← Back to search
Meet Dot: OpenAI's Always On Agent
AppStories · 2026-10-05 · 38 min
Show full episode description
Meet Dot: OpenAI's Always On Agent This week on AppStories, John is back from OpenAI DevDay, so he and Federico cover the history of AI agents, the current landscape, and what OpenAI is doing with Dots. On AppStories+, Federico provides a sneak peek at what's in store for his Reminders, Notes, and upcoming CLIs. Also available on YouTube here . Links and Show Notes AI Agents OpenAI Dots and DevDay 2026 OpenAI launches Dots, its bubbly agentic avatar (TechCrunch) OpenAI’s Dots: A First Look Self Hosted Agents OpenClaw OpenClaw on GitHub Hermes Agent OpenAI hires OpenClaw creator Peter Steinberger Hosted Agent Competitors Meta's Muse becomes App Store's hottest download Grok Bot Grok Bot Brings Always-On AI Agents to macOS and iOS AppStories+ Post-Show: RemCTL: GitHub MacStories: Introducing RemCTL, the Power-User Reminders CLI for macOS and AI Agents NotesCTL: Club MacStories download MacStories: NotesCTL, a Club MacStories Preview of Federico’s Advanced Apple Notes CLI Tool Leave Feedback for John and Federico AppStories Feedback Form Follow us on Social Media Federico is @viticci everywhere John is @johnvoorhees`everywhere Affiliate Linking Policy
✨ Episode Outline — click any point to jump to it in the episode
Problem solved
Explains what
OpenAI's new 'dots' always-on agentic bots are, how they descend from
OpenClaw, and how they compare to GrokBot, Meta's Muse, and
Hermes Agent.
Benefits
- Persistent agents that stay running instead of one-off chat sessions
- Adaptive memory that learns your tasks, tools, and working style
- Proactive suggestions once the agent learns about you
- Messaging-style back-and-forth for email, calendar, files, and web research
- Productized agents without OpenClaw-style technical setup overhead
Use cases
- OpenClaw-style agent reading email, sending email, creating calendar events, summarizing web pages
- Codex projects with their own memory that pick up where they left off
- GrokBot bots each running in their own VM with routines, connectors, and webhooks
- Dots can access your actual computer via Codex, unlike VM-only hosted agents
KPIs / results
- Meta's Muse #1 on the App Store for two weeks and counting
- ChatGPT #2–3 on App Store charts; GrokBot around #97
- Claude now regularly in the App Store top 10
- OpenClaw's first version (WhatsApp Relay / ClawDIS) dates to November 2025
[SPEAKER_01] Hello and welcome back to another episode of AppStories. I'm John Voorhees and with me is Federico Viticci. [SPEAKER_00] Hello John, welcome back. [SPEAKER_01] Hey, it's good to be back. I got, I don't know if I still have a little bit of a sunburn. I was outside a lot. I was out in California where it is always sunny and I spent about six days out there because it was both a personal trip to visit family and also a work trip. [SPEAKER_01] Because at the end of the trip I went to OpenAI's Dev Day, which was a lot of fun. I had a great time there. It was a really neat, really neat event and I talked about it a little bit more on MacStories Unwind last Friday. So if you're curious about what an event like that is like and how it maybe differs a little bit from something like WWDC, I talked about it there and I can, you can listen to that. [SPEAKER_01] But today Federico, I want to talk to you about kind of the headline feature that OpenAI revealed at Dev Day, which is dots. Which, funny story when the rumors were swirling about this, everybody thought it was going to be called O, lowercase O. And, right? And I started writing. It's stupid. It's stupid, right. Well, and look, OpenAI, not known for the best to be the company that's best at naming. [SPEAKER_00] Not the, really not the best. And, I mean, they have released models that were called O1 and O3. Yes. Right. [SPEAKER_01] So there was, you know, it was plausible even though when I sat down and I thought, I'm going to start like framing out an article about whatever this is going to be. I started doing it and I thought, there's just no way any company is dumb enough to name this lowercase O because, you know, starting a sentence with the lowercase O just doesn't work. Anyway, I want to talk to you about dots. The O stood for a dot. That was their text equivalent of a dot. And dots are agentic bots. [SPEAKER_01] And you may wonder, like, what is that? And, you know, you've probably heard there's a lot of these things. This is kind of like the hot thing right now. Yeah. You know, I think you can step all the way back to