Building Websites and Coding with AI

Welcome to Mac Power Users. I'm David Sparks and joined as always by Mr. Stephen Robles. Hello Stephen, how are you today? I'm doing really well. Excited to talk about more workflow stuff. It's been changing just every day, it seems. like So much fast stuff is happening model-wise AI stuff so I'm excited to talk about it with you. Yeah, we've got an AI episode. We're not going to do it every week, but uh we thought occasionally it'd be useful to talk about what we're doing with AI and the kind of the current state of it. We're calling this episode useful AI because that's what we're looking for here. AI that's actually useful for you, what we're doing and how we're doing it. But I got to tell you, Steven, uh this morning was glorious. I woke up early as I do, came downstairs, and as the sun rose, so did my blinds for the very first time. The smart home coming in handy. That's very fun. Yeah, it is. We're gonna do a bonus episode on you helping me get my cause I've been uh investing in home kit . I'm just still gonna call it home kit, sorry. But the uh but I I've got some stuff I want, some automation help from from my uh my Apple Home uh pro buddy. So we're gonna talk about that on a bonus episode. So if you're a more power users episode, uh more power users subscriber, stay tuned. You get the ad-free extended version of every show plus some of these bonus episodes. That's going to be coming in a few weeks. Uh and uh I'm really enjoying the setup when it doesn't make me crazy. That's good. I know you still been working it out, but it's been solid for a number of days now. So you Yeah, no, it's good now. It's good. Okay. It's it's a journey. I had I had to reset my home because uh the twelve year old setup just needed a kick in the pants. But uh and we'll talk about that. And I will say smart shades, it is one of the more fun smart home devices uh you know scheduling those because it's one of those things where I don't know about you but for a long time I just wouldn't open shades just because you don't think about it the manual prospect of it but when you can automate that it's nice you get to see out your windows more. So yeah and and the one of them is behind my my studio desk. So getting to it to raise it and lift it's not trivial. So it's kind of nice now that like when the sun sets and it starts coming in. I just push a button or if I want to record and control a light in a room again, just it's real easy. Um so that's something that folks should be checking out. Uh but that's all we're here today. Uh we were going to talk today about uh uh AI. I guess we should also mention for the bonus segment of today's episode. Steven is, I think, going to try and talk me off the ledge. You know, we got the news this week that the uh the Mac studio is further delayed. Um th for those keeping score, I'm on the M2 studio. I've been kind of pining away for the M5 studio. And now I'm thinking, well do I just get a laptop? Steve and I are gonna talk. So that we're gonna talk about that in more power users today. Yeah, and it seems like I mean the supply chain issues and all of that, it's you know, a lot of people are like, Well buy now because who knows when it will change even SSD and hardware prices . I someone asked I posted a photo because I was editing a video on my patio and I mentioned I've edit everything off external SSDs. I use an OWC Envoy Pro FX. I actually have a four terabyte one. And that's where I put all my final cut stuff on so I can take it wherever. And they asked, like, can you send me a link? And the price was just astronomically different from when I had purchased it. I think I purchased it. It was five or six hundred dollars and now it's over a thousand. And I'm like, this it's a tough time to whether you should buy now, wait or if it's gonna be worth waiting. It's it's tough. I agree. I really don't know what to do. I'm gonna talk to my buddy Steven and maybe make a decision in more power users. So stick around for that. Yes, let's do it. Okay. Um, today, though, we are talking about artificial intelligence and where it stands. And I thought we would start, let's just take a survey of the landscape, right? Um of where AI is. And of course, I think it starts with the foundation models. Um uh Claude from Anthropic, ChatGPT from OpenI, and Gemini from Googles, those are the the big three . And um those are kind of leading the story here for server based AI. I mean, that is like one category of this. And when most people think of AI, that's what they're thinking of. Yeah, and I think local AIs are becoming more popular, but I think for the mass market, I don't know about you, but I don't particularly have local models running on any of my devices just yet. You know, I've don't I would like one of the major players to get into that game before I start experimenting with that. Even Open Claw, I tried it for like five minutes, but I just wasn't about that. And so when it comes to local models, I don't know if we also want to include Apple Intelligence. That is some of the local stuff that you can do on your Apple devices right now without installing like a third-party LLM, but it's it's nowhere near the capabilities of these other foundation models, like you were saying. Yeah, and I have deeper thoughts on this. I'm gonna talk about that later in the episode about kind of where Apple's play is. I think that they're trying a different game here. But you know, in terms of the survey, this is where a lot of the action is happening. But I feel like something that's interesting to me is the performance is becoming less of the distinct distinguishing factor. You know, for the longest time, every time a new model dropped, it was how does it benchmark and you know which one's the most powerful and that's one I should use. But now these people are uh these companies are distinguishing themselves on features more than performance. Like uh Gemini did really great with a nano banano model for graphics presentation. And then now they've just done that over at OpenAI with their most recent release. They've got even better graphics, you know, generation . Uh, cloud and anthropic are really focused on the harness, being able to tie your AI to other apps. That's the whole reason I made the robot field guide, is because uh Anthropic made a way for me to actually turn AI Um so you know it seems like one of the things in this point in time is we're starting to see um the frontier models spread out on more than just performance. Yeah, and it's interesting to me how fickle people are too about how they perform, because it's so easy to switch. You know, most of these AIs, you're just interacting with an app on your iPhone, maybe on your Mac, and most people will just go to whatever has the coolest new feature or might be the best. And I actually see a lot of TikToks and Instagram reels of people saying, like, oh, if you use ChatGP T in twenty twenty six, what are you even doing? Like you should be using Claude. I just saw a reel yesterday where a guy, young guy, he looks like maybe he's in his early 30s, he was like, actually, I don't even use ChatGPT to do my work anymore. I ask ChatGPT for prompts that then I give to Claude because Claude right now is far and above giving me better responses. And so it's interesting now their competition is so fierce, I think, because people will so quickly switch uh between all of them. And I even find myself doing that just today. I actually switched one of my workflows from ChatGPT to Claude, and we'll talk more about our use cases later, just because I started getting better responses, and so it's super easy to switch. Yeah, and I do think Claude is is getting a lock-in on us uh if you're paying attention because because of the harness, frankly. I mean, I built this course on it, so of course that's something I'm spending a lot of time with. But the skills get better, the connections to your apps get stronger. Um, like they have this great feature called scheduled tasks. You and I were just talking about before we start recording and scheduled tasks run it's kind of like open claw but safer and not quite as powerful. But you know the the delta between safety and power is is totally acceptable. And like, okay, so now you've got these scheduled tasks. Are you going to really move all those over to Gemini or OpenAI, you know, on the drop of a hat? Maybe not. That's that's a bigger deal. So, so they are, it is mobile, but also I think they're all working on trying to find ways to become stickier. And I d I do want to mention too, there is a unreasonable or difficult to parse hype when new stuff comes out in regards to AI. For instance, I know Codex uh we had a big update recently with computer use, which is OpenAI's kind of co-work style app. And so I downloaded it on multiple computers, tried doing some of the workflows I was doing with cowork in code x and maybe it's great for developing. That's an area where I'm not doing a lot of work personally. But I did not find it that good. And so I went on YouTube to search like, all right, well what what can you use codecs for outside of development work, that's actually so great because everyone seems to be really espousing how wonderful it is. And when I watch a couple of these videos, I'm like the hype in the titles and the thumbnails and even in the news articles, it's just way above what these things can sometimes actually do, unless you really know like what you're doing. Uh, but also like anthropic with the mythos. You know, the anthropic recently said we developed something so