Future of Prosumer AI Tools

Hello, and welcome back to another episode of App Stor ies. Today's episode is brought to you by Vitally. I'm John Voorhees and with me is Federico Vedi ci. Good morning, Federico. Afternoon or evening. It's uh almost evening, I would say for me. It's already afternoon here too. Yeahah. Ye. Uh how are you, John? I'm doing well. I'm doing really well. How about yourself? Hanging in there? I'm doing well. I was just uh I was I was messing around um for MPC. I was playing around with Game Hub uh with my Red Magic 11 Pro Phone. And I I'm gonna talk about this on the show later this week. I started I I tried to play Resident Evil Records. It's not gonna happen, man. It's not Resident Evil on that phone, it's not gonna happen. What are people doing with these videos? How do they do these things? We'll have a conversation about these. Okay. I'm starting to not believe those videos on YouTube anymore. I d look I've tried everything. I don't know how these people are getting that image quality. Look, I don't think I'm stupid with these things, right? I think I know how to tweak, how to tinker around with things. I'm just I am just not getting that performance out of this gaming phone. Don't they tell you the the things to tweak in the videos too, like the programmers? So Yeah, I hear yeah. Oh well But anyway, we're not here to talk about gaming. We have a separate show about that. It's called MPC and Exportable Console on Mac Stories and on Patreon if you want to get extra content. We're here to talk about something that is very frustrating for me and also I think for John. Oh yeah. Which is making sense of all these new features and this new terminology for AI apps especially on desktop and we are specifically going to focus on Claude and ChatGPT today or I guess Claude and Codex , which is OpenAI's separate system/slash application for programming and also kind of some agentic use cases. This is very frustrating to us because not only are these things changing on a daily basis at this point, but these companies, Anthropic and OpenAI, they are adding feature on top of feature on top of feature, and they're breaking things in the process and they are renaming things in the process. If you thought that Apple renaming the efficiency course to performance course was frustrating, that's nothing. That's nothing. It's really nothing. No, I mean obviously I think the problem that we have here is that these companies are moving incredibly fast and they're not wait you know, I think traditional software development, you would kind of expect things to be pretty smooth by the time they're launched. There are always bugs. But what's happening here is, you know, ChatGPT will come out with some some feature in the in the chat bot. And so the anthropic people will jump all over code and they'll release whatever they were working on that might be similar right away, whether it's really ready for prime time or not. And so you've had a lot of things like you know, a lot of things issued uh released as previews or betas. Research previews. Research previews , which is essentially an alpha, I would say, because when like cowork first came out, and we'll talk about cowork in a minute, cowork was super rough when it first came out. But it got better very fast. I mean it's only been in research preview for what a month, three weeks, something like that? It's not been very long, but it's gotten a lot better. So yeah, that that's that's kind of what's causing it, I think. But I I feel like they need to like pause for a moment and rethink like all of these features, how they fit together, what they're called, what apps have them, what apps don't, and how the apps are designed even, because I've got some ideas for how the Cloud app on the iPhone should look, for instance. Yeah, let's start with that. Let's start with Claude and Anthropic. So I really spent some time over the past couple of days playing around with this. The easiest way to describe this is that Anthropic has stuffed Claude for Mac, Claude for Mac with the three C's, that's what I'm calling them. Chat, co-work, and code. So these are three separate modes that are now co existing in the cloud app for Mac. I'm not talking about the website. I'm not talking about cloud coding the terminal. I'm talking about cloud for Mac. Chat obviously used to be the default chatbot mode with Cloud where you can have your projects and you can have your used to be your MCP connectors. That's a separate conversation now. Uh but now you have these three segments at the top of the screen. You can actually move between them by doing command one, two and three. I discovered you can move between chat, co-work and code. So chat, we don't need to talk about the chat bot. You know what cloud the chatbot is. Cowork and code. This is where I think most people get confused. So Cloud Code started as a terminal-only affair . Last year. In fact, it launched about a year ago. Then they added, if I'm not mistaken, Cloud on the Cloud Code for Web, horribly named, but Cloud Code for Web is Cloud Code, but instead of working on your local computer in a folder, you point it to a GitHub repo and you work in that virtualized location. It creates a little virtual computer in the cloud, you point it to a repo and you say, work on this project for me. Right. Very different from cloud code on your computer. Cloud code on your computer can now be accessed and I'm gonna give it to you, John, because you use cloud code much more than I do. You can use cloud code the old fashioned way with a terminal