Hyper-Personalized Tooling: The Next Era of Software

**[00:00:00]** Hello everyone. My name is Armando. Welcome to the Web Talk Show. We're doing a live episode today with Mike Carlo. How are you Mike? I'm doing well. Thank you very much. Ah, it's great to have you. As always, we're going to be talking about hyperpersonalized tooling, the next era of software. Or at least that's what Mike and I think at the moment. We'll see how it goes. I think it's it's uh it's going that way. What do you think, Mike? Well, if our predictions are true based on what we've kind of comp, you know, comp uh come up with in the past, right, we kind of predict this era of hyperpersonalization in general. We've been feeling this for a while now. I build exactly what I want when I want to. The agents are doing things. And so, I'm finding myself, again, this is just how we build. We're building really personalized custom tooling. And we said months ago that throwaway tooling is is going to be here. You're going to build a tool, you're going to use it once, you're going to throw it away. it's going to be so easy to create things with these models that you're going to have disposable tooling. This feels like an adaptation of that, right? So, let's not throw that code away uh in your business. What is that thing that you're doing that just is a lot of overhead work or I mean I'm seeing a lot more communication on social media around people building entire businesses around I want to build a brand new business and it will be solely built around the concept of agents and how do you build agents to run aspects of your business? What does that look like? You know I have a neat idea. I have an a web application support help all these kind of things. How can I lift as much of that reprogramming out to an adjuncted experience? And now you can get that one soloreneur that can actually run substantial large elements. So I think this is kind of where we're going with this is, you know, for us as we build personally, we're looking at this really hyperpersonalized area, but this is also **[00:02:00]** transferable to your organization as well. Yeah. Yeah. I I found myself recently like we find a tool that doesn't do exactly what we wanted and we just rebuild it and then we use it and then from there it's like well do I really want to rebuild the whole thing and have the stack and have the UI and all that or do I just like you were saying just do agentic and just have my cloud and I'll just tell it hey can you trim this video for me can you get the transcripts out of it can you create thumbnail for me and it's just just talking through it and you can do it from your phone you can do it from your computer So it's really nice. Now what we've been seeing and this sparked today's earlier conversation offline is when we look at the scope of projects that we currently have with clients, we're looking at it's not so much, hey create a workflow automation for me like an NA10 thing or a make.com or even a coded um solution for automation. It's it's more of a I have these problems. Yeah. of time loss because we're constantly doing the same thing and we're trying to interpret all these invoices that get sent to us via email and we're trying to organize our logs to make sure that when we get accounts payable or accounts receivable that nothing is lost in transit. There's a lot of skew mismatches and we have to do that by hand. That sort of stuff. And so we're seeing that we're building hyperpersonalized tooling for these companies. So it's more of a you have a business, you have these pain points where you're losing data, you're losing time, you're getting um failures, things like that. And we go in, we analyze and say, "Oh, well, we can create a tool that does exactly that for you." So you don't have to think of Salesforce or Oracle or SAP or even Monday.com or **[00:04:00]** HubSpot or whatever. It could be distinct separate tools for each department that just do exactly what that person needs. It solves that person's problem and their team's problem. And then another tool solves that other team's problem. And you might get to a point where they're all interconnected and talk to each other, of course, but at the very beginning, you don't have to think of this monster and try to fit everyone into a box. It's more of a here's the thing that you need now that you can start using today. And it's incredible because people are just seeing so much time back. Yes. and efficiency and like I didn't know we were losing or keeping losing track of like $16,000 in inventory that was just on the shelf and we were not moving it and we could have gotten rid of it a lot sooner if we had known that it's got like 90 days aging or whatever and but with AI it just surfaced like that and so it's not build a whole AI tool it's like build these little tools some of them don't use AI at all it's just we use AI to build because it's going to be a lot quicker, but some of them don't need AI in the workings. And so this is this is another term I want to start coining here. I love this point. I want to lean heavily into this concept of like the creator agent, right? So I I think this is the the term I'm kind of making up a bit now, which is we use agents to build things that run on call it normal hardware that doesn't require require token usage, right? So use the expense. This is this is happens everywhere, right? If if you want to make something efficient, you buy the expensive machine to to optimize costs on how you run the individual process, right? Um Let me give you an example here, right? For example, um you could send a image to a **[00:06:00]** large language model and say remove the background from this image. Okay, it could do it, right? It'll happen. It will it would send the image to the token. It'll grab it. It'll tokenize the image. It'll do the the items. It'll then do whatever it is to do and then returns back to you the image, right? There's also a lot of known regular models that can run on your machine, on your CPU, on your computer. like you've already purchased the computer. And so one of these micro tools that or throwaway tools that I was thinking about using but now feels more like a hyper personalization tool, which was I've got 80 images and I want to remove all the backgrounds. Traditionally, I'd have to go to Adobe or Lightroom or some larger application to use their algorithms to remove the background images. Well, I didn't want to do that. So, in 30 minutes, I had talked to my agent and say, "Here's what I want. I'm going to give you a folder with raw files. I want you to go build something. I didn't even tell it what language to use. Build something that removes the image. I may want to have some controls. I may want some settings. And I want you just to process everything in this folder. Use the same folder structure in the processed folder. And then just remove all the backgrounds. Make them a PNG. Done. End of story. And it it built the tool in 30 minutes. So now I have this tool that I can throw any number of images at and it will then autoprocess all the backgrounds off these images. I'm not using any shady website. I don't have to pay any subscription. This is a tool. This is a like a smaller hyperpersonalized tool to something I was I was working in. Now the next step of this is like I want to be able to now lift this task or have this to be able to be easy to use. And so this is where I think the line for me gets a little bit grayer between like the creator agent and like well now I want the agent to help me do and use the tooling. So I think there's to me now I'm looking at this going okay well let's say I build a you know again I'm heavily on social media so I'm doing a lot of media things right I want to remove thumbnails I want to get p people people's images maybe I want to go get **[00:08:00]** information from like a website or a LinkedIn page or a lot of things right I may want to pull a bunch of stuff together all of those may be using scripts in Python or non-agent related things but now I want to start stitching multiple of these like micro tools