powerful we can't even give it to people. And there's a bit of overhype about that that AI companies are using to appear more valuable and to appear like they're on the cutting edge where it's hard to parse. Like, is this actually that good or is this just kind of hype language? Yeah. And I think that is the conversation on AI writ large, right? There's there's the hype cycle and there's the doom cycle, right? And there's very few people talking about, well, what is this stuff actually useful for me today . And like I feel like the things anthropics are doing are are quite useful, but they're not trillion dollar useful. Like you know what I mean? And and that's what they need to keep feeding the machine, building the ser vers and satisfying the stockholders. So they're gonna hype, you know, that it does so much more. But but I would what my advice to people listening is, no, there this does there is useful stuff here and that's not going away and you should figure that out. But don't get caught in the doom or the hype. You know, I don't think that the an LLM-based AI is going to be the death of our species. I also don't think that next week it's going to cure cancer . And but there is something something useful in between those two extremes. Um but yeah, so I think the foundation models, of course, are important. Uh, but there is derivative technologies that are also quite useful now. And the first one that really stands out for me is AI dictation. And there's so many uses for making a transcript or dictating these kinds of things. I mean, even how I edit the video version of this show, having the transcript and being able to search that makes editing process much quicker. And I've like you've been trying to get in the habit more of dictating things to my phone, sometimes even my watch , and it is much faster the voice to text as opposed to typing sometimes. And so I'm trying to get in the habit of that and I know you have shared many apps that have made that easier on the different devices that sync across. And so it is really useful. I'm trying to get in the habit of it. Yeah, I so I've been talking about this for a long time because as a baby lawyer, I started doing it. But I really think that we're at this point now where AI has made this a solved problem. And there's so many great tools out there. The one I want to call out, and I I know I talked on the the watch show about Whisper Memos , but the one that I really recommend for using on your Mac if you're going to uh just have easy dictation anywhere is Whisperflow. It's another subscription, but it is the absolute Goldilocks app of this series. There's some that are way more powerful and more you know tedious and fiddly and there's some that are less powerful and probably free, the local models or whatever. But whisper flow just crushes it. You can have a custom, you know, dictionary. It you know I,'ve got like two hundred thousand words dictated into it. And I spent a bunch of time testing them all for the labs and Whisperflow was the clear winner. It kind of reminds me of the old days when people uh would find text expander and it would change their life. Now it's whisper flow, right? Because you suddenly do it. And I uh I just really like that one for it. I'm a subscriber. Does it have an Apple Watch app? Do you know? It I may, I don't know. I haven't even really gone down that it does have an a phone app that works fine. Although with a recent change Apple made, you have to swipe into it. Every time you click it, it goes to Whisperflow enables. You have to swipe back to it. But it it brings all your your dictionary over your custom vocabulary and like if you talk in bullet points it makes bullet points so it applies just enough AI sauce to make it a little better but not so much that it changes your word s. And I use it on the phone. I use it on the Mac. I I use Whisperflow like constantly. And uh if you're looking for an easy way to get to that point, just sign up for a couple months and and you'll be sol d. Well, I'm gonna talk about some shortcuts later because I use a lot of dictation directly in shortcuts using Apple's transcription kind of built-in action. But one other one I want to shout out, which is something that you turned me on to, which is the whisper Memos app. Mostly because I this has an Apple Watch app and it's actually really good. I show this in a video and you know being able to just tap on your watch, talk to it, and that note instantly sync across all your other Apple devices is really cool. And I've had a lot of people give me some positive feedback once they try this one. So yeah. No, I I s I subscribe to that one. Those are the two I use. And Whisper Memos for me is like it's on the the button, you know, on my watch. You know, and so I just literally push it anywhere and say whatever like I'm thinking, oh, I need to email Steven about that thing. Push the button, dictate the email, then I get back and it's sitting in drafts waiting for me and I can send you the email. It's just yeah it's just so frictionless. But for but they don't but Whisper Memos doesn't really drive on your Mac. So I I think you know like a a stream of consciousness consciousness diction is also very useful. And if you want useful AI, that this that's probably the one button you can press right now and be very happy. 'Cause like I said, it's almost a solved problem now. You just get a good app and start talking. That's really good. And another area where I've moved a lot of my work or uses to AI is search, namely web search. And you know obviously I was Google by default for a long time . The AI overviews that Google has done now for the last several years actually I think has discredited some of the Google search reliability. There's been AI overviews that have just been inaccurate or wrong or not the information that I'm looking for, plus the amount of sponsored links and things like that. I moved a lot of it to AI apps. I use Gemini uh for some general website search. Um perplexity I, used for a little while, and perplexity is a little different. You know, we talked about foundation models earlier. Perplexity uses all the models. It's basically like a model orchestrator where it might pull on Gemini, it might pull on Anthropic, it might pull on ChatGP T, but I've moved a lot of my web search to either Gemini or Claude, uh, believe it or not, because I don't like super verbose answers. Like if I'm trying to ask, like, oh, what was that movie so and so-and-so was in you know I could go to the call sheet app but a lot of times if I just ask Claude sometimes Gemini I just get the answer I want faster with a little bit of context that's nice and then that's it. Yeah, you know, so it's interesting for me because I I also was a huge perplexity fan in the early days of perplexity. And then they were doing some weird things with the way they were collecting data, and I got kind of turned off of them, and then I went back and looked at it again and I just didn't find it all that useful for me. Um, the way I do web search is largely KAGI, uh C K-A-G-I, which is kind of like a clean search that you pay for. You know, it's kind of you pay extra , you pay for it and you don't get a bunch of garbage. Right. And I still largely do search that way. When I do search through AI, it's usually in addition to something else I'm working on. Like if someone writes me an email with a question, I don't know the answer to it, and I'm working on a draft, I'll say, Well, this is my draft answer. Please check it for for veracity. You know, like, you know, I don't want to give someone bad advice and it'll do a search in the background and figure out if I'm full of it or not, you know. Yeah. And but I just don't really I don't do you know, a lot of people say AI for them is just Google Plus. I don't really do that too often. Um well and I was speaking of search like when you're talking about research , if I'm trying to aggregate either a lot of sources for something or multiple links for something, that's where I'll use AI as well. So when I made the video about John Turnus, once it was announced he was going to become CEO, I did go into Claude and I asked for just a bunch of research about him information. And I don't look at the information Claude gives me as like the final word, but I'll use the references it gives me, which it found me like his commencement address that he gave at the University of Pennsylvania. I didn't know he had done that. I didn't know to search for that, but because I used an AI tool, it provided that context and it knew I wanted it. And then the other thing I did for that particular job was I wanted to pull a bunch of clips of John Turnus in Apple keynotes to use his B-roll in that video. So I asked Claude , get me the YouTube links for every Apple keynote where John Turnus appears. And that's something that you can ask Google, you can try to search in YouTube for just like John Turnus keynotes. And you might get stuff, but I specifically asked Claude for the Apple YouTube links to their live events and it just gave me a list of links. I used apps that we've talked about in past episodes to grab all that material and then I was able to quickly use it in a video. And so when I'm trying to aggregate a bunch of information or find multiple original sources, I find AI is useful for that. Yeah. Do you now do you we should have asked earlier, are you paying for any frontier models? What are you paying for? Yeah, I pay for them all, David. Uh I pay for Chat GPT Plus. I pay for Gemini AI Plus, which I pay for like Google, like the one