Disposable Tools: Build It, Use It, Throw It Away

**[00:00:00]** Hello everyone. My name is Amando and welcome to the web talk show. Today with us is Mike Carlo. How are you Mike? >> Hello. I'm doing well. >> Excellent. So the topic for today is going to be disposable tools. Disposable tools in the sense that you can build tools and dispose of them once you're done with them. So that and many other things we might chat with them, but this was a topic that Mike uh was talking about uh a few episodes before and it's a lot of fun and nowadays it's even more valid. So, what do you think, Mike, about disposable tools? Are we in the era of just creating something for the heck of it and then throwing it away? >> I'm 100% on board on this one. I we predicted this one probably about two or three months ago. I was a bit of time ago and we kind of unpacked this idea of like what if what if I'm going to build things on my machine that I've just used for me this one time and I've coined this term on the podcast that I run called explicit measures, which is this hyperpersonalization. I I I really think everything is going to be hyperpersonalized. I need a tool that does XYZ, like whatever that thing is. >> And I've been actually um evangelizing this a little bit to other people. I had a a co-orker um they're a different company and I said, "Look, have you contemplated using like GitHub Copilot to help you just kind of like vibe code your way into a solution?" They're like, "No, I haven't considered it." And we spent about an hour and a half just kind of together noodling out the idea. And then **[00:02:00]** I come back a week later and she had already built four tools all vibecoded and was like, "Look at all this this amazing production I've got out of this." And I'm like, "This is the point. The point is you have a problem. The AI is going to be way more knowledgeable about whatever that thing is you're trying to build. And it can easily digest it and make something that's functionally usable. And if if the time it takes me to build something like that only takes me 15 minutes, 20 minutes, 30, an hour, right? And you get hundreds of hours of saved time out of it, like okay, great. And the neat part about this is why I say disposable to some degree is because if it the reason it has to become disposable is because the language models are getting so much better every couple months, you're going to want to like revisit it. >> And I don't know how we do this. This is what I don't understand quite yet, which which is, you know, my prompts that I used originally could probably be reused again to build the same app. >> Armando, we've been doing some experimentation with this. We've already built one app. Mhm. >> You've built a second or third app and we have since rebuilt our first app using a different tech stack and just basically saying go we kind of stubbed out the idea initially threw away the whole MVP version one of the app rebuilt it on a new architecture and said hey go AI go document everything you wrote before list out all the features be comprehensive. So, we built the app, we had the AI look at what it did previously, and that became like the **[00:04:00]** definition of features for the next app. And I think this is a better development pattern as you become better at prompting, as the models get better, you're going to want to fully rebuild apps because the time to do so is minimal. And now that we have Clawbot or so, OpenClaw, I keep forgetting the name. >> Um, OpenClaw, which is the current name, it's an AI assistant. And you can say, "Hey, I'm gonna build this app." And now you're using Opus, you know, uh, Claude Opus 4.5, right? But tomorrow, what happens when Claude drops five or Chatbt6 shows up or whatever? Who knows? >> So, you want to have the agent go back and say, "Okay, go look at this other app and code library. Write down all the features that was there. I want to rebuild this whole thing overnight." And you just say, "Rebuild it. Have me a prototype in the morning." and it did it. It's like this is insane. I'm so excited about this. >> And the thing is like some of the barriers that you would see, we had this conversation yesterday. I was I was actually on the road when this happened. So Mike is talking to Alfred um the the open agent, >> our open call bot. >> And so they're talking, they're building their something. And then I I see something where Alfred asks, okay, so should we keep the super basease? We were deciding between we're moving sort of the Azure world, right? For my >> that's my world. Like I'm the data guy. So Armano is like more the front end persona, right? You've you've got much more experience there and the WordPress and like a whole bunch of web de webdev stuff. I'm super data **[00:06:00]** and I get anx I get anxious about not being able to like see the tables and I can't if I can't see both sides of the world which superbase probably could do it as well like just not as good at it and I'm an Azure stack so like can like Armando was gracious enough to let me switch my back end out a little bit which is >> yeah it's fine and and because you're experimenting with these and it's like I don't mind using superbase I was saying this this morning like superbase is amazing to get you up and running and it works and you have a free tier and whatever so hobby projects amazing and you can scale If you want, you could always also do it on Azure, AWS, Google Cloud, but for a lot of people, getting just the