Ep 874: Scheduling AI: how to easily make AI work for you in Claude, Gemini and ChatGPT (Replay)

Welcome to the Everyday AI Podcast. My name is Jordan Wilson, and for the past 3 and a half years, we've put out more than 800 episodes. Yet one of the most common questions I get, I didn't really have an answer for. Where do I start on the Everyday AI Podcast? And that's why we started the Start Here series. And with fall now back in full swing, the Everyday AI Podcast is going back to school and playing back the entire Start Here series. From front to back, we've hit pause on our normal Monday to Friday programming to run back our most popular series ever for the next 30 days. We made the Start Here series for beginners and AI champions alike. So whether you're just trying to get a grasp on large language models or grappling with the best coding harness for multi-agentic workflows, the Start Here series covers it all. Plain language, no jargon, and easy to follow along each day. So make sure to subscribe to the podcast and. check back each day for new insights, day by day. The series is a culmination of spending more than 10,000 hours covering generative AI over the past 3 and a half years. So you don't want to miss a single episode of the Start Here series. Let's get into it. Welcome to the Everyday AI Podcast. My name is Jordan Wilson. And for the past 3 and a half years, we've put out more than 800 episodes. Yet one of the most common questions I get, I didn't really have an answer for. Where do I start on the Everyday AI Podcast? And that's why we started the Start Here series. And with fall now back in full swing, the Everyday AI Podcast is going back to school and playing back the entire Start Here series from front to back. We've hit pause on our normal Monday to Friday programming to run back our most popular series ever for the next 30 days. We made the Start Here series for beginners and AI champions alike. So whether you're just trying to get a grasp on large language models or grappling with the best coding harness for multi-agentic workflows, the Start Here series covers it all. Plain language, no jargon, and easy to follow along each day. So make sure to subscribe to the podcast and check back each day for new insights. Day by day, the series is a culmination of spending more than 10,000 hours covering generative AI over the past 3 and a half years. So you don't want to miss a single episode of the Start Here series. Let's get into it.

Maybe you aren't fully agent-pilled like me yet, and that's okay. It could be because your company hasn't given you the green light to go full desktop agent yet with Codex or Claude Code, or maybe you're still just getting comfortable with the web interfaces of Google Gemini, ChatGPT, or Anthropic's Claude. But don't worry, You can still get a taste of that sweet proactive agentic life where you start your morning and a good chunk of your work is already done for you. Well, how? If you can't be using these autonomous desktop agents, how in the world can you get that? Well, they're called scheduled tasks. So think of scheduled tasks in today's large language models kind of as the gateway drug to being fully agent-pilled. So if you haven't or can't, make the big desktop leap to running agents around the clock, you're not out of luck because these scheduled tasks are a legit secret weapon. So secret, in fact, that sometimes it's hard to even find it, and the capabilities aren't even fully known or documented. So we're going to be diving into that on today's show as we put AI to work on Wednesdays. Yeah, It is our day where we go hands-on, but here is the big picture. Scheduling is one of the most underrated features in today's large language models, at least when it comes to being on the web. Because most of your routine use cases that you're using large language models for right now, well, you're probably repetitive and you're probably doing a lot of the same things over and over and well, they can just be done for you. Instead of you just being a button pusher doing the same type of task, whether you know it or not, You can actually just have those be done automagically and wake up and have them done. But I think scheduling tasks has gone through so many different name and interface changes in all the platforms. It's sometimes a little confusing, or you might not know how it works. And there's actually a handful of hidden features when you're scheduling inside of specifically ChatGPT and Claude. So on today's show, that's exactly what we're going to dive into. And here's what you're going to learn. You're going to learn the basics on why scheduling is so important and how to easily do it in the main platforms. You're gonna know what features work well in scheduling and which ones don't. And I'm gonna tell you the best use cases for scheduling to take a step toward a more proactive and agentic workflow. All right, let's get into it. What's going on? My name's Jordan Wilson. Welcome to Everyday AI. If you're new here, we do this every single day. It's your unedited, unscripted livestream podcast and free daily newsletter helping business leaders like you and me Keep up with all the feature updates, the advancements, the UI changes. How the heck do you do this? What's going on? I do all that work for you. I tell you what works, what doesn't, what you need to pay attention to to grow your company and your career. So if that's what you're trying to do, sweet. Starts here with the livestream and the podcast. But if you haven't already, please make sure to go to our website at youreverydayai.com. We're gonna be recapping not just the highlights from today's show, but everything else you need to know, especially the other AI news and developments to stay ahead. And hey, Actually, a quick little announcement. It is back to school time. Yeah, here in the US, it's that time of the year where everyone goes back to school. But maybe you're not a student, or maybe you don't even have kids in school. But guess what? We're all going back to school together. So we're going to be doing some learning together on Everyday AI. So there's actually this thing that we've been doing for a while called our Start Here series that we kicked off earlier this year, and they've been some of our most popular shows ever. It covers all of the basics of not just, you know, yesterday's AI, but today's AI and where it's going. So we're going to be running back the entire series starting this Monday with the first volume. So a little break from our normal Monday to Friday scheduling. But I guarantee if you stick with this series from volume 1 through volume 30 and you put the work in, you'll know more than about 99% of the other people in your company when it comes to AI. So no jargon, no fluff, no computer science degree required. Class starts Monday. Don't be late.