window. Yes. And that's the OG cloud code. Or you can use desktop cloud code inside Cloud for Mac . You go to the code section of the cloud app, and there you can use cloud code, which is kind of similar from a design perspective to cloud code on the web, but it behaves more like cloud code in the terminal in that you can point it to a folder and you can work in a local folder and you don't need to use GitHub at all. Although you can point it at GitHub. Although you can but you can also point to the other. Which is why all this gets confusing. It's like the focus is not on GitHub as much when you're in the app, although it's still possible to do it in the app. And of course, there's remote control, which allows you to work in the app , like one of the mobile apps I'm talking about specifically now, the iPhone or the iPad, and you're connecting back to a terminal session, which is not the same as Claude for Mac. It's something different. So so you wrote about remote control. Remote control lets you remotely control a cloud code session that is running in the termin al . Yes. Not inside the cloud app. No, you have Claude actually running in the terminal already, and you type in a command. This is already confusing. session or you can make it the default so all your sessions are remote controllable, which is what I do. Okay. Because I never that way you don't have to be thinking about it in advance. Because otherwise you have to start your cloud session in remote control. So you like you go to the terminal you're not in claude yet you type claude space remote dash control and now you've got a remote session that's no good for my purposes because I never know in advance whether I'm going to be remotely controlling a session or not. So I just have it as the default. And then what you do is what I I would say the best thing to do is to rename your session, which is just a slash rename and then give it a name. You can like a lot of times all I'll do is I'll rename it for the computer that it's running on, like my Mac Studio, and then just use it for whatever comes up when I'm away from my desk. Or you can name it, you know, for a project if you're in the middle of a project and you have to leave your desk and you're not, you don't have the terminal available anymore, and then you can use it from one of those apps. And I wouldn't say I use it a lot, but I do use it. I mean, for one, there, for instance, there was one night I was working on a project and I went off to bed. I had my iPad with me and something occurred to me. It was like, oh, I should try to do this. And I had it in a remote session on my desktop computer. So the set so the session the session was already running on the computer. I went to bed. I thought, oh, you know, I should try this. I had this idea. I went into the iPad app. I went to the code section. I found my listed remote session. Went into it. And once you open it up, you actually see what you would see in the terminal. I mean, it's formatted more like, you know, for an iPad or iPhone app, but it's the same words are there. And you can just type commands straight into the remote session from the cook from the code tab. So I did a couple of things and you know it was nice to be able to do that while it was still on my mind. I mean the downside is of course is that you're not this is not like uh the screen sharing because you don't have anything else available so if like you're working on an app you you could build the app I guess but you wouldn't be able to see the app or a web app or whatever you're working on you've only got that insight into what's in the terminal. You don't, you know, you don't have actually the other things on your computer, which is why I use this kind of sparingly, because a lot of times if I'm on my laptop or something, or even on the iPad, using screen sharing on the Mac or something like jump desktop gives me access to the other things too. So I can take screenshots and feed those into Claude, you know, other things that you might want to do beyond the terminal window itself. So that's Claude and Cloud Code. Yeah. There's a third. Can I I before we before we get to that, or maybe this isn't the time to raise this, but I want to talk about the difference between co-work and projects because I think that's confusing to to a degree. Okay. So projects are folders for the chatbot. Um the projects are chats in the chat bot that have a common system prompt that can use remote connectors but don't have access to the file system. Right. And they have their own memories. They have their own memory that's that are separate from each project and separate from chat and you can attach documents for reference materials and things like that. And so the way I think about these things, and I th they're related but different. I think of projects as long-term things, like you know, the name suggests a project, where you're going to be coming back to it over and over, whether it's just a thing that's going to be a and you don't want to type the same prompt every single time. Yeah, it it's either it's something that's going to take you a long time to complete or you're gonna ask similar things and want a similar prompt over and over again to do that. So I have a bunch of projects. Whereas cowork is more of like a a one-off thing where you 're kinda so see that's where it gets it gets confusing sometimes to to decide which to use. So I think co-work is the most confusing of the bunch so far, also because it's the m it's it's the latest one and it's labeled as a research preview and it's