Schedule AI tasks that run on their own

It is another week of the AI Academy, which means everything has changed in the world of AI. Steve and Dustin here. How the heck you doing, Dustin? Good. Just trying to stay cool on these hot summer days. Yeah. So summer's hit you. You're in Bakersfield, right? Yes. Yeah. So summer has hit with a vengeance. Yes. Quite hot during the summer months. So what is a hot day?

A hot day for me, I find it unbearable as soon as it gets above 85 degrees. We start to get into 90 degrees, this is the worst for me. But in Bakersfield, what we have to endure for most of the summer is 100-degree plus days, and that's in Fahrenheit, obviously. I don't know what that translates to quite in Celsius, but- Mid-30s. Yeah. It's hot, and unbearably so. So we not only have central AC, and it's a fairly decent

unit for our house, we also have a separate AC unit, like a portable AC, that we set up during the hot summer months to help cool down the house because it gets dangerously hot sometimes. We're talking 110, 115 I've seen- So if you had a power outage, you could be in trouble. We could definitely be in trouble. At that point, the emergency plan is go to a mall or go to somewhere indoor.

I guess a lot of people do that, yeah. As long as they don't-- I guess they have generators, wouldn't they? They do, yeah. Most of them do. A lot of them have solar setups now, so they're in pretty good shape. Well, my long-term prediction, my AI-based-- It's not AI-based, but kind of along the world of, like, I saw AI coming, which I didn't, so my predictions are worth nothing.

But I think we are going to see, within 20 years, an entire city underground. Especially places like the Midwest where it's really hot, where the storms are so severe, that they're going to be building the houses lower, which is far easier to cool, far easier to regulate the heat, but also not in the path of things like tornadoes and storms like that.

I just think that. And then you'll have most of your infrastructure underground. You'll still have your parks and stuff above ground, but the houses people live, there's a lot of reasons to live underground. There is, yeah. Absolutely. That's my prediction. We should probably get to the AI stuff, but one last little tangent here. I've been seeing this concept of concrete dome houses.

Oh, yeah. Geodesic domes? Oh, man. Those seem so cool to me. I don't know why. I just think it's super cool. And I also recently saw a guy built a home inside a cave, and I thought that- Inside a cave ... was super cool. Yeah. For sure. That would be cool. All right. Let's jump into the world of AI.

Steve is an Earther. Oh, is that what it's called, is an Earther? An Earther. Eep inside. Yep. Okay. So what's up? You've got something new, exciting happening in the world of AI in your world. Yeah. So, lots of news last week in the world of AI. Lots of drama that, thankfully, Magai users don't really have to care about. But this week was big for us.

So for the last, I would say, five or six months, we have been building towards this moment. Yesterday, we released something that we're calling Scheduled Tasks, the ability to schedule things in Magai to happen on a one-time or on a recurring basis. And this was always the goal. From day one of Magai, before agents were a buzzword,

I saw that eventually AI, from this chat-based conversational interface, would basically be able to run most of our businesses and most of our lives through autonomous and hands-off-- At the time, I didn't have the word agent. We didn't have this word agent. But through AI autonomy.

But up until this point, it hasn't been good enough. It hasn't been reliable enough. It hasn't been trustworthy enough. The connections didn't exist. It was too complicated. And so we've been slowly, over the years, waiting. And then OpenClaw hit, and I immediately saw the dangers. I immediately saw all the things that were going to go wrong with it and have gone wrong with it. And so we've just waited for the right moment, for the right connectors, the right technology pieces, the right capabilities, the right safety, the right reliability to all come into alignment.

And finally, it did. And so for the last five months, we've been building this feature out so that we can let our users now create AI workflows with natural language. No specialized language needed, no specialized technology, no specialized interface needed. They could just tell the AI what task they need it to be done, at what time and what date, and on what interval, and the AI can just build it, do it, and you can be completely hands-off.