All right.

Let's get into today's episode though, when we're gonna be talking about scheduling AI and how to easily make AI work for you, right? In Claude, Gemini, in ChatGPT. Yeah. 'Cause I don't know. I think there was a certain time before this was a feature, I was like, can we get this please? Right? I remember seeing back in like 2024 when there was rumors, right, of these kind of scheduled tasks or, you know, scheduled actions. They've gone under a lot of name changes. I remember when these were just leaks and rumors and I'm like, oh my gosh, this will completely change the way the world uses large language models. And guess what? It didn't. Mainly because I think a lot of the companies kind of changed things around so much, or they buried these and made them hard to find, which is actually Interesting, because I think that the, uh, kind of the agentic adoption was slower than it maybe could have been if not— if, if these, uh, you know, kind of scheduled options or features would have been earlier and front and center. Uh, because if I'm being honest, I was a very early adopter of these, like the day it came out. I remember, I think it was ChatGPT Tasks that was the first to come out. It might've been in like late 2024. I started moving all of my workflows over to ChatGPT Tasks. And I think that process, you know, I always talk about this concept of unlearning and rebuilding. I think that was actually one of the most important steps for me to prepare myself for, you know, moving most of our processes and even my thinking, my creativity, my brainstorming into an agentic workflow because, you know, you really have to start deconstruct, you know, to deconstruct your entire process when you're like, okay, wait, I can wake up and have all of this stuff done for me, right? You start, I think, thinking less in terms of what do I need to do next? Instead, you start thinking, what is the end goal, right? And then you start working your way backwards from that and saying, how can I have this end goal waiting for me in the morning? All right. So let's go through the 3 main platforms. All right. And, you know, if you don't know, the 3 main platforms anyways are Gemini, Claude and ChatGPT. So yeah, technically Copilot, it's just too messy to do this, right? I can show you the online version of Copilot, but almost every single one of our listeners, if you're using Copilot, you're using it on the desktop. The reason why I'm not doing Copilot, well, number one, I'm on a Mac. But number two, even if I was on a, on a PC, it is entirely too difficult in terms of permissions, in terms of jumping through all of these different hoops. So it's almost like To show you these things, it would kind of be a waste because you would go out and be like, "Oh my gosh, this is amazing!" And you would try and do it, and it wouldn't work. I actually did a show previously on the online version of Microsoft Tasks, which I think are really good. So you know, I'll make sure to drop that one in today's newsletter. But let's get into Google Gemini. So here's the quick rundown on Gemini's version, and these are called Gemini's scheduled. Actions. So let me just be honest, not that good. They're hard to find, they're buggy, it's extremely limited in terms of features and the user interface. And you also can't schedule via normal chat. And unfortunately, when you try to do that, you're actually going to get some false positives. So the other downside is if you— once you do get them scheduled, if you don't use them or check in on them, they will be automatically Disabled, which stinks, right? So yeah, a lot of downsides. So here's how you're going to find them. So if you just go to gemini.google.com, make sure you're logged in. So if you're looking on the left-hand panel, they're not going to be there. All right. So, you know, fortunately, ChatGPT and Claude did move these scheduled tasks for them. That's what they're called, front and center on the main navbar. So Google Gemini, No such luck. So you have to actually look in your settings and there you'll see these scheduled actions.

All right.