the beginning of the year in January when OpenClaw came out. OpenClaw is really what set this off. [SPEAKER_01] And I think before we really get into the specifics of dots, I think it's important to kind of understand a little bit about that history because and how these sorts of things differ from just using an agent in a loop in something like Codex. Because, you know, this is something that I've been trying to do with Codex for a while where I'll have a project that has its own memory that has an ability to kind of pick up where it left off. [SPEAKER_01] But these bot agents are more than that because, one, they're persistent. They don't go away. They're always there ready to do something for you. Two, they have their own memory and their memory can adapt over time to the kinds of tasks that you ask it to do. So it kind of gets a feel for the kind of work you do, the way you like to do things, what your tools are, that sort of thing. [SPEAKER_01] And they are, I think, more than they're proactive too, which is, you know, they really try to, that gives you suggestions about things that they can do once they start learning more about you. And that really comes back to the original OpenClaw that Peter Steinberger, who now works at OpenAI and still works on OpenClaw as an open source project, invented essentially in early January or over the holidays. [SPEAKER_01] And I know you obviously spent a lot of time with OpenClaw. I did a little bit too. But, you know, what we're seeing today, I think, is the friendlier version of what OpenClaw was back in January. [SPEAKER_00] Right, yeah. I spent a lot of months getting really into OpenClaw. And I feel like the, so the big invention back then was this idea that you're using, the idea was that the harness was the product, not the model. The model was not the product. [SPEAKER_00] The concept was what if you could have a conversation or multiple conversations in different channels with a model and the agent behind it had the tools to do things on your behalf. That was possible before with the likes of CloudCode and Codex. But they were tuned for programmers. [SPEAKER_00] And Peter's idea was what if you took that system and you made it more for generalized productivity in addition to coding? Coding was still possible. Coding was still possible. [SPEAKER_00] But what if you could have a back and forth conversation, just like text messaging, with a model that would act as an agent thanks to the hardness behind it, to do things like grabbing your file from your desktop, reading your email, sending an email, creating calendar events, managing calendar invitations, going to a web browser to read and summarize a web page. [SPEAKER_00] Coding was still possible. [SPEAKER_00] That feels like forever ago. It was just 10 months ago. And the first version of OpenClaw, back when it was called WhatsApp Relay and then ClawDIS, was November 2025. So it's not even a year. That first version was really, really primitive, right? It was a WhatsApp Relay channel based on running a model, pointing the model to a bunch of markdown files. [SPEAKER_00] Like the heartbeat and the crown jobs and the soul.md document and creating a little harness for the model, right? To perform and do things. Now that was the... But the moment you saw it, if you tinkered with it and if you spent some time playing around with it, you could see the future, right? You could see the future of... [SPEAKER_00] And you could draw a line and say, I'm projecting myself a few months into the future. This line, this trajectory is only going to continue until we get to an abstraction of this, to the productized version of this that is capable of doing the same things, but does not require the technical overhead to get it going and to maintain it going forward. [SPEAKER_00] And if you draw that line from January 2026 up to now October, so it's been 10 months, we have the abstractions are here. And the productized versions of these agents are here. And we have, I think, arguably the first serious contender in the space was probably GrokBot. [SPEAKER_00] I think SpaceX AI kind of beat everybody to the punch in the last two months. Quick parenthesis here. Do you think, John, that all of these people in Silicon Valley actually talk to each other, even if they work at different companies? How is it... [SPEAKER_02] Absolutely. [SPEAKER_00] Right? How is it that everybody starts coming out with the same ideas at the same time? [SPEAKER_01] They all know each other. They're all moving between these companies constantly. I got this