And so we're going to show you this new feature today. This is a big leap forward in everything you do with AI, and it's going to require a lot of people to just see examples. It's not going to be super obvious what you might use this for in your day-to-day work, but I want you to start thinking about what are the repetitious things I do every single day. What apps do I have to log in over and over again that could probably be offloaded or automated in some way by AI? And I'll walk you through some of my examples today, and hopefully you can maybe

get some ideas for yourself. And so let me just jump right into it I'm going to open up Magai here. And so if you are a Magai user, you may have noticed that there is a new button on the left, and that button is just labeled Scheduled. And this is, just like your chats list, is a list of the history of conversations that you've had. This will be a list of the ongoing automations you have, the scheduled tasks that are happening that you have created.

And before I jump into that, I'm just going to first show you how easy it is to schedule something. So I've got a brand new chat here, just like any other time I've had a brand new chat. And let's say that every morning I want the AI to send me an inspirational quote into Slack. I want it to message me in Slack with something inspirational. So all I need to do first is go to the Attach, go to the Integrations tab, and make sure that I have Slack turned on. So right now, I've got Slack Bot turned on. Okay.

So I know that that integration is live, and I can simply say, "Every morning at 10:15 AM, I want you to send me an encouraging, motivational quote." Okay. And I've got Claude Sonnet selected just because that's the one I typically have selected by default.

Claude's going to look up the current time, and it's going to schedule the task automatically. Didn't have to touch any buttons, didn't have to configure anything. I just literally told it what I wanted it to do, and it understood what to do. So daily morning motivational quote, June 18th, 2026, 10:15 AM, daily. "You're all set, Dustin.

Starting tomorrow morning, you'll receive a fresh motivational quote every day at 10:15 AM. Each quote will be thoughtfully selected around themes of leadership, innovation, and personal growth." Obviously, it knows a little bit about me because of our workspace settings, which we've talked about before in the academy. "Along with a brief reflection on why it hits different. A great way to kick off your day with the right mindset. If you ever want to adjust the time, change the theme, or pause the reminders, just let me know."

Now, what I could also say here is I can give it some follow-up. Maybe I want to change something about it. Let's say I want the task to send me the quote with a Hulk Hogan-style commentary to get me hyped for the day.

I'm not sure if many of you know, I was a big professional wrestling fan when I was a kid growing up, and Hulk Hogan was one of my favorites. So what it's going to do, it's going to find that task. Looks like it failed this time, so it might have just been an error there. But typically, this will work. Maybe I'll just retry it and see if it finds it this time.

Yay for live demos. Okay. Well, we can go and view this task anyway. So you can see this little element here denoting that a task has been created. I can click the View Task button, and it's going to take me over into my Scheduled Tasks list, and it's going to show me in the list here that task. Now, the middle section here hasn't changed. That's because there's actually nothing here. There's no tasks that have been completed, so no new chat for that task has been completed.

But what I can do here is I can click on the three dot menu, and I can click Edit Task. And now what's going to happen is it's basically going to show me the instructions that are built into that task. And it's fairly simple. Share an encouraging quote, blah, blah, blah, blah, blah. It's got the AI model selected, the persona selected. It shows me the date, time, and I can make changes to all of this if I want to. I can change the instructions here.

It's basically just a prompt, right? A prompt with connected integrations, which I can double check and see it's got Slack selected there. So this will run at 10:15 every single day until I decide to turn it off, and it will send that message to me in Slack.

And there we use whatever you can think of that is something you want to happen on a regular basis can potentially be done here through this integrations feature. Sorry, something is happening on my screen. Okay.

Sorry, it tripped me up there for a minute. So let me give you some more examples here. So- How does it get the message to you? So, again, I've connected Slack. Okay. Oh, yes. That's right. You mentioned that it was in Slack. Yeah. But you could tell it to email it, you could tell it to text. Yeah. So any of the integrations that we have here in Magai are usable by the AI. So anything you've connected in here, your Gmail, your Google Drive, Spreadsheets, Outlook, really anything. Any of these tools are now accessible to

create and build into these scheduled tasks. And so I'll give you another example here. One that I had set up for us is the... Where did it go? It's in my communication. Sorry. Email triage. So this was sort of the first one I thought up Where I want it to, every single day before I get to my desk, I want it to go through my Gmail and get rid of all the newsletters, get rid of all the promotional stuff.

