Go back

Vibe Working Is Here: Agent Teams, Claude Code & the Future of SaaS | EP 124

57m 27s

Vibe Working Is Here: Agent Teams, Claude Code & the Future of SaaS | EP 124

The discussion centers on the rise of "vibe working," where AI agents handle tasks like coding, document generation, and analysis through natural language commands. Tools such as Claude Code enable teams of specialized agents to collaborate, significantly boosting productivity. However, fully replacing software development with AI prompts is still limited by complexities like security and integration. The AI market is in fierce competition, with companies like Anthropic, OpenAI, and Google offering heavily subsidized access to advanced models, making it an opportune time for businesses and individuals to leverage these tools. The conversation highlights a shift from rigid SaaS products to AI-driven platforms that assist throughout entire workflows. Despite rapid progress, challenges include managing AI agents efficiently and adapting to potential future cost increases as models evolve and subsidies change.

Transcription

8879 Words, 46623 Characters

English
When your agent is finished and waiting for you, you get stressed because you didn't give them a job. So you want to make sure every night when you're living or the weekend, when you're living, you want to keep your agents busy so they're like, they keep working. But that's agent stress. Your agents are waiting for you. So whoever I'm talking to, they're saying, "Cloth code is handling everything. Cloth code is getting the job done. We are not writing code anymore." Hi, my name is Demetri Benici and I'm a content creator, agency owner, and AI enthusiast. You're listening to the AI Agents Podcast brought to you by JotForm and featuring our very own CEO and founder, Ida Kintank. This is the show where artificial intelligence meets innovation, productivity, and the tools shaping the future of work. Enjoy the show. Hello and welcome back to another episode of the AI Agents Podcast. We're going to be talking about vibe working and it's going to be really exciting how to vibe work and manage your AI agencies is something that a lot of people are talking about right now. I'm working in 2026 is something that we both been trying to do and we're just even fiddling with some stuff before the call to try to get you guys to be more in touch with what we're doing and I'm just excited to kind of show the stuff that we're doing here. Yeah, it's so I mean why why coding has been like a thing since last year but then recently, especially with cloud codes improvements, why coding became so good like whoever I talk to like I had friends they have like their own startups and they're like they're seeing like okay we are not writing code anymore and and one of my friends said like basically he actually has these team of agents like and team of AI agents that he's basically constantly running to like right code and then he's he's kind of like you know constantly like you know one of the agents finished the job and then he quickly goes and like guys like agent to do something else and there's even this new new stress called agent stress I don't know if you heard of that term it's kind of when your agent is finished and waiting for you you get stressed because like you you didn't give them a job so you want to make sure like every night when you're leaving or for the weekend when you're leaving you want to keep your agents busy so they're like they keep working I don't know if you need them because it's like you can even run agents in product now and we are going to showcase it today but that is that's you know even like this agent stress like your agents are waiting for you so whoever I'm talking to like they're seeing like okay cloud code is handling everything cloud code is getting the job done like we are not writing code anymore and I mean at chat form we are also using cursor and cloud code I don't know if it like we actually have some people who are just you know developing stuff like internal tools especially by just wipe coding but usually like when we are developing our product it's just a helper it's not like you know taking over the job yet but and this has been happening recently like last couple of months something is happening and that was even this article about like something big as happening let me try to find that article to share as well yeah yeah you saw that right it's like a couple of things called something biggest happening it's just literally the person was holding a piece of paper right so what's happening is that all these tools developed for wipe coding are now like you know someone actually thought okay hey like this what we built to code it can actually do other things it can use the browser it can use any software like your spreadsheet your you know document editor whatever you need to so if you are actually doing like office work you know it can actually do those things and then and then we you know we showcase cover recently that was amazing last week we covered open cloth it constantly changes them so open cloth was the last thing I believe yeah and that's also amazing like this just this looks so fast and crazy and like just feels so insecure like feels so it's so fast and like just does everything so quickly that you know you know you feel like you have to always watch what it's doing because it's kind of scary