together and this is where I'm debating Like, does that make sense? I don't want to really build a full application. What I'd like to do is just, hey, Instagram, here's five images I want you to remove the backgrounds from. Agent, download them, put them in a folder, remove the backgrounds, upload here. Like, just it uses my tool like so I'm I'm actually providing like the custom tooling that I built to the agent, but now it's doing more of like the logical thinking like, okay, I know what to do with all this. So, I don't know, Marando. I'm still trying to get my head around like where's the blurring of the line between like when do I want to use agents to save me time versus do I really want to spend the tokens every time to process images? Probably not. Right. So, this is a balance I think we're starting to find with this new creator agent space. Yes. What do you think? No, I agree. I we were recently doing some matching for transactions. So, this client has thousands over thousands of transactions coming in through Stripe that then go into some database or spreadsheet or whatever you want to call it. And then they based on an identifier for the person, let's call them a student, they can then surface those payments to the students if they want to check a log of their payments, they want to see their account health or whatever. And so seems straightforward, but the problem is maybe some of these payments came in through an old integration or they were created before they had the new system. So it doesn't have the students ID or it doesn't have the email or it doesn't have something or there's a mismatch. And so when you're using AI to fix something like this, let's say you're using an A10 or make.com or whatever, what it would **[00:10:00]** do is you have to go through all the different sheets and places and try to find those matches and you would have to specifically tell it what to look for, right? Yes. That would take time and it would take tokens and credits and all that stuff. However, if you bring that down, it's a very simple step. You just bring that spreadsheet down, give it to something like cloud code, talk to it and say explain the situation. Hey, I have these these are all the payments. Some of them you'll notice don't have matches. I have these other sheets might have matches there might be some relationship based on the email and date and time and some other things. And so what cloud code will do and it did it amazingly well. It was able to build a script, a little Python script. Yep. in like 20 seconds that it then ran for a few items just to check and then did a did a trust exercise and was like, "Oh, look. Yes, I did find 15 matches for the 16 that I tried." Yes, let me fix some things here. And it went away and then it ran. It took like 3 seconds to run. Yep. A 55,000 row spreadsheet. Yep. and it did all the matches. This is the creator agent. You used it to create a process that you wanted to reuse. Mhm. And this is where the hyperpersonalization thing I get so exciting for me like this is what I want to build. We we want to we want to start thinking this way. And this is an entirely mind shift I think difference here is that you we I was just talking with um Matias on our agentic thinking podcast again focusing a lot more on agents and building things here. This is the same concept, right? Our work isn't the AI isn't replacing our work. It's just shifting what we work on. Right? The internet came out and then everyone's worried about oh it's going to take away all this paper print. We're going to have any more jobs. Like it's going to everyone's going to sit around eating bon bonss and not do anything. The internet has just **[00:12:00]** shifted everyone's work to now staring at screens and looking at computers all day long. I think this agentic era is doing the same thing again. But we're not going to be staring at screens. We're going to be talking in chat using voice a lot more video. Like the proliferation of shorts and video editing clipping things that are coming out from like anthropic and everyone right now. Yeah. My feeds on X are inundated with everyone's building some sort of clip clipping system for short video. Here's my long form video. Build me a whole bunch of shorts. Armando, we are actively developing this same like this is a hyperpersonalized tool that we want to build for what we're doing to make our brands unique in this way. So, how do we build something that's a micro tool that takes long form video, takes it down to shorts. Oh, speaking of which, we talked about this a while ago. Um, I just saw in Teams, Microsoft Teams, their AI system. Remember how we talked about, hey, would it be ne to have shorts? Yeah. For the meetings for it's there. They were listening to us. So, amazing. Keep listening, please. Yeah. Keep listen. We'll keep feeding Microsoft ideas about they must be scanning the channel or something like that. So, um, we had, so Armando and I had this idea a long time ago. I was like, man, how interesting would it be to like have a long form video of a meeting and it does highlights of the meeting and so you can go back and watch the clips of the meeting. Dog on it. It I was in Teams. It showed up. I was like, "What is this new button here?" It was like Teams clips or something like that. I don't even know what it was. It walked through the video. It had me describing and communicating on the video. It had a narrated voice. Mike is talking about blah blah blah. like it literally narrated like an like a video intro trailer and then it said listen to Mike talk about how the semantic model was doing you know how interacts with. So then it it literally audio dodged out **[00:14:00]** the narrator voice dodged in my audio and it clipped to the section of my section and it said it literally watched me explain hey here's how we do this here's how we do this how we do this. So it it used the transcript enough to identify this was a key moment that Mike was describing or educating or teaching something and people were like I don't know how it got the sentiment around that but you know dude if you think about like what we could possibly like you could look at people's like facial expressions you could see like their feedback are they doinghm right it can have all that as feedback to say oh this part of the video is actually really important we need to highlight that I thought wow this This is where AI is going. Yeah, that's Are you sure you didn't dream this though? I I could have been dreaming it. It seems pretty amazing. It sounds pretty amazing. Okay, so after this video, I'll have to go see a video with Armando and I and I'll have to go look at one of the ones we had on Teams previously and I'll have to double check. I will confirm Armando that this is like a real feature. I think it might have been actually out in preview or something like that, too. Like it didn't show up. It hasn't shown up on every single video that I've done. It just was like one time I recorded a meeting, the post recording had this thing and it it had this like clips thing and I was like this is blowing my mind because now it's taking two other forms of learning. So typically everything we do is reading and not everyone loves to read and you're losing a lot of other audience when you just only transcribe things. So having short form video or summaries of things like that was really really really neat to me. Yeah. So anyways, and it's not hard to do. That's it's like the the moat of difficulty in creating these things is gone. And so it's it's how you execute. So for them to have seen, hey, there is a market for this, let us execute quickly, see if it works out, that's that's talking **[00:16:00]** really well about what what Microsoft Teams team is doing. Yeah. So let me I So I just pulled up a really old meeting like like two days ago, right? And it has two really old two days. Um but it needs it needs some time to process. It's not like right away, right? So um the first item here is it's called audio recap which just recaps the whole meeting. It's basically this summary notes sent to like audio so