Go ahead and archive that for me. I don't even want to see it. And then anything that is from a human that needs my response, I want it to summarize that, flag it, and prioritize it, and then send it to me in Slack, because that's what I check first, and kind of give me a brief of my email inbox for the day. And so if I open this up, you'll see that it is a task, daily email triage, morning brief, and it runs at 8:15 a.m.

daily. And now I didn't write any of these instructions. Okay? I want to preface that with I did not write out these very complex and step-by-step instructions. I just had a conversation with the AI, and it built this task for me, and I'll show you exactly that process. But here's the task that it built. You are performing a morning email triage for Dustin. Follow these steps carefully.

Review inbox. Pull all emails currently in Dustin's inbox. Archive newsletters. Archive promotions. Identify important emails. Look for emails from real people that are not newsletters or promotions. Leave them in the inbox exactly as they are. Do not apply any labels, stars, or flags. For each important email, determine a brief one to two sentence summary of what the email is about, whether it needs a reply from Dustin or is just FYI, no action needed, a priority score, high, time sensitive, or requires prompt action, medium, should respond within a day or two, or low, informational or no urgency.

Six, send a Slack message to the bot comms channel. Now what I did here is I actually created a Slack channel specifically for my AI to communicate with me in, because what this does is it actually sends it as an app. And so I can have it send it to the general channel. I can have it send to whatever channel I want. But I wanted one for me personally to have it all in one spot where I can view all my bot communication messages.

And so it's got the channel ID here with a clean morning triage brief formatted as follows. Morning email triage brief. Archived X newsletters, X promotions. So it's going to be the number of newsletters and promotions it archived. Inbox, emails needing your attention, sender name, subject, summary, reply needed, yes or no, priority, high, medium or low. Repeat for each important email. If there are no important emails, say, "No important emails at this time. Inbox is clear." So that's going to run every morning at 8:15 like clockwork. So it can basically trim out my

email inbox for me and let me know which things are important for me to do. Now, also, I basically just repeated this. So I have another one that runs every single day at, I believe, 3:15. Yep, 3:15 PM. So it'll do that again. It's a separate task, the same exact job. So like twice a day, my AI is doing email triage for me. Now how did I build that?

Well, let's go into our... Let's see. Was it this one? Yes. Okay. So I'll ignore this first part because I was trying to understand how the Slack bot works. Let me go right to where I started prompting it. Okay. Okay, here's the prompt. I want to set up a daily email triage process using the AI.

Every day at 8:15 a.m. the AI should review my inbox, archive any newsletters, archive anything that looks promotional, identify important emails from real people, flag them, summarize them, each one briefly, send a Slack message listing what's in my inbox for me to review, send a message about this to the bot comms Slack channel we use. Okay? So plain English. I just explained exactly what I wanted to do, and look at that.

It created the email triage task. You're all set. Here's what it does. It's going to run at this time. Email integration, flagging behavior. Now as I read this, I go, okay, so right here it says, "Different email providers handle flagging slightly differently." In my mind, by flagging, all I meant was I just want you to put it in the summary. I didn't mean to actually tag it or flag it or organize it in any way. So I wanted to clarify that.

So I said, first of all, I didn't actually mean to flag the emails in any way apart from just writing a summary about them and sending them in a Slack brief. The task shouldn't attempt to put any labels on anything. Just leave them in the inbox. Now, here's an important thing. AI can do exactly what we tell it to do, but if we just tell it exactly what we want it to do, there may be things that we can't think of that it is capable of that we aren't aware of. So that's why it's important to have this kind of a follow-up question. How else would you suggest we improve

this daily email triage workflow? I want to spend as little time in my email inbox as possible. Now, you might remember if you were here for our prompting lesson, I used this phrase at the end of conversation, what are we missing? That was my magic phrase that when I think I'm done with something, I ask the AI, what are we missing? And it can uncover things that I didn't catch or I didn't think about.

This is just the same exact thing, except applying it to tasks. How else could we improve this? How would you make this better? And so it said, "Good catch. Let me update the task." And it updated the task. Now, here are my suggestions for improving the workflow to really minimize your time in the inbox. Suggested improvements.

Add an end of day sweep at 5:00 PM. So it suggested that second email triage task for the day. Include a needs reply indicator. I hadn't thought of that. Auto unsubscribe from newsletters. Now, I'm pretty sure it cannot do that. I don't think there's a method for it to do that, so just know that the AI still isn't fully aware of all the things.

