WEBVTT

00:00:00.000 --> 00:00:40.820
There's a new way that the best Cloudco users are 10 x ing their output, and almost nobody is talking about it. Because once you've done the basics, dialed in your agents' skills, and optimized your memory setup, then what comes next? Well, it turns out that the real power users are building tiny personal apps that make their AI agents seriously more powerful. So in this video, I'll show you why micro apps may be the next big unlock for your workflow, share eight ideas for my setup that have genuinely changed how I work, and give you the playbook for when to build your role. And if you're new here, my name is Jay. I spent over a decade working with brands you probably know. I've been in AI since my masters in data science, and now I run our AI solutions practice in one of the largest AI communities globally. Let's get started.

00:00:43.940 --> 00:01:06.200
Before we dive into some examples, I think it's important to realize why these micro apps are becoming more and more useful, especially when it comes to doing agentic work. Well, the two core reasons in my view is that they essentially let you communicate with your agents just much better versus just having to stick to a chat interface where you're simply just limited to text and the amount of visualizations

00:01:06.200 --> 00:01:07.320
and interactivity

00:01:07.320 --> 00:02:04.730
that you can convey from just one chat window is honestly quite limited. And the other thing is that it lets you visualize things better. So if you're dealing with any sort of design or any visual type of work, then creating micro apps specific for your use case is going to help you a lot instead of having to go through a wall of text every time, which with a lot of these AI models, they can be quite verbose. And, of course, the key thing that allows us to create micro apps in the first place is because these AI models are now so powerful that you can pretty much easily spin these up and iterate to create personal applications that would be good for your own line of work. And so just to give a bit of guidance, because obviously not everything needs to have its own application. Right? But if you think that a particular task or job to be done that you have in your work needs to be more visual or needs to be more interactive and is not that complex to build, then I think you should consider making a micro application in order to make that work easier,

00:02:04.890 --> 00:02:05.770
more efficient,

00:02:06.010 --> 00:02:28.335
and also more fun. And, of course, the best way to discuss this is to show some examples. Now what I did personally is to just store all of the micro apps that I created in this console that I'm calling Rubrik just to give it a familiar name. Now some people call this their operating system, maybe their command center or mission control. But at the end of the day, basically, what this is is an HTML page that stores or centralizes

00:02:28.335 --> 00:02:30.530
all of the apps or interfaces

00:02:30.530 --> 00:02:46.455
that you can create in order to work with your agents better. Now I'll be showing you the ones that I use daily personally because I'm sure they will be useful in some shape or form to you as well. So the first application that I use almost on the daily is this app that I call Rubrik Generations.

00:02:46.535 --> 00:02:56.695
So Rubrik Generations, basically what it is is it's similar to Google Flow or Higgs Field. You know those tools. They have this view where you can actually see and centralize

00:02:56.930 --> 00:03:00.370
all of the generations like images as well as videos

00:03:00.450 --> 00:04:53.130
that you generate through models like Cling or Cdance or GPT image too. Commonly, if you're just using a vanilla instance of ClaudeCode or Codex, you usually get these images in line in the session, which is not always the best experience because if you need to go back to a certain image or let's say you're trying to find a b roll that you created from a few weeks ago, sometimes asking your agent to grab it for you or source it from your files is much harder versus just seeing a wall of the images and the generations that you did and just scrubbing through it in order to find the one that you were actually looking for. And the way it works is quite simple. So let's say I want to generate an image infographic explaining Google's announcements at Google IO this year in 64 bit pixel style, and I also ask it to use GPT image too. And then from here, you can see that I also invoke this master generation skill, which basically just teaches Cloud Code for how to reach for the APIs of these image and video generation models. If you want to peek at that skill in particular, you can actually find it here in our skills finder view, which, again, is another sort of mini app that I created just to find our skills better so that I could show it to peers and clients. And then if you scrub through this, you can see under the hood, if you're interested, the services we're using is Google AI Studio for the Google models, KeyAI and File AI, which we've featured a lot in this channel. And these services just aggregate a lot of other models like Clang, Cdance, GPT Image two, which we have all connected to Cloud Code through this skill. And then the rest of the guidelines in here are just continuously being updated by my Cloud Code instance depending on my preferences. And then what I just instructed my agent to do is that whenever it generates a media, it just goes to this location, this folder called Robo Generations, which is where this mini app of ours, this rubric generations is linked to. So to go back to the example prompt that we sent earlier, you can see the Cloud Code said here that it should now appear in the dashboard generations tab, which if you go to the images, you now have that 64 bit infographic style of the Google IO announcements.