uh core work is much slower like it's slow it's very safe and uh you know it's too sometimes too slow too safe on this uh but you know the demo I will do today is it's gonna be on cover because I wanted to showcase like parallel agent teams like how agent like how agents can be run on product uh so that you can actually do the job uh much more quickly and get things done done and you could even like have these like different um specialties you can have different agent specialties like if you are building a website you can have this uh designer and then the developer and then the researcher and then the you know compensate the researcher and all these agents can actually work together uh communicating with each each other and this is the clouds uh agent teams features and uh you know so it's it's all really cool stuff that has been happening um so we were gonna try to and and the and the term wipe working right it was quite an on-linked post I think you're gonna share that um thank you share that yeah the the post starts off I say and I need uh this is power by it's today is the february 19th february 6th and I needed to brace ourselves because this was the weak AI stopped being a conversation it became earthquake 285 85 billion dollars that's how much market value to vanish from the global software industry in today that because of recession now because of the war because of a plug them and drop it launched a new capabilities for the AI agent co-work tools that can reduce contracts that's how the A's manage compliance workflows and draft legal documents well straight direction with some immediate and brutal Thomas Raiders dropped 18% which is insane uh it's worth staying the company's entire history uh relics the parent company of lexas nexus fell 14% its fortune since 19 to 8 legal zoom which I'm sure you're familiar with is like a company that helps you but you know uh legal documents pretty quickly plunged nearly 20% and the head of equity time a ship head of equity research at lpl financial ask the question now loud why would I continue paying for software when developing these systems is becoming faster and more efficient with AI good question it's not a prediction it's a geology um I'm trying to find that quote that you just uh you mentioned earlier from scott white here it is so with this article kind of brings more into the conversation what anthropology is doing um staying how it's taking a different path partnering with firms like the wait and cognizant and embedding engineers directly at place like old man sacks which is crazy I'm actually kind of shocked with the doing that the quote that relief stuck out to me that's about this is yes it brings us into the phrase of the week got white said it was that in terms of the head of enterprise brother he told CNBC I think we are now transitioning into five working you remember by coding or developers could build software by just describing what they needed by working takes it further knowledge workers and to go up entire cast AI agents using nothing but natural language financial analysis legal research documentation generation and then an open AI chair Brit Taylor dropped what might be the most important question of the year the real question is what software still matters that's a good question I think I think in what software even still matters um yeah uh so the question is like uh what what's what's what's going to happen to the software like sars businesses um I think it's it's going to be really challenging for many sars businesses probably including us because you can actually do lots of things with this software so it's kind of also good because it's going to ask you know push push outs towards being more valuable for our users so you cannot just replace it with some prompt I think that's that's the part you know the good software they don't just do stuff for it they actually help you tell you like what's the best way to do something and and guide you like it's it's not just about like you know building forms for example but with just form like with especially with our new AI stuff it's more about like really making sure that whatever you are trying to do like we help you like we guide you to do the right like you know we help you build the right workflows like you know uh like the emails or the you know all the like the how you use the data how you analyze the data like all these you know the not just the form but also all the other additional steps that you follow after you fill out the form. So it's gonna, I mean, if you have a business, like if you have a software business that's just one prompt, like can be copied with a one prompt, I think it's just, that's right. Like it's just very easy to replace. But I think we are still far away from like really building reliable software with a single prompt. I think it takes a long time. I mean, it's just, it's not just about like creating a prototype. Most of the time, like when you create an app with AI, it's just, yeah, it's a good prototype. But when you try to like advance it, it just gets harder and harder because like, you know, you have to think about all these other stuff like, you know, compliances or like security or integration with other, other stuff, other software. Like there are lots of things that are not that easy. But yeah, I mean, I think it's, it's, it's interesting times. I think what's gonna happen is more likely is