Justin, I just got somebody downstairs that I have to deal with. They're doing some work on the townhouse. No worries. I'll keep going. But you can keep going, can you? Yes. I'll be back in a moment. Okay. The next thing it said, add a waiting on tracker if you've sent emails that you're waiting for a response. AI could flag threads with no reply. I didn't need this. I often don't need this for me and my email.

I send an email, it's in their court. If they don't reply, that's on them. Priority scoring. I hadn't thought of that. Rank important emails by urgency. That was the AI's idea, not mine. Weekly inbox zero report. This is another one. I don't necessarily need it, so I ignored it. So I just said, I like the end of the day sweep, so let's go ahead and create a second daily task for this, but make it 3:15.

I also like the needs reply indicator and priority scoring, so add those as well. I don't need the other suggestions. Be sure to bake all of these into both triage tasks. So it did. It made the update to the morning one, and then it created a second afternoon one that's identical. And then it gave me a sort of a breakdown of all of those things that it built into that. So that's one simple example.

Building out an email triage system. If you're like me and you like to keep your inbox tidy, this might be a great workflow for you. Now, here's another one. So let me see, which one should I do next? So we did the motivational quote thing. It's basically just for you for fun. But you might have something like maybe you want to do industry research.

Maybe you want to on the regular, maybe every week, you want the AI to go scour the internet for news relating to your industry or your business and then bring you a brief. So this is instead of you having to go scan all the news articles and sort through all the news yourself, why not have the AI do this automatically once a week for you and just kind of hand you a brief, just like a high level personal assistant would hand you a brief.

So it's something I wanted to do, but I wanted to do it a little bit differently. And I don't know how many of you are Saturday Night Live fans. I'm not particularly, but I do love the Weekend Update. I love the idea of taking breaking news stories and kind of putting a punchline behind them. So I wanted to perform this task, but I wanted it to go out and find all the AI news and then put a SNL Weekend Update spin to it. And so I had it create this task.

I want to create a weekly task that will search for the week's biggest AI related news. I'm primarily looking for stuff that directly relates to Magai or any of the AI providers that it gives users access to. Summarize the main points of these news stories. Creates a Saturday Night Live Weekend Update style version summary of each story that is hilarious. Finding a way to poke fun while delivering the actual truth of the story while being funny.

Delivers this SNL style summary to the Slackbot integration to post to the Botcoms channel. It should do this task every Friday at 3:00 PM. So I'll have a little news brief that is kind of funny every Friday at 3:00 PM delivered straight to me via Slack. And so it created this, gave the details. Now, here's something that I highly recommend with this.

You can schedule the task and just trust that it's going to run as intended, right? But what I realized is it might be a good idea to do a dry run. Let's just try this first, see what the AI does, and see if there's any improvements we can make to it. So I literally just said, "Let's do a dry run just to make sure it works." And so I saw that it looked up the available tools that it had.

Great. Slackbot is connected and active. Now let me get the current date, search for this week's AI news, and then find Botcom's channel ID all at once. So it looked up the time, used Slackbot to find the channel. So it found the Slack channel it needs to go to, got the channel ID. Now let me search for this week's top AI news. And so it searched the web here. Got plenty of great material. Now let me compose the full SNL Weekend Update style message and fire it off to Botcoms. So it sent that message.

Dry run worked perfectly. Here's what just landed in Botcoms. And then it gave me a quick summary of what went in there. Every Friday at 3:00 PM, it'll fire automatically with that week's freshest material. The format is locked in and ready to go. Want to tweak it? Blah, blah, blah, blah, blah. So now that I've seen how this ran, I can see here's the steps that it takes. Now, my follow-up here is based on what I observed from how it executed that task. This is good, but there's some things I didn't like about the output in Slack. So I went over to Slack, I saw the message, and I saw there were

some things I didn't like about it. The format of having the punchline in an inline code block, because that's how it did it, styled it like the headline was code, and I was like, I don't want that to do that. Format of having a punchline in inline code block isn't pleasant. Let's be explicit about the format that only uses plain text and H2 headings for the headlines of each story. Number two, there are no sources cited for these stories, and that means we can't fact check the output.