00:04:53.370 --> 00:07:12.175
And this is quite useful. Right? Because normally, if you are just talking to Cloud Code through this chat interface, it is not very visual. Usually, what it would do is give you, like, a link that you need to open in your file folder in order to preview that image or video. But if you have a generations tab like this, then you can instantly view your images and even store them in one infinite canvas. And the other thing that I like to do in case I'm iterating on a image or a video, if I go to the reference tab in here, that just includes a lot of the logos and a lot of the icons and some images of myself because, obviously, I sometimes need to put myself in our thumbnails. But what I can do here is just click on this, let's say, anti gravity icon. That would now copy the path to that image. And so if I just paste it here, and if I say to our agent to now use this anti gravity logo here in this file path to replace the icon in the anti gravity section, and I'll just fire that off, that is just a good way to give instruction to your agent on any specific edits that you want applied to, let's say, this image. So, ideally, the result that we would get is it would understand that we are referring to this section here, this number two, to replace this icon of a robot with the image of the anti gravity icon. And there we go. So the second iteration of that image, because we sent it the anti gravity logo, it was able to understand just by natural language that we want this particular icon edited and replaced with that anti gravity logo, which was the reference that we just gave it. And lastly, the other things that I use this application for is if I click on this top right models button, that just gives me a nice overview of what are the different aspect ratios, resolutions, and modes of these different generative models that I link to Cloud Code so that I can have a nice reference. And also, if I want to add a new model down the line, let's say the new Google Omni, then that's easy to do as well. I just ask Cloud Code to research for Google Omni's APIs, which right now it's not yet available, but when it is, then that's something that we can add here. And it will take care of mapping out all the technical details so that we can use that model into this mini app that we have. The other thing that is also important is if I go to the styles here, you can see that I have a few creative styles that I constantly use here. So there's a few pixel styles that we use in a lot of our b rolls. This cinematic liquid glass, um, I've used quite a lot as well. And let's say I want to use one of these styles, then similar to the reference, can just click on it, send that to Cloud Code. And when Cloud Code has submitted that, we now have this infographic but now in that paper craft style.