terabyte Google Drive tier anyway. And then I pay for the tier above that for the AI stuff. I think it's like twenty or thirty dollars a month. And I do pay the hundred dollars a month for Claude because I do find that to be I'm using a lot of cowork as you know that uses a lot of tokens and it's one that I go to most often so I'm I'm paying for a lot of them. Yeah me too. I'm at the entry price for chat GPT and I get Gemini because I have a Google Work account. Oh. And um I just I don't pay extra for that, but they're getting plenty of money from me. And the um uh with uh with Anthropic, I actually just upgraded to the $200 a month plan. And that's just because I'm doing so much with it right now. And I'm doing two workshops a week with the AI thing for the robot thing. And what I don't want to do is have, you know, 300 people on a call and have me run out of tokens. So I'm you know, so that's probably a temporary thing. I'll probably I think I'm about a hundred dollar a month guy uh when it's done. But even the you start thinking about this, you know, couple grand a year that you're spending on this stuff. Uh and and that's that really leans in something I want to talk about later about with the future of this. I mean, is this permanently a a a server thing or is this a local thing? You know, and um and I don't I don't really know I don't think anybody really knows the answer to that right now. But but yeah, so just to I think you and I both have skin in the game. We're trying to keep up with it so we can teach people about it. And you gotta frankly, you just gotta pay them all. Yeah, that's true. Another thing that I've used actually a lot is AI browsers. And you know, this can be a difficult topic because, you know, if you want to use a browser and have all of your logins and payment information autofilling, well then you have to trust the company that you're giving it to. And I was using Comet Prop from Perplexity for a while. It was kind of the first AI browser that came out that was really useful. Chat GPT Atlas is now out there. But I never gave either of those browsers like all my information. Uh, they can you can use the iCloud keychain extension to actually have your logins and passwords, but not save them in the browser. So I did that. But now you can have Claude control your browser in any Chromium browser using the Claude extension. And so now that's what I've moved to because it can do all the things that the other AI browsers can, but as an extension and I just trust it inherently more uh than the others. There's a lot of useful things you can do, like even with YouTube videos like I mentioned before, I could say go to a YouTube video of an Apple keynote where Steve Jobs talks about the intersection of technology and liberal arts. And it will literally load a YouTube video, scrub to that timestamp in the video, and I'm watching it in a few seconds. And that's again a benefit of a browser that can actually take action for you as opposed to asking a chat bot, it telling you the timestamp, maybe giving you the linked timestamp, you can kind of cut down on some of those steps. Yeah. I find uh the one that is most appealing to me is Dia, D I A. Yeah. And uh I think that's a nice one. But it it's like a utility that I don't find myself using as much because of my integration with Cloud Cowork and its integration with the with Chrome browser. It it kind of There's exactly one task that I still use Perplexity Comet for, and that's promo codes on websites because I still find that it will try the hardest and the longest. And so if I want to buy anything online, I will add it to my cart in the perplexity comment browser and say , try promo codes until you find one that works. And that's where taking action in the browser, it can literally put it in the promo code field, hit apply, see whether it was accepted or rejected. And if rejected, keep tracking until it finds one. And I still have the best luck with Comet uh of all the AI browsers, which is also now an iPhone. You can literally get the AI browser on your iPhone. Yeah. And really I think that conversation with AI browsers leans into the other kind of developing area of AI right now, which is agentic AI. And that's what we're talking about with the harness and these browsers. It's AI getting outside of the chat and actually doing work. for you And I think AI browsers were one of the very first kind of implementations of that. But now they have the model context protocol, the MCP, which anthropic uh designed but opened up, and a lot of companies are adopting it. The rumor is Apple's going to adopt it in June, which is the harness that allows you to say, okay, chat, I want you to go into numbers and make a spreadsheet or go into Google Sheets or you know go, into Fantastical and create an event or do all these great things. And uh, I'm hearing from a lot of my developer friends that are if not curious about it, they're actively developing them. This is uh this is a thing, man . It's happening. And this is something where I've been a little leery of computer control specifically. Uh this was a big feature of ChatGPT Codex. Uh Federico Fittia Mac Story says it is the best at computer control, uh, even better than Anthropics computer control, which is literally clicking around on your screen, typing things in anything on your computer. And this is a reason why I have Mac Minis, I say plural because I had one Mac Mini and now I have two. Where if I want to try computer control, I have a Mac Mini not signed into my iCloud, not saving, it doesn't have any passwords or payment info, and I'm experimenting with computer control on that because I still don't want it clicking around on my main machine where all my stuff is and like it could maybe buy something without me knowing. Yeah, Anthropi calls it computer use, but it's effectively the same thing, mouse and keyboard control . Um, I teach it in my robot course, but I do not actively use it. And I one of the things I strongly recommend is if you use it, you are there watching it when it is turned on. Like don't walk away from your computer with computer control turned on. I think now you're getting into the more of the open claw style risk. So like I I haven't turned it on in probably three weeks. Like there was something I wanted to demonstrate to somebody, so I used it. But I find that the agentic AI with MCP and more traditional control, uh uh most of us can get anything we want done without having to resort to giving it, you know, the mouse and keyboard control . But it's there if you need it. And um and for some people that's really useful. But the overall agent thing is just we're just on the cusp of it. It's just really starting to explode. You know, almost all the apps that we like and love are getting MCPs. If they don't get them, people are making MCPs for them. Um and and this is really changing the landscape of artificial intelligence because it's gone from a thing where you have everything contained in a single chat window to something that's actually actionable does work for you. I mean, we talked about this in the robot field guide episode, but um it's shocking how much stuff it does for me. I got an email last night. Um somebody said, Hey, I want to reset my course. I want to start over. And that's actually not trivial with my vendor, but now uh the robot does it. You know, it it uses the Chrome plugin . It goes to the finds the user, finds the purchase, finds the specific thing, clicks all the necessary buttons to reset them, and it's done. And so I don't do that anymore. That is agentic AI in the way that I like to use it. You know, I'm I've always said this stuff doesn't need to replace us as humans, it just needs to do donkey work.ork W well. We're there, and it does it now. This episode of the Mac Power Users is brought to you by Mercury Weather. Go to Mercury Weather dot appslash MPU for forecasts beautifully done. Download now for free. Mercury Weather is a thoughtfully designed weather app that shows all the essential weather details at a glance. It has a gorgeous, beautiful interface that dynamically adapts to the conditions, with a warm orange palette on a sunny day, icy tones on a cold day, or deep blue on a rainy night. Mercury uses a glanceable chart layout to present an hourly and daily forecast in a way that feels intuitive right away. And this is a really cool feature for frequent travelers. Mercury's trip forecast feature automatically shows the weather at your destination, right in your daily forecast timeline. So you always see the weather for where you'll be, not just where you are. What I appreciate most about Mercury weather is the simplicity. It gives you just what you need without a lot of extra noise and their widget is amazing. With it you can instantly grok the weather and the forecast, just no trouble. And when the weather gets serious, Mercury offers storm and hurricane tracking. With maps , live positions, forecast paths, cones, and intensity, plus widgets so you can keep tabs on a specific storm or the closest one right from your home screen. Mercury Weather's gorgeous interface makes it a delight to check the weather every day, even on gray and rainy ones. The app's business model is simple, no ads and no selling of user data. Mercury is available on iPhone, iPad, Apple Watch, and Mac . You can download it and use the standard features for free and upgrade to Mercury Premium to unlock all features. Go to MercuryWeather.app slash MPU to download Mercury Weather now. Use that link to let them know you came from the Mac Power users. That's