It's pretty important to do when you're working with AI. Add some instructions here that will ensure we have sources cited for story verification Based on this dry run, is there anything else you can add to the task to make it run more efficiently? Maybe include the channel ID already, so the AI doesn't have to run a search for it. So we noticed that it had to run a search to find the channel ID.

Why would we waste tokens or waste words or waste usage every single time this runs if we can already provide that channel ID within the instructions, right? We're saving tokens and we're saving usage every time we run it by having it not have to do that step. Right? So it said, "Really solid feedback, Dustin. Let me pull up the task and current time." Now, just for those who might be wondering, why does it have to check the current time every time?

Most people take for granted, they just assume the AI knows exactly what day and time it is. The AI does not have that information by default. Any product that you use where you ask it what day and time it is, the app itself or the people who are making that app have dynamically inserted the date and time. They've either done that in the persona or in the system prompt, or they've kind of hidden it in your message.

This is adding additional tokens to every single message, in order to make sure that the AI always is aware of what date and time it is. We found it to be far more efficient to just have the AI run a quick little run when needed. So it doesn't always need to know exactly what day it is to fulfill your query. So we've built this little tool that says, okay, if it's a time related query or task or something the user is date-time related, then we'll run that quick check to check the current date and time. So that's all that is.

That's why it looks up the current time and it doesn't automatically know this. No app that you use, the AI never knows the date and time. It has to be dynamically inserted somehow. So that's just a little side note there. "Looks like the scheduled task isn't showing on the list. Yes, possible," blah, blah, blah. So it didn't find the task at first, and then it finally did. It had to check again. AI does fail. It's a computer, it's not perfect. Got it. Found the task. Now I have everything I need to update with all three improvements in one shot. So it updated the task, and then it told me what improvements it made.

Hard-coded that channel ID so it doesn't have to run that check every time. Task fires every Friday at 3:00 PM. So every Friday, this task is going to run automatically for me. Now, these are examples of recurring ongoing tasks, but you don't have to make everything a recurring and ongoing task. Sometimes you just need to schedule things once. There's a thing you need to do, and it doesn't need to happen now.

It needs to happen a day from now or two days from now. Let me show you an example of something that's sort of a one-off. If I can find... It might be in here. Here we go. So one of the things I like to do is I like to create Instagram content with myself as the subject in these pretty absurd AI visuals.

And if you go to my Instagram now, you'll see pretty much all of them are these weird AI images. And the weirdness of the photo is the attention getter, right? You're scrolling through and you're like, "That's kind of odd." And it catches your attention. And then whatever the content of that post is, usually something of value that I'm sharing that's related to the visual.

And so what I'll do is I'll batch create these visuals in Magai, of these absurd photos, and I'll have the AI help me create these captions based on my thoughts or what it is I want to communicate the message. And so what I did is I had a brainstorming session. I came up with all these weird images, and some of them I'll use, some of them I won't use. But I went through and just kind of batch created them. Once I was done creating them, though, I wanted to get them on my Instagram, right?

The next step is content's made, now we need to get them over to Instagram. And typically what I'd have to do is I'd have to download the photo or send the photo over to my phone, then open up Instagram, upload the photo into Instagram, then come back over to Magai, copy the caption that we refined, and then paste it over into Instagram, right? So that's a little bit of back and forth.

But if you'll notice, you go into your integrations, Instagram is one of the integrations we have, which means Magai can post directly to Instagram for you. And so that's what I had it do. So I started with the first one just to test the Instagram integration. I said, "That's perfect." Or actually, I think we did the caption first. Okay.

What was the post content for the first image where the water was filling the bottom of the room? And then it reminded me I didn't want to scroll all the way up. Here's the copy that it had written for it. I reviewed it and I was like, "That's perfect. Post it with the attached image to my Instagram." And looked up the tools, found Instagram, Instagram connection's active. Sent the post to Instagram, publishing it. It's live.

I never had to leave Magai to post this to Instagram. It posted immediately. And then I thought, and this happened literally a few hours before we released the ability to schedule things. And so I was like, "Okay, well, now we've got the ability to schedule things. Let's schedule the next Instagram post with the image of me sitting in a restaurant with a bunch of robots."