this software business will also change. So this software, like the sales products are gonna become less of this like rigid, you know, tools, but more like tools that can actually understand what you're trying to do and actually accomplish, help you accomplish those things. So with the help of AI, it's good. They're gonna be like, you know, they're gonna help you the whole way. They're not just gonna have, they're actually not gonna give you a tool, but they're just gonna, because the, you know, the sales companies are also using AI, adding AI to their products. And I think that's, that's an interesting, uh, development and like just, you know, we have been doing some cool stuff with JodForm. So we can probably cover it in another episode. Yeah. And now getting into that, that thing that you were just referencing was something biggest happening article by Matt Schumer. Oh, yeah. Super capital. He posted this the other week, which got a lot of people thinking, I went really viral on Twitter and on LinkedIn. It was, it was pretty interesting. I'm actually, I'm seeing a lot of these types of posts and a lot of people I think the hard thing is that they don't really quite see the value and return for the business. They think about it individually. It's doing really well. Um, so I think what would be, would be good to kind of mention here is, I want to acknowledge kind of how fast things are moving and this is a pretty good thing that he mentioned. So how fast is the improvements actually going? So I'm trying to do AI could do basic math reliably by 2023. It could pass the bar, right? The, the like legal exam by 2024. It could write working software and explain graduate level science. By late 2025, some of the best engineers in the world said they handed over most of their coding work to AI and on the February 5th, new models arrived that made everything before them feel like a different era. We, I mean, I know you just worked with sub agents for a little bit on your own and parallel agents inside a lot of little co-work and in cloud code. I'm doing a lot of the same stuff. So I would be curious to kind of hear and see what it is that you think is as a business owner and individually the best thing moving forward for you to kind of do because I think I'm noticing that there's a lot of work that I'm trying to offload that can be by offloaded specifically in the last like six weeks specifically opus 4.5 opus 4.6 these models are absolutely game changing for me. I need to use that phrase because they has real that phrase. But it's it's after it's an absolute game changer because I'm able to open up my computer and just do stuff. I know you were trying to work on some stuff. I'm going to show real quickly what I think is important to note because as I mentioned in the previous episode, I bought a Mac Mini for open claw. Now did that go okay? It went okay. Let me explain why. I'm decently technical. We know this. But I am not an engineer. So I think I'm actually an interesting spot to be in because I can I'm very smart with APIs and very smart with workloads and stuff like that. I'm very comfortable using AI. So I'm not going to be overly like never going to give you some sort of reference point that's like out of touch I think with with the average like go getter business owner. I don't think and I noticed that it takes a lot of work to actually get it going in the right direction. Like you do need to know what you're doing with these different things like open claw. In order for it to not issue off and make you spend like hours and hours and hours like in an infinite loop. Best to go around with with it. There's the security issues. There's all this type of stuff. But for me, I just want to start the thing by saying as an individual this can be really powerful if you if you spend the time on it. But as a business owner, it can be very frustrating to try to find the returns with the cost of their ad. We were talking about this before. A lot of this is actually pretty heavily subsidized. What does that mean? What plan are you using per claw code or cloud? Oh, just max, max plan. Okay, the max $200 plan. Yeah. Okay. That plan right now for those of you that are unaware. If you use it to its fullest extent, it's worth $20,000 a month in token. Put that into perspective. It's worth $20,000 worth of tokens. I'm gonna mic drop again. It's something that we're not gonna see forever, people. So I want to call this out as an urgent call to like learn how to do this type of stuff and utilize it to set up work loads in the short term. Because in the long term, it's kind of cost of decent amount of money to do the same things. But what we're gonna notice, I think, is that and I saw this yesterday. I was doing some research on it. The frontier models, right, clawed, open AI. They're gonna keep getting better. But I don't know if you've noticed as well. Local models are also improving very much so. Like open source models are getting better. KMK 2.5, 1 3.5. They're all way cheaper. They're on the cloud. But they're also getting easier and easier to run locally. There is like a ridiculous thing people are doing that are nerdy. They're buying max studios and chaining them together in order to locally host KMK 2.5, which is a very smart model. Is it worse than Opus 4.6? Yes. Is it going to be worse than Opus 4.6 is right now in six months? No. Right. And that's the thing that