definitely from being at Dev Day. They all know each other. They're all going out at night to the bars together. They're, you know, they're all working at these companies. They're all moving. They all went to maybe college together or they're moving from one company to another. And I think these ideas just permeate from one company. [SPEAKER_00] That was my suspicion, which is why I brought it up. Anyway, and then, of course, we got Muse from Meta, which as of right now, still the number one app on the App Store. For two weeks and going. Then we got, obviously, Dots by OpenAI. Right. And, of course, I should also mention Hermes Agent became a really popular alternative to OpenClaw. [SPEAKER_00] But I would say, you know, zooming out a little, you can probably see the two macro categories for these kinds of agents. The self-hosted, do-it-yourself-because-you're-a-nerd type of agent. Right. OpenClaw, still going strong. Hermes Agent, very popular, arguably more stable and developer-friendly than OpenClaw. [SPEAKER_01] They have OpenClaw for enterprise now, too. They have OpenClaw for enterprise. Which is funny. [SPEAKER_00] Yes. There's also Hermes for teams, by the way, John. But anyway, you have the do-it-yourself kind of agent. And you have the cloud-based hosted agents that tend to work in a VM, in a virtual machine based on Linux. Although, arguably, the OpenAI version of this, which, you know, we can talk about in a second, also has access to your actual computer thanks to Codex. But this is what's happening. [SPEAKER_00] And I think of all these companies, of all the simplified versions of these, they're all based on the same concept. Right. It's a model that has been given a harness, has been given the tools to act on your behalf and to do things on your behalf, going outside of programming and writing code with the capabilities of just doing productivity work, knowledge work, regular office tasks or research for you. Right. [SPEAKER_00] I think Meta has, I mean, the numbers speak for themselves. They have the most successful consumer version of this. But this is sort of the big picture is this. Yeah. [SPEAKER_01] And I wouldn't count out OpenAI on this because I think, like, the problem that XAI is going to have is that I don't think that their model is as good as what either Meta or OpenAI have. [SPEAKER_00] So Theo and Ben from the NerdSnipe podcast, and also Theo is an excellent YouTube channel, makers of T3 code, my favorite way to use Cloud and Cloud code without using the Cloud app. Mm-hmm. Mm-hmm. They had this exact conversation, I think, in the most recent episode of their podcast. [SPEAKER_01] I've only listened to part of that. [SPEAKER_00] And I subscribe to this theory. So GrokBot is actually a cursor product, which became part of SpaceX because of the acquisition. I think SpaceX AI or Cursor, whatever you want to call it, they have the best products out there. Like, if you just take a look at the app, take a look at the product, it's arguably the best one. [SPEAKER_01] Because Cursor was making good products. That's the thing, right? [SPEAKER_00] Cursor was making good products, but they have the worst models. Yes. And so they have the best products. I mean, GrokBot is, you know, set aside the fact that it's called Grok and it's, you know, the name ruins it forever. However, I actually wouldn't be surprised if the whole Grok series is eventually renamed to something else. The app is actually really solid. [SPEAKER_00] The idea that you have these multiple bots and each bot has its own VM and you can set routines and there's support for connectors and there's support for webhooks. Like Cursor is a really well-designed and well-integrated product, but it's powered by subpar models. Right. Whatever Elon wants to say that, you know, Grok is now approaching Opus 5 levels, it's not. It really is. [SPEAKER_01] Yeah, no, it's not. And the thing that I think the most surprising entrant here really is meta because meta felt like it was very far behind. But their models, but their most recent models, by all accounts, are a lot better. I have not really used them myself, so I can't really speak to that myself. But Muse obviously is both leverages their ability to make good products, which, I mean, there's a reason people have used things like WhatsApp and Facebook and things for years. They're good at products. [SPEAKER_01] And now they have a model that stands behind it. That's pretty good, too. I think OpenAI, though, I wouldn't count them out only because I think that right now, some of their