Now, here's what was really magical for me. It said, "Let me grab the current time." So it found the current time. Before I schedule it, here's what Sonnet asked me. It had a couple of questions. When do you want it to post? Date and time, e.g. tomorrow at 9:00 a.m., Friday at 11:00 a.m. Which caption do you want to use? I can use the one I already wrote for the concept, the breakfast with robots image, or do you want me to write a fresh one to pair it with? So it already remembered, it already knows.

Remembered is the wrong word. It already sees what it wrote before. I don't have to go back up myself and copy and paste it somewhere. It sees it. So I said, "Schedule it for the optimal time based on your knowledge." I don't know what the optimal Instagram time is. Maybe the AI has an idea. And yes, use the original option you created for this concept. So again, looked up the current time.

Based on the best practices for Instagram engagement, the optimal posting time is Wednesday morning at 9:00 a.m. Pacific daylight time. Midweek mornings consistently hit peak engagement windows. Now, I don't know if that's true. I don't know what the optimal times are. It's 9:00 a.m., that seems fine, right? The idea that it maybe had an idea, I didn't have an idea. That looks fine to me. Good enough.

And it scheduled it. I, again, never had to open up Instagram. I scheduled this yesterday. I don't know what time it was that I did this, but 9:00 a.m. It was supposed to go out today at 9:00 a.m. Sure enough, I checked at 9:00 a.m., I got a little notification that it published without me having to be there. This effectively replaces all of your social media scheduling tools.

And it can also replace- A $30 or $40 subscription replaced with the AI subscription that you already have. So, the other thing that those social media tools do is they listen for us. Mm-hmm. You'd be able to have it pull your social media on a regular basis and flag anything that you should be paying attention to.

That's a really good question. Now, the tools that are connected in Magai, all of them have a set of actions that they can perform. They're not opening the app like you would. They're not going and viewing everything. They don't have all access to the application. Kind of like a Zapier or a Make.com. They have certain actions. Webhooks have been designed for it.

Right. So what you can do is, if you're unsure of what an integration can do in Magai, all you have to do is go to your integrations, turn that integration on, and say, "What actions can your Instagram integration perform?"

Because the AI can see all of the actions, and it can tell you what those actions are. And so it's very easy to just go in and ask it that question, "What can you do?" It's going to look it up, and before you create any automations or schedules or anything, you can just get an idea of what is it actually capable of. So, it can publish content. So it can create a container. This is just how it posts an image, apparently. And it can publish media.

It could get your media, so it can go search and see what's on your Instagram. Get a specific media object. It can get carousel children, children meaning all of the images in the carousel. It can get posts you're tagged in. It can get account insights. So, I don't know if this is true.

Some of this might be hallucinated. Let me just push it here. Double- Yeah, that seems- And only provide- Robust ... what tools and actions the Instagram connection has.

Let me just push on this a little bit because I don't think those are all- I don't think it replies to all. It'll generate replies automatically. Do you? No. No, not on that. That would be bad. Okay. Get media, get comments. Yeah, that's what I thought. Seven tools. So get user media, so it can retrieve your content.

Get comments, so it can actually fetch the comments from a specific Instagram post. You just have to know which Instagram post you want it to get those comments from. It can get the replies, meaning it can go read and deliver you, show you what the replies are. It can post a comment on an Instagram object, but again, you have to give it the exact one that you need it to comment on.

It can't just go find stuff to comment on automatically. You can reply to comments. So if you find a specific comment, if it goes and finds that comment, it can reply to that comment. And then it can delete a comment. So if it's posted a comment and you're like, "No, delete that," it can delete it. So those are the things it can perform. You can do this with any of the tools inside of Magai.

You just say, "Tell me exactly what actions you can perform with this tool." And again, we have over 130 different tools in here that you can go through and sort of brainstorm and figure out- And this is the baseline of what they're going to be able to do. So, in your example of replacing a social media dashboard, it'll replace some of the features in the social media dashboard.

It won't replace all of them. It won't replace the listening skills yet- Right ... at this point. But it will be able to at some point in the not-too-distant future. You can say, "Please audit Twitter or audit Instagram, and any time there's an at mention of me, let me know." Correct. Yep. Exactly right.