barrier to entry to just starting using those tools is big. So, but if you have someone on your team, like in this case, Mike, that already knows that infrastructure part just just getting access to it and having the security piece uh built in, then great because then the tool will help you regardless, right? >> Yes. Correct. So we had the first MVP in Superbase and it and it did what what we needed to and then we were saying okay so now use this other stack and I asked it hey are you are you going to stick with superase or are you just moving everything to the database over there in Azure and said well we have option one we can first move it up the the human way right first move it over and keep the database the same way that way we we sort of know that it **[00:08:00]** works and then migrate later on and I was Yeah, but we might run into some iss lose some time like figuring out some things between Superbase and Azure. And I mean just just do it again, right? It was like okay and it just did it. >> Yeah. And it was it was no faster and it did it right. And it's it's amazing >> and I'm literally Armando I am floored by how well this works. So what what's so disposable tools is part of this. >> Yeah. Let's talk about just a little bit of just I want to touch on a bit of the security around these bots. >> How do you get them to build projects with you? I don't I think a lot of people are just kind of like thinking that I got to go into clawbot. I'm going to go say or oh dang it I'm going to say the right name. I'm going to go into open claw. I'm going to go give it my credentials and it just will post things or grab stuff or make things or talk to my git repo. I don't think that's the way we should approach these. I think we should approach openclaw or AI assistants as a single identity and you stand them up almost like a a separate user. Treat them like an onboarded employee. >> And again, I come from the Azure world. It's very easy. Go to enter ID. You can create a user. There's a username and a password. And that way the bot can manage its own thing and you can limit it to surface area of what it has access to. So Armando and I were like, "Well, how do how does Armando and Mike talk to **[00:10:00]** a a bot that both we that Armando and I can both see the Git repo? We can both see the database. We can both see like so the developers, us, the I guess the human side of the equation needs to be able to see things. >> Yeah. >> Independently of what the bot's doing on its own machine. And we don't want to have to log into the bot every time to see anything's going on. We we need the bot to kind of like operate on its own. own take its own notes and language or whatever and then we have to have these tools that we hook into. So to do that we gave our bot identity on GitHub. We made a brand new organization. Welcome to our new org. We've we've called ourselves Forge Meta Studio. So that's what we're doing now. So we have a we have a we have produced our own AI company with AI building solutions for us. So we're just using it in selfishly for us right now. But who knows, these may turn into like legit products at some point. Um, but that I just want to touch on that security part. I think that's a better approach >> is to treat your bot like a new employee and build identities where they need to be. >> Yeah. And keep the security because that's a mistake, right? That many people do, oh, log in. Yeah. Let me log into my GitHub. No, don't do that because then it'll have access to everything >> in your GitHub and all the permissions you have. Like that's not what you want. You want to least per least privileged permissions is like the mantra you want to go by. >> Mhm. And also **[00:12:00]** the computer like everyone, oh Alen is telling my computer. No, no. >> If you don't have an extra computer or you don't have a cloud service, then use a virtual machine at the very least host it there so that it's completely isolated. Openclaw is an amazing tool for anyone, but keep it separate from your stuff. Like don't immediately go and plug it into your computer or to your email. Like give it another email account. You want to give one of your employees access to your email account and like oh you have a VA that's helping you with your think well you might have it also in a certain way where you protect your data so they cannot just take over your company >> so same deal right I think uh in that sense but it is so helpful and so funny too like literally I'm doing something else and I just see the messages coming in and it's you and Alfred talking about tables and schemas and things and then I was like I don't care just ignore you keep going. And then I I I dip out and Armano's like, "That doesn't look right. Let's fix this. Let's add this other dashboard. Let's add these more things." So like we it's almost we've got two PMs on top of the same project, which is kind of fun, too. >> And it's it's so fast. >> It's so fast. >> Super fast. And And then you can And also, it's not like we think of, oh, it's working on this thing. >> No. Like Mike was working on something at the same time that he had it >> Yep. >> working on something legit complex. >> Yes. I told it in another channel, >> hey **[00:14:00]** by the way, I would like a dashboard. You think you can build a dashboard? It started building this whole web app at the same time because really it's not it's it's just using cloud code behind the scenes or you codeex or whatever. So >> crazy fast