So once you do that, um, then essentially you are met with a very simple and straightforward enough, um, scheduled action box where you just essentially type it in. So you give it a name, you give it instructions, you say how often, and then a time. And that's it. Right? It's an extremely simple interface. No bells, no whistles. And the downside is, well, it's not that useful, if I'm being honest. All right. So next up, and I am gonna— I do have an example, you know, one kind of simple prompt that I did. And then at the very end, after I show you these examples. So yeah, this is one FYI, and this is one is technically prerecorded because it's gonna be— it was a little difficult to, you know, go through all of my different systems without accidentally doxxing myself. And also these are scheduled, right? So I didn't wanna schedule these for, you know, oh, 7:45 and then, oh, I'm done before then, right? 'Cause, you know, the livestream kicks off at 7:30 AM Central Standard Time. So it was gonna be a little tricky to try to get all these to schedule live and in real time. So yeah, this is one of those shows for AI at Work on Wednesdays, normally very hands-on doing live prompts, all that. So this is more of, you know, hey, I'm gonna show you how to cook, put it in the oven, and guess what? It's already done. Got the cake ready. All right, so next up, let's talk about how ChatGPT's scheduling works. So theirs are called scheduled tasks. And again, I actually did a dedicated show on ChatGPT's scheduled tasks. Let me see what, what one that was, because it was actually fairly recent because they went through some major changes. All right, so this was actually on July 1st, so episode 8. So I'm going to go over the high levels, high-level points for that. But it was actually a pretty in-depth episode. I remember I spent a good amount of time on it. So if you are a heavy ChatGPT user, like a lot of people are, make sure you go check out episode 810. But here's the gist. So tasks now support more agentic app actions. And that's the great thing. They also work inside of projects and you can use memory for more personalized automation. That part's Huge. So you can access them directly from the sidebar like I talked about before, or you can go to chatgpt.com/scheduled and you can just create one by describing what you want ChatGPT to do in natural language. Like, you know, you can say, scan my Gmail daily at 7 AM and tell me what's important. All right. But there's a lot more to uncover here because this is where ChatGPT really stands out. So a little Um, kind of hack. Uh, you can go through the normal, uh, interface to set this up, or like I said, you can set it up conversationally. But what a lot of people don't understand is if you set this up conversationally, you actually have, um, some options that you don't have if you were to set it up in your normal scheduling routine. Uh, that's because— and I'm kind of looking on my, on my other screen, uh, right now— so if you go into the scheduled section, um, you're gonna see it's, it's gonna look like a fairly, uh, a fairly normal prompt box inside of ChatGPT, except there's one or two things missing. And the big thing here is the model selector. So yeah, it's, um, I believe what happens in these scheduled tasks, it just is going to default to the last model you use, which may or may not be a good thing. So I recommend just doing scheduling a task via natural language in the main kind of new chat box and just say schedule this at this time and ChatGPT is smart enough where it's going to do that. The main benefit there is, well, you can choose the exact model you want. So maybe you want to use, you know, GPT-5.6 SOL extra high, right? At least you have the ability to configure the exact level of reasoning that you need for something. maybe it's a little simpler, right? Maybe it's more of a personalized search and you don't exactly need the most, you know, the smartest model. But if I'm being honest, it's one of those things where unless you're running out of limits on a ChatGPT Pro plan, which is pretty much impossible, it's— you might as well just use the most powerful model at this because it's going to be scheduled. It's going to be happening while you sleep. So throw as much juice as you can at these things. So that's a little tip there with the scheduling tasks inside ChatGPT. The other one, a little hack, and for whatever reason, this doesn't always work because maybe it is truly a hack. But if you go into the new Work tab inside of ChatGPT, okay, and I know that there's a lot of confusion about this, so I'm not going to dive in too deeply. But essentially, there's a new version of the ChatGPT desktop, which is called ChatGPT Work. Which is kind of the same thing as Codex. But then if you go to chatgpt.com, you'll see at the top now there's a new chat and a new work section. So the easiest way to tell the difference, because people are like, wait, when I go to chatgpt.com, do I use chat or do I use work? Well, the simplest way is, well, use chat when you just want to chat or talk to ChatGPT. But if you want something more like an agent, use work. So The cool thing here, and this is again, I've never seen or heard anyone else doing this and it doesn't work every single time, but if you get it to work once when scheduling your task, for me, it works over and over. But the cool thing in the work mode, it actually has its own virtual and dedicated computer and browser, which is awesome, right? So if there's anything that you normally have to do that requires a ton of interface clicking around, right? It could be from a website that you rely on every single