MercuryWeather.app slash MPU. Try it out and get all the standard features for free. Our thanks to Mercury Weather for their support of the Mac Power Users and Olive Relay so when it comes to creative workflows, I'm still again hesitant to introduce it in some of the important tasks, and I view those as the ideas I come up with, the brainstorming, how I'm going to present something. But there's lots of great apps and tools to help me with the later parts of the process and Gling.ai, which we've talked about, I think you started using it as well, where it's using both the dictation transcription part of AI. So I give it my A role, it transcribes everything I said, but then it also has the intelligence to eliminate silences and bad takes. So recognizing when I start and stop multiple times and then just keeping the last one. And it does really well at that. And so that I'll that put every video that I make through Gling first. And then sometimes I will use AI tools to help me with a thumbnail. I don't do it often. It's usually with my shortcuts thumbnails because I've you can check my playlist. I've literally made 71 videos just about shortcuts on my channel and there's only so many thumbnails I can think of to having me and shortcuts floating around me. Like I just don't I don't know what else to do. And so I It's like you just dream of shortcuts every time I see this. I just imagine you. Yeah, it's kind of like that. Like a cartoon. If someone hit you on the head instead of stars, you would see little shortcuts flipping. Like wily coyote with the cloud, yeah. So so I to So I'm right now in this mode of like, all right, well, I'm gonna make every shortcut's thumbnail, either a marvelo superher or an X-Men or something like that. And so I have historically given a thumbnail image, so I'll take an image of myself, I'll give a bunch of shortcuts like cut out as cards to the AI, and I'll say make a thumbnail in the style of Storm from the X-Men or Gambit or Professor X. And I will have done, I'll do this across multiple models to see which gives me the best. And Gemini's Nano Banana was ahead of the game until recently. It was the best at not changing the text on the shortcuts. A lot of times I would give it these shortcut cards, you know, and it would garble the text and it would change it and it would look like nonsense. Yeah. And so Jim and I was the best at keeping me photorealistic and then just adding some elements. Uh but now just with ChatGPT images two point oh, it's ChatGPT has actually gotten a little better than Google's nano banana. And this is where like that switching cost comes in where I have no allegiance to any of these. I don't care what which is the best. I'll just use whichever one is the best. And so I do that. And then I transcribe my video to upload for like subtitles and things like that. Yeah. So uh and you didn't say earlier when you asked Claude to give you all the John Turnus videos, how did it do? It did extremely well. Um, it actually and one of the things I wanted to find was the oldest or first instance where John Turnis spoke in a keynote, and it found me. I think it was a keynote from 2016 or 201 7, and then all the events after that. I think there was one keynote where I don't think he was in, or he must have been in it very briefly because I scrubbed through and I never saw him, but it gave me 10 to 12 keynotes all, the way from the Apple Silicon transition to recent keynotes, the Mac Pro, the last M2 Mac Pro that he announced, gave me all the links. And it expedited that process a lot. Very fast. So just a little follow-up. I like you, I am very hesitant to use it for anything in the creative process. That's the part I enjoy. Uh but I've been making my thumbnails by hand with Pixel Mater for ages and listening to you makes me want to try to do something to make them better. The other thing is Gleanai , this is a really niche product. Most people listening don't need to do this, but for people who do record, even I use it for audio. Um, like every Friday we do a a podcast in the Mac Sparky Labs, which is kind of like the new summary. I try to keep it about 10 minutes, and this is what happened this week. But I record that, and of course occasionally I have to do a second take, and I would go in and manually edit that out. You know, take out the bad takes. And the first time I tried Gling, what it does is it transcribes it, it reads the words and says, Oh, I can see he's he read it twice. I'll delete the first one. And once you figure out the settings in it it,'s actually pretty great. It is pretty great. And I I'm gonna put a link uh to my Mastodon post, but I do want to show you if you're watching this. So as an example, this is one of the images that I'll give the AIs. And so this is me with a bunch of shortcuts floating around me. And I will in my Mastodon post, I show you the prompt that I use and I also attached some like comic book images of Gambit just to kind of give the AI a starting point. And this is what Jim and I did first. And you can see like things start looking weird, text is weird, like icons on the shortcuts get a little messed up. You know, this message is bubbles is a little messed up, and I look AI generated. Like I'm a little bit more weird. I look weird, I'm a little too smooth. But then ChatGPT Images 2.0 did this version, which is markedly better. And you can see all the shortcuts cards, the icons are good. They're untouched. All of the text is accurate. It didn't make up any words. And it looks very much closer to the actual photo of me, just with a little glow around me rather than this weird AI version of me. Maybe there's future Max Sparky thumbnails where I'm a Jedi. You should totally do it. Yes. If if if if for nobody else for myself. Exactly. Yeah, exactly. Exactly. And internally. And I try to be careful because people are sensitive. Uh, you know, when I share on social media, people are like this is AI slop and this is whatever. And it's like I try to ride the line very carefully. I don't generate things out of whole cloth. I try not to go too far in a direction where it's like I obviously couldn't have made this. Like I'm not a graphic designer. And that gambit gambitification of that thumbnail maybe I could have done with like hours of work. And so it's something that's within reach, but man, that was done in like 10 seconds and it saves me a ton of time. And that's a real tough thing as a creator. Like when I did the productivity field guide workshop series last year, I made a whole series of slides using AI help that were like national parks themed and kind of like camping theme. And that was kind of of the idea you know you being on this journey, this this trail hike as you figure out what's important to you. And I got a bunch of email from people saying those slides were awesome. They really helped me understand as you were talking through it. But I got a few emails from people that were just outraged by them because that it was clearly not my skill set. And I can't afford to hire an artist to create a national park series of a hundred slides. And they're like, why would you use AI? You know, and they really offended a and it the number of people that liked it greatly outnumbered the people that were offended by it. But at the same time, I don't want to offend everybody anybody, you know. So it's um as a creator, it it is really a challenge right now to figure this stuff out. And one use case we actually forgot to mention, but I do want to uh to point it out here is the gener ative version of the voice. And I've uh used Eleven Labs for a long time for some use cases. And I want to point out actually, Jason Snell announced on Six Colors that he's making a podcast version of the newsletter, which I believe is an AI version, namely it's AI reading it. And Casey Newton at Platformer also does a similar thing where he has an audio version. You can listen to his newsletter, but it's not him recording into a microphone, it's AI generated. And I actually trained a professional voice with Eleven Labs by giving it three or maybe thirty hours of my voice, which I've recorded podcasts for years, so that's not a problem. I could just give it that. And 11 Labs trained a professional voice. And I will include a link to an audio clip because uh in the in the show notes, it does sound a lot like me. If you listen to it, you know, it's like 90%. There's some like weird phrases and pacing where you could tell, like, oh wait, this is AI. But for my primary tech daily podcast, which is like a member uh benefit, sometimes I just can't get to a microphone. And I want to still be able to do it. And so when I release an episode that was generated with 11 labs, I say it at the very top of the episode, this episode was generated by 11 labs. And I make sure to disclose it. And you know, is that worse than not doing it? Is it better than not? You know, I I don't it's it's still hard. Like we're in a place we're still navigating it, but it's been a balance where I'm just doing it to a smaller group, making it clear that it's generated and trying to still make it the best possible by having it trained on my voice. Yeah. That is something that repels me though, the idea of my voice, uh giving it over to AI. And like I think about that too, because we go on vacation for a week every year, and I mentioned the lab report. That's the only thing I do that's newsworthy. Everything else, like I can record Mac Power Users ahead of time, I can record labs videos ahead of time. But when I do this lab report, it's news-based, so I have to record it a day