelley: the [01:54] Richard Webbe: we [01:54] Dave Pengelley: first [01:54] Richard Webbe: want [01:54] Dave Pengelley: four [01:54] Richard Webbe: you to digest [01:54] Dave Pengelley: hours. [01:55] Richard Webbe: it. Yeah, yeah. [01:56] Dave Pengelley: And I'm like, what the heck, LinkedIn? Like, I I'm I swear I've been shadow banned. There's Something wrong with my my LinkedIn account because I've not posted anything offensive. If anything, I've just posted too much. I treated a bit more like X, and I was like, Oh, here's some thoughts. And it's like completely rate limited all of my posts. And it's really annoying because you're trying to actually add value and not just be one of these spam bots just putting generic AI crap all over the platform. And what's the point? You just get punished. Um that's that's my rant. Oh in balance in, taking off the shower. [02:33] Matt: You deserve it. You deserve the rant. Uh, I think everyone can tolerate the rant. It's um it is a weird thing. The whole algorithm stuff, like what do you post and you know, who cares about your stuff and why? Um, how do you show the platforms that people do care? You know, like it's a weird thing. It's a constant journey. And it's funny because, like you say, you want to stay away from the hype stuff. And I agree, that's kind of like what this show is. Um, but the hype stuff is actually what creates virality. [03:00] Dave Pengelley: But even [03:00] Richard Webbe: Exactly. It is the [03:02] Dave Pengelley: when [03:02] Richard Webbe: algorithm. [03:02] Dave Pengelley: you try and jump on the bandwagon, like open AI Dev Day, that's a bandwagon thing. And I'm I'm in there with the news cycle. Um, I'm not trying to be oh super contentious or anything. I'm just saying, hey, here's some thoughts on this stuff, and it doesn't get picked up by anything. And you see other people that have way less followers and and people than you, and they just manage to get thousands of views on stuff. I'm like, the math isn't mathing for me. But [03:25] Matt: Deeper analysis is needed. [03:27] Dave Pengelley: anyway, let's let's do a show. [03:42] Richard Webbe: What [03:42] Dave Pengelley: Welcome [03:43] Richard Webbe: you [03:43] Dave Pengelley: gentlemen. [03:43] Richard Webbe: should do, what you should do, Dave, is send Grokbot to go do some analysis for you across [03:49] Dave Pengelley: I have. [03:49] Richard Webbe: the left [03:49] Dave Pengelley: I've uh [03:50] Richard Webbe: here. [03:50] Dave Pengelley: I've had deep therapy sessions around my LinkedIn performance with GrocBot because it's got access to see my LinkedIn. Um and it tells me no, this is normal and you just need to get people to comment on it in the first hour and stuff. And I'm like, I've tried that. I I seem to have no friends because when I send it out to a dozen people to get comments, I get zero. I get cricket. So I think I'm just in this black hole that no one cares. So I just move on with my life. I'll just I'll just keep fighting, fighting the system. And um, it is what it is. I'll just come on here and have a winch to you guys occasionally. And I and you both of you guys commented on my open AI dev day stuff. So thank you. I'm not counting you and the people that cricket me. But uh anyway, um, people aren't here to hear me whinge about social media, people are here to hear about the news and the AI and and all the things. So um yeah let's let's let's let's get into it and and let's talk all things open AI. Right. [04:44] Dave Pengelley: I [04:45] Matt: Well, [04:45] Dave Pengelley: am I am a little bit tired. So if I seem grumpy and not my usual like bouncy self, it is because I uh the technical term is [04:52] Richard Webbe: Dave, [04:52] Dave Pengelley: um I'm [04:53] Richard Webbe: grumpy you is happy everyone else, do you know what I mean? [04:56] Matt: I a good idea. that's think [04:56] Richard Webbe: Like your grumpy is, you know, [04:58] Dave Pengelley: at [04:58] Richard Webbe: all [04:58] Dave Pengelley: least [04:58] Richard Webbe: still in [04:59] Dave Pengelley: my [04:59] Richard Webbe: the [04:59] Dave Pengelley: hair [04:59] Richard Webbe: happy [04:59] Dave Pengelley: looks [04:59] Richard Webbe: zone. [04:59] Dave Pengelley: good. At least at least [05:00] Richard Webbe: Sorry. [05:00] Dave Pengelley: my hair's on point. Like there's that at least. I got that. Um, so you guys um obviously I watched the whole thing and live commented and have lots of thoughts, but have you guys caught up on it? Where are you guys at with the dev day announcements? Um the community is fairly mixed. In fact, on X, people are pretty ticked by the overall, [05:20] Matt: It's [05:20] Dave Pengelley: from [05:20] Matt: fascinating, [05:20] Dave Pengelley: what I've seen. [05:21] Matt: isn't it? Like the the I guess it's like where we we've got this acclimatization to certain releases, and you know, they were like oh Well, this was pretty boring, you know, um pretty lackluster, disappointing, whatever the case may be. But I personally think, like similar to what I put in your in my comment on your post, Dave, I think they just rushed too many actual announcements. Like so it's like you're just reading titles with small summaries rather than the actual dot points use cases, like you mentioned, you know, what this translates to for everyone individually. Like, because I think we we miss uh we misinterpret people's. Comprehensions of what these things are actually for, um in general. Yeah. Like [06:06] Dave Pengelley: I think, yeah, I I I think there's a few things going on. And look, I come from Salesforce background, and Salesforce are known for putting so much care and detail and rehearsal into their live demos and their keynotes, especially for their big events. Like their attention to detail is insane. So I come from that pedigree of having been on the stage doing the demo stuff on the demo desk in large audiences [06:29] Matt: I'm I'm not sure what [06:29] Dave Pengelley: for [06:29] Matt: saying. [06:30] Dave Pengelley: For these sort of corporate events. Um, and so when I see them go in there and I've got these these half-cock demos that don't really tell a proper story, they don't really nail what the value of the thing is and the problem they're solving. And you can hear the people presenting it doing the so then and like those filler like transitions, which is just like we do that on this show because we're live at Living It. But these guys, seriously, you think have rehearsed this multiple times and they know their script down and they don't need to go so then now. And I've got these obvious like brain transitions where then they're not quite sure what they're doing, um, let alone the demos failing. And like probabilistic demos are not something you really want to test live that much in that kind of environment. You want to go the Apple path where you've got your pre-recorded demo reels [07:17] Richard Webbe: Yep. [07:17] Dave Pengelley: a little bit, a [07:18] Richard Webbe: Yep. [07:18] Dave Pengelley: little bit more, I think. [07:19] Richard Webbe: It's I [07:20] Dave Pengelley: Um [07:20] Richard Webbe: reckon it's it's value versus clickbait and it's pure and applied marketing, even across those algorithms you were complaining about earlier. My father was an Archibald Prize. Anking fine artist and was listed as one of the World Uh Australia's best uh fine artists, particularly in watercolours. And he told me about a particular very famous artist who I won't name, who when he had his first show, if you've ever been to an art show, they put all the artwork around. And if there's a yellow sticker on the uh on the painting, it's under offer, and if there's a red sticker, it's sold. So this guy from day one went around and put red stickers on all the paintings. Like the [08:00] Matt: Oh. [08:00] Richard Webbe: whole thing was sold out before anyone walked in the door. The [08:03] Matt: Wow. [08:03] Richard Webbe: perfect envy clickbait done in real time. And people would walk up to him and go, and this is a very famous artist, but that you know, and people walk up and say, Oh, I really like that, and someone's already bought it. He goes, I'll tell you what, I've got one just like it out the back. I'll sell you that one. And it was the same painting. Now, when I look at the demos and the releases from the AI companies and the software company, they're fighting between demo value and Real clickbait, real short term, here's a great thing, come and get excited about it. And they're going for those algorithms soaking up that you were complaining about earlier, Dave. [08:40] Dave Pengelley: Yeah, I don't know. I mean today's demos were just wasn't even that. Like it was just it wasn't even all glitz because it wasn't that glitzy and it's just these weird use cases with these demos. And I made the point, like they're just oh let's build another game, let's have another 3D model environment. Like ninety percent of the world's users Aren't interested in making 3D environments or even video games. Like, even the even the their spaces collaboration tool, which should be their new thing, which is going after your notions and your Google, Google Docs and all these kind of things, and sort of this all-in collaboration space. That's the one where they really should have landed the plane on business value and had a really cohesive story around, you know, running a project plan and collaborating with others to deliver a show or something. And that was so botched and just looked like a bad. Notion clone or something. Like it's just that I've I'm being harsh, but I mean that