desktop only. So co-work has no mobile presence at all . Co-work, the technical description is that they took the agentic guts of Cloud Code. The way that Cloud Code responds to you, calls external tools, comes back, compacts the conversation, continues. They took w that thing that John loves to call the agen tic harness of cloud code my favorite word and they turned it into something for more general productivity. That's what co-work is, but there's a difference. So co-work, it doesn't organize your chats in projects. You can choose to work , you can choose to start a co-work session in a folder on your Mac or not. Which is w which is a little bit like how clo cloud code works, right? I mean because you're doing it inside some sort of project folder. Yes. Typically. If you work in a folder and if that folder contains a claude.md sure instruction file, it will read that. Yeah. Even if your cloud.md is sim linked to an agents. md, because openai and anthropic cannot agree upon a naming standard for this instruction document, it will read the se lamink ed document. Right. Okay. So you can you can basically pre-inje ct some system instructions into cloud cowork. For example, in my markdown document that I have in my agent folder, there are some preset instructions that say I'm Federico Bittici, I use to doist, I use Notion. This is the Notion database where I keep my daily notes, like basically preset information for cloud c owork to get started. So that it doesn't have to search for my daily notes every time, for example. Right. The thing about cowork that I really like is that it behaves like cloud code but presents you information like like regular cloud. Right. So it's got that sort of sequential tool calling nature without showing you that it's running terminal commands, without showing you that it's doing like programmary things. And that's sort of the pitch for cowork, right? It's it's an agent, it's a general purpose agent, and they have done this because people were using cloud code, which is meant for developers and programmers, for general productivity, which is not designed for. Well you know like 'cause there are bash commands to move files around your computer and and you know there are scripts being written in the background to maybe export uh markdown as a PDF using Pandoc and there's all kinds of things that that can be done. Yeah. So I am going to uh I think I mentioned all the things there's three more things I need to explain. Okay. First is a series of complaints. Uh then there's a a series of things that anthropic are doing right and third is a final feature that I want to mention. So the complaints is uh I call this section in our in our document, John, the great fragmentation of tools. Yep. There are too many words anthropic. There are too many terms that you're asking people to learn and keep up to date with. Yeah. Just to go in succession here, John. Skills, connectors, mcp, plugins, hooks, slash commands, agents . Um sorry I would add artifacts to that. And artifacts. Sure. Technically a tool, but. So let's try and simplify this for our listeners. Yeah. Skills are uh those uh skill. md, they're markdown documents that basically teach Claude how to do something. Uh they're packaged instructions, they're basically prompts as markdown files. And they can be combined with a script, for instance. They can contain scripts, they can contain image assets, they can contain whatever you want as long as you store them in a folder. Connectors and MCP, they're basically the same thing now. Uh Anthropic is trying to rebrand the MCP standard as the English word connector, they're the same thing. You are connecting to to doist, it means you are using the to doist MCP server. And then this is like a step in the right direction, I think, in the sense that people don't care if it's technically built on the MCP standard or whether it's like some one-off tool that Anthropic built for Notion or Gmail or something. Because that's the difference, right? Is under the hood some of these things maybe not not may not be strictly MCP servers. They may have some additional code that's working on the anthropic side. Slash commands, I never use this, but these are basically ways to come up with your own text shortcuts, I think. Like you can create a slash command where you type slash, I don't know, uh J O and you have a and you have a slash command for John that does, you know, sends an email to John or something. Yeah, I use a few of these. I for certain Yeah, nice. Yeah, for certain things, for certain things that I've built where I want to like clean up a transcript or something. I have I have one that that calls some scripts that's like called uh transcribe cleanup and it will call some things and and then basically abbreviations. Yeah, they're just they're just easy to remember ways to kick off some sort of thing that you don't want to have happen automatically. It's like a manual process, basically. This episode of App Stories is brought to you by our friends at Vitally, the AI powered workspace for customer success managers. Vitally's purpose built AI is for scale customer success. It's there to help you better understand customers to reduce churn, unlock growth, and always stay a step ahead. Basically, all the things that make you great at your job. Vitally delivers the clarity, automation, and AI-driven insights your team needs. So you can move faster, stay a ligned, and drive customer outcomes at scale. Vitally AI services insights from every customer interaction to unlock another level of CS productivity, visibility, and