00:07:12.610 --> 00:10:47.865
And if we want to add new styles here, again, it's very easy. For example, this is actually a style that I wanted to add to this sort of dot framework that the Cloud Code has in their YouTube channel. So what I'll just do is send a screenshot of that, and I'll just say to please add this as a style in rubric generations to analyze the image and give it a proper name and design framework. And when that is done, it basically just added that style into our dashboard in here. And whenever we want to use this specific design system, then we can just copy it and send it to Cloud Code just like what we did earlier. And by the way, if you're interested in going from just using AI to getting paid for it, then check out the Roboneggots community down in the description. We've got founders in there who landed their first client in weeks, live build sessions where we create this stuff together, and the actual templates behind what I just showed in this video. The community is also the reason these lessons get made, so see that below if that's for you. Now the next application that I'll show you is this one that is called Rubrik Docs. And what it does is very simple. I just have it linked to my workspace, and it just searches and parses for all of the markdown files within my workspace folder. And this is quite useful for any type of knowledge work really because in some shape or form, you need to write text, maybe articles, maybe scripts, and so on. So to give an example, let's say I have this simulated prompt where I'm saying that I'm doing a podcast about Google IO's announcements with Google CEO, for example. What I ask Cloud Code to do here is to do some deep research on their announcements and based on everything you know about our channel, let's say, give us top 10 questions we can ask for this interview. So when it did that, it now saved that to rubric docs. And if we try to find that markdown file in here, it is really fast because this whole application is anyway very lightweight. And there's a couple of things that we can do here. One is we can, of course, edit it ourself if we want. But the main thing that I use almost on the daily with this rubric docs tab is I actually give comments in line to my agent straight from this rubric doc. So to give an example, let's say for this model's bullet points that it gave us, I wanted to expound further on what these are about and also to explain them in layman terms so that we can also communicate it even for nontechnical folks. So if I just highlight the whole thing, click on comment, and we'll just give a comment here to give a paragraph for each point here just explaining what these are in layman nontechnical terms. And so that comment is now saved. And let's say we go through each of these questions and what it thought so far, and we have a particular build for, let's say, number 10. We can just give a comment in there. Let's say we want to make the phrasing of this question more casual. We can just save that, and that comment should now be available in this document. And back to our Cloud Code session, I'll just say that we made some comments in there. Please revise. And here you can see that it read the comments that we made as well as where we made it, and it was able to revise both of those bullet points. So if we just hit refresh here, you can now see that for the models, it gave us a nice paragraph to also explain this in layman nontechnical terms. So it's it's saying here that Gemini 3.5 Flash is Google's new everyday brain and so on and so forth, which is exactly what we asked for. And if we go to number 10, our comment here was to make it a bit more casual as a question. And what it did here is to revise it to say, you've been at the wheel through mobile and cloud now agents. What's something about the job today that just wasn't the thing ten years ago? So I think it got the comment there properly and revised it accordingly. Now I know what you may be thinking, which is if I already have Obsidian, then why do we need to have, like, a separate micro app just for viewing markdown files? Well, I also tried Obsidian myself, and even though it has this flashy force directed graph view that sort of shows your second brain, Let's be honest, a lot of people don't really use this graph view except for this fancy timeline animation that you can start through this wand, which, uh, shows sort of your second brain forming,

00:10:48.105 --> 00:12:29.930
which is just good for showing it in YouTube videos really. But in terms of utility and usefulness, this view doesn't really do much. The other thing that I don't like with Obsidian, and maybe this is just a user error on my part, but it does always seem like whenever I search for my markdown files in here, it takes a long time. And it might just be the size of the software maybe because with Rubrik Docs, it's really lightweight. But you can see here that I'm trying to search for that markdown file, and it always takes a long time, at least for me, before it appears. And the other important thing is if you want features that should fit your own setup like that inline commenting that I showed earlier, you can't really fully change Obsidian directly in order to introduce those features. And also, I think it's just much simpler to have one less software that you need to rely on, which is why I defaulted to just using rubric docs and this tab for viewing all of my markdown files in just one nice dashboard. Now this next app that I'm about to show you is a bit different because what it actually does is allow us to use all of the apps that we created right on our mobile. So you can see I have the dashboard for this service, and, yes, this is entirely free. And what I have here is my laptop, which is this one, as well as my Pixel 10 Pro, so my mobile phone both connected. And so because of this service, I can actually access all of the tabs that we were viewing earlier, like rubric generations right in my mobile phone. So if I'm on my mobile and I need to generate some images while on the go, then that's something that I can do. Or if I want to look at a rubric docs, then that is also available from my mobile phone. And that is all enabled by this free service that is called Tailscale. And Tailscale, in its essence, it is basically built upon zero trust networking. So it just builds a tunnel basically from your device, let's say your laptop, to your phone. And you can read more about it in tailscale.com,