and fast. I wanted to touch on fast because the reason I think also disposable so disposable in the sense we're talking about now like you can do something do an MVP throw it away do another iteration of it grow that way right but also like many tools that you need right now like oh you're doing your taxes you need to put this information from all a bunch of stuff right now well create a tool >> just locally for yourself that does all that you're finished you're not you don't need to use you can use next year if you want but like the other day I built a tool for a call was going to have a scoping call with a client and I have my questions and instead of using a Google doc or a sheet and then having to do a bunch of steps I said well just talk to cloud code told it exactly what I needed and it built out a web app >> that had all my questions with checkboxes so I could mark them as check and then it had sections with progress of how we're going and then >> went through and it had the pricing and everything so based on what we discussed then at the end of the call the thing could then create >> the full statement of work. Yep. >> Right. Right. During the call with everything and I captured all the information. It gave me a **[00:16:00]** summary et because then it could pipe the output into itself to generate everything. So >> no creation of anything, no expenses, no no I just needed it right there and it goes away. >> It was super helpful. I think thinking like that will help a lot of people >> explode the capabilities in their business. >> Yes. >> And I think Yes. And I'm going to say also in addition that what you described I think to me is again this concept of like hyperpersonalization right >> I am working this is the workflow that I've identified that works for me I need to do it this certain way now >> if you if you stand back and like zoom out on like an entire organization there's probably like very large tasks that are across department or very common things right you you look at the IT organization and team right there are some very large tasks they're adjusting to I need to make sure that these servers are running I need to make sure people have access to this data I need to take the data from here and I got to move it over there. These are like very large level tasks. But if you go down to like the individual person level, each person understands and works towards those larger goals. This is why you have OKRs and objectives in your company. But down to the person level, how each person does their work might be just slightly different based on what they know, what their knowledge is. >> And so I may be like a really big fan of PowerBI and and using um analysis services. That's how I analyze data. another person may not be comfortable with that. They're like, I am an **[00:18:00]** Excel person. I want to analyze that way. So, in both situations, we're basically getting to the same answer. It now becomes a little bit of a a race to the finish to who can do the thing the best. And, you know, my design might be a bit more complicated, but it might be more robust. Like, I can refresh data and bring it back in and be more repeatable. The Excel version might be a bit more ad hoc but quicker and you can deliver short-term insights very quickly but longer term it's more difficult to maintain and update. So there's trade-offs to everything that everyone's doing. This is where the disposable tools I think really shine. >> Mhm. >> Because now you can really step into I am I'm going to get this downloaded data. I'm going to build a little mini HTML page local on my machine that's going to do XYZ things. To your point Arando like >> I'm going to type in a bunch of things. Here's some information. Uh, one thing that I've been really interested in here recently is building skills for AI agents. And I use a lot of VS Code. That's how I do code things. So, what I'm doing now is I'm talking with different agents and saying, hey, I want you to accomplish this. And then I work with the agent to build something and then the output of that is, okay, before I like, for example, let's talk the statement of work thing, right? How I would maybe materialize this in my head. >> I'd start with a, okay, I'm going to do the statement of work. going to have it build a little tiny mini app or whatever that is. I'm going to have it um **[00:20:00]** write down a bunch of documents at the end of that before I go into like the next conversation. I'm going to say write yourself a skill or go back to the agent and say document what you did make this a spec or a standard that we can then reuse over and over again. So I think there's this concept of and this is what um open claw maybe opened up in my mind is the work that you're doing with the agents is capturable. you can capture the process and that's the fundamentals of these disposable products dis disposable software right build it once but if you can get rid of the software if the software just becomes an output the real value is what are the features of that and this is where I think um so random other side note here this is a this is going to come back I promise So, I've been building a lot of music on Sunno AI recently, and it's been super fun. I love it. I'm actually preferring writing my own music as opposed to writing music from uh or listening to music from Spotify because the lyrics are someone else's ideas, someone else's thoughts, right? When I build with I build everything for me. It's