or two before it goes out. So every year I bring a microphone and during one part of a vacation, I I hole away in the room and I record it and I edit it. You know, the whole thing takes me an hour or two to to kind of get together. Um, it would be faster if I just fed it the script and let uh an AI generate, but I just can't imagine myself doing that. And uh but you know, uh this is the world we live in. We're all trying to figure out our way around this stuff and and that's one I've not explored at all and and just not that interested in. But I guess I don't like the idea of my voice being generated out there outside of my control, which is a risk, you know, when you train at that, but being a YouTube creator, being a podcaster, anybody could literally No, you're right. I mean and you and I have so much public footage, someone could make a very convincing version of us. Yeah. And um, you know, it's like that's one of the, you know, there's downsides of all this AI stuff, and that's certainly one of them. You've heard these stories about people who uh commit scams by like reproducing a loved one's voice and calling you and asking for money or saying they're in distress or something that's can be very believable, especially to a senior citizen that doesn't really know what's going on here. Um so uh yeah it's just really scary. And I have been deep faked, to be clear. I don't know if I mentioned it on this show, but there's deep fakes of me on TikTok where people have made videos of me saying things I never said, and it's it's not super bad. Like I'm not saying anything inappropriate, but it's not me, and it's not things that I would make. And so it is it's a risk now that we have to to deal with. So all right. Yeah. Uh but in this survey, this is that's taking us a while, but it's good. I think we're getting a lot of good info out. Uh we we're finally getting to Apple intelligence. That's right. And listen, Apple intelligence gets a lot of fleck and and a lot of it rightly so uh as far as like being behind or whatever. But there are a lot of benefits to Apple Inellitgence one being built into your operating system privacy and security first and local there's a lot of local stuff you can do with Apple Intelligence just in different apps and also through shortcuts and you have the writing tools and notification summaries. I think those things are not as useful kind of peripheral. But when it comes to transcription, when it comes to summarizing, and I actually have a video ready I'm going to be publishing. It might be live by the time this releases. But several AI Apple Intelligence features that people might not know of. Simple things like you can auto-categorize reminders lists with Apple Intelligence. Press a button and it will move it all into sections and that list will auto-categorize ongoing and you can always go back into that. You can now make Apple Music playlists, uh as and just give it a prompt, base it on artists, give it genres. So there are some some beneficial things you can do there. Yeah, I think the thing with Apple, so uh you know, with the John Turnus hire and just the the general state of AI, I think they really see themselves for one reason or another, they did not pursue the frontier miles. And maybe that was a mistake. Maybe it was deliberate. I'm not sure. You know, I would kind of think it was they got caught with their pants down a little bit, but you know, I'm not sure. But at the same time, when you look at the business model of these frontier models, we talked earlier how they have to hype it up because of the stock price and to keep up with these trillions of dollars that are flowing, but they're not making that much money. And I really have a question about the economic viability of these, you know, frontier server-based AI models. Um in the future. Like at some point, investors are going to say, Well, what am I getting for all this money we're putting into this? And they're either going to have to really raise prices to make it profitable, or they're not gonna be profitable. You know what I mean? Or or I could see them turning these server farms into like very high-end AI that you can rent that companies get for thousands of dollars an hour or something like that. But you know, this consumer model, I could see it breaking, frankly, because they're not, they're losing money. I think Sam Altman said on the $100 a month uh chat GPT, they're losing money on every one of those because the cost of feeding the tokens to the people for the hundred bucks. And I I suspect you and I are getting the same deal from Anthropic. I I don't think what we're paying, we're getting more value than we're paying for for in terms of their cost to do the token generation. And I could see a real future where this splits and the big high-end server stuff is really expensive, very customized, you know, high-end developers, companies, etc., buy that. And the rest of us use local stuff. And, you know, one of the things I've kind of learned in my journey with AI in the last six months is the stuff I wanted to do is not super hard. Like the stuff the donkey work stuff I talk about is is you know, name files and move this and go to the website and click that button. It does not need a PhD level AI to do that, right? Right. It's like the uh what was it in that um in the uh Douglas Adams book, the robot that was like a genius, but he was a butler and he was constantly depressed because like he wasn't being used to his capacity. Yeah. Marvin, I think his name was. You know, um that's what that's the way I'm treating a lot of us are. We're getting these really high-end models, we're giving it donkey work. It doesn't need, you know. So so my point would be maybe Apple's play here in the long term is look, we're gonna really kill it with the hardware and we're gonna make good AI, you know, chips that can run proper models local. The M5 is exhibit A here. It really is a great chip for running local AI. Um just imagine what it's going to be like at M7. And and then if Apple can get the harness bit that Anthropic did into the software stack, like using MCP would be really smart here. Um, suddenly people are able to run their donkey work style AI off of a Mac. And suddenly Mac be just like everybody's buying Mac minis out, just you just buy a Mac and then you don't have a subscription to a big um you know company or a frontier model. You're getting enough AI that for your buck. You pay for it once with the hardware. You don't have a monthly subscription. I think there's a play there for Apple Intelligence overall. I'm I'm gonna be really curious to see what happens at WWDC. I know Apple's very conservative the last decade. Um are they going to see that opportunity? Are they gonna jump on it? But I feel like there's really something there. I agree. And I was actually just gonna say and and you said it first, the local model play where companies that might be using clawed code and their developer teams are gonna pay a lot of money because it's going to save a lot of time. Unfortunately, might obfuscate some jobs because Anthropic will be doing the coding as opposed to real people. And so there'll be that bifurcation of the super high end where people pay a lot of money for it or companies specifically and then the low end where there is a race to just the fastest model and one that can run locally and it's funny Apple has benefited from this AI revolution or whatever you want to call it, because they make the hardware that everybody wants. You know, all the Mac minis sold out, the Mac Studios, the chips, and they've benefited either way. But I think when Gemini comes underneath Apple Intelligence, if Apple can not only deliver on the features they talked about two years ago when Apple intelligence was first introduced, but push forward into that, you know, the sky, I forgot what that app was called, which was now purchased by OpenAI, but like the Sky app on the Mac, uh Federico Fetici again did a feature that on Mac stories. Yeah, yeah. The one the uh Sky AI, I think it was called. But one that can you know, do control your computer, have that constant memory, but it be local , private, and secure, I think people are going to just be like, oh, this is built into the computer, done. You know what I mean? Don't have to pay a subscription. It's over. So many people going through my course right now are like lawyers, doctors, people who have private data who just you can't put it, even if it's on private cloud company, you can't say I'm sending your data to some server somewhere. You just can't do that. And there really isn't a product out there that makes it easy for people to do that. And I know there's a lot of open source stuff that's trying to do the anthropic harness using local stuff. But really, you need someone like Apple to just step in and say, okay, we're going to use the MCP harness , it's all gonna tap happen on your silicon, it's not gonna go to a server, and you're gonna get all the benefits of that stuff without the monthly fee, which is great. People don't wanna pay the monthly fee, but it's also gonna be private and you know it's going to be better for the environment because you're not like burning through all these servers and uh I just feel like man it's such an obvious thing that could happen and they're so well placed. The dream if I could command space for spotlight and say where did I see that post about this and that? Yeah. And because the LLM is just on my machine and it just knows everything that's been on the screen forever, to be able to pull that up in a moment, like that's