models are right up there with the likes of Anthropic, who is not in this space at all right now. So they have probably the best models of the three that are doing bots right now. [SPEAKER_00] It's kind of strange that Anthropic is not in this game at all. We should stop here for a second. [SPEAKER_01] Yeah, we should talk about that because I think that while these are very much, in a lot of ways, consumer-friendly bots, and Anthropic is very much pointed at the enterprise, I do think that there's room for these kinds of agents in kind of the workplace, that the same reason they're more approachable to your average consumer will make them more approachable and usable for people who are just doing their everyday tasks at their job. [SPEAKER_00] A couple of things. So just looking at the App Store charts, Muse is number one. ChatGPT is number two or three. It's been fluctuating this week. GrokBot is, like last time I checked last night, number 97. Oh, wow. In the top three top charts. So pretty clear discrepancy. Again, the Grok name is ruined for a whole lot of reasons. [SPEAKER_00] I think it's going to be an uphill battle for SpaceX with this product, but we'll see. [SPEAKER_01] They just have a lot fewer users in general too. I mean, if you look at kind of the size. [SPEAKER_00] The models are not great. Grok has been doing questionable, illegal arguably in many states things for a long time. Not the greatest brand, you know, in the AI space. But which is a shame because again, I do think the cursor team makes good apps. It's good software. Objectively. It's good software. [SPEAKER_00] But I was going to say, lately, in talking to friends in real life, I noticed that more and more people are aware of Claude. Oh, interesting. Which was not the case. Was not the case a few months ago. I think what actually happened, at least in Italy, when the whole Dario versus Trump thing happened, it was discussed a lot on Italian TV and social media. [SPEAKER_00] And I think because of that drama, a lot of folks, like regular people, were exposed to the brand of Claude. Like, what is Claude? And I think this has also been happening elsewhere. It happened here too. [SPEAKER_01] I remember that they moved up the charts in the United States when all that happened. [SPEAKER_00] Claude has been and it's now regularly in the top 10. So, obviously, trailing ChatGPT, but it's up there. And the brand recognition has gone up and up. Second thought. [SPEAKER_00] I think a lot of people ask me and you, John, like, why do I need to care about these agents when I can just have a project in Claude or just create a new chat for whatever I want to do? [SPEAKER_00] I think what's fundamentally different here and what these companies seem to think it's the future of the product is all things being equal, right? In the sense that you can use Claude, you can use ChatGPT without bots. There are memories, right? There are plugins. [SPEAKER_01] There are automations, all kinds of things. Scheduled tasks. [SPEAKER_00] The difference is that, I guess the idea is that you created these agents. I think we need to underline these points because these companies do not do the best job at explaining this difference. [SPEAKER_01] No, they don't. [SPEAKER_00] You can continue to have a conversation in Claude Cowork or in ChatGPT and it's going to use your memories and it's going to know your preferences and it's going to have your connectors and it'll be fine. The difference here is that you're creating these quote-unquote personalities that are agents and you can define for each of them a role, a model, depending on the app and the ecosystem that you're using, a thinking level. [SPEAKER_00] You can give them specific plugins. Again, depending on what you're doing, you can have them communicate with each other. This is something that is possible in Hermes, in OpenClaw and in Rockbot. The difference between an agent and a conversation is that think of the conversation as more ephemeral, right? [SPEAKER_00] It's the regular chat, but now with tools, whereas the agent is more persistent and because of that persistency is also capable of doing greater and more flexible things. This seems to be what these companies are after right now. Maybe it's a fad and it'll go nowhere six months from now, but the money is now flowing in this direction. [SPEAKER_01] Yeah, and I don't think it's a fad. I really do think this is going to work because part of it, even for someone like me who is used to starting up multiple threads and codecs, keeping track of them, having