they're just who's asking for these things? Why are they scatter getting this stuff? And then and the and this is the consensus on X. So like then they're saying, oh, but uh everything's more expensive now, and we're gonna give you a $500 plan that's only slightly better than the old $200 plan. Um it's it is weird, and they're saying, No, but we're giving you more, we're giving you more, but it's really unclear what you're actually getting. As far as you know, Matt, you were just talking about your dot before we went live. They're saying dot usage is included in your plan. But then Matt, you said your dot was posting messages in your Codex in the cloud, which means now it's actually eating your codex quota. It's no longer using your dots quota. And [10:15] Matt: That [10:15] Dave Pengelley: so [10:16] Matt: was requested. That yeah, so the the entire like usage model that they've got going is that the current access to dots, you know, so when you're communicating with your dot, it's doing things. Um Basically, it's you know for free, which kind of sounds crazy. But what Thibaut's actually [10:35] Richard Webbe: Prince. [10:35] Matt: expanded on this. I could find his post [10:37] Dave Pengelley: I'll [10:37] Matt: if you want. [10:37] Dave Pengelley: find I'll [10:38] Matt: He [10:38] Dave Pengelley: find [10:38] Matt: posted [10:38] Dave Pengelley: the tweet now. [10:39] Matt: he posted like a hundred messages today. Um, he expanded saying basically that there's going to be upgrades that you can do that's based on faster interaction, more bandwidth, you know, like just dots plus, I guess. [10:54] Matt: Yeah, [10:54] Dave Pengelley: Unless [10:54] Matt: well, [10:54] Dave Pengelley: you pay [10:55] Matt: true. [10:55] Dave Pengelley: more [10:55] Matt: Yeah. [10:55] Dave Pengelley: money. [10:56] Matt: Um, but then yeah, if you're gonna get work done through codex, of course it's gonna use codex to. That just makes sense. Yeah, I I've enjoyed it so far. I've used it for like an hour. [11:06] Dave Pengelley: Yeah. [11:07] Matt: It's been an interesting experience. I I would say it's a it's a different experience to Grockbot for anyone that's actually used that so far. It feels like it's not the same tool. It feels like they don't replace each other exactly. [11:20] Dave Pengelley: Wow. Interesting. [11:21] Matt: Um [11:22] Dave Pengelley: Because [11:22] Matt: yeah. [11:22] Dave Pengelley: I've been using my Grokbot a little bit because it's pretty good value that is on the thirty dollar plan. I'm getting a lot of Grokbot. Um, and that's good. I mean, the fact that DOT, and as I've talked about like recently, just using chat GPT.com has been a huge unlock for me to get a lot of extra usage and chat and collaboration, integrating with third party tools like Airtable and GitHub for some projects without using any of my coding quota has been awesome. So I'm interested to see where the dot fits in on that. Um but yeah, I mean, this is this is Tebow's post just ahead of Dev Day, where he talked about the bringing back the 200 plan, and he's saying, look, the old 200 plan. Was the the 20x, right? So you had the the the 5x and the 20x. And he's saying this one's actually now 10x. But he said, we're gonna number ship a number of things that increase what you can do across the Probless and Pro plans. A lot of computers online for this increase as we increase the floor, we're changing the relative difference between the plans to be this. So he's saying, we're giving you more because we're gonna include all this extra stuff in your plan. Um, but that means the coding quota is gonna come down and not have as much of that. But the coding agents are better, so it's kind of good anyway. Anyway, it's it's this weird mixed messaging they've got around what you're getting, considering they were the ones going, no no, twenty X means twenty X um the other [12:38] Matt: Yeah, [12:38] Dave Pengelley: day, like uh [12:39] Matt: a little [12:39] Dave Pengelley: only [12:39] Matt: while [12:39] Dave Pengelley: recently. [12:39] Matt: ago. I um I quite f I'm I you guys know that I I don't look at Theo Brown as a as a a person to be hated upon, like apparently the rest of the internet believes. Um I he he has some interesting takes, of course, but his take on this, like you asked at the beginning, Dave. You know, what how have we caught up on these announcements and whatnot? Like my when I woke up this morning, I just I put the keynote on on 1.5 speed, and you know, I just watched it in bed quickly before getting up. Um and I noticed that Theo had also posted a video too, and it was on um 6.1 soul. And it was his experience with that. He had had a little bit of extra um access to it leading up to the day. And he had a couple of really, really interesting takes that I don't think a lot of people will actually think about. And if they do, it may not make sense to them. unless you sort of go a bit deeper. But 6.1 soul, I don't think was ever actually meant to be a thing. I believe this is their attempt at a Astra light, you know, it's a small version of Astra in order to compete with Sonnet 5.5 and the these these cost changes. So, you know, now basically 6.1 soul is your is your go-to. It's your [13:55] Dave Pengelley: Yeah. [13:55] Matt: main and you would then use Astra only as a sub agent of 6.1 soul. Or in those really specific use cases where you know it's going to be better. [14:03] Richard Webbe: Excuse [14:04] Matt: Bless [14:04] Richard Webbe: me. [14:04] Matt: you. So yeah, I think we've been lulled into a sense of like um oh, I don't know, comfort with the idea of us getting all this crazy value essentially for free. Like the two hundred dollar plan before the times twenty, regardless of you know what it equated out to be, it was basically like twelve K worth of API costs, you know, which is crazy. And to think that you're getting that for two hundred Like, yes, they've got big margins on things, but it's still insane. Like the value is still insane. So, yeah, like this $500 plan, it's not for everyone. It's going to be for those that of us that are can use it can use it for expenses. Um, you know, or you can just straight up afford it because whatever. Like I said in my post, uh, sorry, my comment on your post as well. Like, if you can't figure out how to get a thousand dollars uh return out of a $500 investment of this tool, then you know you've got bigger problems. [14:59] Richard Webbe: I saw that. [15:00] Dave Pengelley: Yeah, but I mean we've we've been talking about tokenomics for months around how these subscription things can't last forever. It is interesting to see the rebalancing and I feel like that the the mints the communications up a little bit. Uh and you know, Sam's up there going, uh, yeah, but you can now do ultra fast as well, which is gonna like burn your quota at six times the speed, but you'll get eight times the processing power. And he's like, Good, look, it's worth it. We made a rocket ship fly off in five seconds instead of thirty seconds. It's it's worth it. You should totally, totally spend the extra. credits. Um [15:31] Matt: Yeah, that was a weird demo. Yeah, that the definitely uh a revenue generating use case right there. [15:37] Dave Pengelley: right, [15:37] Matt: Maybe [15:37] Dave Pengelley: right. [15:38] Matt: Elon [15:38] Dave Pengelley: That's like [15:38] Matt: should use it. [15:39] Dave Pengelley: yeah, I mean he's the rocket guy, right? [15:42] Matt: Yeah. [15:42] Dave Pengelley: Um yeah, so it was it was interesting morning. Um I yeah, they what was the other big one? Uh the decisions API. So they're coming after Jeff. We've obviously obviously talked about Jev a lot and they're saying they've got their new decisions API which is based on Luna, which means it's got vision. You can actually pass the pictures and it'll make decisions on the picture and give you that that fixed classification. [16:06] Matt: Did you see [16:06] Dave Pengelley: No, [16:06] Matt: the [16:06] Dave Pengelley: sorry, [16:06] Matt: uh [16:07] Dave Pengelley: I'm [16:07] Matt: the [16:07] Dave Pengelley: just [16:07] Matt: type [16:07] Dave Pengelley: dying [16:07] Matt: safe [16:07] Dave Pengelley: here. [16:08] Matt: AI guy, the Jev guy, comment on that? [16:15] Dave Pengelley: You talk. [16:16] Matt: Oh, sorry I didn't notice you I didn't notice you dying. Um yeah, it was he had quite an interesting uh take on it. Um where he looked at it and he was like oh Okay, so it's it's basically Luna with some sort of different different uh classification structure. Um, you know, he was quite diplomatic. He said something like, you know, like hope it works, you know, good luck. Um and there's already a couple of really good benchmarks and people like playing with that already. But yeah, it's fascinating. I I love I just love it. I I love all this stuff. And you definitely can get a little bit overwhelmed and you know, do I need to know about this? You know, if you're like if you're out there, somebody listening to this and you don't even have a an open AI subscription, you know, you don't you're not even you don't live in that world, you're a you're a clawed person or uh you know God help you, you're a Gemini person. Um you don't need to worry about it. [17:11] Dave Pengelley: Apparently [17:11] Matt: Just [17:11] Dave Pengelley: Gemini [17:12] Matt: say [17:12] Dave Pengelley: 4 is the killer one. Gemini 4 is gonna wipe everyone, apparently. [17:15] Matt: Yep, as they say. [17:17] Dave Pengelley: Um yeah, [17:18] Matt: Yeah [17:18] Dave Pengelley: but but [17:19] Matt: I [17:19] Dave Pengelley: like