hyperpersonalized. So, this idea of being able to hyperpersonalize things and it's the barrier to create has gone so low >> for me to get into creation. It's like this is this is so exciting to me and and this is we're going back to the things that matter now are the prompts, the requirements. >> Right. >> Right. So in technology where everyone says, "Oh, technology will make you easier. It'll be faster. It be less less effort." That's something that's changing **[00:22:00]** now. What are your thoughts, Armando? >> I I completely agree. the skills part of it is and by the way the other thing went away so we can keep going. See the >> I really like the fact that you created something and I did this the other day with uh what was it? I think I we're doing the outreach thing and the first time I had to explain like oh this is an email I got where it bounced please go into column XYZ mark it bounced write the date that it bounced write the message and then make it inactive right >> once then any other time something comes in like that like oh bounce it does exactly that >> so it learns right >> yes yes >> and that's where the The self learning part of these is where it gets really interesting, especially once we get into the open claw. >> Yeah. >> Environment or deal, right? Because now you can tell it and like like Mike was saying in the skills, you can say, >> "Oh, by the way, anytime you learn something new or we tell you to remember something, this could be in the instructions. Yes, >> create yourself a skill or do something like and so next time something like this comes up it's like it knows >> what to do because at the beginning >> we've we've seen it right it's like I sent it a video it's like sorry I got a video I really don't know what to do with it and you're like oh >> can you transcribe it and it's like no I don't have the tools but let me check and and then yeah I have the tools now and it gives you the **[00:24:00]** trash >> yeah it's just crazy it's amazing the other thing I will say around that particular comment around open claw. >> Yeah, >> you're exactly right. I love that comment. There is a little bit of a weirdness where it kind of drops its memory sometimes. So, sometimes I'm learning to talk to OpenClaw or Alfred bot a little bit differently now >> because I know what it should know and then sometimes it'll say I'll give it a a task and say, "Hey, I have a little side project. Why don't you do this thing?" Today I was saying, "Go to this YouTube channel and go find this playlist and then give me all the views and the links for every video in the description of the list." Pretty easy. And I said, "I can't do that. You need to go use the YouTube API." And I said, "Wait a minute. Time out. We have a project that did this. Go read up on this project and then come back to me and tell me what you found." And then it gave me the answer. Oh, yes. I see that I have a skill or something around like how to go get information from a playlist. I'm going to use this scripting thing. I'm going to do D. And then it came back and gave me like the readout. Here's my little markdown. So that was something I had to be just a bit aware of is the memory portion of what it's doing. It's not necessarily always checking its memory all the time. And I would argue it probably shouldn't >> because even in humans, like if we're doing lots of projects together, I have to kind of like trigger your memory. Hey, remember that project we **[00:26:00]** did this time? Okay, I want you to reuse something from that. Right? So that's again that same way you interact with humans. It seems like that's the same with the bots experience as well. You kind of have to like trigger it a bit and I can't always assume it knows everything. And yes, and there's there's people working on this. I saw some some examples. So what happens with humans is we have long-term memory. >> Yeah. >> And short-term memory, but then we have these sort of this is my way of viewing it. We have these sort of meta tags for which we can go and draw information quickly based on triggers or even smells and things like that, right? And so that's that's why for large knowledge bases with LLMs, you use vector stores and things like that so that you can very quickly based on context grab information and not have the full context window of the thing filled up. >> Yeah. >> Because it's not it's not very large really. They say it's large. It's not really large. So >> what I think there's going to be a point where we're going to be able to give it a good instruction set so that it knows to have this long-term memory and then sort of a shortterm memory and then a way to access those. So I saw someone talking about the the memory earlier today. So we're going to try that later with Alfred. But but I think they're not touching on the like meta tags or things to wait to sort of access because this was something it's very funny that uh one of my clients said the other day >> he was laughing because he he's like >> I I **[00:28:00]** don't know how you do it. He said we're talking about this completely random other thing >> software related in a sense but like he sees me interacting with other providers. >> He's like how do you keep track of everything? There's so much. this is completely abstract. >> I've got the same comment, >> right? >> And so I'm like, and I made a video on this. I'm like, yeah, I mean, software engineers, you or people sort of in this space, there's so many abstractions. I