collaboration. It's why Vitally is trusted by more than 600 leading B2B SaaS customer success teams. And how is this for a great offer? Vitally is offering a free pair of AirPods Pro for every App Stories listener who takes a qualified demo call. So if you're a customer success decision maker, schedule your call by visiting vitally .io slash app Again, that's Vitali.io slash app stories for a free pair of AirPods Pro when you take a qualified demo. Our thanks to Vitali for their support of App Stories. Hooks. Have you ever used hooks? Maybe once or twice, but not really. Hooks. So if I'm not mistaken, hooks are basic. So we're we're talking about cloud code here. Is hooks into the guts of Cloud itself. It's it's they're basically triggers and events for things that can happen automatically when something happens in Cloud Code. You know what I use them for, TC? I use them for notifications. I hooked up Simon Stovering's new beta app called Burr B R R R and I send myself notifications because you can do pretty rich notifications with that beta and that's that uses a hook to different events that happen within Cloud, such as like when something's pushed to GitHub, I get a notification that something has been pushed to GitHub. If Cloud I don't ha I don't have this beta by the way. I sent it to you over the weekend. You did? I did. I sent it to you on Saturday. How many weekends ago? No, just like last Saturday, I think . Uh maybe Friday afternoon. But yeah, it uh so I I know when GitHub happen when things are pushed to GitHub, I get notifications when Claude needs my attention on my remote server, that kind of stuff. Hmm. What's it called again? I'll s Burr B R R R scroll up to Saturday. There aren't that many messages. We we exchange too many links. It's not a question Yeah we do. I'll send you I'll resend it to you. It's it you should check it out though. It's very good. And he's building a CLI for it too, even in the time since I installed it. So you can which which allows you to use it, I think, without necessarily using there's advantages to the CLI versus hooks. I've got it installed with hooks. Okay. The part that is most confusing to me is plugins , which I also think is again, it's the latest thing that Anthropic has done, and it's the one that they have explained the least. So I pulled some links from the c loud documentation and the cloud code documentation. In cloud code documentation, they say plugins extend cloud code with skills, agents, hooks, and MCP servers. Plugin marketpla ces are catalogs that help you discover and install these extensions without building them yourself. It's a bundle as well as it's a bundle of so a plugin can contain an MCP server, some slash commands for that MCP server , maybe some skills for that, and it's all prepackaged as a as well. And it's on a marketplace where presumably you can trust it more than if you just pull something off of a random GitHub GitHub rebut. The problem is that these plugins in the cloud app for Mac , they're very new. There is a gallery right now that only contains anthropic plugins, no third party plugins. I tried to install an anthropic one. Yeah. Uh so I clicked a little. I clicked a there was one in the anthropic web directory called uh Cloud Cowork Plugin Creator. Okay. So it was like a w a plugin to create plugins, very meta. I thought this is interesting. And I clicked a link in my web browser and he redirected me to Cloud for Mac and it said the plugin didn't exist. So Oh I you can get at them easier, I think, through Clog Code. I haven't tried them on the Mac app. I've only tried them in Clogg Code where you can do things like get a um a code review, cart code review plugin, for instance, which I think is made by Anthropic. They have a design one, I think. There are too many words. There are too many words. Um but the one thing that Anthropic I think are doing right now is they are unifying a lot of things. Uh for example, something that I've noticed uh lately is that skills now sync all across your cloud apps. Right. If you upload a skill in clo on the in the cloud website, it shows up in cloud for Mac on another computer. I just booted up my Mac mini and I open Cloud on the Mac Mini and it had all the same skills from Cloud on the MacBook Pro. So they're now syncing all your skills with your Cloud account. That's the right thing to do. They're also now properly syncing connectors. So anything that you connect to from cloud.ai or cloud for Mac will show up in every other instance of cloud. And not just that, but something that I'm sorry, they they seem to have stabilized too. For instance, like the Notion one and the Todoist one, which would get disconnected all the time, now seem to be at least and this is literally only over the last week or two that they've seemed to have stabilized. Yes. And something that I've noticed a few days ago and I told you uh is now when you fire up cloud code in the terminal and you do the slash command slash mcp to see a list of your connectors, it'll show you all the existing connectors that are tied to your account that are coming from cloud.ai , which means you previously connected from either the cloud website or cloud form ac. So you don't need to reset up all of those connectors in Cloud Code for terminal again. They all come, they they all follow you around as long as you're using your account. Right. Finally, I wanna talk about scheduled tasks. Yeah, I want to hear about this because this is actually something I've been meaning to try and I haven't gotten around to it, and it's