I said, [17:19] Matt: just [17:19] Dave Pengelley: they're just the messaging, I think the demos, they didn't land the value. I don't know that this is gonna convert any new customers to open AI. It's certainly got some people saying, nah, I'm off back to Claude. And the people are fickle, people have flipped. Anthropic's gonna screw them over in two weeks' time and they're gonna be like, oh I'm on going back to Codex. Um but you know, even with the pricing, that whole 20x, 10x thing, that was Tebow, the one I showed you coming back 24 hours after his first announcement, because he was talking, that's uh apparently that's like 50% sense of the dollar of the API, and it was just so confusing. No one understood what he was even saying. So that's why he had to come back with that post. Uh, and then today announcing dots. Uh work dots work 24-7 for you. We got community note. With people going, actually, you know, that said that they don't count towards usage, but it does. Uh, and then in these open these articles, it even actually says, you're gonna get the first month for free, and then we're gonna detail pricing after that. And so Tebow's going, hey, I've got community node, but the node's wrong. The primary dots included in your plan. If you ask it to create a codex, it will draw your codex usage. But when it does work directly on its own computer, I assume that uses nothing that is on top of your plan's usage. Um [18:30] Matt: Yeah. [18:30] Dave Pengelley: Yeah. [18:31] Matt: And that's that's [18:31] Dave Pengelley: So [18:32] Matt: what I've experienced this morning. And it actually works pretty seamlessly. But you know, like all of these AI tools, you kind of need to know how to, you need to know their language. You need to know how to speak to them. Because, [18:42] Dave Pengelley: I'm if not sure you're [18:42] Matt: you [18:42] Dave Pengelley: able be do going not that. to to [18:42] Matt: know, I have a little bit of awareness of how these things work. So I'm able to say when it gives me a response in some kind of like failure, I'm like, well, hang on. And did you actually do X, Y, Z? And then it goes [18:53] Richard Webbe: Yeah, [18:54] Matt: back and [18:54] Richard Webbe: how [18:54] Matt: it's [18:54] Richard Webbe: did [18:54] Matt: like, [18:54] Richard Webbe: you [18:54] Matt: oh [18:54] Richard Webbe: count to [18:55] Matt: yeah. [18:55] Richard Webbe: that answer? [18:55] Matt: Yeah, [18:56] Richard Webbe: Yeah. [18:56] Matt: you're right. You know, maybe, maybe I I've I missed that the first time or something. So yeah, it's fascinating. I don't think it's as user-friendly as Grocbot. I mentioned before. Like there is different aspects to this. I see this a lot more like PUC in AMP. If anyone is familiar with AMP code out there, um puck is essentially like your it's your concierge of the whole AMP ecosystem. So you can get it to spawn threads, you can get it to ask for context between [19:27] Richard Webbe: I [19:27] Matt: threads, [19:27] Richard Webbe: don't think that's [19:28] Matt: you can get it to give [19:29] Richard Webbe: a good [19:29] Matt: you basically [19:29] Richard Webbe: thing. [19:30] Matt: This lay of the land approach of your entire ecosystem in AMP, that's kind of like what your dots are. Um, the Slack integration works really good. I've also already set that up and played with that a bit. The phone call works really good. If you guys haven't experienced like having [19:44] Richard Webbe: It [19:44] Matt: a [19:44] Richard Webbe: makes [19:44] Matt: voice, [19:44] Richard Webbe: a cold foy. Does it [19:45] Matt: yeah, [19:45] Richard Webbe: make a [19:46] Matt: like [19:46] Richard Webbe: cold [19:46] Matt: if [19:46] Richard Webbe: foy? [19:46] Matt: you you can you can speak to your your agent that's doing things. So it's not just a voice chat. Like all of us are used to voice chats now, whether you're wanted to be on one or not when you've dialed a phone number. Um When you're actually on a voice chat with an agent that can do work. So GruckBot recently had their phone call feature. Fantastic. This is the same thing where you can you basically call it, you engage in a real-time voice conversation, you talk about stuff that you just need to get done. You know laying the land out. And it's already started to put stuff in my spaces. Um, I've got a whole bunch of like documents and everything that it's built out for me for things that I just mentioned this morning. Um so this [20:26] Richard Webbe: So [20:26] Matt: kind of [20:27] Richard Webbe: can [20:27] Matt: feels [20:27] Richard Webbe: any of these [20:27] Matt: like [20:27] Richard Webbe: agents take control of your phone and mobile? My calls on your behalf. [20:31] Matt: Yeah, of course you can. Not of not con not take control of your phone, but you can give them the ability to call through other web phone dialers and that sort of thing. [20:43] Matt: Yeah, well, a lot of them. Like there's there's many. [20:46] Dave Pengelley: But [20:46] Matt: I [20:46] Dave Pengelley: you [20:46] Matt: wouldn't [20:47] Dave Pengelley: you [20:47] Matt: be able [20:47] Dave Pengelley: gotta [20:47] Matt: to [20:47] Dave Pengelley: set up another tool. You're not just saying, here's a phone number, go call this thing. The bots aren't doing that without connecting other services, though, right? That's [20:53] Matt: Muse [20:53] Dave Pengelley: what I think [20:54] Matt: can. [20:54] Dave Pengelley: Richard's trying to understand. [20:55] Matt: Yeah, by default, Muse can because they always give you uh like a free phone. number as part of meta services so you could always do whatsapp um phone numbers and stuff like that through the [21:06] Dave Pengelley: Meta [21:06] Matt: whole meta [21:07] Dave Pengelley: Meta Muse, [21:07] Matt: system [21:07] Dave Pengelley: we haven't we haven't even touched on meta. [21:10] Matt: and there's another one called uh instinct as well which you guys probably haven't heard of [21:13] Dave Pengelley: I've I've [21:14] Matt: um [21:14] Dave Pengelley: that's that's been blowing up on my socials. They've got some kind of underground campaign going on promoting them, and I'm seeing them on LinkedIn a lot about [21:22] Matt: yeah [21:22] Dave Pengelley: because they just raised a billion dollars or something. [21:24] Matt: yeah so you got instinct muse grokbot dot um then like i mentioned you This morning, Dave, it'd be fascinating to see what Claude comes out with or what Anthropic comes out with. Um, because we've mentioned in the past that they're the ones that sort of sleep on things for for quite a while and then they release their you know really polished version of the product, um, which usually is uh met with quite a cult-like uh fanfare. [21:53] Richard Webbe: Is MetaMuse restricted to the Facebook meta environment, or can it step out of there and and interact with other applications? Applications and data that you have. [22:04] Matt: It's only in a US and Canada right now, so I can't give you any first hand experience. [22:11] Dave Pengelley: But apparently it's it's pretty fast and loose with personal details and security stuff, so um, [22:16] Matt: Yeah. [22:16] Dave Pengelley: like because it's all Facebook and meta and you are the product. [22:19] Matt: I saw a guy this morning. Um the the post was, you know, I let my muse agent handle my marketplace, you know, so Facebook marketplace. And apparently it or it already was organizing pickup things, you know, giving people his address and um rejecting like offers or giving low ball offers on things. And yeah, he wasn't [22:40] Richard Webbe: So [22:41] Matt: happy [22:41] Richard Webbe: it's [22:41] Matt: in [22:41] Richard Webbe: becoming [22:41] Matt: that post. [22:42] Richard Webbe: the Wild West, and so we're not that surprised that when Anthony Albanese went to the US to act like Mr. Social Media and Mr. Digital AI man, that he learnt in the US that uh uh Medibank and Medicare had been raided by a number of agents from Open AI. [22:59] Matt: Did you [23:00] Dave Pengelley: Let's let's let's clarify that because I I did see this article from OpenAI. How we will do better for Australia. Because there's a lot of there's a lot of misinformation and and you know fear-mongering reporting about this sort of stuff. Um but this is the open AI story that in June, during internal training and evaluation of our models across Australian government websites, um ac they accessed Australian government websites in ways they were not authorized to. Uh we we should have handled our response better. But uh Basically, they said, Go find some information. So we're reviewing earlier training evaluation activity after the hugging face thing. And so they tested by getting it to go find information. And it went and found non public access to the service, ranked commands, retrieved internal files, credentials, no patient or client records. But it basically went and looked at the website, and the website said, Oh, would you like this extra information as well? That I'm not actually publishing on the website, but I've got it here. Here for you and went, sure, that'd be great. [24:02] Richard Webbe: So [24:03] Dave Pengelley: So [24:03] Richard Webbe: that's not really OpenAI's fault, that's Medicare's [24:06] Dave Pengelley: it's [24:06] Richard Webbe: data [24:06] Dave Pengelley: yeah, [24:07] Richard Webbe: publishing [24:07] Dave Pengelley: it's that's that's [24:07] Richard Webbe: and [24:07] Dave Pengelley: kind [24:07] Richard Webbe: security [24:08] Dave Pengelley: of our government [24:08] Richard Webbe: fault. [24:09] Dave Pengelley: services