see doctors and I'm like, how can you learn that stuff? I just I just I see the med exam. We work with some doctors and it's like what? There's no way you can memorize all that. But it's the same thing with us. It's just a different context. I didn't see it that way, but then I was like, yeah, of course. So I can magically think of very weird things that I worked on in 2008, right? And like yeah, yeah, of course this is the way it worked because whatever you you remember because you tap into that long-term memory. Yes. >> And so I think as we enhance it such that it has the long-term memory and can draw from it when necessary apart from the skills and all that then that's going to be cool. >> This is interesting. I I I like So one thing is I like what you're saying here. I'm also trying to inter I'm trying to also think about interpret what you said about the doctors. >> One thing I've heard and I don't know if this is true or not but I've heard a lot of doctors Google a lot of things. Now there's fundamentals what they learned in school or **[00:30:00]** whatever that is. But I don't assume that every doctor knows every symptom to all the things. And I think now especially with the health care system again thinking of this the same way as I'm looking at code. So code is very complex. There's lots of languages. There's a lot to know about it. Right? There's no one way for one human to know it all. >> The same way a doctor can't know every single symptom or every single thing from a person anywhere in the world. And they they're the stakes are much higher because if they get it wrong, they get sued. There's people dying. Like this is a big you got to be right. You got to be 100% right. You can't have any hallucinations or weird stuff. Now, however, you could also do an area and I think there's probably programs like this that exist already. doctors can go tap in, do a knowledge search and go find things. And so there's terms that they know that they can relate to like Arando and I in programming like while loops, for loops, you know, um context, front end, back end, there's like terms that we understand, there's known objects, but you tell me build an entire backend in Rust, not going to be able to do it. I understand what Rust is. I just haven't written in it. uh Armando, you know, we pick pick some other random front-end thing that's brand new or comes out like you know what the front end should look like, but you're not building on that anymore or don't know it potentially. So, this is where I think the analogy I think fits really true with like the AI parts of this thing is you can just have **[00:32:00]** conversations with it and pull out extract out the information that you need. So, in the same way a doctor would leverage a search tool, an AI to help them get symptoms down to what possible things are, the doctor still needs to reason across the answers that are there and say, "Yeah, this is right." I see the same thing. I've seen a couple tweets around people using X-rays. Same concept, right? Send the X-ray to the agent and the the AI and it finds exactly what it is and highlights it on the thing and says, "Oh, this is the problem." Well, then the human comes in. Yeah, that's the same thing. But now you don't need to pay an hour's of time for a human to go look at an X-ray. You can just send it through a bot first and have it get a good first pass. And then you need the human to kind of bless it because again stakes are high here. You need some human to be liable or system to be liable for for things there. >> But this is the same pattern I think and we're going to start seeing evolving everywhere. And it just seems to be happening in in software technology right now. This is going to look really different in 6 months. Maybe not even that. in three months it might actually look really different. >> That's right. Right. That's right. Because we see the the AI big wave coming mostly in the software side first because that's where it's more immediately apparent. But no, I mean at the end of the day it's going to it is touching everything already, but it's going to touch many more things. I I don't so I was looking at the **[00:34:00]** flow of what like open cloud does and it doesn't need to have everything in context but the skills are important to be to have something that's small right so you have your knowledge that it knows and then you have the context of the conversation and then it can compact some of that conversation but it can keep key points somewhere and then it has its skills so when it needs to do something it does not quite sure it could just read the skills and it's like okay Perfect. I should use this skill and then it'll read a little thing and now it knows how to do that specific thing. So that's amazing. As you give it more skills, it'll be able to do more things. It doesn't. So right now what was happening with us is that many of the things that we've shown it what to do, we haven't created a skill for them. And so that's where when it loses context for whatever reason, it's like, hey, I don't know how to do that. But if if that skill is there, as soon as we add the skill for that thing, now it's going to say, oh, for sure. The same way when you tell it go search this on the web, it doesn't ask you like how do I search the web because it has the skill. It knows how to do that. >> Exactly. Yes. >> I like this. >> This is >> this is where things are getting pretty crazy. And I I think you're I think