automations and scheduled tasks and all these things, it simplifies that. That's one of the things about it that I think is important is that I can just pop open my dot and I can ask it to do something and it keeps track of that. [SPEAKER_00] So you've been using dots. I have been. I cannot because I'm very disappointed actually about the fact that they, I think they kind of botched this rollout by not launching the EU. Yeah. Especially, I guess meta is the same problem. I'm just. [SPEAKER_01] I wonder how it's different though than just, you know. [SPEAKER_00] I'm very surprised by the fact that Grokbot is in Europe and maybe the European Union doesn't care about SpaceX. [SPEAKER_01] Well, I think they do, but because they've absolutely gone after Grok before. But at the same time, I do wonder why, I do wonder if it was just like they figured they'll get to it because I don't know how, what these agents, while they're packaged differently, they're not really doing anything fundamentally different from what already exists and what is already available. Right. [SPEAKER_00] It's all about the presentation. I think you're right. I think the, not the visuals, but like the ergonomics and like how the product is designed is very important here because it's, I think there's a, there's a, there's an argument here that it's the design of these agents that makes them fundamentally different from the, a thread. Right. And if, I mean, it's the harness. [SPEAKER_01] Well, and you can, and you can feel it when you read some of the articles that are doing, what we're seeing is the, the hand wringing of over OpenClaw coming to the service again, all over again with these bots. You know, you see articles about how, oh, you know, it went off and it did this thing and it did that thing. And I didn't mean for it to do that. And I think that there, I mean, when this is packaged for an everyday user, two things, I guess. [SPEAKER_01] First of all, I think that the concern there, especially with government regulators, is that it's a little bit like advertising cigarettes to kids with cartoons in the sense that you're making these things friendly to everyday users. And now people who don't know what they're doing are going to use them and get themselves in trouble. [SPEAKER_01] And that, but I think that that also, but personally, I think that's an overly paternalistic viewpoint of people that people are not as stupid as they are sometimes, you know, treated by the tech press and that people can figure this stuff out. There is a burden on the companies to make sure that they're just transparent and explaining to people what's happening and what they're doing and the consequences of what they're doing. [SPEAKER_01] But I don't think that this is like, this is, this is not like handing a loaded gun to a baby. It's not that it's, it's just not as bad as I think sometimes it's made out to be. [SPEAKER_00] Can you tell me more about DOT? Yeah. So, so it's not a multi-bot product. Yeah. You have the one DOT for you. Yes. Okay. [SPEAKER_01] But they very heavily hinted more than once, not just on stage, but with engineers that I talked to that multiple DOTs are probably coming in the future. I think that they are concerned about given the size of their user base of having the compute to handle all the DOTs because DOTs are running on Astra, which is a very proactive LLM as I'm, you know, I've, [SPEAKER_01] a little annoyingly so at times, but Astra is in the background. And it's all very simplified in the sense that it's Astra. It runs very fast for Astra. It doesn't have any thinking levels or model choices or speed choices. All of those things are just gone from the, from the product. [SPEAKER_01] You simply have what is, it looks just like an iMessage thread with your DOT, which you can, you know, customize, pick the shape, the color, the eyewear, the headwear, you know, all that stuff. They're, they're very much looked like, uh, they're very geometrical looking. Uh, they're not like little people or anything like that. Uh, and yeah, I mean that, that's how it works and it's got access to all the plugins that open AI has deals with, which is a lot. [SPEAKER_01] I think they said like 4,000 or something like that. So that's an advantage because you're going to get up to your Gmail or your calendar or Slack. [SPEAKER_00] And it becomes like a standalone section of the Chagipity app. [SPEAKER_01] It does. It's right at the very top. And it's, uh, because it runs in the cloud first and foremost as a VM, the nice thing about that is that that