So, I've got some more examples here. We're running low on time. So it's got my Instagram post scheduled here. So just so you understand how to navigate the Scheduled tab here, what you're going to see in this tab when you have scheduled a task is the top level item here that shows you the task. And below that will be the individual runs of that task. And so I don't have many here, but if I look at my daily motivational business quote one that I did just for fun. So you can see run number one ran yesterday at

9:03, and if I click on that, I will see exactly what ran. It will show me what happened. So think of these as conversations that are started without you having to be there, and you can jump into that conversation at any time. So if I were to have a follow-up question and I wanted to continue this conversation, I could jump right into this run, and I can start talking to the AI about that task or whatever is involved in that task. And I can continue that conversation just like it was a normal chat inside of Magai.

And then you can see run number two. And as these runs start to accumulate, you'll just continually see the newest one at the top underneath the sort of the hierarchy of its original task. You can click here and, again, you can edit the task. So if there's any changes you want to make to those instructions, you can. If you want to change the AI model, if you want to change the persona, change up the integrations, these are all fully editable at any time. You can run it now if you want to hurry up and run it now, even though it's scheduled for the future. You can rename it.

You can pause the task if you don't need it anymore or you want to hold off on it for a while. Maybe you're going on vacation, you don't need this to run. You can pause the task. You can pin the task, just like you can pin chats, and it'll stick to the top of the list here. You can add it to a folder. So tasks also have folders. And so you see I have a folder here for communications and I have a folder for business ops, and this is just a way for me to keep them a little bit more organized.

And then you can share them with your team as well. Just like anything else in Magai, you can share them with your team members, and you can also delete that task, which will obviously automatically pause it, stop it from happening in the future, and then completely erase it from your history and from the Magai database. Folders, just some simple options here. You can rename it, pin it, and share it. And of course, share permissions, if you share this with your team, obviously anything that goes into this folder will then be shared with your team as well.

So very similar to how chats are managed, but just a different type of chat. They're all chats- When- ... just different kinds. When will this be rolled out? It rolled out yesterday morning, to my surprise. It's on our systems now. So let's- Everybody has access right now. Let's buzz through a couple of quick questions. Alex is going back to your email synopsis. "What would you suggest for the email procedure if I don't have Slack?" So he's talking about where it's going to report to.

Yeah, so there are several communications tools in there. Pick your preference, really. We're working on getting WhatsApp support and text message support. We don't currently have those, but eventually you'll be able to have the AI text message you with a summary or WhatsApp you with a summary. Good. So that, once again, that's an evolving list for sure. Yeah.

Um- It's really all about what do I want to... What is my ideal scenario? How do I want to communicate with this AI? Now, you'll also just be able, if you want, you don't have to have it delivered to you anywhere. If you just want to go into that, you can go into your scheduled tasks and just check them every morning. When there's a new one, there will be a little indicator there that, oh, this is new and you haven't seen it yet. So you can do that- You could use... It attaches to Keep or to Google Tasks or to Apple.

Google Tasks, yeah. So you could just have it create a note in your notebook. You could have it create a summary almost anywhere. So you can be as creative as you'd like, I would say. Oh, how... Oh, sorry. "How does one determine urgency?" So when the AI was reading your email, how does it recognize what is important versus what's not?

Good question. I really don't know. How do you determine urgency is, I think, a better question, and because the AI is really unpredictable. It's a machine and it makes decisions. Yeah. So I imagine over time- It's hard to know, I guess ... over time, you could train it. If you received an email, you could say, "This was really important. Make sure all of these emails make it through onto the top of my list next time."

Yeah. Yeah. It's similar to how Google does it, really, in how Google sorts and determines what's priority mail versus what's updates, et cetera. Mm-hmm. I'm not sure what Agatha is referring to here, but she asked if you can go over the scheduling tool. Is it an additional charge? I didn't quite catch the answer.

Nope. The feature is available to everyone. The usage is consumed, just like usage is consumed in a normal chat. These are really, if you abstract it, if you kind of back out, it's just a conversation like any other conversation. They're just scheduled. They just happen on a schedule. Okay, we've got a couple of questions in the regular chat.

For future reference, folks, it's easier if we have all of the questions in the question chat area, in the question and answer area. But Robert is asking, "What about videos? Can you upload and schedule videos?" I don't know. I haven't actually tried videos for Instagram.