you're on to something here, Armando. Short-term, long-term memory on these kind of things in addition to the skills pieces, right? That's I think going to be really helpful here as well. And I **[00:36:00]** think there's also right now it feels like a lot of this Clawbot stuff is being managed only with markdown files and like context windows. I think there's something to be said here for like a knowledge graph that's going to have to start appearing on these machines where these machines where these bots are running, right? We're going to need the ability for it to like graphically have like an ontology or, you know, here's short-term memory things. Here's tags that relate to their short-term memories. Here's every project I've worked on and here's summary of the project and tags that go along because that's I think how I in store information, right? I don't know all the details, right? Someone asks you about a particular thing I built, I have to go back. Oh, yeah. I remember vaguely the summary of the project. This is kind of what we're doing because some keywords you said sput out to me, sput out, spit out happened, whatever. It triggered some memory, right? And then from there, I'm able to dive into, okay, let me go pull up the project. And then I go look at the oh yeah, now I remember. And it kind of like then it brings in like your long-term memory. Bring it. So, I think that is it's very interesting that we're seeing a lot of these models and things now starting to mimic a lot of what we experience as humans because it's very similar in the same way, right? You don't know everything at all times. You have to be able to >> that's >> pull from that >> and pull back in. >> That's how we're going to do it. I I have like we can bring up a knowledge graph relatively easily **[00:38:00]** because we've we've done it. So >> either knowledge graph or vector store but maybe knowledge graph makes more sense have it attached to it such that correct >> that long-term memory y >> is there so everything we're doing instead of keeping it you have the markdown files as it's built but then have the knowledge graph as it's long-term memory accessible so that it can like oh I found something grab it if not it's fine but but it has the ability to in milliseconds grab from that and and do the whole um >> information comes out >> yeah And that's and that's where I think it really excites me right now. So one of the projects I've been working on is again I'm a big data platform piece and so when I work in Microsoft fabric data platform I've got a whole bunch of tools on my I have like an ontology I have graph databases I have blob storage for files I have SQL databases Cosmos DB like I have all these I have real-time tools with custod I have so many disposable tools at my my my fingertips. I just need the agent to know that they all exist and then have it be smart enough to know like which tools it needs to live off of in order to persist things, right? So, how does that work? So, to me, I look at this going like, wow, this is really neat that all this is is happening, but I'm of the opinion of I want to start consolidating all of like the data information to a single platform and then inside that platform leveraging bots to kind of help me, you know, rip through my tasks or daily things. And I I fully **[00:40:00]** envision that instead of bringing more people onto your team, you're going to take the people you have and highly skill them and say, "Look, we're going to help you become more productive by really going through your tasks." And I think at the again going back to hyperpersonalization again with disposable tools, bring it back home here >> is yeah, that's really where it's at. the more you can build these disposable interconnected tools, the faster you become and you start moving. Uh H Jensen from um Nvidia, the CEO of Nvidia, I saw a short on him and it impacted me greatly. I've said it multiple times now, >> but he's kept saying um if I could have all of my engineers never write a line of code again, I would do it in a heartbeat. and he said, "I would switch that out for problem solving, problem identification and stating what needs to be done." And I thought he was very insightful on that. And we don't even understand all the things that need to accomplish in our business. We we're not there yet. We haven't figured we haven't asked all the right questions yet to even get stuff out the door. And I think that's the pin for me here is I want to stop doing task based things. I want to start managing and directing and solving problems with disposable tools or tools that maybe persist a bit longer but you know those disposable tools maybe turn into a full-size product a full scale something right and then we can address it and I think the barrier to rewriting your whole tool >> is almost next to nothing now it's almost like >> yeah who cares start now build something and if it doesn't work **[00:42:00]** switch it out do another one >> yeah also for those who don't know that would take like two Yeah, that's this is this is what's blowing my mind. Like, hey, I want to rebuild this whole thing. Switch out the whole backend architecture. And again, I took a little bit longer because I had to get all the connection struct set up, >> right? >> But to your point, it rebuilt the whole app, brand new everything. We said we sw tailwind figuring out the new front end, switched out the back end, and we just gave it a prompt with five minutes of