very, very new. It's like what, like a week old or something. Yeah. This is very new. Now this, leaves in the sidebar of Claude for Mac, which I think you have some thoughts to share about the sidebar. It's uh it's a new entry there, and it's basically a way for you to schedule tasks that cloud co-work will run for you. Which means you can work in a folder, you can take all of your existing connectors. Yeah and you can schedule a task with a time and a prompt. So here's what I did John. I took a previous project from the Cloud chatbot it was a very detailed prompt that told Claude how to process my email, look for important messages, and turn those important messages into deep linked to-doist tasks. I used to have to run that prompt manually every couple of days or every morning. I took that prompt, created a new schedule task in Cloud Code work, running in the cloud app on my Mac Studio, not on my MacBook Pro, not on my Mac Mini. I chose the Mac Studio because my Mac Studio is always on. And I can keep the Cloud app open all the time. Right, which is because that's one of the limitations of scheduled tasks. Right. You they need the cloud app to be always running. Otherwise, they will not trigger. That's very that's why like with the remote control I'm doing the same thing on always on Mac because the terminal window has to be open. Let me ask you where this is because I don't see it. Huh. Have you updated Cloud for Mac? I just did while you were talking because there's another complaint right there is that sometimes you re realize after a while that you're out of date because uh you have to restart the app. Try and go to co work and you should see schedule tasks. Oh, it's under core work. I was under chat. No, it's in co work. I see. I see. Okay, no, I got it. I got it. Yeah. Yeah. Uh your schedule tasks don't sync across devices. So right now on my Mac Mini, I do not see the schedule task that I created on the Mac Studio. Okay? So more fragmentation. However, I took that prompt, turned it into a schedule task. Yeah. And it just worked. I did I did a first test. It sus successfully found my Gmail and to do his connectors. Ran a first pass of the prompt. And I can even tell you schedule tasks in co-work can use your existing skills because as part of my prompt I added one more step that said every time you're finished processing my email and creating tasks in my to-do list send me a notification with pushover using your pushover skill which I personally created. And it did . So your co work schedule tasks can use connectors, can use your file system, and can use your skills. That's nice. And they can be scheduled. So now, every day at 3 30 PM, Cloud Cowork running on my Mac Studio should be able to process my email for me. All right. Well, so let me ask you about I want to ask you, this is like an aside, what about your email processing? What kind of give me an example of the kind of things you you tell it to flow to the service for you? Do you s is it like people who I've responded to before and developers who are pitching apps and that or somebody from uh with an Apple.com uh email that kind of I mean you're you are answering yourself. Okay. Uh those are just the things that I kind of already have set up in a way with like splits and superhuman, right? But you're doing it all as one task and it becomes I do have my splits, but the thing is if I don't create a task for me to nag me, I'm not gonna Yeah no, I'm the same way. I'm terrible about email these days, and I wanna set this up myself. So , email emails from Apple PR. It excludes things like newsletters, uh junk, you know. Marketing emails. It tries to look if that's a frequent contact. And obviously it only processes emails that I have not responded to. So uh emails were the last message, threads were the last message is not from me. Uh yeah, that's the kind of thing that I do. So over all, I will say anthropic is obviously as obviously nailed the productivity space and the programmer space, I think. I agree. They have too much going on in the cloud app. Way too much, especially when you're on something like an iPhone. I would at least spin off cloud code into a separate applic. Oh see, I don't think they need to have three apps, right? I don't think there needs to be cloud chat, cloud co-work and cloud code. Co work and code together, maybe? Co that's what I was gonna say. Maybe c co-work and code together and just keep the chat bot the chat bot. Yeah. I don't know, but three products into one application is too much. Three separate applications is also probably too much. So maybe the middle ground is two. Yeah, which is surprise is what we're gonna talk about with open AI. Yeah. Yeah, no, because it's i I mean the thing about cowork and claw code is they're very similar in the way that they operate and so I think they make sense together and you could just do it as like a uh you know a a tab bar or uh or uh some other kind of UI element to to separate the two. The thing that drives me a little crazy about the Claude app is what you're saying is that the way this app works, especially on the iPhone, is that it's got a it's got a little menu button at the top that opens the sidebar. And this sidebar just goes on forever because it's chats, projects, artifacts, code, and then it's starred things, which can be both projects and chats, and recents. And to me, it feels like this, first of all, I think that this should be a tab bar app because you have this is right Right? Because it should be a tab bar that has chats. Uh I think artifacts should be combined with