When I did try to upload an image to my prompt, it seemed like it couldn't find that image link because it basically needs to find a URL for the media object, and then it needs to Use that for the Instagram integration. What it could find was the image that it generated. So it generated the image, and it has that link to where it's stored in its database, and it was able to use that.

Maybe if you have it generate an image for you or a video for you, it might be able to find that. We'll have to do some investigation there to see exactly where the disconnect is. Tom's asking if this ends up using a lot of our credits. Again, it's just like a conversation, so it eats up just as many credits as you would if you did that in a real-time conversation. It's no different.

So for the average stuff, very small amount. Especially because most of this stuff is one-off. Every task that runs is individual. It's not building on the previous parts of the conversation. Every new run is a brand-new conversation. And so you're not building these very long, high-context scenarios, unless of course, if the task builds a bunch of context, obviously.

But so yeah, just think of it like brand-new chats every time, and whatever that chat consumes is what it would normally consume if you were to do it in real time. Avi got on a bit of a roll. He says, "Does this work both ways? Can Maga send you a message every time a specific person posts new content?" And there's several questions that Avi's asked in this area, but you answer and then I want to add a comment.

Yeah. So you have to understand the difference between reactive and proactive. Okay? So saying I want it to react to something that happens somewhere else is far different from a scheduled action. Something that is scheduled is the AI says, "At this time and at this place, I do this thing."

Has nothing to do with the other platform and whatever happens on the other platform. That is something completely different and something we haven't built out yet because it's very complex and nuanced and will require the user to do a lot of technical stuff that most users don't want to do. And so until there's a way to do that easily, we won't be doing that. What you can do is say, "Every day at this time, go check for this thing," and it will check for that thing, but it cannot receive signals from another place that will trigger an action.

And really, this is where we already do have technology that serves. It's having a subscription. It's subscribing, and if you're following somebody in the feed, you should be getting their content. Certainly for things like Sub Stack, you will. Some of the other algorithm-based ones, you might be missing the odd thing, but you can just check their feed. But I think your explanation was on point for that. And I think people are, again, Bob's asking, or

Brad is asking, sorry, "Can the app move things from a to-do list from an email or website?" Yeah. So it has a to-do list connection. But that's not a scheduled task. So we're talking about different things. Well, it can be. Well, I guess you could. Oh, well, yeah. You could create that summary and make it a to-do list. Right. Yeah. So part of that triage could be if there's anything actionable in here that I need to do, add it to my to-do list.

So Keith is asking, "Does the scheduled task have any memory? For example, if I tell it to scan a Google sheet and perform a task based on any new lines on the sheet, will the task know which line is new from the last time it ran?" Currently, no, because these are all brand-new conversations. But what you can do instead is I think the Google sheet itself, maybe if you had a column in there that would say a date and time in it of when that row was

added, then it would be able to say, if there are any new rows between this day and this day, then do it. But it has no recollection of past runs because obviously, if we did that, the amount of context building in every single time would consume more and more and more and more usage. So every time you run it, that run's going to become more expensive. Day one, it costs you $0.05.

Day two, it costs $0.10. Day three, $0.15. Day four, $0.25. And exponentially, that cost will start to rise every single time it's run. So it's not a good idea to have that. So it's better to have some way of in the sheet that will indicate what day something happened. Avi would like to know if you can duplicate a task and then modify it, so you keep the original but also have another task.

Yes, you can. Actually- And Brad, I answered this to you in text. Is there an Airtable extension? Yes, Airtable is on Dustin's list of tools that are available. So we don't have the duplicate feature in here right now, but like I said, you can edit the task. It's literally just a prompt. That's all it is. It's a prompt. So you can hit a new schedule, paste the prompt in, set the integrations, and set the times, and then the schedule, so you can set the date, and date doesn't really matter if

it's repetitions of 5:00 PM, repeat weekly. You can recreate them in a matter of seconds. Excellent. And with that, I think we'll call it a day. That was exciting news. Congratulations, Dustin. That's going to be something that I know we will find and discover new, unique, and exciting applications for.

So it's yet another step on the AI journey. So thanks to each and every one of you today. And that's it. Stay warm. Stay cool, Dustin. Stay cool. I can stay warm. Cool is the hard part. Okay. All right, folks, till next time. Have fun storming the castle.