we're noticing. So point being, it's going to fall off. There's going to be a point and I'm sure you know this is a business that are like, and their investors are going to want to recruit their investment at some point. They're going to be like, all right, we've been given away. We've been using this and you might be asking by the heavily they subsidizing it so much. One, you and me and everybody else are going to champion this thing because we use it. We like it. We're going to be ingratiated with it. So we'll use it as a business owner with the API costs. But two, they're also using us as training data, right? I'm sure. So point is, I want you guys to learn how to do this now because you can build a lot of great infrastructure. You can build in-house software. You can build in-house workloads and cron jobs and stuff like that. And then when they hike up the price, hopefully you could just shift on to like a local model or something that's cheaper in the future. But for now, we are really being subsidized at like a nerd group here that like pays attention to this stuff. So I'm just using this as a call to action to actually take action with the advantage we have right now because in three to three years when they say they pull the plug on this, they're going to be we're going to not be able to have the same type of access as they do now practically to this much value for dollar. And it's after the third reason. And that's market share. So there's this huge competition. So cloud is gaining so much momentum recently, especially with cloud code is like becoming like the standards. And so CHGPT has been growing especially in consumer. And it's really good like 5.2 is great. So and then Gemini is also doing great. So there's this competition between those three products. And I doubt that any one of them will jack off the prices because because they want to gain more market share. So I think for a long time, we are gonna still have these prices. - Right. - How long do you think I'm curious? - I think as long as this competition continues between those three, so all of the time you own the events, just CHCBT, HSI consumer, they have like a billion users, they became like the name for it. It's just in AI, HHPT. But when you go to like the best users for AI, like the biggest users for AI, people are actually writing code, like the old developers in the world pretty much. They kind of pretty much switched to cloud code enough. So that means they have like the most important segment in the market, even though they are much smaller, cloud. And Gemini has this huge like Google's user base, like the Gmail and Google Drive, billions of people, pretty much everyone is using them. So they are not doing a great job. Like every time I try to use like this Gmail, the Gemini on Gmail or Google Drive, like I get disappointed. I'm thinking like, okay, Gemini is this good. Like, like, and I go to Gmail and I, I ask you a question, they are still like, if they could just get those two things like Google Drive, Gemini and the Gmail Gemini, if they could just get them working properly, I think they could just like build a momentum from that as well, but they're not doing a great job on that. So I think this is still like a big fight. And they are all these three, you know, top contenders, they are actually doing a great job. But, you know, the winner is not clear. I don't think it's gonna be clear for a long time because just as soon like everybody was thinking that Chichi Pt was winning and then Cloud was like, came up with like the Gemini, not Gemini Cloud code and now Cloud Core Work is amazing. And so I think it's great time for us because now we can actually use this, you know, products and do lots of great stuff. - Absolutely. And life. So for example, I just wanna show this to you because I'm somebody who hasn't been messing around a lot with it. But there's something as simple as this. Is able to be made in AI with AI right now. If you know what you're doing with like something like Cloud code and Cloud Core, the sound effects, there's lighting, there's all this type of good stuff, right? It's incredible. I wanna just call this out because I spent like the day putting together the visuals or some ideas. Well, I spent a day putting together static images and saying like, oh, this is my new like LinkedIn style. But another day building out the video editing or video creation edge and based off that. - It would have cost me it in tokens. I don't like $2,000 to put together all this stuff, but it's instead cost me, you know, a day of the plan each time, which is incredible. So what I have to call out is the fact that there is a distinct rartinity to think of different things like that for yourself individually to do and to try it out. Because that's something that I literally cannot imagine doing a month ago, a couple weeks ago, a day ago, like it feels like right. And that's the stuff that I don't know what you've been working on with co-work. But for me, it's like the idea of finishing a meeting, I mean, this hate ads proposal put together for a client or something and then them us aligning on what next steps are, then me telling my log code to analyze the transcript and to do the work. And all I'm having to do is check on it, add more prompts, work, add more MCPs, right? And all these little things to order connected to different tools. Now even you have tools that