it summarizes it. I haven't used that one. And then it has video recap. And video recap was the one I was checking out. And that's when it narrated through portions of the meeting, highlighted areas that were interesting, and then um it even audio dodged in me to talk about things and then explain stuff. That is amazing. Amazing. That's good. That's good. I I see a lot of more things coming in that space where for us it's very clear because we're doing content and and you like you said in your feed, that's something that's being surfaced a lot. people creating correct tools for creators. Why? Because people real if they're posting obviously they're do some sort of creation and so they realize that the existing tooling that you needed to pay for before like Opus clip and dcript and all these other tools are no longer required really. I mean they have their their place of course and but you pay for credits and you run out of those credits and you have to pay more and more more where if you spend the time a little you might be able to use some of these other open-source tools or create your own that will do what you need coming back to what you said at the beginning a local version of it which is something I really wanted all the time because you have these massive 6 n 12 gigabyte recordings of shows and then you have to upload it and they'll compress a ton out of it to be able to create these clips. And so you lose a lot of fidelity where if you just have them and and it **[00:18:00]** takes a while to upload and download and process and but if you have a local version of it then it the trimmer that that we're working on what it does is it it just strips the audio makes it smaller just to send to the transcription service if you want to use an off like a cloud transcription service or you can use an opensource local one if you want. And then once it has that, it can use AI like claude or you could use a local one if you want as well. Check out what all the main good parts are of it. And then it uses local again CLI tools like ffmpeg that that everyone uses behind the scenes to do all the actual cuts. You can use something like hyperframes or whatever you want to actually work with the wording and captions and things. And then it outputs the final product. But it works on the real not a proxy. It works on the actual like high resolution, high quality 4K video. Yes. For those cuts if you wanted to. So the output is also much better. Right. And and I would argue like when you build custom cling like this, like if you're using So we use candidly opus clip to help do some video clipping. Yeah. We're we had an episode on this one earlier burning down the software as a service solutions right now just just replacing it building what we want in this scenario of building these kind of clipping type tools I'm not paying for subscriptions and they're not even using the best model to detect the best phrasing around what's going on and it and now that we're taking control of this we can then publish them how we want when we want better scheduling Again, we're I'm looking at a lot of other tools and going, "This is nice. I don't really want to pay for them. I actually want to control it on my own because the barrier to make the software, the barrier to maintain the software is now I just **[00:20:00]** talked to my engineer, which is my agent, and it fixes everything for me." Yeah. Oh, speaking of which, we should talk about agentation. Oh, yeah. Yeah. That's a neat little tool. So anyone who's talking with their agent or claude or whatever you want to call it codeex or you just chat directly whatever you're using sometimes you will want to like screenshot the website that you're working on send it some information like oh change this part or this doesn't look right or whatever and you typically have to send the screenshot and explain what you want to tell it or you like Mike did take the screenshot then point a little nice arrow on top and tell it this is what this is what I I and I'm really I do a lot of presentation stuff, so I use like um for my computer, I'm on a Windows computer, so for those of you on Windows, I use um uh ZoomIt and and Zoomit is installed with this really cool Microsoft Store application called uh I think it's called Toolbox, I think is it called? It's um no, Power Toys. That's the that's the name. So, it's like a it's a whole collection of like, you know, very neat tools. They don't not built into Windows directly, but they're like adders or extension things you can add to your Windows machine. Well, this one has a ZoomIt program built into it. And so I can zoom into the screen. I can make a little square around it and I can make a little arrow pointing to what I want. And everyone's like, "Wow, you're such a great presenter." Like, "Well, what is that program you're using?" Like, "Oh, just Zoom it. Microsoft makes it." But if you can get good at using that program, it really helps drive people's attention to like partions of screens. So, I use the same thing and you can use ZoomIt in concert with control shifts, which is save a screenshot. So zoom in, draw on the screen, and then you can then screenshot that drawing, and then you'll get like an image that you can then send to the agent. So I found that to be really effective for telling the agent, I'm on this website page of the app that I'm developing, and I want you to look at this text box in this location. So it's probably doing some kind of like, okay, it's expensive to do that because **[00:22:00]** now I'm sending images to an inference, which an image token is way larger than text tokens. Good point. Yes. So while I was doing it to be effective for me to communicate to the agent, it's actually way more expensive on token usage. So moving away from that and so that's this program here and I I've put it on um the PowerBI tips YouTube channel chats and and Armando I think you can put it on your channels as well. I will. Yes. Agentation which I can't say the word right and I kept having this mental block of like what are they trying to say? It's annotation but the annotation portion is the front of it is replaced with agent. So instead of annotation it's agentation is how I think you say it. So really useful tool already my team members are showing me videos of hey look how I'm editing our apps we're building now using agentation. I'm also adding it into my production apps. Mhm. and saying if you add a URL parameter of agentation equals true, you can tell your agent only turn on this tooling when you need it in the URL parameters. So if I'm building in production or looking at a production website and I want to capture like something doesn't look right in the production website app, I can just add my URL parameter. Agentation shows up. I can comment directly on the screen. It gives me all the divs and everything. So helpful. Yes. Yes. And it it shortcuts it makes your token usage less. It makes your development cycle go faster. And the neat part about this too is if you copy this out, every single it's almost like you're commenting on the page itself. If you have five comments, you can copy that text out and send it as text to your agent. And it's it understands there's actually a skill. There's a cloud code skill called agentation. And so if you add the skill for agentation, you can say agentation and then you can call off the code and it knows that skill interprets the code **[00:24:00]** from the website to go find each item and then each comment becomes like a fix and then it fixes them one at a time. So then dude, I'm telling you it Wow. Wow. This is so helpful. We had heard of it what maybe four or five months ago. It wasn't that it wasn't that long. Two. I would say two or three. It feels like four or five months. Three for sure. Yes. Yeah. Three for sure. It it I don't remember. I was trying to look at the syntax website. I thought I had heard about it from West Boss and their podcast, but I think he probably just tweeted it cuz I think it came up on X for me too. Like I think I think it came up on X feed. I was like, "Hey, look at this." Or you