day. It could just be a lot of, you know, web searching that you can't just do from reading a static page, right? Because a lot of times when these large language models, you know, quote unquote, use the web, they essentially just scan the code of any given page, which is helpful in most cases. But what if you have to, you know, scroll down to load more? What if you have to, you know, click around a bunch and add filters, right? So think of all the web browsing that you do that maybe you haven't yet handed off, um, to an agent. And there's probably a reason for that. And this is awesome. So, you know, one quick example of something like this. I love Google Trends, right? So there's something in Google Trends, you go, you can put all these filters on, and then you have to scroll down. I like to see like what's trending in big tech, in AI. And I realized, well, why am I Doing that, right? So that's something I went into ChatGPT, clicked the work, did just a natural language, gave it the URL to that. You know, once I put all those filters on the Google trend, and then I said, "Hey, every day at 9 p.m., go to this URL, scroll down to the bottom, see, you know, use everything that you know about me and the Everyday AI audience, and see which of these trends that." our audience would care the most about, then go out and research all those trends. Tell me why they're trending, right? Like yesterday, something about the iPhone was trending and I didn't know why, right? So it's gonna go and it's gonna, you know, help me stay more in touch with things that matter to me in my career without me having to go through and click and then go out and research. I just wake up and I know, right? So that's actually something inside of ChatGPT that is very unique to it because You know, Claude can't do that. You know, we're not talking Copilot. Copilot can't do that. And obviously Gemini's scheduled actions can't do that. So big hack there. Another one that's very unique to ChatGPT is having these open-ended tasks. So this is something that OpenAI didn't even really highlight in, I think it was technically late June that this came out. We covered it like the next day. I think it was like July 1st, right? But you can now just say something like, You know, check my competitor's pricing page and tell me when it changes. You don't have to say check it every Tuesday or at this PM. I don't know how this magic works, but it just does, right? Uh, so that's really cool. You can also just give it open-ended tasks. All right. Um, so now I'm showing on my screen how that works. So on the left-hand side, you have your, um, uh, scheduled tasks there. Then you can go into the scheduled tasks, and like I said, you still get, uh, you know, you can use all your connectors, your skills, all of those drop down. But the downside is you don't have that model selector. But you can still create a scheduled task from the scheduled section, or you can just click new chat and just type it out in natural language, give a time, a frequency, etc. Then you can go in and edit them. So if you— when you do click that scheduled portion, or if you go chatgpt.com/schedule, you can go in and edit it. So if you do give a prompt in natural language, it's going to turn that into a set of better instructions for the model to follow. But then from there, you can make any edits. So if you remember the kind of old school GPT builder, you could just chat with ChatGPT and it would turn your chats into a set of custom instructions. Kind of the same thing happens here under the hood. So you can go in and update the end scheduled task. And you can also— there's a dropdown menu for when you want it to repeat, the frequency, the time, and an end date. All right, let's keep going and talk about Claude's. So Claude's are powerful but a little confusing. Okay, and here's why. Because you have your normal scheduled tasks. So again, this is all just Claude on the web. Okay, so I'm not talking, you know, Codex Desktop. I'm not talking about you know, Claude Desktop with Claude Code and Claude Cowork. I am talking about Claude on the web. But if you didn't know this, Claude on the web also changed a little bit, just kind of like how ChatGPT did, how you have Chat and Work. So now Claude has Home and Code. All right, so if you go to Home, you can do scheduled tasks that way. So you essentially can do a scheduled task in normal chat or in Claude Cowork, even more confusing. So there's technically like 3 different ways you can set this up. But the other way is a routine. So if you click on the Claude Code button, you can set up a routine. And I'm going to quickly tell you the difference here and the pros and the cons. For Claude's scheduled tasks, that's just the— it is very much the equivalent of the ChatGPT version, which is like chat and work, right? So if you need something that's just more straightforward, you know, just use the, the home. If you need something that's a little more technically agentic, you can click the Cowork tab. But you can also, just like you can in ChatGPT, you can create a scheduled task just conversationally with Claude, or there is a manual setup. The cool thing about Claude is you can work in a project. You can also set an approval level per task because as an example, All right, if you're giving it send actions for the first time it runs, you might want to approve those. I don't fully understand if you're trying to schedule, right? I kind of scratched my head on this because I'm like, okay, why would you actually want, uh, you know, to like manually approve everything if it's on a schedule, right? Like if you have 10 of these things running