something called documents. Just call them documents. And your documents are either reports you get in like Word or Excel or Markdown and Artifacts, which are just they're just web apps is all they're like little mini web apps. So it's just HTML or React or whatever usually. So there should be a document section, a chat section, and a code section if you were to keep code in here. And then you have the the search button which they could use the new tab bar thing where search shrinks down to the far left, right? Uh and then you would move the new chat thing up to the top to the upper right. I'm just doing this is just like a really down and dirty uh design review here. Instead of like opening it up and you've got just got this long list of chats which you know, I don't know. I it yeah. It's it's too much. It's overwhelming. It's too much. So and I think we agree that one app is uh is too confusing. Three we probably don't want three apps No. I mean I'm I'm a big fan of Claude, b all the apps actually. I mean I find what we've just been talking about confusing and hard to keep up with. And I think it could be simplified a lot , but I like the product in general and most of these different things, features we've talked about. Yeah. Now OpenAI is doing things very differently. And I think I I use uh ChatGPT and openai m a lot more than john. Yes. Yeah. So open AI, they also have the confusing thing where they have a bunch of things called codecs Yes, it too. If I'm not mistaken, codex started as a separate model . Then they did the codex on the web, which was basically cloud code, but by open AI. Then they did codex CLI, which is code x in the terminal. And isn't Codex both open source and proprietary ? Codex code the codex command line utility is open source. Okay. The whole thing is open source. Uh codex for Mac, which is uh and for Windows now. The codex app uh is not open source. Right. But it's based on but it's using the open source tool. It's using the open source hardness and uh and yeah Codex is entirely open source. No. I really like Codex. I I really do. It's very good. It's very good. I really like the new GP T 5.4 model, but it's a very different experience and it's a very different vibe from Claude. Um, so first of all , I would urge everybody to use the Codex app in stead of the terminal. I think OpenAI has done a pretty good job with the Codex app in the sense that it takes all of your pre-existing instructions or MCP connectors if you were using codex in the terminal before and it gives you a UI for everything. Right that happens because the settings for the codex app are based on the same settings files for the codex CLI. Everything lives in a hidden folder on your computer dot codex where inside you will find a doc a file called config dot is a fancy JSON file that contains all of the settings for codecs. That same file can be configured both with codecs in the terminal and the codex app. The codex app is more intuitive because it gives you a UI to manage those settings. So that means that if you use codecs before, once again, this is totally separate from ChatGPT, right? Uh ChatGPT as a website, ChatGPT as ChatGPT for Mac, which I think is a pretty good mac app actually is pretty good they have adopted liquid glass they have the feature called work with apps which lets chat gbt control some Mac apps like Notion and Apple Notes or Text Edit, um Codex is totally separate. And Codex is right now primarily for developers and programmers for doing things with code. In fact, when you open codecs and you point it to a folder, it says let's build. All right. Design for design for coding. And and if you are primarily on cloud like I am, I think one good way to get a sense for codecs is to take whatever you've been doing building in Claude and dump it into Codex and have it do a backstop review it. That's one of the things that I do all the time with some of the web apps and things that I've been building is I just point it at the same folder and say, hey, I'm working on this thing. I want you to dispatch some agents and do a thorough review of the code and tell me what you think could come up with recommendations. And then you you can either fix those things if it finds bugs or whatever, right there in Codex, or you can take that output report, bring it back to Claude and say, Hey, I had Codex do this. What do you think of these recommendations? And then you can play them off of each other and kind of come down. Because one of the things I think in a lot of code reviews that these these chatbots all do, that all these LLMs all do, is that they will sometimes suggest things that are more theoretical problems than real problems . And so it can be useful to kind of see, all right, here's a list of things and then ask, play one chap out off the other and see which ones are the ones that are actually real meaningful bugs that need to be fixed versus things that are just kind of either hallucinated or just minor issues. That is an excellent point also because codex by nature uh loves to analyze and read a code base upfront. Its spend a lot of even if you set the thinking level to low or medium, it spends by nature a lot of time reading and analyzing up front. And then it executes and it's fast. But it's very it's a very different sort of hardness from cloud code. Yeah. Co uh cloud code usually just springs into action, sometimes to a fault. Yes. Um codex loves to pre-think and pre analyze everything. So very different behavior. Anyway, the codex app I mentioned it picks up all of your existing MCP servers if