sent it to me and I looked it on X. I was like, "Oh, this is really interesting." And I I try to put my brain for weeks. For weeks, I've been asking you, Armando, what's the name of that dog on app? I think I want to try it now. What's the name? And he was like, I don't know, man. I can't find it. I don't know, man. I'm like, I could literally agent, you idiots, please put the word comment somewhere on your freaking site or somewhere that's searchable because I kept trying to say, I'm googling agent, comment to agents on website, comment to agents on web page, and the dog on website would not show up. I could not find it. So, your marketing sucks. That's what I'm searching for. You need to go buy those keywords ASAP because that's what I was looking. It's such a cool tool. Yeah. And it's rock solid. It's really good and it's a it's a it's a paradigm shift, right? Because you the only reason you take the screenshot is because you want what you wanted to visually see the thing. But what a lot of people don't don't understand is Yeah. the agent, what it does is it first it processes the image to read out what it thinks it's seeing. It's not really seeing it and then it receives it. Have you seen what it does? What it says to you? It says like user sent me an image. I think they want me to look at this text box because **[00:26:00]** there's an arrow on like I've seen I've seen it like reason about my image I sent. So it's doing all this tokening and reasoning just to get to the idea that like I sent it in an image. It has to interpret the image and figure out what that is. Yes. And with this, you write your comments and then it grabs the whole DOM part of it. So you don't have to actually explain to it this LI item that's right beneath this UL item that has a title on top of it. It just it automatically gives it that as context along with whatever you're commenting. So that that's a really neat tool. So if anyone's Vive coding or just AI assisted coding and and you need to show your agent stuff, look at that. Just go to the repo, go to the website and you can see how how to install it. Basically, you just tell your agent like look at this use it right and and it'll figure it out. Don't worry, it's not hard. No, it's not. Um but but it is a very very nifty tool. But I really like your idea of putting it on an actual app. So yeah. Yeah. depending on the surface. Like maybe you want it on a production app, maybe you want it like on a de environment. It really depends on what you want. But but in reality, I think it's a solid idea because any app can Yep. can want feedback, right? You would always want someone to provide you good feedback about something. So if you have that and you know your crowd or your users are developers or developer oriented, why not give them an option like hey did you find the bug? Let us know. Like and and it it works really well when you're on localhost. So localhost is actually the secret sauce of this because what local host does it allows you to run a local MCP server where you just say here's the local thing that I'm building. You turn on agentation and then you can literally comment what you **[00:28:00]** want and then you say you go back to your agent and say fix the comments like that's all you say and it knows oh we're using agentation. It goes to the comments on where the page is. It finds the so it literally caches them in the browser or something like that. It's got some kind of browser cache for it. It finds the items, builds them, fixes them, resolves them, and then when it resolves them it deletes the comment from the page. Oh, nice. Yeah. So, so it actually like you can like make the comments and then as it's fixing in real time, the comments are getting removed and the page is getting rebuilt in real time. Like this is like next level development stuff with my agent because again I would rather just say this text is wrong, this is formatted incorrectly, drop this header, add this thing like it speeds up my time to development immensely. So, while it's building the solution and working on that, I'm actually back on the page back testing some more things, figuring out what the next thing is I want to build. And the reason I'm I'm pointing this out is so important. This is not just good for websites. Mhm. More and more people, again, what I'm building right now, and Armando, we've talked about this in the past, and actually I'm be Armando's changing over here. I'm I'm going to be very honest with you with with the audience here. Mondo is is going through a transformation. He's metamorphosizing into a fabric and PowerBI developer to some degree. So I'm pushing Armando hard. It's like never touch Microsoft stuff ever. And then after talking with Mike's like I might touch a little bit of the Microsoft stuff just a touch. We're starting to test fabric as a as an option. My toe. But I I'm and you're asking great questions. You're like what why do I need to use fabric? Like what's the advantage here? Like how can you help? And like what does this what does this give me that makes it easier for me to develop on top of right I'm firmly in the camp of fabric as a backend to whatever operational or reporting layers that you're building. So there's two main data points here that I'm using. um I want to use SQL databases, GraphQL **[00:30:00]** as like you know API calls to to data systems, right? And then I maybe want to use a semantic model, right? So semantic models are this really hyperfast visualization layer that queries data quickly to get reports and and visuals built on pages, right? Well, now if I unblock myself, PowerBI has its own reporting solution. It's okay. But now I'm building full-blown apps that have my own reporting in them. And I want hypers speed on the analytics side, but I also want the operational data right next to it. So I want to be able to edit and modify and manipulate data, but then right next to that, I need the reporting and analytics side as well. So this to me, this is where again, I've been super Kool-Aid drinking the Microsoft and fabric space for a while now. So I love it. And the fact that I can now bring this agent space, hyperpersonalized agents and and building hyperpersonalized experiences directly to fabric. Oh, dude, I'm like in heaven. Like, this is so fun for me. It is a it it's very interesting. I we, as you know, I I mean, I I love the Cloudfl of the world. Like for me, if I can host everything on Cloudflare and just use everything they're they're they're pumping out, it's amazing because for many it's great because it's free mostly, but then also it just scales to the world and it has all their edge stuff and and it's it's it's remarkable. But as we've been talking like there there are some things that Azure has and fabric has that that are very interesting especially in the BI space right there's yes there's not too many players in the BI space and for those who don't know some of the big ones are Tableau and PowerBI Yep. Pricing wise, I want to extend two there too. In business tools, extend your extend your like the Tableau top I would argue are like business tools. Extend that to like Word, Excel. **[00:32:00]** Yeah, for sure. Email, but then you're but then you're still talking like there's only a couple big players like we're very small and Microsoft has got that business space pretty locked down. They do. They do. And when my first foray into PowerBI was when when a client of ours a few years ago was about to get like a $120,000 Tableau license just because they needed embedding. Right? Coming back to I'll I'll come full circle here just because they needed to do to do embedding. So it was it was just and so it didn't make sense at all. So then PowerBI, thanks to a conversation with Mike, made much more sense because they could do embedding 100%. And pricing was ridiculous. Like nothing nothing like literally peanuts compared to what what it was. And so then now with all the fabric stuff, it's not only because I wasn't really a fan of the PowerBI aesthetic. Now, of course, Mike's Mike's templates and all that are amazing, but but just out of the