thread is persistent everywhere. You can open it on your Mac. You can open it on your iPhone, your iPad. [SPEAKER_01] It gets around one of the biggest problems that you and I have had forever, which is how do you kind of pick up and let off with a project and a thread that maybe is on your Mac and you're going out for the day and you want to do it on your phone. Now they open AI has, you know, remote and all these other things, but those tools, although they've gotten a lot better can be fiddly. [SPEAKER_00] And this is so, so can the dot running in the cloud reach out to a computer that has codecs installed? [SPEAKER_01] Yes, it can, it can run codecs. It can run codecs. So, and I ran into this because I no longer need to have my custom tunnel. [SPEAKER_00] I have, I have a custom thing that I built for myself that basically exposes my Mac via MCP. [SPEAKER_01] Yeah, you don't need that. But because what, so I ran into this because I was on the flight home last night and I wanted to do something on my Mac studio back at home. And I was on the American Airlines Wi-Fi on the plane. Okay. And I had to, there's one step you have to do. You have to go onto the Mac, go into your conversation with your dot and say, make this computer available for this thread. [SPEAKER_01] And it'll think for a minute and then it'll give you a button, you click it, and now it's available. So now what that does is while the dot is running in the cloud, it then has access to your file system on your local Mac and can run things on the Mac. Which, you know, I mean, they don't give you that by default, which I think is a safety thing, obviously. The, you know, OpenAI is doing a pretty good job, I think, with this of doing a progressive disclosure of permissions. [SPEAKER_01] So that, you know, by default it really doesn't have access to much of anything. And as you start asking it to do things, it'll be like, oh, you know, I need this to do that. Are you okay with that? And you give it the permissions you need. And so over time it starts learning, it starts, and you can then get comfortable with what it's doing once you get the feedback of whatever actions it's taking on your behalf. It gets to the point where you can, at least, you know, they hope, trust it more and do more things for you. So I immediately set that up. [SPEAKER_01] I went into Jump Desktop, remoted into my Mac, authorized it, and then I went back to my iPad and just used my dot on my iPad doing things on my Mac at home. For instance, one thing I did was I updated the new beta of Golden Gate 27.2 remotely flying over the United States, flying over the Rockies, updated it. [SPEAKER_01] And when Jennifer picked me up, she said, your computer made that noise it makes when it starts up. And I said, yeah, that's because I was updating my computers while I was in the air using my dot. [SPEAKER_00] Huh, man. [SPEAKER_01] Which is pretty cool. And that's just like a little thing. You know, there's a lot of, because it seems to do a very nice job of tracking all the different things that you're having to do. I was doing things like I was reading on the plane and I'd find a link and I'm like, oh, I want to put this in the newsletter this week in our interesting link section. And I just told my dot, save this URL for Friday when I do, you know, when I put together the newsletter. [SPEAKER_01] And it did. And this morning, just to make sure it was working, I said, give me a list of all the links I've given you for the newsletter on Friday. And it still had them. So it does, that saves me the pain of having to have a special shortcut for that, a place of notion where it's stored. I just give it to the dot and then I, on Friday now, it will automatically, when I get up in the morning, have that list for me ready to go so I can start writing that section of the newsletter. It's just like a lot of it's little stuff like that. [SPEAKER_00] See, the way that I do this is either with Hermes or GrokBot, I share from the share sheet a link with a specific bot. And the bot, this is like the power user version of, I think open, from what I've seen so far, OpenAI seems to be going after like more muse than maybe GrokBot. [SPEAKER_01] Probably. I would say that's a fair way to look at it. [SPEAKER_00] But I have each bot with a role and each bot has skills so that I, when I share a link with my weekly, Mac Stories weekly bot, and I say, save this link, it knows that it's an interesting link of the interesting links section of the Mac Stories weekly newsletter. And it knows that those links