you were using Codex in the terminal before. If you were not, you can connect to uh MCP uh servers right in the codex settings. They have some recommended connectors, things like linear notion, but you can also connect to some um custom MCP servers with a UI. And it's confusing. The UI for connecting to a custom NCP in codex is too confusing. Like uh there are like fields for like headers and token values. They try to give a UI to the different types of MCP authentication that exist, like STDO and and streamable HTTP. Right. But in the process they made j everything less intuitive than just doing what I did today, which is I Googled how do I add the to-doist mcp to codex and I copied a terminal command and I opened my terminal and I connected. Yeah. So that part needs to be rethought a little. They did a good job, I think, in Codex with the skills directory. They have a built-in visual gallery with some pre-baked examples of skills that you can add to Codex. And not just that, but Codecs is also able to automatically read skills that you have in other agents folders. Meaning it can read skills that you have in your cloud slash skills folder on your computer. So they're doing that sort of like, oh, we see that you're using another agent. We can also sort of import your other skills for you. OpenAI, since Codex is open source, these cur ated skills that they make, they're also open source. You can find them on GitHub. They have an open AI codex skills repo that you can browse. Funnily enough, John, they used to make a skill for ChatGPT Atlas, the browser that they have semi forgotten about. And they used to make a skill to control atlas and to like fetch your atlas tabs. In a recent commit to the repo, that skill is gone. They have replaced the ChatGPT Adas skill, which I still have on my computer because I didn't delete it, but they replaced it with a generic playwright skill. Now playwright, if you're not familiar, is a virtualized Google Chrome um MCP server thing that you can install to let a chat bot or an agent uh test web apps that you're building on its own. Right. So I don't know what OpenAI is doing with Atlas, but the basic. I wonder if it just is because Atlas is but I mean, uh playwright's been around so long, it has probably a lot more features than what Atl the Atlas one had. Probably. You can as of a few days ago in codecs enable fast mode. This is another fragmentation problem. Fast mode for codecs is not the same as you may recall Codex 5.3 Spark. Oh, right. They're not the same thing. Codex 5.3 Spark is a special, smaller, less capable model that OpenAI released on Cerebras the fast inference provider that they entered an agreement with Fastmode is a setting that is exclusive to the new GPT 5.4 model. Parentheses here. It's just called GPT 5.4. It's not called Codex 5.4. Right. So it seems that they are unifying the model names. They no longer they no longer have a separate codex model. I thought that the Codec the Codec Codex models were optimized for codecs and that that they're I also thought so. I also thought so. Okay. But then why did they launch everything and said oh GPT five point four is the new coding model also. Yeah I guess that's true. They did say fast mode fast mode goes, if I'm not mistaken , speed is increased by one point five X , but your credits are consumed at a two X rate. Great. So you're going 50% faster, but you're spending twice as much. Yeah. Right? Right. 100% more. Yeah. So it isn't it didn't seem it didn't seem that much faster to me when I tried today. It's a little bit faster, but it's not Cerebras faster. You know. So you're but also mostly just burning tokens for no reason then. Unless you're in a real hurry. Unless you're in I suppose the accuracy is less too, though, isn't it? No, no, no. No. No, no. I don't think they are dumping down the model when you're they're just asking you for more money. Asking you for more usage. But it's the same model. It's it's uh because I Spark is a completely different thing. Yeah. I would be curious to see if eventually we get a pro per GPT 5.5 on Cerebros model. Yeah, I mean that it's weird to me that they would both they would combine a dumbed down model, which would presumably would be dumbed down to make it faster with with Cerebrus instead of putting five four on Cerebrus and seeing what that chip can actually do. It must not have been enough for it to be I think since Serebrus I guess yeah, I guess Serebras is is a very small company. I don't think they have the resources right now. Because Serebrus they're not like renting NVIDIA GPUs, they're actually making their own chips. Yeah. And I think it's a the the problem is that they don't handle the load. Yeah. Yeah. Yeah . Uh fin ally I want to mention automations . Automations in codecs are kind of similar to schedule tasks in cloud , but they are more geared toward programmers . So they have a similar UI. You can s you can create an automation in codex for Mac to say every morning at 10 a.m. do a review of all my open PRs in this GitHub repo and give me a summary. The problem is that codex you can use it if you want as a general agent, but it's very slow. And the the the agentic loop inside of it as clearly not being designed for general productivity. I tried today to connect my notion and my to do with stem CP to codecs. And I tried asking, hey, can you give me a br iefing of what's going on in my Notion Daily Notes and also what's going on in my task manager today? It took forever, really. I spent five minutes thinking about it. Even with a low thing level. Oh