did not have their their data security lockdown, and then they're blaming [24:12] Richard Webbe: Yep. [24:12] Dave Pengelley: OpenAI for hacking them. It's like, no, you guys left the door open. What are you talking [24:16] Matt: Yeah. [24:16] Richard Webbe: That's [24:16] Dave Pengelley: about? [24:17] Richard Webbe: exactly right. Do you know? I was thinking exactly that day. That's a great, great way to phrase it. There was no hacking. Someone left the door open or left the lights on so people could look in. [24:30] Matt: Another article where these people were saying this has been open for uh for a long time now. You know, this has already been scraped all of these times. You know, why are they worried about it right now? They've just pointed the finger at this particular instance. [24:45] Dave Pengelley: This is not a political podcast, but this government has shown a tendency, and the previous one to a certain extent, on both both sides of the aisle, to want to control data and communications. And we've seen like the social media. Media sort of access stuff coming through. There's a lot of misinformation, disinformation bills. There's a lot of focus on controlling narrative. And big tech services like this challenge that a little bit potentially. And so if they can, you know, I I think there's some merit to some of the sort of false flag sort of theories where this is a let's make a big boo hoo hoo around how dangerous all this technology is, because then we can put more stuff in for protection, safety. [25:30] Richard Webbe: Definitely politically motivated, and we've all seen the articles recently, not on one side but on both sides of politics, on the shutting down of what we would all call freedom information. I mean, even a minister said, who's going to decide what is appropriate information and what isn't? And they said, I will. [25:49] Dave Pengelley: Yes. And and [25:50] Richard Webbe: It's [25:50] Dave Pengelley: we we [25:50] Richard Webbe: a [25:51] Dave Pengelley: we [25:51] Richard Webbe: bit [25:51] Dave Pengelley: saw, [25:51] Richard Webbe: scary. [25:51] Dave Pengelley: you know, not not that long ago, there were a lot of other things that we uh a lot of other freedoms we gave up under the name of safety. Um because Of a of a certain virus that was going around. Um, and so people sort of have this trust in institutions, and if it's for safety, especially if it's for safety of children, people don't actually question it too much. We talk about AI and people giving up their critical thinking and just going along with whatever the robot tells them to. People have got a tendency to do that with government in many ways as well in this country. [26:24] Richard Webbe: And yet we [26:25] Dave Pengelley: So [26:25] Richard Webbe: all surrendered all our private information to social media without even batting And oil it. [26:31] Dave Pengelley: Yeah. [26:33] Dave Pengelley: So this [26:33] Matt: What [26:33] Dave Pengelley: is [26:34] Matt: you just said, Richard, about um I forget your exact wording, about them ramping up and then saying that they want to close down on this kind of sovereign stuff. [26:43] Dave Pengelley: a good idea. [26:43] Matt: This literally happened in Anthropic, like this week, like in the last 24 hours. So they they announced their IPO and then they released this massive blog post bashing on GLM 5.3, which is from ZAI, a Chinese lab. And basically they want to get GLM 5.3 banned, but the only reason why they know that they want to get it banned is because nothing else was able to actually do the analysis, but 5.3 was because it's it's open, you know, like surprise, surprise, like an open weight thing can actually be inspected, and you know, the people can find things out about it, and that's the point of open source. So um, [27:18] Richard Webbe: Absolutely. [27:19] Matt: yeah, I found that fascinating. So yeah, [27:21] Richard Webbe: I [27:21] Matt: brace [27:21] Richard Webbe: yeah, [27:21] Matt: yourself for [27:22] Richard Webbe: no, [27:22] Matt: GLM [27:22] Richard Webbe: I agree. [27:22] Matt: 5.4. [27:24] Richard Webbe: The whole [27:24] Dave Pengelley: Yeah, [27:25] Richard Webbe: freedom [27:25] Dave Pengelley: I a good one. that's think [27:25] Richard Webbe: of information, what we can and can't share, privacy, there They're all going to need to be rewritten as this stuff changes very fast. And it is the balance between convenience and privacy. [27:40] Matt: Yeah, [27:40] Dave Pengelley: Yep. [27:41] Matt: true. [27:42] Dave Pengelley: Yep. [27:42] Matt: Yeah, it [27:43] Dave Pengelley: So [27:43] Matt: is funny, like how much convenience do you want without you know being able to give away some of your details? Because it it's like signing up to um signing up to anything. Yeah, you need [27:55] Dave Pengelley: Well, [27:55] Matt: to give [27:55] Dave Pengelley: I [27:56] Matt: your [27:56] Dave Pengelley: mean, now you got the digital ID stuff, right? And then you don't have to get a digital ID You can still do everything with your paper and your plastic, but you're gonna have to go down to the office and do this and sign this bit of paper and do that versus going, [28:06] Richard Webbe: Exactly. [28:06] Dave Pengelley: oh, hang on, approved, done. All right, now I can pay my registration or whatever. So um people are you know trading one for the other. [28:17] Richard Webbe: Hey, I did a quick LLM audit on my data during the week. I thought that would be an interesting thing to play with and then talk to you guys. And the result I recommend anyone watching, um, the result. are quite astounding of where your data appears, where it doesn't appear, how it can be accessed and how it can't be accessed and who has it. I invite everyone to get on one of the LLMs, type their name in, all the details and say, where does my information appear? And it is astounding. [28:47] Matt: You know that Dave, that uh that opens up a uh a nice opportunity for uh for a sponsor segment that we don't have. Um there's there's a really good company that does that really good product, it's called Egan Cogni. And they will literally find all that for you. You can pay them a service to just wipe yourself from the internet. Um, that kind of service has existed for quite a long time now, where people [29:10] Richard Webbe: It has, [29:10] Matt: have [29:10] Richard Webbe: yeah. [29:11] Matt: lived in that kind of white hat security land where they will go find your things and and remove it for you through various gray areas of legality. But yeah, that that print that product that is a is a good thing if you're somebody that needs that. Maybe they should reach out to us and sponsor a segment. [29:30] Richard Webbe: It's a free aid for them. [29:32] Matt: Hmm. [29:34] Dave Pengelley: Yes. Um all right. Wow, we're we're flying flying through stuff. Um [29:40] Dave Pengelley: I'm I'm [29:43] Dave Pengelley: brain is [29:44] Matt: Sleepy. [29:44] Dave Pengelley: frazzled. [29:45] Matt: Yeah. [29:45] Dave Pengelley: Yeah, I'm I'm [29:46] Matt: Tell [29:46] Dave Pengelley: just [29:46] Matt: me [29:46] Dave Pengelley: trying to [29:46] Matt: you haven't [29:46] Dave Pengelley: I'm just trying [29:47] Matt: slept [29:47] Dave Pengelley: to connect [29:47] Matt: quite [29:47] Dave Pengelley: the dots and work out where we go next. We've we've flown through so much stuff so quickly. Um and uh Matt, you've been sending me some links. Is there anything uh specific there I should be bringing up that you want to [29:58] Matt: a [29:58] Dave Pengelley: talk [29:58] Matt: bit. [29:59] Dave Pengelley: about? [30:00] Matt: Depends really like who's going to be listening to this particular episode. So there's a there's a mix there. I I have a very um I had a very interesting response from that that root the last link that I sent you, the article, that I think would be worth us trying to share somewhere, maybe in the in the video description here, because that was originally shared when I saw it, it was shared by Sam Altman. Um this David George guy, I didn't research into who he is. uh whether he's an open ai employee I'm assuming he is but that article goes into I think [30:37] Dave Pengelley: Ah, I [30:38] Matt: a [30:38] Dave Pengelley: read [30:38] Matt: bunch [30:38] Dave Pengelley: that article. [30:38] Matt: of you did [30:40] Dave Pengelley: Yeah, I read that, yeah. [30:41] Matt: yeah how did you feel like what was your reaction to that [30:44] Dave Pengelley: Um, it doesn't match up with what I saw today in their dev day thing. Um, because I mean this is talking about inventing new customers, right? And creating new products to create new product categories and [30:59] Dave Pengelley: that It has been true to a certain extent, but also um I didn't see that today. I and and part of the issue with today is they overhyped it. Like, and this is one of the one of the things. If if like it's like when you when you hear this this movie is the best thing ever, the greatest movie, and you you have to see this movie. It's oh it's so good. And you get it, and you watch it, and you're like, that was that was alright, I suppose. Like, because we got so massively overhyped. This dev day, they're like, 20 plus releases, oh, we've been cooking. So hard, we've had Astral, we've got just so much. Oh, we can't wait to show you guys. Oh, it's gonna be the best, the best. Wait till Tuesday, wait till Tuesday, wait till Tuesday. And then we got the hour of stuff we got, which was a bit of a mishmash of like confused value messaging. And I was like, ah, that what? What? It's like that little white monkey