box thing, I I don't I don't really love it. But now with all the fabric stuff, it's like what you were showing me the other day. You can build your own type of interfaces for data visualization using components. And it's no longer I'm embedding a dated dashboard. Yeah. It's it's like it's extracting the the data just and and having the the layer that that you need to provide that. So I that I liked and then I really like the Asher um AI what's it called? Uh foundry foundry where you can have all the different inference tools. So if you want to use Mistral or you want to use OpenAI, you want to use Enthropic whatever you can you can sort of pick pick and choose and you can do that in Cloudflare as well. But you do everything and then with that connected to your other tooling and you can have vector stores and you can have all sorts **[00:34:00]** of stuff. So what I'm finding I'm not like Kool-Aid here at all yet. I'm just trying to test some of the things and compare because I'm more of that Cloudflare like Tanstack superbase kind of kind of builder, right? And so what I'm finding is in some of these tools, if you're developing the way we develop, right? Typically, so maybe you get a $20 plan for something like Oh, here we go. Superbase, right? Or or Nex.js or whatever, and then you need another site, you get another one, and then you got and so you have all these like separate accounts or sub accounts or whatever. Now I what I've really started to see while you looking at some of the things you guys are doing y is if you have this PowerBI workspace already set up and you're licensing and you're good you can start to spin up like oo I need a database area container whatever you want to call it that I will use for this company and I have this other one and I have this other one and I can have all of them there in the same sort of visual space and organize myself and connect them amongst them if necessary or have them completely separate and it's all since it's all enterprisegrade software if you need to do HIPPA compliance if you need to do sock two if you need to do anything like that all that's already sort of thought of from the start so because Microsoft's always been sort of on the boring side of thing like they there it was this slow and steady doing it right kind of thing and and So now that it's becoming cool, sorry Microsoft people, it's like I know it's and it's being done on purpose. Like it's literally they're doing they're doing it on purpose. Like fabric sounds fun. Yeah, fabric sounds engaging and and using it starting to use like oh it's not that hard to use compared to what you would think Azure would mean or AWS or Google Cloud for anyone who's dabbed a little bit their toes in. And **[00:36:00]** with a couple really well-crafted skills, agents are good at it. And agents are good at this. So, so this is where I'm like, so again, Marmando, our version of content nudge that we're building right now. So, Armando and I have been working on this project project, which again, admittedly, Armando, we should be a bit more public about what we're building and getting more people like we're building this amazing like this is it is a monster of a software piece now. I'm constantly editing new features and things to this thing. It's amazing for content creators. I think for me personally, I'm using it for everything I'm doing content creation wise. But the idea of this is like I needed a robust back-end system to do all this stuff and I knew the fabric world so well. With a couple well-crafted skills, I was able to make my front-end static web app from fabric from Azure talk to my SQL server inside fabric. Great. Now I've got a database. Same thing as your superbase thing, right? So the reason I think we kind of all used superbase as a as an agent developer was because it just kind of knew how to use it like just kind of built themselves for agents and that was how they were working. Well, you know, Microsoft is built more for APIs and developers, not necessarily for agents at this point. Now I think they're transitioning more towards agent level building, but there isn't really a simple skill yet that just says I can get what I want and I can I can connect to things to a production app with that. So a skill helps me understand okay I got to create this app registration I can simplify some things if you sign in here it can it can just start creating a couple things for you so it removes the friction and I think that to me that's the that is the moment where we need to reduce friction for the fabric and power that's on me right that's me the developer in that space I need to build skills that assist with that that allow you to then bundle them into whatever we're doing here I want to touch on that because I've been building a lot in my **[00:38:00]** traditional stack which works really nicely with all the components that I have. Sure. when we started building, we have a a client in common and oh yes, I need to build a a dashboard for set client and our stacks, visual stacks might align, but the back end has historically been different. In this case, I needed to be able to push this into Azure because this client's infrastructure is all in Azure. And so this was the first time I I was experimenting with push it's the same code right but pushing that into that type of infrastructure. So I'll give you my experience with with that and cloud code. It worked like I was surprised with how easy it was. I went in I didn't do the fabric thing but I I you saw everything it set up right? It's I explained everything I wanted. I told it, okay, because I was running it locally, so we needed to do a demo for a conference. So I said, okay, so we're going to put in Azure. Let's put in my account first. Let's see how everything runs and then we'll move it over. So I explained to it, I give it access to the docs for Microsoft. And then I just tell it, okay, so what do we need to do this? And so it started asking me, oh, do you want to be do an S SWA? Do you want to use functions? Do you want to use a container, etc., etc. And so we went through that little interview process and then it asked me to create the only thing it needed, right? It asked me to create the right access credential to start with only the first one like only a way for it to have access to the the system, right? The system. Correct. And then with that, it was able to and language wise, I'm just going to add the language of Microsoft to this. Yes, please. Please. It was asking for the app registration. No, not yet. Okay. Not yet. Higher up. This is very interesting. So, **[00:40:00]** it asked me for what did it ask me for? I'll I'll look in the logs, but it asked me for a higher level um access point. What was it? uh I don't I don't want to mess up the naming here but basically to give it access so that it could work on my behalf on on the tooling and this is my own personal account it's not a business so so it was fine so I was like oh yeah yeah let's play with it and so it was able to create the app registration and it was able to correct create the like set up the tenant to make sure everything was fine it was able to set up foundry it was able to create AI search inside foundry It was able to create a knowledgebased store. It was able to create with my assistance a plugin that would because this is for a WordPress site that would grab data from here and constantly be push it over to that vector store, right? And then it also created a container that we needed because it had to run some CLI commands. It created the proper authentication methods and uh now here's where we're going to need the actual good language but the the right credentials that the app registration was going to need to act on behalf of the tool etc etc. So it built all that just talking to cloud code and it did it very fairly well. made some mistakes or assumptions like putting some stuff some stuff on central and some stuff on east and like like baby mistakes that you would do if you're never used any any sort of like and and that can also be fixed too by saying like again just going well and