are controlling your computer. So practically speaking, you know, you have Sonnet 4.6 just throughout, right? It's continuously improving its computer understanding capabilities. So you're going to see more and more throughout this year. Vive working is going to be possible because it's going to have better computer understanding with every single model, better UI, UX understanding with every single model. Better thinking, better Google Sheets creation, better spreadsheet understanding. And for her, what you just saw, take a wild guess how long it would take to make one motion graphic like that if you were a video editor. I don't know if you do much of it, but it takes like, it takes like three hours. It makes it like that. Yeah. It took, it made 500 iterations and about eight hours would be working on it. And I was also doing something else. I was also doing other stuff. That's where we're at right now with Vive working. That's pretty crazy. Did you have it right? Free peak. Free peak. Like share your screen. Yeah, I mean, we just, we just, let's see where is it. Okay. We just announced platforms like 20 year anniversary. Like we did a lot of stuff like videos. And our team has been using this, like, it's amazing. Like just, they were so impressed. And I was like, okay, did you actually use nano-banono? And they were in actual, they were using this free peak. And like those, you know, motion graphics, things like that. And just, you know, doing those kind of motion custom graphics and then adding them to their videos. And the videos look so amazing. So just, just something to share. Yeah, no, that is good to share. Yeah. Yeah. I appreciate that. You know, there's a lot of, there's a lot of, because the stuff they've done with this product is amazing. Yeah, maybe we'll cover it in any episodes. Absolutely, that'd be cool. I like the lot of the, I think a lot of the creative stuff that's going on with, with, oh, right now, it's actually a little bit unterrated. There's a lot of people I think in the, you know, granted, I appreciate it. But there's a lot of people in like the communities of like, of media that like seem to want to, the, who, all the AI stuff. And like, I watched a clip on Twitter. I'm not sure if I could find it. We're showcase, if you're watching Spider-Man. Oh, yeah. Yeah, like they did like a whole AI trailer about like a future Spider-Man movie where Spider-Man said that he would as kid die. And like Mary Jane's like in a wheelchair. And the music's like insanely sad. It shows Tobium Aguire old and with like a beard and, like, he's Spider-Man. Then there's this new kid who's the next Spider-Man. And people are like, and the title says, I'm shocked that anybody thinks this isn't the future of cinema. And then somebody else respond immediately with like, let's just, they'll never make scenes like this. And it's literally a scene and God, God bless Tobium Aguire. But like, you remember the 2000 Spider-Man movies? They were kind of cringy at some points. And like, it's the showing the ending of him being like sad and walking away from Mary Jane. And he's like, and it's like a voice over in his head. He's like, yeah, it's I'm Spider-Man. And I'm like, if you're being honest with yourself, this is cringy. It's just you, it's just nostalgia. Like, sorry, it's a cringy scene. It's not that well shot, but you're just pretending like this amazing thing you just saw was bad 'cause it wasn't the nostalgic thing you remember. And I love media. I'm an media. I'm in this sort of realm. Sure, being human is very important. But I'm like, it doesn't mean that there can't be opportunity in media too. And I just want to like harp on that. There's opportunity in every realm, whether it be motion graphics, whether it be basic office work, whether it be paid ads, whatever it is. There is the ability right now to take your idea using cloud code, cloud cowork, open claw, figure out a way to get it to do it self-manulate. And I will say there's one thing that I was trying to mention with the pricing is important for the short term. Figure out how to get the most out of these plans. And then go from there 'cause I don't know if you saw open claw actually removed access to, or sorry, and profit removed open claw access yesterday. Did you see that? - I think it's happened because open AI should bot open claw. - That's an interesting one. - Yeah, it's an interesting one. Well, I know they bought the guys. right? Like you're about the guy. It's like it feels like one in the same, right? It's like they bought the. So you have you have core work and you have open claw and open claw is like much better core work like because it's just much better but unsafe, very dangerous core work and then open AI goes and buys it and then obviously atro pick is not happy because you know it's now they're made better. They're made better. Got it and then even some out misuse was like you know it's it's weird like he was like I don't remember like he was like we got open claw or something like that we conquered off a claw or something like that just it was weird when they bought it. That's how you report like you know acquisition. It's a very it was a very weird situation but yes you're right and so there's two reasons for it to my understanding. One was that and two was basically there's something happening right now that with the pricing discrepancy and that's where like a libel