wow. And and and so that's because like Codex is designed for analyzing code basis, reviewing code, writing code. It's not designed to be a digital assistant. Right? So I am not gonna use automations in codecs because I am not a developer. I don't need to run code reviews every morning. I don't need to, you know, check out my open PRs every morning or things like that. But if you are a developer, you can do that in the Codex app. And it's the only way I think to schedule something with Codex, with the Codex app. And yeah, that's essentially Codex. Is a the story with Codex right now is a lot less complicated than Anthropic and Claude and Claude because they're not doing personal productivity yet. That I think is the the n big missing item from the open AI offerings. They're doing a chatbot, which is insanely popular. Yeah. And they're doing codecs. It's interesting to me though, when you think about that too, because they're the one that's like angling to be the consumer model, whereas anthropic is a little more business enterprise in general, productivity oriented. You would think that tools like calendar and tasks and things would be a lot more important to open AI that they'd be on top of that. You know, the and this goes back to an old episode of App Stories. You know what's missing from all of this? The prosumer angle. Once again, once again, we have one company that's catering to the enterprise on one side, and then we have open AI on the other that's just doing friendly consumer things. What is the product for somebody like me or you? Right in the middle. No, I've thought about this too. I thought about this a lot. Is it open claw? But it's too confusing and it's not intuitive. Where's is it is a perplexity computer ? Maybe. And what type of maybe? I think it's closer to c I think Claude's closer . I think I think it's something in between perplexity computer and co-work. Yes. But the big problem with these two, and maybe we'll talk about perplexity computer next time, the big problem for these two , there is no mobile version, there is no mobile story. If I'm out and about and I want to do something, or or if I'm if I just got in bed and I for like oh I forgot to do something. I cannot do it from my phone. Well that's why I was so happy when remote control came out. Because it's not like I use it every day, but on the other hand, I'm not tied to my desk if I'm working on something in cloud code in the terminal. So I mean I think as we wrap this up, Federico, let's I I guess we should explain maybe which of the of these tools are the ones we're using the most. And I think for me with Claude, it's the chat interface. I use that quite a bit , mostly for research type stuff. And the terminal using Claude Code for building web apps and some native apps and other things like that. And then I'm using codecs to bounce some of my code off of Codex and then reintegrate it back with Claud Code when I'm finished. I think the piece that I haven't used a lot that I want to try out and use more is cowork because I've really enjoyed using Claude Code as a coding tool. And I think you know there's there's a lot of potential for that to integrate well with productivity tasks. I just haven't spent a ton of time in it. I tend to use projects more probably because I have a lot of recurring tasks that I have to do over and over, or I have long running tasks where I'm going to work on something for an hour or two today and then I might not pick it up again for a couple of days. So it's nice to have a project for that kind of thing. How about you? I mostly I mostly use Chat GPT and OpenClaw, which eventually we'll do an episode about OpenClaw, I guess. And I use Codecs a lot. I don't use cloud code. I mostly use Codex. I just I really like the nerdy, sort of like cold vibe of Codex, which is very different. Codex is very different, yeah. Codex is the the cold pragmatic nerd. Uh Cloud Code is the friendly, funny, approachable n erd. And I don't know. For some things I just I just prefer I really like codecs. Here's my problem. I mostly use ChadGPT and Codex , but I know in my heart that the tool that I want is co-work . It just kills me that it's not on the iPhone. It kills me that it'll get there. It'll get there. It'll it'll get there. I really, really like Clog Cowork . And I hope that they can make it faster, easier to access, easier to use. It's very early, but it's exactly the sort of thing where it's like it's an LLM and a bunch of connectors, but it can also use your skills and it can schedule tasks . Like that's the kind of tool that I want to use for work. But it's Mac only and it, you know. Yeah, well, I mean look, they they brought code over to the iOS and the iPad OS app. And that is even less of a good fit, I think, with those platforms than Cowork is. So I'm sure co co work will get there eventually. Like we said at the top, I mean I think Cowork has been out for like a month at most. It's not been, you know, it feels like forever ago in the AI world, but it really hasn't been that long. Yeah. Boy, I'm sure there'll be some new words next week that we haven't covered here. So we'll talk about the uh the the overhang and the and the harness again. Oh my god. Yes. Sure . All right, everybody. Thanks for joining us for another episode of App Stories and thank you too Vitally for supporting our efforts. And of course, you can find the two of us over at Mac Stories.net and on social media, where Federico is at Fetici, that's V-I-T-I-C-C-I, and I'm at John Vortiz, J O H N V Dou R H D W S. Talk to you next week, Federico.