you can also say when you create this resource group create a resource group where you're going to put everything in and you can actually say add a policy that only has a particular region attached to it and so and that's where the skills come in right so you as as an MVP is like oh I **[00:42:00]** know this People that don't use Azure don't know this stuff. I can create a skill that next time anyone running something, it'll be like, "Okay, docs, look here. You're creating a resource group. Follow these steps. Make sure everything's on the right region, etc., etc. Oh, you make make sure you put limits on budgets, make sure you like all that stuff that a beginner wouldn't wouldn't know about." And then that way cloud code would do it right the first time. Yes, I liked it. It it worked. It worked out worked perfectly. We could even deploy container. It would create it was I was amazed because it would create Docker images on the fly when I clicked deploy on the web app to then send to another third party service and it it didn't time out. It didn't have any issues like that. It's just held really nicely. So, I enjoyed the experience. I'm good. I'm still not like moving over but for this particular project we're building in it and I like it and what I will say which is a like tremendous game changer for anyone doing any sort of scale if you're doing scale in sales or whatever. Mhm. You are going to need very good reporting. Reporting on a live production database can get really expensive and can be dangerous. And so what Mike showed me the other day where you can have that layer that will automatically and Mike can explain this a lot better. It can automatically be mirroring all the data for reporting purposes and then you are using BI stuff. Yep. On that. Okay, let's go let's go into it. Yeah, exactly right. That that's amazing because doing that like in any other system and trying to add that on is going to either be really expensive or it's going to use a lot of consultants or it's going to break very easily. But the fact that you can like **[00:44:00]** one click basically the whole setup. Yeah. And so this is this is the idea of like the data you operate on is in a different format than what you need to report on. and and um I was showing Armando the the back end of Content Nudge, which is where all the data goes. You know, Armando and I have uh you know, web scraping things that it's doing. It's getting like channels and YouTube and looking at our channels are individually, but it's also looking at other people's channels for content. And so, every day, every 3 hours, we're looking for content and finding things and we're finding new insights on the other channels we look up to or the channels that we think are competitors or the channels that are what kind of topics are they uh highlighting. Are we are we interested in that? So anyways, that's a whole another sides another story. But to your point here, um when I'm doing the operational data, I'm usually operating on data rowby row. It's a rowby row thing. So it's it's row based storage. That's what SQL servers do. That's really the way you know SQL databases run. That's that's what they're doing. There's little objects that define a single row of data. Where things get interesting now is well when I want to do the reporting, most reporting is columner aggregations, right? So I'm not just looking at a single row of data. I actually want to be able to aggregate lots of data across aggregations on columns. And so that's where the the row storage needs to shift into columner storage. And that's the transition that's kind of difficult. And and this is why I like fabric because there is now much more and we started with the columner storage format that that's kind of where we began. Microsoft already had SQL but it was in Azure. It wasn't in fabric at any point. So now they've brought this other rich rowle storage system next to my columner storage system and they're doing the hard thing that I just don't want to do. Everyone does it. Doesn't matter what you're doing. Every reporting solution needs to go from row storage to columner storage to make performant reports. Period. So they just **[00:46:00]** fix it for me. And and that's one of the reasons why I like the system to be frankly honest. Armando like there's a lot of companies that are doing this. It's very well known. This is nothing secret sauce that Microsoft's doing. It just so happens that their integration is tighter between the systems than others that I've seen at this point and and I know it really well and I want to stay close to that enterprise stack because those are the guys who are going to need help in supporting this. That's that's why we do this. Yeah. And that's well amazing. You make it sound like it's a small thing, but yeah, hooray for everyone else that's doing it, but it's a third party integration that you have to do and maintain and make sure that they talk to each other. So what I really like and another subscription and another whatever whatever it's the fact that it's just it just does it there and especially these projects that require sock 2 that require HIPPA that require any other GDPR uh compliance that you get into trouble need to to just like figure out policy and things because just because you're using multiple tools. So if you have everything under one roof and it works, obviously you don't use tools that don't work just because they're there. But if if if they work closely together, yeah, I I like it. So I'm I'm going to explore more and see what else can be built there. All right. But uh yeah, I the video you sent me the You're trying you're trying really hard. Really hard. Need to get a hat or something. the the the other day you sent me a video from Jared Freriedman, the managing partner at Y Combinator, right? Yeah. And uh he was talking about the SAS moat going away. And I really like that because many people just because it gets clicks are **[00:48:00]** saying SAS is dead. SAS is gone. And we've even talked about like creating our own tools and all that. Now, creating your own tools takes work a little bit and so you have to have the time and so most CEOs won't do that and then some of their teams might do that but it's like the whole point is not recre like clone a tool to then sell it. That's not I don't think that's that's really where it is. It's more of a like we're talking about today hyperpersonalized tooling that's internal for you that works for you and solves the problem way better. That's one part of it. The other part which Jared was talking about was that since AI has now made it 10 to 100 times faster to build things. Now you see this software and you can recreate it and everyone's going to the main ones, the obvious ones, the content stuff, the client acquisition stuff, like the CRM, all that, right? That everyone's going there. Yeah. because you see and they're all and those are all extremely laborious and extremely click heavy and the UI every one of these softwares you look at it does like 80% of what I want but it's missing the 20% to make it effective and it's like I don't want to do that anymore. I I I just want it to be streamlined more. Yeah, 100% agree with you on that one. Agree. And and they what I think is very valid is if you go into any of those since everyone is doing it then you're going to have a ton of competition. So, it's not necessarily the greatest idea if you're planning on selling like if you're trying to sell that service. Yeah. It might be if like it's very niche, like not a CRM, but like a very specific like a CRM for commercial kitchen hood exhaust cleaning companies. Yeah. Okay. Yeah. Do that because it's very niche. So, the tighter you go, the more valuable it's going to be. So, he was talking about how like think outside the box. Look at like industrial control system software. Like software that has only one or two **[00:50:00]** providers out there in the world and that have always been the same one and no one's touched. Like that's that's where you need to focus. If you're thinking, hey, I need an idea or something or I