00:12:29.930 --> 00:14:17.780
their website. And we're not affiliated with them, but they're just really a good product honestly, and they have a lot of security features built in just so that you can stream whatever applications that you use, including the ones I featured in this video so that your micro apps are accessible in your mobile. And just to give one very specific example and use case that I have, in these videos of mine, I actually don't script them at all. I just have, like, rough bullet points of things. But the intro to these videos, I sometimes read them via a teleprompter. And I used to pay for, like, a teleprompter app separately, and it was a bit buggy. So what I just did is to Vibe Code and just ask Claude Code to create an application for me. And so here, whenever I have an intro, I just ask Claude Code to push that script, let's say, into this teleprompter desktop app. And from here, can find that relevant text file. I can play it so that it scrolls through the feed and even mirror it. And the reason why that's important if you've never used a teleprompter before is because usually I use my teleprompter right here on my mobile phone connected via Tailscale. And from here, I can just play that teleprompter feed, and I can just read through it, and I can also mirror it from here. And, uh, unfortunately, don't have my teleprompter rig right now because I am in The US, but it is quite useful, and it is really simple to create. So point being that if you need a particular custom app or maybe you're relying on an application that you can easily code yourself or create with any of these agentic harnesses, then you might want to create that yourself. And to also remember that whatever it is that you create on your desktop app, you can actually easily stream to your mobile phone through a service like Tailscale. Now one of the favorite apps that I built and constantly used and you see a lot in our videos is this piece called Rubrik Flows. And what this basically does is it allows you to illustrate the automations that you have built with your agents. Because one key problem with these agents, especially these coding harnesses,

00:14:17.860 --> 00:15:04.755
is that they are not that visual. So if you have used n eight n before, what's great about that tool is the fact that you can clearly see how each automation is built. And so if you need to present it to a client, let's say, it is much easier to explain and walk them through step by step specifically how this automation work that step one, this is what happens, and then step two, this is what happens. But with Cloud Code, since a lot of that is abstracted, it just happens with code flying by in like a terminal view or even in a Cloud Code chat session, which is all just text and is not really fit for any particular visually engaging demo. This rubric flows, what it allows you to do is to clearly explain how an agent works that basically it has couple of skills that is attached to it. And let's say you built this particular automation that creates posters

00:15:05.100 --> 00:15:43.620
based on a client's given brief. You can easily visualize it so that your client or whoever you're presenting to clearly understands the steps that the agent will take based on this skill, which right now, because I just invoked it, it is now picking a style of what this poster will be about. It will build the prompt for that poster and then generate that poster. So it is much more visual. It is much more engaging, and you can actually run it with Cloud Code whenever you need to visually show how each pipeline is moving. And you don't always have to do it obviously because under the hood, when the automation runs and you don't need any monitoring or you don't need any visualization,

00:15:43.620 --> 00:15:57.135
then that's completely fine. But for those cases when you need to educate, when you need to illustrate, or when you need to, I guess, impress your client with what you have built, then this rubric flows piece becomes really useful. And by the way, that whole piece,

00:15:57.375 --> 00:16:26.975
that is because of this session where you saw me just kick that off and I said to read the fantastic poster skill, and I said to demo this in rubric flows. And you can see here that for the sake of this demo, what it basically said is that it picked a style, let's say the vintage travel style, It built a prompt and generated a poster using GPT image two. And similar to a lot of these apps, the way that you add, let's say, new flows or automations in here to illustrate is simply via natural language. And when it's done with that, it now said that it's available in generations.

00:16:27.310 --> 00:16:35.550
And here you can see that image that poster already generated. Now next up are these two apps that sort of go hand in hand. It's called Sprint and Backlog.

00:16:35.710 --> 00:16:41.310
So for any product managers out there or anyone who's worked in a sort of a scrum environment or squad environment,