lia alarm bells went off. I'm where I was like maybe there's going to be an issue here right for the foreseeable future but is that they're gonna like own down on these this pricing discrepancy but maybe they're just gonna do it from an API standpoint. So you're having people use their open claw and the claw $100 max plan and getting the $20,000 out of it right. Kill with all this us nerds and what you got to see here that's really interesting and the topic didn't just remove access to open claw yesterday. They removed access to their OAuth token authorization for wad plans or anything not only open club but even their own agent SDK which is fascinating to me but long story short we'll say there are abilities that you can do to kind of I don't want to say work around it but as an individual I want to keep highlighting this as an individual you can get a lot out of it like for example right now open claw. You cannot use quad max plan in order to through open claw build it up right now. What is open claw except for an agentic software sort of. You know what you can do since open claw is like a software and has files like anything else as next year's like anything else and all these sort of things. You can if you know what you're doing build up your open claw through the clawed code terminal and get the $20,000 value. It just takes an extra step of the technical understanding. Oh you can just just work. Well yeah no yeah for sure or you can just use coer too that's true. So there's there's there's if you don't want to be a technical and you don't want to do open claw you don't want to work automatically and you can just use coer then very true. Yeah I want to show like how to parallel like how to use parallel agents in coer. Let's share share that. So basically. So like research like this is a prompt that I used to try it and it worked really well and it's going to take a while so I'm just going to show you the output but I want to show the parallel agents and the agent teams. So and what you can do is like you can actually build these agents and you can actually give these agents different kind of like skills teach time different skills. They say one of your agent is like a screenshot taker one of them one of your agents is like a web page builder or something like that. So you can actually build all these different types of agents and then you can get them to work together. In this example I'm just showing how to do parallel use them and research 10 SaaS products is powered by AI for each one launch a parallel agent to research what makes the AI so powerful and I create a slideshow. So it's you know very simple prompt and I'm running with opposite 4.6 and so one of the things it does is it's going to create basically let's first because I will ask for PowerPoint it found the PowerPoint skills and it's going to create like this to do list and basically it's doing a research like finding like picking like 10 products and then it's going to just spawn 10 part of agents to do like research each product on so on. So this this way we say time and the great thing is like you can just say like hundreds part of agents and then you will it's going to take longer but you can always leave it to work right you can just do something yes it's just working for you out there and as you said it's so cheap because it's just subsidized so so basically it's it's thinking right now but let me show you the outputs so basically it found these you know 10 products and these 10 products and for each one it made a research and created a slideshow like this okay here's the part here's the fun part started so it says running agents some multitasking so basically it's it's fun all these like 10 different agents so it picked all 10 different sales products and basically doing research on each one and basically using the browser to you know go to this website and you know research them and then just working on part of and that's the cool part you can actually see see them all you know it has multiple steps grandma AI three steps grandma AI writing assistant tagline description grandma AI technology features how it works grandma statistics achievements user valuations so basically it's doing you know each one is like deep research running there and then finally it's going to make you know it's gonna create the presentation so as soon as these all these product agents are done it's gonna go ahead and create the presentation while it's doing the work I'm gonna show you the I just run this prompt before the show and what's created caspere AI AI marketing content is kale so monthly active users air marketing agents UI pad enterprise automation with agent the AI gong revenue intelligence then this AI is that result tickets synthesis yeah text to video so basically you know it summarized as well like just made these I guess I mean so how I use this usually like when I'm working on on a product I'm like if you're building like a one type of feature let's say we are building a copilot for our product I'm gonna just go here go to copa co-work class co-work and say okay go and like do this research and like find me hundred different copilot different you know products that have copilot and then like for each one basically just you know find me screenshot take a screenshot and then create me reports so this actually like if I gave this project to a product manager that will probably take that