overnight, why would you not want them auto-approved? So I guess in theory it's a nice feature. I just literally don't know point of it. But anyways, it's there. Maybe if you just want to run it once in front of you to see how it works, maybe that's what it's there for. Otherwise, it's extremely counterintuitive to schedule something, but then, oh, it's not going to fully run because you had to approve it. Anyways, you can also choose the model, but you cannot unfortunately choose the reasoning level. You can set the frequency. It doesn't have a custom version like ChatGPT, but you can also use connectors and skills like you can in ChatGPT. But then you have these scheduled routines, and I actually did this. Let me see, this, this show was a little, a little longer ago. Let me see when I did this, this scheduled routines. Yeah, this is live, y'all, or, you know, unedited, unscripted, at least. So let's see when I did this show. So this was Okay, this was episode 756. Um, this was in April, but these are all, uh, still very, very relevant because, um, you know, we went over— this was technically through the scope of Claude Desktop, but it kind of works the exact same. But we talked in depth about the differences, uh, between these scheduled tasks and the routines in Claude Code. But Difference on the routine, right, is well you can actually sync those with your desktop. So if you set up a Claude scheduled task, it's not going to sync inside Claude desktop. So the only way you can do it is with a Claude code routine, and then you have to choose the cloud version. So on Codex, right, on the desktop version. If you're using ChatGPT Work to schedule, it'll all sync. So why would you use Claude's scheduled routines? And I'll just give you the 30-second version of episode 756. I think they're really, really good. They're more powerful. And I think a lot of people, especially non-technical people, are gonna look at like Claude routines and be like, Claude code and be like, I'm not touching those, right? You're gonna say there's enough with skills and all these other things and co-work, right? That's technical enough for me. The Claude routines are super simple. The, you know, you don't even really have to use Claude code to go in and use Claude routines. Um, the, the biggest, um, kind of benefit is you can use, um, essentially an API token as a trigger. So what that means, it's very simple. You can choose that, and I have a screenshot of that later so I can, uh, walk through Oh, I have it right here. So essentially, if you choose that as an option, you're going to get essentially an API token. All right. And then you can go to something like Zapier and you can use that to essentially connect anything in Zapier and have that be a trigger to start something in a Claude routine, which is really cool, right? Because as an example, you know, in ChatGPT, they actually used to have this, I believe, in Codex, but then they kind of simplified and unified The, the Codex and ChatGPT, they used to be called automations, so now they're just called scheduled tasks. Anyways, this used to be available in Codex. I don't believe it is anymore, but this is great in Claude routines. So as an example, you can set anything that happens in Zapier as a trigger to start, to start a routine inside of Claude Code. Again, you don't have to be super technical. You can literally just give it to Claude and be like, walk me through this, right? But that's super valuable because otherwise every other thing has to start at a certain time, right? So maybe, I don't know, you have a super highly valuable form on your company's website, right? When someone comes in, it's a potential million-dollar client and you want to be able to start doing some legwork right away when it comes in. Obviously you can't do a scheduled task on that because you have no clue. But in this case, right, you can set this up inside Claude Code, takes like 30 seconds, choose that as an option. You're going to get this key. Then you go into Zapier, you know, whatever it is, if you're using a Google Form, Jotform, whatever, you know, then you're going to get this key, plug it into Zapier, and then anytime that web form comes in, then that routine is going to start inside of Claude Code once it kind of sends to that API token. So a nice unique feature there inside the schedule All right, so that's kind of the gist of it. And I'm gonna go through now and I'm gonna share my screen for our livestream audience and podcast audience. You can always go to our website at youreverydayai.com if you wanna see the video version of this. Nothing super visual, and I'll kind of explain what's going on, right? And, or you can always check the show notes and we do put a link directly to today's episode. All right, so let's talk about what our test actually was. So this one was simple-ish, right? The downside is because Google Gemini is not good at even using its own connectors like Gmail, I was trying to find something that I thought in theory could be a very You know, apples-to-apples comparison and something maybe realistic. And also, this is something that I wanted to build anyways. So I said, all right, let's just throw a simple schedule prompt and talk about the use case of this. So, you know, I said every day at blank o'clock, right? Because I spaced these out a little bit, but I gave it the same thing. So I said every day at this time, please put together an interactive either site or artifact or canvas because they all call them something different. But this is essentially an interactive visual and in some it worked and some it didn't. And I'll show you that. So every day at a certain