want to build something. Go look at the pain points in these places where they're using this software that's terrible and that just nobody has touched just because they're so big that it's just we'll just keep it this way. And if you can create a better version for your own company or even to sell to others, then that's that's really where the magic is because you're going to make some people really happy by transitioning to your tool. And it doesn't have to be an overly complex tool. There's a lot of these that do relatively basic things, but that before it was so hard to make that software that that's why they're kings and that's why they're there. But they've stayed like that without making changes just because there wasn't any other competition. Yeah. And and competition is always healthy. So that I think was something very good that he was saying. And I agree. I I think there's a ton of new places where we're going to start seeing not only AI use, well, we're going to see AI use everywhere, but specialized tooling that's just going to become better, easier to use. I mean, like, how many companies do you know that would save so much time if their workers could like just talk to the machine while they're doing that%? Like that's just a very small thing. Like, yeah, they have gloves. They're like working with metal and stuff. Like, why can't they just tell it to do something? Why can't they have like an AR thing just showing them the whatever? Why can't they just look at the screen to verify and then just ask it things? The There's so many little things that can be made better with small incremental steps that are very quick to implement. Mhm. **[00:52:00]** I that's where like that's that's what what we're seeing like I said at the beginning what we are actively seeing with clients is that they're not needing a fullblown redo of everything. It's very specific pain points creating tooling around that that they can use on the day-to-day basis whether it uses or not AI. It just makes their lives easier, reduces the number of errors, and makes them their their business more efficient. That's the whole point. Back to the creator agent, right? Use the agents to create. Yeah. Build in systems that are reliable. Like SQL's been around forever. Notebooks have been around forever. Webs been around forever. Like these things are existing well before the agent space has been around. Use the agents to help you create those things. Like I don't to be honest my value isn't sitting down and writing lines of code for a computer anymore like that should just be that's totally absorbed by the agent now the I was just talking with Matias on the agentic thinking podcast same concept he just came out of a really intense work session with his agents right his work isn't writing the code anymore he hasn't touched writing lines of code in months I haven't either my senior engineer hasn't touched code since October. Like we don't touch code anymore, right? That we now rely on the agents to then leverage their knowledge because I can't know more than my agent and how to write code. Period. It knows every Python function. It knows every language. It knows JavaScript, TypeScript, React, all the things. It knows every single thing more than I will ever have time to study or learn. And and if you are I'm sorry I'm getting really passionate about this part here, but like if you think that agents are you're gonna be able to out code like that's the part you will lose on all day long every day. Now what can they not do? They're not good at system design yet. They're not good at stitching together like what is the problem your business is trying to You could just say hey **[00:54:00]** agent I have problems. It doesn't know what to solve. It doesn't know how to get through the issue. It doesn't know the inputs and what the desired outputs are for what your workflow is doing. That's the stuff that you've got to think about, right? How do I stitch three different agents together? That's the problem I'm trying to solve. I've made micro tools for these little hyperpersonalized. Let's remove backgrounds. Let's make a a clip. Let's put words on a screen. Let's all these little like remedial tasks that were very click and intensive heavy. Well, now I'm using AI to do these things. Awesome. Love that. But now I have all these like micro tools that are living together. Well, how do I get an agent to weave that together to a better more comprehensive process and I think that's where we're going right now where I'm seeing this now appear in Foundry. I'm seeing other tools come out with this and had was like very early in this process where I could take you know trigger based actions and then do manipulations to data and there's this little like agent nugget that you could use. Very helpful, incredibly powerful for orchestrating workflows. But now as we think about workflows and this extends to hyperpersonalized tooling, there's another whole big need here that's coming from I'm not spending my time building the code. I'm thinking about what is actually solving the problems. How do I build these um enablers with this hyperpersonalized software? And that's going to the skill to have now is can you aptly wield one or multiple agents together to help you stitch these micro programs together. Again, the hyperpersonalization. That's what makes your business tick. The the reason your business exists because you've built a process around something someone needs. You're solving people's problems. You have a business. People pay you money for it. Awesome. Love it. How are you using AI to streamline and make that process from problem to solution better for both you internally and for your customers? That's the story here. **[00:56:00]** Yeah. Yeah. And many people can do what you do, but if you are doing better than them, then it means that process is a little bit better. There's something there's your secret sauce which won't be taken over by AI anytime soon. Nope. But you do need to use AI because your competitors are going to be using AI. So you either use it We're in an arms race right now. Yeah. You either use it Yep. in production like it does things or you use it to build things like we've been talking about throughout the show. All these micro tools. The great thing is if you're creating them as actual software tools like little Python scripts and things like that or orn net script, whatever you want. Mhm. Rust, right? It's uh don't get me started. We we joke, this is like an inside joke. Armando and I talk about like anytime we see software come out and we're like, "Oh, this is a good project. It's built on Rust." Yes. Okay. Yeah, definitely. Should be probably probably like good engineers there. And the agent knows how to build rust things for you and you're you're trying to simplify. That's that's the other nice thing I like about AI that you can you can build something you can just scrap stuff together too, right? You can you can just like build something really fast to do something and it works and then since it's so fast to do things you can then have another iteration just make it simpler and make it simpler and make it simpler and that's something Karpathy does very well. Uh Andre Kaparthi is very good at like narrowing down to the most basic principles of what the thing should do and he has some really interesting demos if you're like academically inclined. There's there's like he explains the large language model in the simplest way I've ever seen and you can have an actual large language model in like 200 lines of code and he has it all there. You can read it. It's really interesting. And so if you can use these tools to then simplify things, you are going to have so much less **[00:58:00]** points of failure that that can go on. There's there's and it just makes everything more efficient and more things can come out of it. We we talked about the Raptor engines like Elon Musk is building these rockets that are much better than any other rockets that have ever existed and more powerful and will eventually take us to Mars, right? This is great. But if you