need to be saved in reminders in a specific project, in a specific section of the project. Right. For Friday at a specific time. [SPEAKER_01] See? No, I get it. But what OpenAI's engineers would tell you is that you don't need that because the models are so good now that they're generalists. You don't need specialist agents anymore. [SPEAKER_00] That's what they- Yes, they will tell you that. [SPEAKER_01] And so far, I have found that I can get rid of that whole structure of having it in reminders or having it in Notion or having it somewhere saved. Instead, it just pops it up for me in the morning on Friday. [SPEAKER_00] I would lose my mind if I do that. I need to have my structure. I need to have my boards and my projects. No. Just what? Am I going to just dump my links and apps in a bucket and then magically the bucket is going to, you know, appear and reveal its contents on Friday morning? No. Yeah. [SPEAKER_01] Yeah, it works. So far, it's worked fine for me. But we'll see. You know, look, I'm experimenting with it. I want to see how far I can push it. Because to me, this is the solution to a lot of like bucket apps that have been out there in the sense that it's nice to throw things in a bucket, but then you have to tag them, put them in folders. [SPEAKER_00] No, that I agree. Obviously, I also have a bot whose entire job is to scan my Readwise reader every 30 minutes and tag articles. Like, so I'm the kind of user that has embraced these agents. And this is exactly what I was doing with OpenClaw 10 months ago. I was a big believer in cron jobs. I'm now a similar believer in routines and automations, which is exactly what I like about Hermes and Grokbot. [SPEAKER_00] And why, for example, I struggle. See, this is another thing for the kind of user I am. I struggle with the single agent products. I tried Muse and I completely understand why Muse is so successful right now. But for me, the kind of user I am and how I like to compartmentalize things, [SPEAKER_00] having everything in a single thread, it gets very messy with my brain. [SPEAKER_01] Okay, it does. And I would, you know, I don't want to sound like the OpenAI shill here, but... [SPEAKER_00] Yeah, you're a shill, John. Come on. But here's the thing. You were paid by OpenAI. Shame on you. [SPEAKER_01] I was. [SPEAKER_00] That is not true. It's a joke. Don't take it seriously. It is a joke. [SPEAKER_01] And I paid for my own ticket. I wasn't there as press either, okay? I went there as a developer, folks. Anyway, there's a new product feature called Space, too. And Space is a little bit like a Notion clone from OpenAI. And one thing, I had the same concern you did with a single thread. It's like, all right, I had this stuff that the dot pulled up, you know, maybe an hour ago. And now it's like buried. It's like way up the thread. I know. Right. [SPEAKER_01] And that's a perennial problem because maybe it was some research or maybe... [SPEAKER_00] What am I supposed to do? Like scroll through a whole transcript? [SPEAKER_01] No, right. You know what you do? Instead, you just tell your dot. You say, you know those links I found for you an hour ago? You know, those links you found for me an hour ago? Put them in a space. And the space is a document. And the document now has the useful information that you have the dot create for you. And then you can use the dot inside the space to kind of pull that along further. On top of that, Federico, it integrates with codex. [SPEAKER_01] So maybe you start exploring something and you're like, oh, you know what? I really want to build a script for this thing that I've been exploring. This is a much bigger project than I initially thought it was going to be. You just tell it to spin up a codex thread and keep an eye on it. And it goes and does the codex thread on its own, separate and apart from the dot conversation. And the dot will let you know when it's ready whether you need to give some input or something like that. There are ways to peel off what's happening in your thread and put it in other places [SPEAKER_01] where it makes more sense at some point in the conversation. And that's what I've kind of been doing. Like this morning, you know, they've got Amazon Prime big deal days coming up next week. And so I downloaded a CSV file with a million rows from Amazon. And I had my dot look at the CSV, look at my criteria for what I use to determine what a good deal is and what's appropriate for backstories. [SPEAKER_01] And it went through and it found a bunch of stuff for me. And then I had to put it in a