that's in a little box going, wow. [31:52] Matt: Yes. [31:55] Dave Pengelley: Um yeah, so this this might be true, but they didn't, they didn't showcase. That particularly well today, that they understand creating new product categories because I were chasing with dot um new model and and is just to like balance out the pricing changes that they're doing. The fast stuff is fast, is good. You can do a lot with fast, but you're not going to be doing it for very long because you can't afford it. Um yeah, just I don't know. Um [32:22] Matt: I [32:22] Dave Pengelley: what [32:22] Matt: think [32:23] Dave Pengelley: what [32:23] Matt: the [32:23] Dave Pengelley: what [32:23] Matt: the [32:23] Dave Pengelley: do [32:23] Matt: biggest [32:23] Dave Pengelley: you think about their new their their inventing product categories from this article, Matt? [32:27] Matt: I I agree with it. So this is what shocked me. I because whenever I see Sam talk about something, it always makes me feel a little bit sick. Um same, same as Dario. They're both I don't know what it is. What the way that they communicate just makes me feel ill. Um so when he we he shared this article, I was like, well, hang on, okay. Does he agree with it? You know, do did he write it? Did he help this dude talk it out? You know. So as I read through it, I was like, okay, I kinda understand what they're talking about here from like a a big perspective, like a big play perspective. Like if you look at the the split of people that actually have accounts on these different tools. Like as far as I'm aware, the the, you know, I would have to go cite the details here, but the ChatGPT accounts still surpass every other application. So when you talk about like an ecosystem that doesn't exist, like it kind of does. Like so if they can steer people into doing what they do elsewhere inside this platform, you know, the, the, the super app, I forget all their weird marketing terms they'll talk about ages ago. Like it works great. Like, do you guys remember the browser that they did? I think it was called Atlas. Um that that exists in the GPT app now. And you know, Astra on Lite is your AI browser interaction thing that used to exist in Atlas. Um, like you just mentioned before, spaces basically is Notion. There's if you already use Notion, you could literally just move everything into there now. Like it's probably not as polished or as fleshed out, but it's realistically the same thing. And then sites, you know, sites. Is the same as going and using any other website hosting platform now. You know, it's slowly getting better. It was pretty rubbish at the beginning. So you could literally use your dot to spin up codec sessions, to build you your thing, host it on sites, all the documentation is in your spaces. Your team integrates there in the spaces and in your Slack. You know, when are they going to integrate a Slack into the GPT app? Um I think the ecosystem thing is there. And in saying of this, I still feel like I don't need to use it. It's it's so funny how like it's there, it makes sense, but it's just not as much of a value prop for me at this point in time. Like I'm I actually I don't know if I told you guys I resubbed to Claude a while ago and I've been using Opus 5.5 as a as a daily and using that to orchestrate Astra sessions and Devon sessions. I've been using Devon as well. I still think now that we've seen these releases, you know I'm gonna mess around with dots. I'm going to continue to see how it compares to others. But for me to do work for my client work where I'm actually getting paid, I'm going to continue to use Opus. And yeah, I'll use Sol 6.1 as a replacement for Sonnet. I believe that it's, you know, a worthy replacement for that in many cases. But other than that, I don't really know where else they can go other than just making all these things better and better. [35:26] Dave Pengelley: Hmm. Yeah, I mean that they're definitely trying to be the everything app, which is funny because that's what Elon wants X to be. X he wants X to kind of be the everything app. He's talked about that. Um so where he's gonna take that with Groc and X and SpaceX AI and and all that in the future, I don't I don't know. But that's one of the reasons he didn't want to be called Twitter. He wanted it to be to be X, because he wants it to be X [35:46] Matt: I feel like we're [35:46] Dave Pengelley: X [35:47] Matt: still [35:47] Dave Pengelley: tra [35:47] Matt: like [35:47] Dave Pengelley: thing. [35:47] Matt: we're still six months or more away from seeing anything useful [35:50] Dave Pengelley: Right. [35:51] Matt: there. [35:51] Dave Pengelley: But but on on the chat GPD thing, like this is what their second or third attempt. And Sam even said we've tried this before and it hasn't landed probably. Like Integrating third-party apps into their feed. Like they used to have the canvas where you could have integrated canvas things and canvas applications, and then they got rid of that and they created sites. And now they've got spaces where you can have pages in the space that are part of sites. Like it's all there's that I don't think they've fully got their product strategy together yet. They're not quite sure. And again, because they're not sure what the use cases and who their user is and what though they actually need beyond internally, where they're trying not to use anything except their own tools, and so they're creating their own tools. For themselves, which can work to a point, but I think it's just a little bit scattered. And for guys like you and me, Matt, and people that are running like basically solo, some of these team collaboration things don't make as much sense to us as far as when you're really going to get value from a space or or something like that. Um, but look, I think one of the interesting things is the take your chat GPT subscription into other apps. Because I'm not convinced that then we talked about this last week. I'm not sure that the feed. Is the right interface for everything. I think sometimes you actually want to be in an interface for an application to do work in that application. But the fact that now they're saying you can plug ChatGPT into supported applications and get ChatGPT to drive inside that application, I think that's really interesting, especially for app developers, that you don't need to worry about managing the token burden and having some like token blowout. If you let people bring their own Chat GPT into your application, then they can get all that information. intelligence and you're not carrying the bill for them. Um including including Notions one of their signup partners. So you can sign into Notion with your Chat GPT sub and write them pay Notions agent rates. Just use your your your chat GPT sub and [37:42] Matt: That [37:43] Dave Pengelley: extensions. [37:43] Matt: still needs a business account. But yeah, um it is it is fascinating. And I think it goes two ways because they've got their new uh I forget their term for it, whether it's extensions or plugins or or whatever it was. But um [37:56] Matt: yeah, so that it seems to be an extension. Of the plugin system where you can actually build your app inside GT Jap GPT as well. So [38:06] Dave Pengelley: Yeah, which is why I'm saying the everything app, right? Like if you can build applications inside the ChatGPT harness platform. Um, [38:15] Matt: for [38:15] Dave Pengelley: and 'cause [38:15] Matt: me, like I I'm gonna experiment with that with my own tools and products where I'm gonna try and see if I can build my extension in the app and use it and just see how it feels, plus have that authentication system in the other. [38:29] Dave Pengelley: able be going not to to you're [38:30] Matt: The external app itself, [38:31] Dave Pengelley: Yeah, [38:31] Matt: so that [38:31] Dave Pengelley: yeah. [38:31] Matt: you can run your tokens in there and just see how that feels as like a you know a series of API links together and just a product and user experience overall. Because I think that has legs, like that kind of system where just work where you want to work, it doesn't matter. Um, [38:49] Dave Pengelley: Cause [38:49] Matt: there's too [38:49] Dave Pengelley: because [38:49] Matt: many ways [38:50] Dave Pengelley: when they [38:50] Matt: to [38:50] Dave Pengelley: brought [38:50] Matt: access [38:50] Dave Pengelley: out custom [38:50] Matt: things. [38:51] Dave Pengelley: GPTs, one of the things was you're gonna be out of uh sell services and sell or get some kind of subscription cut from custom GPT usage, but then that never really Got monetized. So, you know, people built thousands and thousands of these custom GPTs and never saw a cent for it. And then they've just migrated that to plugins. So if you go to your old GPTs you built, there's now a migrate button. You can migrate them into plugins. And then now there's these plugin extensions, which are full applications, and then they've got a marketplace. So you can submit your plugin or plugin extension into their thing to be listed as a thing in the marketplace. [39:23] Matt: Yeah, [39:24] Dave Pengelley: Um [39:24] Matt: I I noticed that they had a partnership with Base Ten. Base ten was one of the first um inference providers. That I experimented with to where you can essentially host your own custom like open model and you get your own endpoint that you own. So it's kind of like open router, but you're paying for the actual VM that's hosting that inference as well. So for them to serve that through Marketplace too, so you basically can go and use any of the open models that base 10 has