look at the engines, they've got version three, if if I'm not mistaken, you look at it and it just looks like a turbocharger from a car, a modern turbocharger from a car, the whole the whole rocket booster looks like a turbocharger. It's not you see the previous versions or any other version of any or if you ever get been to NASA and if you haven't, please go to NASA. Look at the previous generation rockets. They are impressive and amazing. But I can't fathom somebody feeling confident enough to launch a rocket with that thing strapped under it. There's so many wires and tubes and shield. Like there's so many things popping out of it. Like how is it not going to just break? And you see the new versions of the rocket. It's just so slim, sleek. Everything's it's just like one big component that they were able to iterate on until they got that amazing piece of that's what our software should look like. Our software is gonna do the same thing. Like it's going to have a lot of wires and lines and things that are all bolted together and right what can we do to simplify? Um Elon has a phrase that I I really like. Right? If you're you needed to delete stuff out of your stuff, out of your software, you need to delete things, remove stuff, get rid of things. If you're not deleting enough that you need to go back and add a 10% more back to it because you overdeed, you're not deleting enough. Right? So delete so much that you you can step back and actually add a little bit every time you're doing more deletions and that and this is like the first principles concepts that I really enjoy. I think this is really useful and I think the software space is just going **[01:00:00]** to hyper hyper emphasize the need to move quickly the need to um simplify right and and the fact that we can just keep telling the agent simplify it make it easier simplify it less things that's that's going to be beneficial and the companies I think that are able to wield and yield that kind of mentality into their teams will benefit from this this line of thinking. Yeah, totally agreed. And don't be scared like if you are building something remember how fast it was to build. If you're using an I then use source control of course right have your repositories use GitHub like as well push it somewhere else not just computer and don't be scared to tell it okay it works now make it simpler and plan another and then another day make it simpler and ask another because you never know it's nondeterministic so one time is not going to do the same thing the next time it's always going to do something different so if you keep asking it to make it better it it might actually make better and and also technology is changing quickly too. So, um, again, this is another one I was talking about on Tuesday with the Aentic Thinking podcast. Very similar concept, right? The concept was, um, you probably need to build multiple iterations of the same thing, right? That's really interesting. And when you build multiple iterations of the same thing, you now need better testing to evaluate the performance of the output, right? the so now if again agents collapse all this development time into I don't need to build one thing I can build three things and then either take one of the three or promote a mix of two and three or per promote a mix of one and three because it kind of did a better job there but now the other side of this is okay how do you evaluate what can you automate in a way that says does any of these three solutions deliver the better outcome and and this **[01:02:00]** is where I think things get really exciting that is where we should be spending our time, right? That's where we should need to learn how to do this, learn how to build simultaneous projects, simultaneous features, and then we should be figuring out what is the evaluation criteria of success because then that actually really does enable faster feature building. Yes. Yeah. Yeah. Oh, so much to talk about. There's Well, we'll continue on another show. Uh it's there's there's a lot of little thing like just just you were saying now like why not teach I remember when when uh when I was in college I took the Carne Melon personal software process certification and it was nice. I liked it because it would count time and you would change check like how how many issues you had and defects and things. So why not like grab a curriculum like that and teach it to your coding tool, right? Exactly. Like as a skill or something and then yeah it's very good at many things like you were saying but why not grab like best practices and like tell it for you like my version I want you to do this and it'll become better at the things. So uh this is not this is all very wellnown best practices documented practices these are all things that are existing already. All we need to do is just kind of like how do we correctly steer agents towards using those best practices or even developing new best practices with agents. I'm now building full repos where I describe to the agent, hey, I'm building a repo. This is all going to be about research and I want you to only like build this for agents. This is for you. I I'm going to talk to you agent directly, but you're going to put everything in the repo for research or articles or things that I'm finding linking. This is designed for agents. So make the repo agent ready in consumption. So you can scan and search and find things. So I'm I'm I'm act I'm physically shifting how I build. I now need to think am I building for people? Am I building for agents? And so now my **[01:04:00]** workflows are being broken out into two plat two two verticals now. And I'm even seeing the metrics now. Agents can out build, they can outrun, they can go faster. Um you know agents will need identities. They're going to have to do payment systems. They're there's going to be a whole another economy. So, you know, I've seen some YouTube shorts and people are talking about this, which I think is really true. The next trillion users on the internet won't be people, it will be bots. Yeah, I have multip I'm going to have multiple agents or things that are running around doing things for me. Well, if I can have 10 items per me working on things on the internet, authenticating, getting access, looking at, paying for, doing these things like transacting, like whatever this stuff is, I don't know. I don't even know. We're going to need different pay rails. We're going to need different systems in place because we're going to again we're seeing another elevation of the amount of proliferation of content. We thought we were doubling video and content before agents at I don't know every couple years we were doubling the entire amount of video and content on the on the internet. The side the volume of it. This is just ramping up that elbow to another level. Like now we're going to go even faster. Yeah. Yeah, next time we have to talk about the payments thing. This just Stripe thing and and some other Yeah, they had a huge announcement that we should unpack because that's payments for agents basically what that's what that turn into which I think is uh I I think we have a project we should probably integrate this into actually as I was thinking about this. I think so. Yeah. Yeah. All right. Well, thanks so much uh for coming on the show, Mike. And everyone's that's listening. If you haven't liked or subscribed, please do so so that the channel can grow and you learn about more amazing guests like Mike coming on and talking with us. Please also like and subscribe to PowerBI tips the podcast and the what's the new podcast Mike? It's Agentic. Yeah, our new podcast is called Agentic Thinking and it's still heavily PowerBI **[01:06:00]** and fabric and data, but it's more about where do agents fit in this new world? How do we use agents and leverage agents? We had to learn how they work. So that's that's our new podcast is talking about agentic experiences and how they're going to be used to help you build things primarily on PowerBI and fabric. Amazing. Excellent. Well, thanks everyone. Thanks Mike and we'll see you in the next one.