table in a space where I could look at it and then ask further questions. [SPEAKER_00] I am a little bit skeptical, but I will say this. I don't think Meta is ever going to go after power users. I think rightfully so, Meta has identified a very lucrative space, which is the very simple [SPEAKER_00] concept, very simple idea that billions of people have no idea that AI can be so powerful and do things in addition to answering questions. And Meta has found a way for people. And I think the people trying Muse for the first time, I'll be like, wait, this is a completely different AI from ChatGPT. [SPEAKER_00] Most people still go to Gemini or ChatGPT to just ask questions. They have no idea that they can have a little virtual assistant, an acute one at that, do things on their behalf. And Meta has identified, recall months ago, John, when we made fun of the Fiji Simo previous OpenAI executive talk about the AI capability overhang? [SPEAKER_00] This Muse is a product that answers the capability overhang. Meta has understood that so many people have no idea that the same AI can be a completely different type of experience. I don't think Muse is ever going to fundamentally resonate to a high degree for somebody like [SPEAKER_00] me to the same level of Hermes. Right. I think OpenAI, on the surface, they have meta Muse level aspirations, but they're a bunch of nerds and they cannot help themselves. [SPEAKER_00] So I think OpenAI, this is OpenAI in a nutshell, they will try to make it a commercial product, a commercial product that is well understood by regular folks. [SPEAKER_01] Right. [SPEAKER_00] I guarantee that they will not resist the temptation of making it more capable for power users. Yeah. [SPEAKER_01] And look, it's a hard design problem, but I think that they, what I'm seeing, at least initially, is something that's designed to be, at least in these early days, user-friendly like Muse, but the hooks are there and the capabilities are there for it to become something more powerful. And how you have those things coexist is the whole, you know, that's the whole trick for OpenAI right now. [SPEAKER_01] That's the whole trick of merging ChatGPT and Codex together in the same app. I mean, it's addressing both of those markets at the same time, which is not easy. And it's not at all guaranteed that they'll be able to do it. But I've been pretty happy with the dots so far. [SPEAKER_00] Yeah. The dark horse in this race continues to be Anthropic. The moment Anthropic realizes they can make their little claw characters, the mascot. [SPEAKER_01] They have the best character of all too, right? I like the little guy. It's so cute. [SPEAKER_00] I like the little guy. The moment they realize they can make those little guys agents and sell it to people for a price, I think that they will be unstoppable. But maybe they don't want to enter this space. Anyway, we're going to be talking about some of my recent experiments from this week, John, while you were gone. So we're going to talk about that in App Stories Plus. For App Stories Plus subscribers, you can get App Stories Plus in two ways. [SPEAKER_00] You can just get the podcast. That's $5 a month or $50 a year. And you get every week an extra segment. And the whole episode is also ad-free in higher bitrate audio. And it comes out usually, most of the time, a day early. Or you can just get it as part of the Club Premier bundle. That includes everything from Club Back Stories. That's $12 a month or $120 a year. [SPEAKER_00] And that gives you App Stories Plus in addition to everything from the Club. So we're going to be talking about some of my CLI experiments, including a change of plans for what I wanted to do with my Apple Notes CLI. Oh. All right? So that being said, John, I'm just going to do the outro this week. [SPEAKER_01] I would like you to. I have to say to folks, I got a cold while I was gone. [SPEAKER_00] John got sick while I was gone. It was a whole thing. It was a whole thing. Anyway, look how professional I am, John. You can find us on MacStories.net. We'll be busy in October. We have a bunch of apps for iOS and iPadOS 27 to write about. Still, there's the iPhone Duo coming out. Maybe some HomeKit products. John and I will be busy. You can find us on every social media. I'm VTICCI. That's V-I-T-I-C-C-I. And John Voorhees. I don't want to spell John's name. You know John's name. It's easier than mine. [SPEAKER_00] We are on all the social medias. And we will be back with App Stories next week. Can you say goodbye, John? Goodbye, Federico. Arrivederci. Bye.