product manager but like you know I don't know a month but you know AI can handle it in any couple of hours and then I get this report and I also ask it to like you know compare them to each other find the reasons why people love them like just you know try to find the reviews about these features and then just create me a report and this allows me to like just you know basically do the research for me and when when we are working on our product this is really helpful and it's still working on this so I'm not gonna wait for it it might take a while because it's doing this that search but basically it's creating this PowerPoint here and then adding like all these different agents working separately and then they're adding adding their pages to this PowerPoint and it's gonna complete PowerPoint and just give it to me and yeah I think this is really cool just this is a very simple example I'm gonna stop sharing but basically if you like let's say you have a team and you have this cross-functional team you have a designer and you We have a developer, you have like, you know, maybe product manager in this team. Like you could basically just create the team in clubs, you could give each agent their own skills and just train them. Like, you know, if you need one of them to take screenshots, just like train them how to do the best job. And then, and then what you can do is like, you can give projects to these teams. And you can say things like, okay, do this project in this folder and when you're finished, you know, so I can let me know. And then, you know, you go to sleep and in the morning, you have the product ready or whatever you ask for them. And this is amazing. Just shows us the direction. And it's just gonna, the work is gonna be much better because it's cross-functional, right? Different types of agents. The context are not like mixed up between each other. Like each specific agent has their own skills, knows like what to focus on and, and it's not messed up because of the context. Or it doesn't go out of like, you know, tokens or something, right? Even though like the, I think the latest opus is like one million tokens or something like that. You still don't wanna have like very large, you know, context in your agents. You wanna make sure that your agents are like really doing their job well. And to be able to accomplish that, I think it's agent teams are great. And hopefully, you know, we are gonna see more and more examples of these agent teams. And I will try to, you know, use the agent teams to do another demo in the future as well. Yeah. Yeah, for sure. I'm using, I'm gonna pull up something I'm doing in real life to kind of show you as well. I should do an example of the output right now, but I'm trying to, I'm trying to get this. Let me just get my screen share doing better. So let me, once that, so let me pull this up. So what I'm having to do is I'm having to, you forgive me. It's a little bit janky. This is Jerry. Hi, Jared. Hi, Jared. Yeah, hi, Jared. So Jerry is my open clock. It's a joke because if you translate my name into English, it means Jim, but also Jerry, like a funny way. I feel like I'm saying Jim sometimes. And then his name is Jerry Baker. Baker is how you translate my nonsense into the English show. I'm just being funny. So we got Jerry Baker. And what he's doing over here is this is his quad code. What I have it doing right now, which is pretty cool is if I pull this up, I want to see what happens if I, oh, I actually can do this. So I'll know it until me broke just now. Now we did nothing. Cool. So this is what I've been doing with open claw. It's, it's a mission control that I've built. I kind of make it look nicer. So this is a, the big issue with something like open claw is you kind of have no idea as to what the hell it's doing. Unless you make a mission control. You're just, it's like, okay, so I built all this stuff. What does this even mean? Right. Did you, did you use cloud code or open claw? That's a great, that's a great question. So because of the issues of what's going on with the tokens, right? And it up using a cloud code to build this out. And it has the bill it gives all the crime jobs and all that kind of stuff. They're stored on your computer, right? And sure, there's stuff that's sensitive to server, but things are stored on your computer for, for decent percentage of this. So basically I put this together. Sorry, my stupid things crashing, the sharing their screen thing is crashing. So this is essentially what it is. Maybe I could get Jerry on the call and I could screen share it through my other computer. But long story short, is this guy is a mission control that I made using clawed code work or a quad code. So cloud code has the ability to do a lot of cool stuff. And I can just pull up my open my current one right now. So I'm working on some paid ads right now. And if you see right here, I ended up making, I spent a bunch of time doing a bunch of different things here. Like I'm making a, and this is the crazy thing. So I'm making it a tool right now for a ad dashboard to analyze paid ads that I'm running because look or studio is not that good. Look or studio is Google's tool. And I am setting it up. Do you know what verse L is? I'm just guessing. Oh, yeah. Yeah. So I'm just literally going to, I literally built this out and I'm going to have her verse L posted. Okay. Right. And all it's going