time, please put together an interactive visual that shows the current stock prices for these AI stocks: NVIDIA, Alphabet, Microsoft, Broadcom, Meta Platforms, Taiwan Semiconductor, Micron Technologies, AMD, Oracle, Palantir. Also, please include an up-to-date factual bullet point recap for each that explains the biggest news for each of these companies. So this is something that I don't do it every single day, but I probably every single day end up checking at least 2 to 3 of these companies in terms of what's happening with their stock. Not because I have stock in all of them and I'm tracking it from a portfolio perspective, but from an AI news perspective, a lot of times a company stock will shoot up 5% or drop 5%, which is very rare. And a lot of times it has to do with something that's happening in AI. As an example, Alphabet, the parent company of Google, announced a major AI leadership change a couple of days ago, and their stock dropped like 5%, which for a multi-trillion dollar market cap company is an insane amount of money. Right? So sometimes I just want to be able to see at a glance what's happening with all these big AI companies. And, you know, essentially if I see a 3% up or down, that's kind of my threshold. Then I know that something meaningful is happening probably in AI around news. So this is something I kind of wanted to build for myself. But the thing is, well, it refreshes all the time. So that's the good thing with this being a scheduled, right? This is now technically an agent that's going out and doing this every single day for me and updating it so I don't have to. All right, so let's go ahead and look at the results. Let's share my screen here. All right, so Google Gemini first, and Google Gemini kind of failed. All right, so what's interesting— so I did say use canvas mode, um, it didn't. So, um, I do believe— I could be mistaken, but I do believe when you used to have the scheduled actions here inside of Google Gemini. I thought you used to be able to use some of the other features, specifically the canvas mode, but in this case it didn't work. Um, so that's interesting. It did, however, go out and do the rest. So it actually fired at the correct time. So I did this one at 5:15 PM, right? So after the markets closed. Um, so it got the correct prices for Nvidia, and then it gave me the key news, Alphabet, Microsoft, etc. So it did fire at the correct time, it followed instructions, um, it's, uh, you know, got the correct prices, but it didn't put together a visual, uh, that I wanted to use in canvas mode, right? I could go down here, um, and I could say, right, I'll actually do this just so we can actually just make it at least give Google Gemini a chance, because otherwise I'm just ragging on Gemini this entire time. So I'm going to say, you know, using all this info. All right. So, yeah, unfortunately, I'm going to have to put together a 2-step one on this. So this is where I also wanted to, you know, test out and say, hey, what other features are available when you're scheduling a task? Because sometimes features or modes make all the difference. It's the difference between, you know, using your data and not using your data, right? Like, spoiler alert, one of the reasons I'll— I've never really used Google Gemini's scheduled tasks is because they don't use your connected data, right? So it's like, for me, it's not super useful. And in this case, it wasn't able to use the mode for Canvas. All right, next, let's go ahead and take a look Uh, let's take a look at Claude. So, uh, same thing. Um, I did tell it to use an artifact. It, however, didn't. So it just wrote some strict HTML. But I don't know, I'm not going to say it's a failure because it looks good enough, right? It actually put together a very clean, uh, you know, it says AI Stocks Daily. It's interactive, right? It has a light mode. A dark mode. It gave me very clear ticker symbols, the percentage that they're up and down on the day, right? Like here I see Google is down 3%, so I can click there and see what happened, right? So it said, you know, Tuesday it fell, you know, 3.8% amid a rotation of high valuation mega-cap tech. It says investors remain uneasy about AI spending. Right? As their CapEx guidance was about $200 billion and their CapEx was up 100% year over year. So this, for me, this is all really helpful stuff to read, right? So when I'm talking about AI things, I, you know, can ground it in a little more recent data. There's also this button that it says, you know, expand all news, collapse all news, which is really nice. So it didn't technically make it like a shareable artifact, Um, but overall, you know, it, it did, it did pretty good. Um, so, uh, Gemini, you know, probably failure. Uh, you know, uh, Claude got a passing grade, but not ideal. But let's take a look at ChatGPT. So, my gosh, this one's good. All right. So yeah, if, if you're not looking, um, At the live stream here, this is one of those ones where you know sometimes I'm like, oh yeah, it's cool. This one's really good. This one's really good. All right, for a zero shot prompt, right? I read it to you super simple. Did a fantastic job. So I used ChatGPT sites, which by the way is a freaking cheat code. Like if you're not using ChatGPT sites, it is in ChatGPT now. It used to only be available inside of Codex. It used to only be available. For team plans. Now it's available even on the web and it's available on individual paid plans as well. If you're not using ChatGPT Sites, I'm just saying this, you're literally missing out. I make them all the time, right? I have custom skills to make a lot of them in kind of a templated