my dream, is like to not have to capture and think about every single thing in the moment because I don't know if I'll be able to get to it later, but if just to know that like that constant memory is being held, I've access to it whenever, and then ask it to help me synthesize that information either into research, make an Apple note and give me, you know, a outline of something, that's the dream for me. Yeah, and Apple has always sold themselves as bringing the power to normal people. You don't need to be a programmer to use our stuff. I mean, this is the perfect play for them. And if they did it, I think they would just like own the hardware stack for the next ten years. I think you know if they got ahead of this and really did that, who would buy anything other than a Mac ? This episode of Mac Power Users is brought to you by Ecamm. If you're a Mac user who creates video or podcasts, and especially if you live stream, you need ecamm. Ecamm Live is the all-in-one studio built exclusively for Mac, so it looks feels,, and performs like a native Pro app should. And if you've ever used live streaming apps and tools in the past, you know they can feel a little weird. Some of them are just web apps. Well that's not Ecamm. Whether you're live streaming, recording a podcast, or producing training videos, Ecamm gives you broadcast level control with drag and drop simplicity. Switch cameras, share your screen, cue overlays, and control audio, all without ever leaving your Mac. I've used Ecamm for years personally, and if I'm ever going to do a live stream to my YouTube channel or to multiple destinations, Ecamm can do that too. I do it with Ecamm. You can even schedule an event in YouTube for your YouTube channel, and Ecamm will see that scheduled event, and you can start your live st ream going directly to it. With Ecamm, you can brand your show with titles, graphics, and lower thirds. You can also pull in guests via interview mode or record multi-track audio for perfect post-production. And if you're into automation, which if you're a listener of Mac Power users, you probably are. Ecamm works beautifully with tools and apps like Stream Deck and LoopAck, as well as many Mac tools you already know and love. You can upgrade to Pro and unlock Ecamm for Zoom, letting you feed your polished setup straight into Zoom meetings or webinars, share Zoom comments on screen, and even capture each participant's audio and video separately for easy post-production work. So don't wait any longer, go and check it out now. And to get 15% off, go to ecam.com slash MacPowerUsers, all spelled out, and use promo code MPU15. That's 15% off at ecam.com slash MACPowerusers with code MPU 15. Or thanks to ECAM for supporting this show and all of Relay. Okay, Stephen. So we've been talking about it and hinting at some of the stuff we're doing. Let's get into it. Um,' lets talk about the stuff we're actually using this to be useful AI. You want to go first? Yeah, we'll go first. And I'll actually do a brand new use case, which is Fastmail, which Fastmail just released a M CP server that's first party. They're supporting it. And so I've finally been able to connect my Fastmail to cloud co-work securely, safely, and schedule tasks is a thing in c in cloud cowork that you can do. And that's really, I think, where a lot of the magic happens because you can tell it do this thing on the schedule, and it's just going to do it. And so what I've done is I get a lot of emails from potential sponsors , yeah, and that's mixed in with a bunch of other junk mail and newsletters. And I have Sanebox helping me filter my inbox, which is great. So my inbox doesn't get overloaded, but then I have a couple folders that is just full of a ton of stuff that I don't know what it is. And so I've given Claude Cow ork access to my email, my fastmail account, and I've told it, look in these two folders, one is newsletters, one is my Sane later folder, which SaneBox puts emails that don't look high priority in that folder. I say, look at these folders once a day. Anything that looks like a potential sponsor, I want you to draft an email saying this and just put it in my drafts folder. And so you can give access to Cloud Cowork to read and even draft but not send an email. And this is again what I really like about Cloud Cowork is that granular control of the different connectors . And so just to show you as an example, I'll look at uh my connectors here and you'll see all the things that you can give it access to. And so this is just the Cloud Mac app in the cowork section in the under the customize. And you can see here I've connected it to my my bit.ly account. I've connected it to Notion, Slack, and now even Circle, which is my membership community platform and fast mail. And if you go under fast mail, you can actually say, well, for you know, reading emails, I want to always allow that. But for and reading lists, reading threads, but when it comes to writing emails I want you to have to need approval if you're going to delete an email I'm going to require approval for a deleting an event uh but for drafting an email I want you to actually be able to always allow to draft an email, but not send it. And maybe for moving an email, I want you to have to ask for approval. And so this granular control of all the actions available across all the different connectors is a w in in my book. And that's and allowed me to feel comfortable connecting my email and have that scheduled task run once a day. Yeah, and all the apps that are responsible making these MCPs, I mean the developers that we talk about in the show so often do that. Like I I notice you also are running the drafts MCP. And if you look at that one, it's the same thing. There's like thirty different things that you can approve on a granular basis. And that's why the MC MCP protocol is taking over so big because it gives developers a way to sanely allow you to connect to AI and let the user have a lot of control over what it can and can't do. So email is another thing that I'm also working with AI on. I just recently set up this is uh brand new for me. It's only two days old, but I rebuilt a whole email system. And uh because I keep talking about um to folks in labs and and this this new field guide about how I feel like we're reaching a point where I call it the Star Trek moment, you know, where the software becomes abstracted from us, the the AI becomes our interface. And like when you think about ways like Whisper where we can just talk to it and it just does things for us, suddenly having your hands in the gears of your spreadsheet and your task manager and your email isn't really the future. I think it it's probably you talk to this thing and it says, oh, you got a message from Stephen, do you want to do this or that? And the and it sorts out the calendars and the replies and all that for me. And so I've been trying to kind of like force the hand on that a little bit. So I I bu ilt an abstracted email system. Just uh it's as an right now. I'm gonna call it an experiment. I don't know if it's useful or not. But um what it does is it goes through my email, I I switch over to Gmail because of all this AI stuff. Of course, two months later, now Fastmail has the MCP, so I have egg on my face, but so be it. Um, but what I do is have it say, okay, just go through the whole inbox every day. And then there are things that require my attention, like there, and there's different categories of those, like customer support, um, you know, customer problem, that's an urgent one, software devel oper writing me about software. You know, there's different categories of things. And you know, having an email app, you have to go through and figure out what they are, and that's always a pain in the neck and tracking them. So the experiment is I'm having it create an obsidian document for each one of those emails. So every time there's an email, put it, fill out the and it does all this for me, so I don't have to really think about it. Also put a link in it to the Gmail message and a link to the superhuman message so I can go to either platform to deal with if I want and it just I kind of collect everything. If there's if they ask me a question that requests research, do initial research, just put it together for me. And and then obsidian has this feature called bases, which is kind of like a kind of a poor man's notion. It puts together a database based on what the metadata is. So and it runs this scan three times a day. So morning, after uh afternoon, evening. Uh and now there's other email I get that doesn't require me to really deal with. Like it's somebody sent you a receipt because you bought a new shirt or uh you got the newsletter from that guy you like to read. And so it goes through those and there's just a single document that has all those summarized by who they came from and what what it's about and a link if I wanna read it. So I can just click it if I wanna go read it. But um usually I don't r click many of those links, but I am aware of those things coming through. If it's a receipt, it automatically files the receipt the appropriate place in Obsidian, so all my receipts are tracked and all that's handled. So I've got this new system that is on these two tiers. There's the one tier where it creates an individual record for each one, the other one where it just kind of gives me a list and does all the organizing and sorting for me. Well I've been doing it a couple days, but I've abstracted myself out of an email application by this. And it's working. I think like I am catching the urgent stuff