available in ChatGPT through this marketplace connection. That's actually kind of fascinating. [40:00] Dave Pengelley: Yeah, interesting. Yeah, it's it's it's strange. Um because I saw that um they're allowing you to use uh open weight models in the codex harness now as well. [40:11] Matt: Yes. So that that's yeah, that's kind of um what I meant, [40:15] Dave Pengelley: Part [40:15] Matt: what [40:15] Dave Pengelley: of what [40:15] Matt: I just [40:15] Dave Pengelley: you're [40:15] Matt: said. [40:15] Dave Pengelley: saying. [40:16] Matt: So yeah, it's [40:16] Dave Pengelley: Yeah. [40:17] Matt: funny how all that come all the links together. I don't know where it's gonna go, and I don't think they know either. I think they have ideas and they're just gonna keep uh steering that rudder to wherever the wind takes them. [40:32] Matt: How do you [40:32] Dave Pengelley: Yeah. [40:32] Matt: feel [40:32] Richard Webbe: It's [40:32] Matt: about overhead? [40:33] Richard Webbe: um as as a as a not so technical user of AI, I feel like I'm in the middle of the video streaming war. Where I went out and I got Netflix and I thought, great, I've got all the movies and TV I want at one spot. I've got Netflix and all the free-to-wear. And then all of a sudden along came Disney Paramount. And before I know it, I'd cancelled my Foxtel uh, you know, uh subscription. But I'm paying triple on all these different streaming channels for all this TV. I'll never have time to watch. But I tend to get dragged there for one particular show or one particular movie. And I always rationalise it when I'm sitting there, just using the video as the metaphor. You know, I'm sitting there and oh, that's a show. I'd like that show. It probably once for seven weeks. And okay, it's 10 bucks a month for this streaming channel for me to get access to that show. So I might do that. I I rationalize on the basis, well, if I went and saw this at the movie, so I probably paid $20 or $30. So there's three months subscription to get access to the movie anyway, and then off I go. So I'm using the value of going to the movie as the comparison. It's like you were saying earlier, Matt, you know, getting you know a thousand dollars worth of processing power and AI, you know, for a hundred bucks or something like that. But really, they're setting the agenda of what the value really is on both sides of the equation. Could be costing them one cent, and we're all just suckers. [42:03] Matt: Yeah, exactly. That's another layer to this. Like I mentioned, yeah, you know, the paying 200 bucks, getting 12 grand worth of value, you know, is irrelevant if that 12 grand also comes with a big asterisk on it as well. You know, like it's it's just like a comparison bias um straight away. [42:21] Dave Pengelley: It's it's a um who's on who's on is anyway? The points are made up and none of it matters. Like it's They're giving you here's the comparison against this mythical scale that we just decided to make up ourselves. Like these are all numbers we just invented. And so we invented this number, which is twenty-five X that number. And so believe us, because those are the numbers we've told you, because we're losing money hand over fist anyway, because none of this is actually paying for itself. Um [42:46] Matt: Yeah. [42:46] Dave Pengelley: we're all we're all propped up by by venture capital debt. [42:49] Matt: It'd be very, [42:49] Dave Pengelley: So [42:50] Matt: very interesting to see what happens at the end of um uh end of October, because that's when the the legacy pro times twenty accounts get brought down. To essentially times 10. [43:01] Dave Pengelley: Yeah. [43:01] Matt: Um, you know, because I I I'm on one of those accounts now, and my question is, and I'm sure there's going to be people listening to this who are on a times five or a times 20, and they're like, Do I do I go to the 500? You know, do I go to that times 25 um plan? [43:20] Dave Pengelley: Well, I've gotta say I've been using the new six point one soul this morning, and I was I'm down to like five percent on my GPT, and I've been getting a way more usage than I would have thought of out of that sort of 5% left using 6.1 soul. And so 6.1 soul does seem pretty efficient. And especially because the cached tokens are like really cheap. And so if you're using a decent harness and and I'm using Hermes is doing really well at the moment with the caching usage and it actually shows you percentage of cache usage. So if your harness and OpenAI would say our code their codex harness is very optimized for it as well. But um if If it's if it's tagging, if you're not switching models midway back and forth and just using the one 6.1 soul, and that way it's able to cache the tokens and and give you that massive discount, like 10 cents versus 90 cents or whatever it is for the um for the input tokens, it goes a long way. And so even though you know 5x, 10x are all made up numbers, at the end of the day, can you use it semi-continuously to get what you need done without running out? That's that's the big question. Um and how much does it cost you in order To not run out, [44:31] Matt: Exactly. [44:31] Dave Pengelley: um, with [44:32] Matt: And [44:32] Dave Pengelley: without [44:32] Matt: like [44:32] Dave Pengelley: random resets coming at you once a week. [44:35] Matt: Yeah. Yeah. Because that [44:37] Dave Pengelley: Um, [44:37] Matt: that is exactly, I think, the right sort of question and and mentality to have. And the the people that are out there going, okay, yeah, I've just I've just upgraded to my $500 plan and I've just done a an Astra Ultra session on ultra fast. [44:53] Dave Pengelley: got seven minutes of work, and I'm done. [44:54] Matt: And and they're like, Yeah, this is a scam. This is a waste of time. Like, well, no. Like it's not it's not [45:01] Dave Pengelley: Yeah. [45:02] Matt: optimized for that usage. That's there just as like [45:04] Dave Pengelley: I [45:04] Matt: a a carrot on a stick, if [45:06] Dave Pengelley: I [45:06] Matt: you will. [45:06] Dave Pengelley: want the [45:06] Matt: Like [45:06] Dave Pengelley: turtle [45:06] Matt: it's [45:07] Dave Pengelley: mode. I want the half speed mode. So when I'm going to bed, I can sort of run turtle mode and run it at like half speed overnight where [45:12] Matt: well, [45:12] Dave Pengelley: I'm not gonna [45:12] Matt: they [45:13] Dave Pengelley: be sitting [45:13] Matt: have [45:13] Dave Pengelley: there [45:13] Matt: that. [45:13] Dave Pengelley: watching it. [45:14] Matt: They have that through the API. Like so through agents API, responses API, they have um they call it flex. And flex is usually like half the price of just the standard rate. Um so yeah, you're right. Like if you could do like flex mode just in the app. Rather than three API, like that would be kinda useful, just to batch things together, like as you said, go overnight. [45:36] Dave Pengelley: Right? Because if if you're not in a rush, like f fast is good, and we've talked about that. That that quick iteration, I'm gonna try this, I'm gonna try that, I'm gonna do that. And if you just like that interactive, and obviously they're pushing the voicing. Like if you looked at the dot demo, their little launch video for dots, it was all Tony Stark style, talking to ambient dots, sitting on TVs on the wall and sort of flashy screens. There wasn't people talking on their phones, wasn't people in their AirPods. It wasn't [46:00] Matt: Yeah. [46:00] Dave Pengelley: typing to their dot. It was all here's my my big fluffy creature up on a TV on the wall that I'm talking to it about, fixing my diary and booking an Uber and doing all those sorts of things, right? So they're selling the vision of that ambient personal computer, Star Trek, Tony Stark style. Just talk to it and get things done. Which, you know, they're they're now dealing with um um what's his name? The Amazon bloke. What's his name? Jeff. [46:30] Matt: Bef [46:30] Richard Webbe: Bezos. [46:30] Matt: Jesus. [46:31] Dave Pengelley: Bezos. [46:32] Richard Webbe: F [46:32] Dave Pengelley: Cause [46:33] Richard Webbe: J [46:33] Dave Pengelley: um [46:33] Richard Webbe: Zos, [46:33] Dave Pengelley: I I [46:33] Richard Webbe: well [46:33] Dave Pengelley: I've got [46:34] Richard Webbe: said [46:34] Dave Pengelley: one on my desk, so I can't say her name, but the smart speakers from uh the shopping company from Amazon, they um they've they've got their new plus edition that they're rolling out, which is basically a chat GPT sort of LLM powered edition of the [46:47] Matt: That's [46:47] Dave Pengelley: smart [46:47] Matt: also a [46:48] Dave Pengelley: speaker. [46:48] Matt: dot, isn't it? Like the product's called a dot. [46:50] Richard Webbe: smart speakers, like with our LLMs, will remember who we are and copy what [46:54] Dave Pengelley: Well, [46:54] Richard Webbe: we do [46:55] Dave Pengelley: already [46:55] Richard Webbe: and [46:55] Dave Pengelley: that that they'll recognise voice signatures and stuff, and so if I put stuff on the shopping list on my wife. It puts who added it to the shopping list just with the old the old algorithms. But you know, they were very much, you know, answer response and very not they weren't even a chat bot, it was very much one command, one response, one command, one response. Uh, whereas now they're more conversational because they've got LLMs in them and you can actually go back and forth with them. So that's a that's a stronger play to be embedded in your household because households already have those things and they're all over the place. Like we've got one in every room