way that really work for me. I send them to my wife, right? Because you can make them public. Anyways, let's take a look at what ChatGPT Sites did. First of all, visually fantastic, right? It gave me this breakdown. It says 4 of the 10 advanced, 6 of the 10 declined. It gave me what's called an AI basket pulse, which is pretty cool. So it said overall it was down 0.6% across 10 companies. So I can see overall, how are these things moving? Gave me the winners, the losers, right? I can click on the biggest moves, the winners first. I can create a watchlist, which is pretty cool, right? I can toggle this little star on and off. And then when I click on any of these, it then brings all of the news that matters. So let's find another. We already talked about Google, or no, actually let's click Google and see if it gave me similar reasons that Google was down almost 4%. So it says Alphabet reported Q2 revenue of $120 billion. Up 24% year over year. Search and other revenue rose 17%. Google Cloud revenue grew 82% and backlog reached $514 billion, with management attributing the acceleration to AI infrastructure and AI solutions. So I will say this, it pulled some older news, right? So if I were to redo this, obviously the ChatGPT one did the best, but it pulled in some older news. I don't know if the Google one, it didn't give me the date of the news, but I'm guessing this news was maybe a little more recent. I'm not sure. But from a visual UI/UX perspective, it's just much better, right? I'm actually going to go through and, you know, probably make one or two changes in this and, you know, make it mobile optimized. And the good thing is, is then And all I do, you know, on my phone, I can essentially save this as an app and then, you know, I can save it to my home screen, you know, and every day when I open it, it's going to reload and bring up the freshest information. So yeah, overall, I think they all had their own strengths, right? Let's take a look going back over. Sure enough. We had another error, another error in Google Gemini. That's fine. I really try to give Gemini in canvas mode, which I've always loved their canvas mode. I think it's really good. There we go. So now it's working. Is it? Okay, interesting. So it went a completely different direction, gave me kind of bar charts, you know, prices, quarter 2 growth, CapEx. So, you know, it actually looks nice. Some nice features and functionalities. Not exactly what I was expecting, but then, okay, when I scroll down, I think it does a pretty good job. Let's read Google. So it says executive recap summary, although the Google, it's not right, right? Because it says it's up 1%, almost 1% for the day, which I don't believe it was right now. I have to double-check to make sure. Yeah, Google was down almost 4%. So Google was wrong about the actual day. So that kind of ruins it, right? So I'm not even going to look at the rest because if you get the facts wrong, right, you always got to check these things, right? But yeah, Google was actually down, you know, 3 point, you roughly 7%, which the other models got correct and Google unfortunately got it wrong. So even though, you know, it did finish the dashboard with multiple prompts and get a fail, Google, we get a fail. All right. So let's wrap up here and talk about my takeaways. So overall, Gemini, not useful at all. It struggles using its own connectors. Essentially, the only thing that it's really useful for is is to customize web search for you. But even like you saw here, when in my— again, my one example, it pulled in incorrect information. So I haven't used Google's, right? Aside from when I test these things, I haven't used Google's scheduled actions. I hope they improve them, right? We'll see when we get a— whether it's a Gemini 3.7 Pro, Gemini 3.5 Pro, Gemini 4 Pro, whatever. Hopefully they bring some updates to the Gemini web platform. They did go through a UI/UX update, so hopefully these scheduled actions get some love as well. But overall, I'm not using them, and I would say you probably shouldn't either unless Gemini is literally the only thing that you can use at work and the only thing you really need is like a customized Google search, because otherwise really not useful. All right. My main ChatGPT takeaway is it's really good. But there's a lot of unique benefits like the, you know, the anytime scheduling, right? Just, hey, go check when this happens and it somehow figures it out without you having to schedule it. Using work mode with the virtual computer and browser is really good. The custom schedule the others don't have, like if you say like every other Tuesday, right? You can't do that in Claude or in Gemini. Having an end date is a unique feature as well in ChatGPT because maybe as an example, You want to run it every single day for 2 weeks, but then you don't want it to keep running and clog up your feed. It's great. I think ChatGPT, their scheduled tasks more than any others, are great at contextualizing your history or contextualizing your scheduled task in your data and history, right? I forgot the technical update for ChatGPT's memory. I think it's called like Dream V3 or something. It is so much better, so much better than, you know, anyone else out there. It's not even close. I again don't know how OpenAI does it, but that actually really comes into play when you're scheduling these things, right? Especially when they're of high value. You want to be able to bring in all of that other context and history from everything that ChatGPT or Claude knows about you. So unfortunately, right, like if you're doing the Claude routine, which is really helpful. Claude Routine has