very easily. I just read one base and say, oh, there's two things. And like because the robots attached to it, it's like, oh, this person just, you know need needss their course reset. Okay, please go reset it. And then they reset it. I push the button, pop over the email, write a quick reply, send the reply. By the time I sent the reply, it's already reset by the robot, and I go on to the next one. And that may be the future of our email. You know, imagine in five years where we just talk to it and it tells us, oh, so-and-so, we had a course reset request. Well, please take care of that. You know, and things that I used to spend a bunch of time and effort on now just get handled. And do you still have the robot working on like your second monitor so you can kind of keep an eye on it? Yeah. When it's doing those things? Yeah. I I so I just did an update. I had an extra studio display. It was my wife's, but she wasn't anymore. I mounted it vertically to the right of my main screen. And almost constantly the top is Claude co-work and the bottom is Chrome that Claude controls. And so when it's doing stuff over there, like I'll see it go into teachable and reset the person's course on the side. I'm not really paying attention to it, but I know generally what it's doing right and that that gives me that control issue you know the thing my problem with open claw was it's like if somebody got in it and had it doing things you wouldn't know with this it i'm wa literally watching it and it doesn't do things when I leave, I'll turn the app off so it has no power when I walk away. That's yeah, that's pretty slick. Yeah. Um well I'll mention a something that might be not as deep or involved, but something that I've used Claude uh personally is helping me with my website and some building elements via HTML. And so one of the things I wanted to do was at the top of my website, I always want my latest YouTube videos at the top. And it was I didn't have a great way to do that in Squarespace. I was like basically creating a new blog post that would go to the homepage every time, but I never often forgot to do that. It was time consuming to try and do that. So I wanted to know can Claude just like build something? And basically I knew there was an RSS feed for every YouTube channel. It's kind of like hidden behind the scenes. And you can have a just a RSS feed of new videos. So I went to Claude and I asked it, can you help me build this? And it was a long conversation and I don't know much, you know, about like uh Cloudflare workers and all this kind of stuff, but Claude literally walked step by step. I do have a Cloudflare account that I used for something. And so Claude helped me build a worker for Cloudflare. It literally walked me through deploying it in Cloudflare and then gave me HTML code to just embed in my Squarespace website and then it just worked. And then it's showing my latest YouTube videos on my homepage and I was done. And it it's it's staying up to date. It's working. And I was like, okay, well, that's awesome. Uh, can it also do like can it do more? And one of the other use cases was on my same website, I have a whole page of smart home devices, home kit devices. And Squarespace deprecated the Amazon product block. There used to be a block where you can literally search for an Amazon product. It would link it with the image and title. It would uh use your affiliate link and then it would just have the block on the page. That block isn't available anymore. And I was like, oh no, that was a really big page on my website. What am I gonna do? So I literally pointed Claude to this web page and I said, listen, I need to rebuild this page. I want to make it so I can add some products. I want categories with anchor links. So when I click a link, it jumps to that part of the page. Make HTML that I can just paste in the site . And David, sure enough, it did it. And it just gave me a bunch of HTML. I put it. This is just a big code block in Squarespace, but it made it so when you hover over the little objects actually raise and have the shadow these are all linked with my affiliate links and it's still kept the categories and when I wanted to make changes now I can literally just go back to my conversation in Claude I've pinned it, uh, which you can like pin conversations in in your Claude app. And so I have my uh home kit webpage pinned right here. And if I want to add something, I can literally just give it the stuff, give it the link to Amazon or my affiliate link, it'll update the code and I could just copy and paste that into the code block on my website. And I was like, this one, a bunch of people were like, wait a couple months, you're just going to build a whole website this way, which was tempting, but I like having enough control of my website. I like Squarespace. I like how it works. I like the organization that I have already. So I'm not going to have it do that just yet, but it probably could. But to make these kinds of small changes or big changes, build a whole web page, rebuild it from scratch. It Claude kind of blew my mind on that. Yeah, I think part of all of this is um I like the analogy of the original movies. Like you're a movie guy, and when they first got movie cam eras, the first instinct was to put a camera where the audience would sit in a theater and shoot a movie and they would do the act on the stage. And it took several years for them to realize that they could put the camera on the stage and get different angles. And that's just because of your mindset. You can't you can't kind of break the mindset immediately. And I think as this stuff gets more powerful, we have to do that. We have to think, okay, just because I did it this way all the time, like for me, even with this email system, this may be a bust. I'm only two days in, but I think it won't be because it's doing categorization, everything for me effectively with my instructions. And when it gets it wrong, I can say, oh no, when I get the one from Ben Thompson, I want to read that. So make sure that gets to the top. You know, and so it it knows it's getting my preferences down in a way that nothing else ever has. So uh I feel like we're we're in the days where we're shooting movies in the in the seats, but we gotta figure out how to get the camera on the stage with this stuff. That's that's good. That's a good analogy. What's another big one that you use? It's the back end stuff of Mac Sparky. Is that's the whole reason I made the robot course because like so much of that stuff, like like I said, just basic customer service things. I do a code for lab numbers. They get a discount on field guides. And like I just now, you know, I put that on a scheduled task now. So it just does it. And I don't think about it. And then uh setting the banner for the sponsors on the website every Sunday afternoon. It's now a scheduled task. It goes to the web, it makes the change, it knows what it needs to do. And you know, as this thing learns my preferences and how I want things done, these simple tasks are very repeatable and reliable once you build them correctly. And you get to a point where things that you used to think about and worry about you don't anymore. And you know, I I get emails now from people saying, hey, your newsletters are a lot longer, or you know, your blog posts are getting more in-depth. Uh are you using AI? Seems like, you know, like everything. You know, I'm like, no, it's because I'm using AI I have more time to actually write you know and it's like I don't I don't know how to convince people but the um but yeah so I'm I'm really really leaning into donkey work with this stuff and and finding it it's it's very possible now. I think it's still in the fiddly stage where you've got to learn how to build the skill, you gotta learn how to do it, but it is absolutely achievable. And um I think the next penny to drop is like if Apple gets real about this and it gets suddenly way easier to do this stuff. But man, we're here. Yeah. I want to share two other uh use cases or projects that I've done. Uh if you might have seen on YouTube, I did vibe code an app and it is literally in the app store right now. This was super simple app, very silly, but I just wanted to know like could ChatGPT code an app that actually gets through app review all the way to being able to be downloaded in the App Store? It turns out you can. And this was almost a year ago now, but it did work, uh my co-host Jasonon Ayt , he's making a notes app uh that uh that uses a lot of Apple Intelligence features, integrates calendars and tasks. He's been vibe coding that and with Claude and he might be uh actually releasing that soon, which is pretty wild and so uh that's gonna be a big one. Stephen on that point you didn't I did the same thing. Oh did I release an app too it's called the no list and it's a place to write down all the things you say no to and that was that was generated with Claude. And uh it actually needs an update. It works fine, but I wanna I wanna add some features to it and stuff, but I just you know, I got kind of on the robot assistant uh train and I don't have time right now. But is it the orange icon?? What Yeah. It's like an orange icon. Yeah. Yeah. Oh yeah. We'll put it we'll put a link to our silly apps in the show notes if you want to check out. There's the there's there's the no list. It it is a silly app, but people actually I hear from people actually using it. What's funny is when people write me who have no idea who I am. They just like got in the app store and now they're using my no app. That's so funny. I