of the house. The other one that to watch. For is we we don't laugh about Gemini, but Google has speakers in people's houses everywhere. So if you want that ambient agent computing, they're one to watch. [47:41] Matt: Yeah, [47:41] Richard Webbe: Yeah, [47:41] Matt: I [47:41] Richard Webbe: oh, [47:42] Matt: agree. [47:42] Richard Webbe: absolutely. [47:43] Matt: I agree. [47:43] Richard Webbe: That's why. [47:44] Matt: There's another thing that everyone has though, um, usually on their person as well, which is you know these guys. So [47:51] Dave Pengelley: But that [47:52] Matt: the [47:52] Dave Pengelley: wasn't the demo pitch, they were pitching the ambient devices in the house. [47:55] Matt: Yeah, which is another good point that their their demo probably was not properly thought out. Out because like I've been doing voice chats with my my stuff for a long time now, and since like the Grok bot voice thing came in, I've been giving it a go. But unfortunately, Grok is just useless. Um 4.7. There's certain aspects that are good, but uh, it's just cooked. Like, if we could have um the GPT live voice thing that they've actually got now through through their dots and through um the codex sessions, it's so much better as far as just interacting. And understanding, it doesn't make you just want to hang up on it. Um, like literally, there's a few points recently where I about actually asked the other agent to hang up because I'm like, you just hang up when you're when you feel like we're done. You know, I'll keep chatting until then. But on on these other ones, I'll just be like, nah, that's enough. Turn it off. [48:51] Dave Pengelley: Yeah, I I turned off the um the new live mode and went back to standard model because I found the live mode when they first launched it. Was just more janky. Like it was all like it was working too hard to go, mm-hmm, uh-huh, uh-huh, mm-hmm. And then when I then when it was stopping, I'd just go thinking and then finally deliver an answer versus the old traditional model where you'd say it and it would take a moment to process it and then go, Yes. Blah blah blah blah blah blah blah blah blah blah blah. And then pause for a second as a process the next bit, blah blah blah blah. But didn't have all that fake, uh-huh, uh-huh, agree agreeable stuff. And then when it delivered an answer, it was actually a smoother voice output. My initial experience with the live was That it was much more broken and static y, and I didn't like it. So I was like, I'm just gonna go back to standard voice because for me, though that that's great. Um, but even then, like, even if it's not their live thing, the Chat GPT voice is actually really, really good on mobile, like regardless of which one. Like, [49:48] Matt: Yeah. I'm [49:49] Dave Pengelley: yeah. [49:49] Matt: curious to I I can't wait for you to get access to the dot thing so that you can have like just a voice chat with it because I think so far um it's the best version of this that I've experienced. Uh I haven't experienced Muse or Instinct if they do it yet. Um But like I mentioned before, I'm so curious to see what Anthropic does as a response to this because like they're gonna have their version and um it's gonna [50:15] Dave Pengelley: So [50:15] Matt: be [50:15] Dave Pengelley: so [50:15] Matt: interesting. [50:16] Dave Pengelley: long as the dog can have the maple voice, because maple is my girl, like I I don't want to talk to a different different one. I like maple. Um whenever I've tried switching voices, I'm like, nah, this is not my person. I can't talk to you. Um but that that's the end of the thing. Thing about Anthropic and Claude, right? They are not doing as many things as OpenAI. Like they don't have images. Like they're not like that, they don't have tools like ImageGen over in Claude Land, right? So they're not as capable in some way. They need to outsource more of that stuff to other models and other third parties to do some of those um other [50:54] Matt: That's a really good point. [50:55] Dave Pengelley: tasks. [50:55] Matt: Like 'cause like they don't like if you're in the Claude app, like you don't you don't go and generate an image or generate a video. So it's a really, really interesting point because like how how are people doing those crazy um [51:07] Dave Pengelley: I [51:07] Matt: those [51:07] Dave Pengelley: generated [51:07] Matt: crazy [51:07] Dave Pengelley: this video with [51:08] Matt: videos? [51:08] Dave Pengelley: Claude. It generated every frame and put it together. I'm like, no, no, I didn't. Like it might have used foul and and outsourced it through an API and it made the prompt, but it didn't rot make the freaking images. [51:18] Matt: Yeah, I would love to see like what what APIs were used in the creation of those videos because you can't just code that, I don't think. I don't really understand how that works. [51:30] Dave Pengelley: Whereas like the open AI, the fact that it's got built-in image gen is great. I use it all the time. Like, especially when I'm trying to develop something, I was like, make me some mock-ups, show me some ideas, and I'll work from that. I did that with the Sites by Agents um webinar a few weeks ago. I got it to mock up what a website that looks like a book would look like on brand for the for the for the thing. And it went and built me an on-brand looking book thing. And I gave that to the to Codex and said, build it. And I did. [51:56] Matt: Hm, it's a [51:56] Dave Pengelley: So [51:57] Matt: strangely effective workflow to To uh get an image of a mock-up first rather than a coded prototype of all of the elements in place. Um, because yeah, [52:09] Dave Pengelley: Well [52:09] Matt: you can generate a few different versions and then go, okay, this is good. Go ahead and do this [52:14] Dave Pengelley: I [52:14] Matt: now. [52:14] Dave Pengelley: got I got that idea from OpenAI. I was at one of their ver one of their events here in the city once and one of the OpenAI sort of SEs got up and sort of said, Look, we know that we're not the best at creating interfaces. Like they just he just owned the fact that you know compared to to Clawed or whatever, they didn't make the best interfaces. But he's like, but we've got really good image generation and we're really good at following a plan. So use the image generator and mock up what you think you think wants to look like, and then give us that, and we'll copy that and we'll do a great job of that. So I was [52:42] Matt: Yeah. [52:43] Dave Pengelley: like, Well, that's that's good advice. [52:44] Matt: And for those that are really good at actually steering tools like Figma and whatnot for for design reasons, you can take your designs there and then you know translate that into your building session and they can [52:57] Dave Pengelley: Right. [52:57] Matt: use that. The Figma FCP is apparently really, really good for that. [53:00] Dave Pengelley: Well, Figma is one of the plug-in extensions that's there. So I've got a little Figma thing I haven't authenticated. And Canvas in there as well. So I'm actually keen because using Canvas AI to build things in Canva has been really rubbish. Like it's just it's just not up to spec. It it annoys me because Canvas should be in a place like Framer is apparently doing from the Framer Marketing, where it's really opinionated, knows how to use its tool set really, really well. And so when you say build me a thing, it should be able to look through its stock library of images, find good images, composite things. Together, apply the right colors and gradients, but it's just rubbish. You just get the most the worst looking designs. And I've just I've not been happy with the Canva AI for generating sort of any kind of presentation or assets. But maybe chat using Canva integrated with chat and getting getting GPT to drive Canva, maybe that that's a better outcome. I haven't I haven't experimented with that yet, but [53:53] Matt: Yeah, worth a [53:54] Dave Pengelley: um [53:54] Matt: try. And that Mirage Company that I linked you [53:56] Dave Pengelley: yeah, [53:56] Matt: guys to [53:56] Dave Pengelley: with Tesseract [53:56] Matt: last time. [53:57] Dave Pengelley: and and even with Figma, because one of the things I've We've all seen my kangaroo. Um, the SVG kangaroo that that Astra made me. That was rubbish. But maybe if I'm integrated to Figma, which is a vector drawing tool, maybe I can get it to build better things using Figma than trying to just do raw maths. So yeah, it's an exciting time, but it's also the big question with a lot of these announcements with X is what sort of tell them the price, son, tell him the price. Because it is a little bit confusing. What you're paying for, what What's included, what's not, even the dot. Like it's it's free for a point, but if you do wrong things, it's not. And then the spaces, as soon as you jump into a spaces, now you're using chat GPT work, not chat GPT chat. And so work does use your codex allowance. So be careful if you jump all in with spaces. That's not free, just like using the Chat GPT projects and chat GPT chats that will come out of your allowance because it's work. And so there's all there's gonna be all these little gotchas where I can see people going, hang on, what What do you mean? You said you're giving us all this stuff included, but it's actually pulling out of my work codex quota. Like that's that's not what I thought was happening. So yeah, [55:08] Richard Webbe: Kate [55:08] Dave Pengelley: watch [55:08] Richard Webbe: Kate [55:09] Dave Pengelley: this space. [55:09] Richard Webbe: just made [55:09] Dave Pengelley: No [55:09] Richard Webbe: a [55:09] Dave Pengelley: doubt [55:09] Richard Webbe: comment [55:10] Dave Pengelley: we'll [55:10] Richard Webbe: in [55:10] Dave Pengelley: get [55:10] Richard Webbe: the [55:10] Dave Pengelley: lots of resets out of it, so I'm happy about that. [55:12] Richard Webbe: Kate just made a her friend Kate Fabian just made a comment in the chat saying that uh Claude Design is a is a great product. [55:19] Dave Pengelley: Claude Design gave us our website, AI Operators Pod. We did a live demo on that way, way back when it first came out, and I threw our um threw a few of our assets at it and it designed. Website for us, so then when I wanted to create the podcast website, I said, Here's the design pack, [55:35] Matt: Do you remember [55:36] Dave Pengelley: go [55:36] Matt: what [55:36] Dave Pengelley: get [55:36] Matt: model it [55:36] Dave Pengelley: it. [55:36] Matt: used at the time? Like because that actually probably would be boosted now with you know these five point five releases. [55:42] Dave Pengelley: Oh, it's probably like 4.5 or whatever, like a 4.6. So it was like back when I was still paying for for Claude, which has been some time now. [55:50] Matt: Yeah, which like that's crazy. Like 'cause if you think of Claude four point six, the the you know, one of the most popular open source local models that people run. the 3.8 quen 27 [56:02] Dave Pengelley: Yeah. [56:02] Matt: billion, that actually is technically better than Opus 4.6. Like [56:08] Dave Pengelley: Yeah. [56:08] Matt: that kind of blows your mind. Um so yeah, [56:11] Dave Pengelley: Yeah. [56:11] Matt: anyway, it'd be worth trying design again on these new 5.5 models and see how that goes. [56:17] Dave Pengelley: Yeah, I'm I'm not buying Claude immediately. We'll see. [56:22] Matt: Yeah, no [56:23] Dave Pengelley: Um [56:23] Matt: need to no need to rush as what we mentioned before. No need to like the grass is greener is not actually true. [56:29] Dave Pengelley: right. Um yeah, so all these things, if if you're working and if this has overwhelmed you all and you're like, I don't know what any of this means, then we're all available on LinkedIn. DM us for a chat to see whether any of this is applicable to you and your business. Maybe it is, maybe it's not. But we don't start with, here's the tool, let's find a problem. We want to understand the problem, where you're at, what your business is doing, where your friction is, and then work out what the right tool is. That that that's the that that's the way. If people are trying to sell you the tool before that. Understand your business, then you're less likely to see success. So the move is understand the problem and then determine the tool. [57:09] Richard Webbe: Where is the value? Absolutely. [57:12] Dave Pengelley: Brilliant. [57:12] Matt: True. [57:14] Dave Pengelley: Well, um, we're at the end of the hour, boys. Any uh any last thoughts, comments, things that uh any any mini wins of the week that anyone [57:23] Matt: Hmm. [57:23] Dave Pengelley: wants to share? [57:24] Matt: I [57:24] Dave Pengelley: I won't [57:25] Matt: actually [57:25] Dave Pengelley: I [57:25] Matt: didn't [57:25] Dave Pengelley: won't [57:25] Matt: go [57:25] Dave Pengelley: play [57:25] Matt: and [57:25] Dave Pengelley: the [57:26] Matt: look back [57:26] Dave Pengelley: video. [57:26] Matt: at my my wins as such. Um I think I'm just I I'm pretty happy with my workflows now. That's probably a win. Um yeah, just getting comfortable with the tools. You know what actually gets you c more comfortable with the tools and and you start to expand or go a bit more um like vertically in your workflows is when you stop looking at all the other options. You just kind of master one thing and and go down deep in one. Um [57:54] Dave Pengelley: Yeah. [57:55] Matt: actually another win that I've had, sorry, is just really good conversations with people. You know, out outside of my work as such is just really good chats about visions, um, you know, where things could go, um just beneficial chats. I I want to have more of those. That that's definitely a win. It feels good. [58:16] Dave Pengelley: Excellent. [58:18] Matt: And you [58:19] Dave Pengelley: Excellent. [58:19] Matt: guys. [58:22] Matt: No wins. [58:26] Richard Webbe: My win is clarity. All of this stuff and running businesses and reaching out to customers and that, I tend to lean on AI and not just LLMs, but the broad spectrum of the tools to help me refocus where I'm going and why I'm doing it. And that's a bit of a win for me. [58:48] Richard Webbe: Kate made a comment nicely that we tend to drain out a bit of the noise. [58:52] Dave Pengelley: Yeah, Kate [58:53] Richard Webbe: And [58:54] Dave Pengelley: Star. [58:54] Richard Webbe: that's that's [58:55] Dave Pengelley: Super [58:55] Richard Webbe: what [58:55] Dave Pengelley: fan [58:55] Richard Webbe: I find. [58:55] Dave Pengelley: Kate. [58:56] Richard Webbe: Yeah, that's what I find. Just you know, you sit back, there's all Stuff coming at you, and you go, Will I use it for that? Will it help? And then, okay, let's clear the machine, as Kevin Costner would say in his baseball movie. Clear the machine and refocus on what are we doing right [59:12] Dave Pengelley: I [59:12] Richard Webbe: now? [59:12] Dave Pengelley: thought he said if you build it they will come in a baseball [59:14] Richard Webbe: He [59:15] Dave Pengelley: movie. [59:15] Richard Webbe: said that as well. There was two, it was two, there was two different movies. [59:20] Dave Pengelley: Ah, okay. I think Cosner [59:22] Richard Webbe: He [59:22] Dave Pengelley: Baseball [59:22] Richard Webbe: loves [59:22] Dave Pengelley: Cornfield. [59:22] Richard Webbe: baseball, and he made two different baseball movies. And one of them was uh one was a movie about a famous uh baseball. Player who did what they called um was it a no-hitter or a no-out or whatever. And uh he was just standing there looking, and they created this great focus where everything else disappeared, and it was just him and the guy throwing the ball at him. And he went, clear the machine. And that's what I I find I use my LLMs and my stuff for, just clear the machine. And then I find that the all the chatbots and that are getting better at remembering and cross-referencing. previous discussion and previous data and pulling that data up. They were pretty crap at the start. Now I find them accessing that data and cross-referencing that, which is why we're here, right? That's what it's all about, cross-referencing that data for real information and decisions. That would be my wins these week. [01:00:18] Dave Pengelley: Nice. My my win is the new operator stories alive on the channel. So we've got a new show type uh that launched. I think I put up Tanya's video last Thursday. Uh So our first interview with Tanya Ferguson from uh Georgia, Atlanta. She was lovely. Uh really interesting chat with her around her AI sort of enablement enablement um and services business that she's running over there. I had another chat uh this last Monday, which I'm looking to get up on the channel this coming Monday. So I'm gonna aim for a cadence for the channel for for our regular viewers that watch this sort of stuff. So we'll have an operator story on Monday, we'll have the live show on Wednesdays, and we've got the field notes on the Fridays. And maybe one day we'll get like I can get Matt to make a field note, so it's not just me. But um one day for me, you suck with me for field notes. [01:01:04] Matt: Give [01:01:04] Dave Pengelley: Um [01:01:04] Matt: me an idea and I'll I'll just do it off an idea. [01:01:08] Dave Pengelley: but uh but yeah, so we want to get regular and have that have different kinds of content, different lengths, different styles. Uh the operator stories is really good. The the chat I had on Monday, again, very interesting. Uh guy from Toronto in Canada. And just interesting hearing how people are doing these things in different markets and their focus and how they got to being involved in this sort of stuff from very different career paths. So yeah, interesting if you like that. Kind of personal interview stories that's sort of AI adjacent and aligned, then uh check out the operator stories segments on the channel and make sure you like, subscribe, bell icon. And uh thank you for uh joining us today. Thank you, gentlemen. [01:01:45] Matt: Thank you. Thanks, Lavinus. [01:01:46] Richard Webbe: Thank you. [01:01:47] Dave Pengelley: Apologies for my uh my coughing fits. This cold and tiredness and talking a lot has just been like making my throat a desert, and I'm like, uh dying. Um Apologies for that. Hopefully I'll be um up to spec next week. Enjoy your long weekend. If you're in a state that has a long weekend this weekend, we're away camping, and apparently it's gonna rain. So hey, no AI and rain. Be [01:02:12] Matt: How [01:02:12] Dave Pengelley: good for the [01:02:12] Matt: good. [01:02:12] Dave Pengelley: quota. All right. Thanks, boys. See ya. [01:02:16] Matt: All right, catch up.
Related episodes

Keep going

Chapters

Jump to a section

0:00Cold open and the LinkedIn algorithm
4:00OpenAI Dev Day: demos and business value
9:30Dots, pricing and Codex quotas
15:42Fast modes and agent tools
19:30Voice agents and personal data
23:00Government data, privacy and convenience
30:00ChatGPT, Spaces and the everything app
40:33Subscription economics and value
45:36Ambient agents and voice experiences
50:16Image-led design and building workflows
56:29Start with the problem, then choose the tool
1:00:18Operator Stories and the channel cadence