work, and we just let it eat. We walked and came back, and it had it all done, and it was so fast. That's that's the type of production I think people need to start getting their head around. Like that's how things are going to change. It's going to be more of that direction. Now, >> I think I have to touch on something real quick. Like there's there's this thing of oh because you said it and people are going to jump out and be like no but I don't want to manage people. Like I don't want to be a manager. I want to it's very different to manage agents than to manage people. >> Agree. >> So you don't necessarily have to have the manager mindset to manage agents. I I am I like to do things. I Yes. I like performing. I like actively doing things. I like the whole software development thing, but not I've realized >> not because I like writing the code, but because of what I can do with it. So wielding code I think is a sort of alchemy because you can have an idea **[00:44:00]** and create something out of nothing. It's really nice. That's what I like about software. However, it's not that I love Python or I love JavaScript or I love C++. No, I don't mind what it is. And sometimes I'm like, how did I do this other thing? And sort of, no, I I like to have the idea and say, okay, computer, do this, which is what you're doing when you're coding. >> When you're coding, you're saying computer do this, but in a in a very impractical way for a human. So if I can tell the computer do this and it actually does it. You know how we used to in the in the photography world, we used to always say like people we would laugh at people who don't know about photography and they like you take a picture of them like, "Oh yeah, just just Photoshop me, right?" Like they they would just tell you that and and you're like, "Dude, that takes like an hour work." And now literally you could just do like remove cables, fix blemishes, and it'll do it in a second. But >> yes, >> they were they were far ahead. Um, so >> I think it's still going to be a lot of fun. I'm having a lot of fun as I think you are >> telling the thing. >> Yep. >> I do this and it builds it. Cool. Okay, now let's do this other thing. Okay, great. Now let's do this other thing. Or if you if you are thinking like an engineer, then you might say, okay, this is the whole thing we want to build. But we break it down into little steps. This is all the things. Hey, think to your point in the **[00:46:00]** business. Interview me. >> What am I missing? and it can ask you a bunch of things that you didn't think of. >> And then okay, perfect. Now you have your document of tasks, etc. And now you can tell it go forth >> and build and it'll do and it'll test and it'll see because now you can have it go in the browser and actually test the whole thing. So it will be do a much better job if you prompt it correctly >> and give it the right tools than what you would do in most cases. And and people who defend their no but it's me. No, like no person can code better than the LLM in my opinion because they know everything. But a person can plan better and >> can tell what so if you tell it exactly what you need in the correct way, it'll it'll do a great job about it. So I don't know. That's what I think. I love that assumption. The everything you everything you Armando, we are like the same like we're having the same aha moments. This is super fun. I'm really loving unpacking this and we're going to keep keep pushing on this stuff. we're going to keep building things and um I'd like to continue just kind of investing in understanding like where all where all this intersects and again I'm I'm expanding like my knowledge like I'm looking at the data space of other MVPs in my space in data and the next logical step for them is to explore development tools to build these disposable type tools like that's the next logical step like I already understand the data structure I understand how to articulate business problems and get it to something else **[00:48:00]** so I'm switching out the medium of, you know, working with business teams now and saying, "Okay, well, instead of working with them to get stuff done, I'm taking the requests and then I'm having agents and things develop and build things that we want." And I think I think we're in a really interesting world right now and it's going to continue to get really fascinating as we go here. So, I thoroughly enjoyed this conversation. A lot of fun. I know Armando, we probably need to wrap here. We're about at the time of doing these things. So, stay tuned. You'll see a lot of these things on shorts as well. We're starting to play with the formats that we're doing as well. Uh but that being said, Armando, thanks for having me on the show today. This was a lot of fun. >> Oh yeah, anytime, Mike. It's always fun. >> Absolutely loving unpacking this. Please let us know down in the comments. Do you like this kind of conversation? Is there more things you'd like for us to talk about specifically to AI open claw? Like what is interesting to you? Give us some feedback and we'll do some more shows. Do you want to have this weekly? Do you want this multiple times per week? This is pretty lightweight for Armando and I to get in and talk about what we're experiencing. If you like this kind of conversation, let us know in the comments below and we'll see what we can do to to meet some objectives there. Awesome. Thank you so much. >> Thanks everyone. See you in the next one. Cheers.