00:16:41.755 --> 00:19:21.530
this may be familiar to you. But, basically, what this is is a way for me to track tasks and any projects that I am doing on a weekly basis. So these are actually demo tabs because it's, uh, much easier for me to just show the full interface without blurring it. But, essentially, the way I use it, because I have my Telegram connected to Cloud Code as well, sometimes when I'm on my phone and I'm on the road, I think of tasks or projects that I can do. And, naturally, what I do now is to just have that added to our backlog. So for example, here, I said to add to the backlog to write bullet point scripts for this rubric dashboard intro video. And if we go to our backlog, you can see that that has now been added as a separate line. And from this backlog view, essentially, it's just an ongoing list of things that I thought of that I think would be good to be tackled sometime in the future. So I don't usually view this myself because it's actually easier for my agents to just parse through this whole list and suggest, like, the top priority things that I think we should be doing. But you can, let's say, add in epics, which are sort of big projects in the product management world that you can filter through in here. You can see which ones are in progress, which ones are to do, which ones are done, and also which of the agents are the key assignees, which at least for my case, you've seen my other videos before, I just named them based on the work that they are meant to do. So Devo is like a developer agent. Bizzo is like a business agent for clients and such and so on and so forth. And you also have prioritization in here as well as an ability to search for tasks and even add new ones. But again, like I mentioned, I don't usually look at this backlog list because it's mainly just a list for my agents to parse through. It becomes important though when it's time to sort of look at the sprint for that particular week on what needs to be done. And the way the sprint works is, let's say for a given week, let's say for this time period, March 24 to March 28, the idea is you pick from your backlog on what are the key tasks that need to be done during that week. And then that just gives you a nice overview of what is already done, what is in progress, what is coming up next, and what is still to do. One of the things that I like personally, which I never saw like the other product management apps like Jira or Linear do, is an ability to be able to add in, a nice progress bar here in the bottom, which at least in my view is just a nice way to see the sort of completion rate for each of these tasks. So this one, I use it every day in order to just see, let's say, which ones still need work and which ones are already done. And because this application is agent native, there are a few interesting things that you can do with it. Right? For example, if you go back to our Cloud Code agent in here, I told it to look at the Sprint demo tab and tell me which ones you can do now. And what it basically did is it looked through that file, and it suggested

00:19:21.530 --> 00:20:31.855
which tasks it can actually move to the next step. And we can either decide which ones would be good for it to tackle right now just so that we can take some work off our plate, which we can just give direction to this agent on how we want these tackled. Now a big reason why a lot of these agents cannot be fully automated to just complete all of these tasks in one go right now is because we are living in a token constrained environment. So you'll easily quickly hit your rate limits if you just let the agent go through all of your backlog and try to complete all of them without any steering. We're just simply not at that point yet in terms of the availability of tokens in order to complete that task as well as the, let's say, intelligence or steering of these agents to make sure that whatever output that they create is aligned to what it is that we want. So long story short, we do still need to steer them a bit, but there are a few things that you can definitely do in order to fast track your work. And having a sprint tab like this, and let's say you are on mobile again with this Telegram agent, you can simply just knock off things on your list just by talking to your agent, asking what are the things that you can do for the sprint this week, and, uh, just direct it and steer it on what it is that you want for each of these tasks so that by the time that you get back to your laptop,

00:20:31.935 --> 00:20:43.695
you'll have some ready made artifacts and materials that are ready for your review. Now just to cover these two apps together, what I have here, this agent's view and this skill tree view, I usually call them illustrative views because

00:20:43.990 --> 00:21:08.185
they are meant basically to illustrate sort of the scope of your workspace as well as your agentic setup, and it's just a nice visual way to communicate how your entire command center of agents is set up. So for example, here, what I've set out to do is to have six agents so that they represent six roles or six portions of the work that we usually do, where Robo is sort of the chief of staff all around agent. We have a workspace for development,