did see uh my app has four point seven stars, which is frankly much higher than it deserves, but it's a four seven. And the last one I'll mention so Chat GPT for a long time has had custom GPTs where you could literally train it on some tasks. And so I have one that I call my YouTube sidekick . And basically what I did was give it a bunch of transcripts, titles, and other information about videos that performed well on my channel. And what I'll do then is give it a transcript . So I use the app transcriptionist to transcribe all my videos. I actually export an MP three first, get a transcribe to do this process, and then uh the video will begin processing . But then I have a prompt through text expander. So I just expanded this prompt right here. And when I run it, you know, at the end of the process of making YouTube video, like my creative juices are pretty drained. I've I usually record , edit, publish, or at least schedule to publish all in the same day. So getting to like thinking about titles and descriptions and tags to put on YouTube, it's a bit of a a lift for me. And so this now gives me a title idea, gives me a description that I might massage a little bit, it gives me the tags, which I basically just copy and paste this into YouTube. YouTube has said tags don't really matter a whole lot, but I like having them, so I just put that there. It gives me some alternate titles, which I might test A B style, and also gives me some ideas of what text to put on a thumbnail. And then if I mention any videos at the end, you know, on YouTube, you've got to program the end cards . I tell it, uh, you know, tell let me tell me what videos I mentioned at the end of the video. So this little blurb right here is now giving me a huge head start in actually getting the thing in YouTube with the title and the description. And then a lot of time what I'll do is if there's one title that I like a lot, you know, I'll take that one title that it gave me and I'll say give me 10 more like this. And it will just go and now generate 10 more ideas that'll be slightly different. And maybe I'll like one. Maybe it'll inspire an idea that I of an original title. But it's basically that creative partner that, you know, I don't have a team, I don't have an editor, I don't have anybody else doing anything in my process. I make the thumbnails, I do all the things, and this allows me to bounce idea the ideas off something and get some ideas back that might be useful. And I will say I just started using that doing that same process in Claude because I created a project. And in Claude, I think of projects like custom GPTs where if I use a project over and over again for the same kind of task , it will slowly learn from my input and what it gives me. And I do actually like what Claude is giving me a little more as well. And so I might move it's funny, I've might move from chat GPT to Claude for this process for YouTube videos, and I might move from Gemini to ChatGPT for the image side of the thumbnail part. So it's interesting. Yeah, with projects in in uh cloud you can also upload assets to it, which you can also do with chat GP T. So you could get like a list of all of your prior titles so you don't bump into a prior one. If you're going to make that many shortcuts videos, that's a possibility. You know, yeah. So you could um you could also kind of integrate that way. A couple of ways I use projects that are kind of funny is um one is I've been really working on my my uh my health uh the last six months and so much so that I I gave it a couple of my blood test s and just said, you know, what you know, what does this mean? And I'm like feeding it data from the calendar and my exercise routines. And like when I get stuck, you know, I've been trying to lose weight, but I've been the same weight for the last like five days and you know, I say it'll tell me, Oh, it's okay, you're as long as you're doing the you know. It's like a little coach, you know. And I know it's an analog, it's a it's an algorithm thing, but it it actually I find a little useful. A a funny one that I did that I shared in the robot field guide was I um my wife always complains that I wear the same clothes all the time. You know. So and so I I thought, well let me let me see if I can use AI. So I took pictures of all my clothes. Well, and I just fed it to it. And I said, You're gonna be my style assistant. Your your purpose in life is to get my wife to compliment me on my outfit once, you know. So, you know, and and I did it as a project so it has memory of what I wear when I tell it. And I d I tried it. And and the the first day I I picked a a shirt and pants and belt that it told me to wear and we went out to lunch, and my wife's like, You look really nice today. And I thought, oh my gosh, this worked, you know. And then uh I shared it in one of the workshops or robots , and then it got back to my wife, and then she comes home and says, Did you use AI to figure out that outfit ? And if it's good, what does it matter? Yeah. Then she thought it was kind of cool that I had done it. But the um but yeah, so there's all kinds of things you can do that. Uh for coding, you had mentioned, you know, cod ing, a very cheap version of coding I did just last night is I got an email from a a student who had built a daily brief with their robot, but they did it in the style of Wes Anderson. They named it Atomic City . And I was looking at that and it looked really nice. And I thought this would be really fun. And and I'm struggling with obsidian. You know, like everyone, I don't like the way it looks, but it is so good for what I'm doing with it that I'm I'm sure I'm gonna s probably stick with it. Um uh so I I said, make me an Atomic City obsidian theme. And I just I showed it to you before we started. It's got you know the, Futura font and kind of a paper look to it and I'm like, you know, this would be more fun to use Obsidian if it looked like this every day. So I I've got it writing themes for me for Obsidian. That's really good. Uh that's I I said last, but I actually want to mention t one slash two other use cases. I'm s not super handy. I'm a little handy. I've changed out my own dishwasher. And uh you know, I know how to do some things, but some things elude me. And so there's been several times where previously I asked ChatGPT, I'm not sure which I would use today, but one example was I got we got this trampoline with like a hose attachment, so like those sprinklers on the trampoline. Fine. And uh it had one of these weird quick connect things and I wasn't sure like what to do with it. So I literally just took a picture, gave it to ChatGPT, and it gave me the instructions of what to do. And it helped me figure it out. And I've done it with that. I did it with my mom's tub was like dripping. And so I asked it for what part should I buy? How do I replace it? And it walked me through all of that. And for that kind of uh like part stuff too, I've had to get like uh the the uh door plastic holder parts of a fridge, the trays or whatever you want to call it. Yeah. I've not if I've needed to be a replacement for that, I had to buy a replacement ice maker for my fridge at one point . You know, that process can be pretty laborious just trying to Google your model number and see where to look for this. Well I will literally just take a picture of the label inside the fridge that has the model number and say I need an ice maker for my fridge. And Chat GPT, I've done it with Claude too, does great. But here's where here's the part. Here's the one you need. Here's several places to buy it. And it just expedites that whole process. And so even little things like that, where you just want to show something what you're looking at and say, what do I do now? It can actually help a lot with it. Uh the one last one I want to mention, because I haven't called this app out, is um is my transcrip tion preferred app is Mac Whisper . I just they do such a great job. And I bought that thing when it first came out like I don't know three, four years ago and he just keeps updating it. He just added a command line interface , which means he's on his way to MCP town. You know, and like having your uh your transcription done locally makes a lot of sense. And for my kind of work, I do it daily. I mean, I'm always transcribing something. And uh and Mac Whisper is kind of my weapon of choice for that. But but that is in effect i you know, Mac Whisper is using a local AI, a local model to do it, but you know, getting really effective and useful transcripts enables me to to do more with the AI and also just to have more use for the kind of work I'm doing. So that's a big deal for me. Yeah, well I'll put links to all the apps we talked about and of course the robot assistant field guide, which will help you actually learn and implement all these things. All of those links will be in the show notes. And I don't know, should we go talk about uh your upgrade crisis? Yeah, let's do that. Uh yeah, let's do that. We we are the Mac PowerUsers. You can find us at relay.fm slash MPU. Uh go there, check it out. Thank you to our sponsors this week, Mercury Weather and Ecamm . Um if you're a more power user subscriber, you're going to get that monthly ish bonus episode and you're going to get ad-free episodes and extended episodes every week. So we'd love to have you join. Today we are going to talk about my upgrade crisis. And I am I think I'm going to make a decision, Stephen. You're going to be very influential now. So let's do that. Thanks, everybody. Have a great week, and we'll see you next time.