no clue what you've been doing in Claude Chat or in Claude Cowork, so you can't wrap it in that extremely important business context. All right. ChatGPT, you can also use your connectors and plugins and skills, which is really cool. And then like I said, being able to tap into the full power of ChatGPT. So both in the reasoning levels is extremely important and helpful, or being able to use the sites feature, right? Having that sites feature that just updates all the time. Amazing. All right, so now let's talk about Claude, the home or the chat scheduling takeaway. So more straightforward, but it works, right? Unlike Google, you can use all your connectors, which is good. Being able to set the permissions, even though I don't fully understand that, if you want to run it once as a test, maybe that's helpful. Overall straightforward, nothing, you know, there's no standout features in the kind of home or cowork scheduling. But there is in the other version in the routines. So in the routines, there's a lot of, you know, I already went over the kind of the 2, well, 2 unique features. I didn't mention the other one. The one is being able to kind of have that API token and have that act as a trigger, which is very unique in terms of what you can do to schedule right now inside of Claude Code routines. Like I said, don't be scared of Claude Code. You don't even have to use anything else in Claude Code. Right? You can just Think of it like a normal chat, but you can kick off a routine based on literally anything that happens on the web. Just go on Zapier, you know, take that API token and say anything that happens in the 6,000 platforms that Zapier connects with, send it to this API token, and then you can kick off any routine. A great, a great use case and something unique for all of the others when it comes to scheduling your work. And then the other one, which is a small one, Inside Claude, Claude Code routines, you can also set routine-level messaging alerts, which sounds like, okay, why does it matter? You know, in ChatGPT and Gemini, you can set notifications for all of them, but not at the individual level. So one of the reasons why that's helpful, right, is you can, as an example, for a certain routine that runs, you can send a Slack message, you know, which is Pretty helpful.

All right.

So like I said, there's pros and there's cons. I think overall, um, ChatGPT is the winner with a close second to the routines scheduling of Claude. I wish Claude would just be able to unify those somehow, um, and bring some of those better features in routines into their scheduled tasks. But, uh, hey, if you split it up, you can still get the power. But yeah, unfortunately it doesn't kind of have that cohesive memory, which I think is so important to wrap your scheduled tasks up in. Otherwise you may spend more time going back and adjusting them and making them better anyways. All right, so what should you schedule as we wrap here? Well, I would ask Codex or ChatGPT work first, right? Uh, unfortunately, Claude Code doesn't have cross-chat or cross-cowork code memory, um, but, uh, unless it commits something to memory. But I would go through and spend time seeing what you actually do manually inside of these platforms. From there, Find those repetitive tasks. That's what you should be scheduling. The other thing, kind of a little hack here, is look at your browser history or give it to an agent and see what your most viewed websites are, and then tie those websites to a deliverable and then schedule a task for that, right? The biggest thing is being able to do these things that are time intensive, whether it's research, um, for the ChatGPT work hack that I told you, clicking around on a bunch of websites, right? See where you're spending your most time either manually pushing the buttons in Gemini, in ChatGPT, in Claude, turn those into scheduled tasks, or just where are you spending a lot of your time on the internet? The internet's a wasteful place. Don't spend time there. All right. And then last but not least, the other one is start with a simple triage dashboard, right? That's updated every single day. It's obviously only going to work in ChatGPT and Claude. But, you know, bringing in your, you know, connectors for mail, calendar, Slack, Teams, your CRM, your major KPIs, your most used connectors, MCPs, etc., being able to automatically have an agent go through, pull all the most important things, and essentially have a start here dashboard, right? It's a very, you know, simple use case. Maybe that's the 2026 version of like the 2024, you know, improve this email, right? Probably in 2026, it's essentially having a custom dashboard and triaging everything and telling you where to start. But if you are newer to the game and you haven't already done that, I think that's a great place to start with a scheduled task or action. All right, that is a wrap, y'all. So remember, if you haven't already, we're gonna be kicking off going back to school with our Start Here series starting Monday. So don't be late. I'll see you then. And if you haven't already, please go to youreverydayai.com, sign up for the free daily newsletter. Thanks for tuning in. We'll see you back tomorrow and every day for more Everyday AI. Thanks, y'all.

And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going. For a little more AI magic, visit youreverydayai.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers, and we'll see you next time. And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going. For a little more AI magic, visit youreverydayai.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers, and we'll see you next time!