00:21:08.265 --> 00:21:09.385
one for content,

00:21:09.545 --> 00:21:11.945
one for business, one for community,

00:21:12.105 --> 00:22:29.125
and one for just a personal assistant that I have. The other thing that is quite useful at least for this view is you can see that they have a status of what they're doing right now. So you can see Rowboy is doing some deep work in there. Como because we just talked to him a while ago in Telegram and says that it was active a minute ago. And, uh, this is just important because especially with these agents where you constantly need to talk to them via Telegram, which seems to be becoming more and more common especially with, uh, tools like OpenClaw and Hermes. When I send a message here, usually, you don't get much signal if this particular agent is still working on a hard task or not. Essentially, what I did here, if I type in hello world, you can see that this particular agent said that it is active and it is actually doing some work. And then the skill tree view, if I go to that, this is, again, just a nice visual way to represent your agents as well as the skills that they have access to. So I don't usually spend a lot of time in this view, but let's say if I need to check if my context is clogging up and I want to see what particular skills are connected to these agents, then this is just an illustration of that that you can also use to communicate it to your clients, for example. But similar to the Obsidian view, it's more just for illustration and for visualization purposes because if you do really want to clean up your skills list, for example,

00:22:29.445 --> 00:23:03.865
you wouldn't really do it here. You would just talk to your Cloud Code agent and ask for recommendations on how to do that right from within the chat. Right? But it is here because I do foresee that once AgenTek setup services become bigger in the future, visualizations like this will also be important in order to communicate the value as well as the architecture that your agents are structured against. Now those are the micro applications that I personally use. Even though you may find value in them, what is actually most valuable to you is if you can create your own views depending on the work that you do. And what's nice about having a Canvas similar to this Rubrik console

00:23:03.945 --> 00:24:33.375
that you can just centralize all of your applications in just this one command center is that if you open Cloud Code or whatever agentic artist that you're using, you could just reverse prompt and recommend to you three new views for rubric, for example, based on all it knows about you. So here, I ask it that question, and it recommended to me couple of new views, like, for example, today, like a daily briefing kind of thing, a pipeline view for content as a Kanban across the several steps that I use to make these, as well as this deeper agents view that just provides, uh, more details around the different activities that our agents are taking. But point being that you're actually not limited to just the applications that I'm showcasing here. If you have a nice canvas like this and if you want to be a bit more playful with it, you can also add icons to your agents just to make it a bit more fun. That is definitely within the realm of possibilities now with these agentic platforms. Platforms. Plus, Plus, it it also sort of gets you in the mindset that you can actually create apps for other people, for clients as well depending on what their needs are. To give some example that I won't fully cover because this is highly specific for content creators anyway. But one of the things that I do use a lot is this YouTube outlier dashboard. So this just has some logic around the engagement metrics in here, the maturity of the video just to see if it has been out for a while or not, the number of views as well as an outlier score. This is just a nice way for me to be updated around some of the videos of my peers without having to go through the whole YouTube homepage.

00:24:33.375 --> 00:24:39.680
I also have this content shortlist tab where this just shows me sort of the priority content ideas

00:24:39.840 --> 00:25:22.805
that I have and also continuously add, like, a running list of things that I think would be valuable for our community and for our audience. And, usually, how I add to this list is, again, on Telegram, if I'm outside and, uh, a couple of ideas come in or I'm browsing Twitter, let's say, for, like, a new content idea, I have this simple skill to add on the three star atrius. Let's say this called new knowledge work plugins that I was reading about earlier. And you can see here that it identified that it's already there, which is this one. So again, probably not very practical for some people, but helps me personally. And it allows me to just move things in here just to be more focused on some of the stuff that I think I would tackle next in these videos. But there you go. Those are some of the micro applications

00:25:22.805 --> 00:25:40.910
that should get you started and should give you ideas on what to create next. I just think that with the capability of these AI models and these argentic harnesses now, if you just stick to talking to them just via chat or via your IDE and see a wall of text every time, number one, it's not really that delightful or exciting.

00:25:41.150 --> 00:25:50.815
And number two, there is a lot of communication and meaning that is sometimes lost or is glossed over that I think can be represented well if you have a separate application

00:25:50.895 --> 00:25:54.095
that can visualize that better, that is more interactive,

00:25:54.415 --> 00:26:06.490
and is also a bit more fun to use. I hope that was useful. And if it is, then consider subscribing because that helps me a lot to put out more educational content like this. And as always, appreciate you guys for sticking to the end. I'll see you all next time. Thank you.