to be is literally Google sheet data that's piping in like my YouTube's ad performance, my video ad verse, my Google and LinkedIn ads, my LinkedIn company page and YouTube performance. And it's going to be AI powered. It's going to analyze how the ads are doing. It's going to report to them without me having to do work. It's going to make budget pacing changes without me having to do anything. And all this comes from log code. Now, if you guys like cloud code work, which I love, I have to call this out because I didn't realize this because I and not have technically sound as developers. Log code in cloud code work are actually really similar thing. The only difference is that this, you want to build stuff, log code can let you do that, but you don't have to be technical. You do not like, I'm literally just prompting that whole video thing that I showed you earlier. That like motion graphic was cloud code. Mm hmm. And all I did was I said, go on YouTube and find me how to create this type of stuff. And it's like, okay, I will give you a sec. And it did. And then from there, it found the emotion and it did the terminal implementation and all of that stuff. Whereas cloud co work is limited by things that are outside of the terminal that are outside of coding capability, which is fine. But if there is that next level of thing, you're not going to have to learn how to code. It's going to do it for you. It's going to figure it out for you. So if you're like really enjoying cloud worker, called code work, play around with it, play around with cloud code, because that dashboard, I just showed you was literally me saying, Hey, can you make this visual or what? And it was like, yeah, for sure. What do you want? And I said, do this. And it made the dashboard. So it's not like it's anything more than that for cloud code inside of the cloud app. Just that one people. So if you added something to your dashboard, can you show us the demo? Like adding to it. Yeah, for sure. So let me let me screen share again real quick. So I got this guy. Let me just cancel the screen share restart it because it sometimes bugs out. I'm going from computer to computer here. So bear with it. Let's let's screen share. Jerry is so busy. It's doing everything. Yeah, shout out, Jerry. And he has his own. You're coming here with it by email. Telegram. Yeah. You can do telegram. What's that? All this type of stuff. Okay. And all those things they were working with open cloud. They still work with manual switch to clouds code. Yeah, basically what I'm doing is I'm you can through telegram. Be like, hey, do this, right? That's fine. What you should do though is spend the time to build stuff out using cloud code because it's cheaper. So it's like, I'm going to build out the crown jobs and the plan for the crown job in cloud. Oh, and then say to open claw. Okay. Now do this on a recurring basis after I built it the skill. That makes sense. Yeah. All right. I'm sharing again. There we go. I got started. Yeah, I'm going computer to computer here. All right. So let's take a look at this, for example, right now. What is it doing? I gave it a huge prompt. Or basically, I'm like communicating between the two things, right? So I'm going to share my screen. I, yeah. This is how I'll literally do it like in real life. So if I pulled this up, this is my max mini that I own, by the way. So I go here, I go to this telegram. Right. And I say something like, hey, um, I want you to make it look visually nicer. Here's a screen. Here's a shot I've seen of somebody making an office, which is really cute of like the agent's working. Right? So can you have that and give me a bunch of more visual improvements to my dashboard? Then, since it understands its own back end, it can give you props. It's like, okay, found it. This is the next JS mission control that I'm working with that you're working on. It's going to understand a little bit more intuitively. It's own back end than Claude will. Now, you can train a skill in Claude code to be like, understand the back end of a Claude. Totally fine. But what I basically am having to do is make me the props here. And then the actual coding thing, I literally copy this a minute ago. Went into my Claude code and gave it all of these props. So as I had a convention, you remember how he said that there was what's called parallel and/or agents that are sub-agents. Right? So what is it working on right now? Let me give an example. So it says, this is crazy. You can see kind of see it's doing all this stuff in the background. She's just wild. So it made it, I asked for it to make a heat map. Right? I mean, refresh it because it's coding. It's going to be weird at times. First, every time it goes through a coding cycle, it can get a little bit finicky. So you just have to repress it. And also, this is the open-claw dashboard that it comes with, like that lets you kind of understand open clause, I guess, back end. And you can chat with it directly here, but I just use Telegram because that's like, what was recommended for a lot of people. And it's the easiest setup from my perspective to like talk with it on the go. Okay, it's go load. It's like, I do you edit or cut this out? This is ridiculous. Sorry, the timing on this is so bad. It was literally working a second ago. So let's give it a minute. That's the thing, like whenever we try to do a demo, things fail. - Literally so fucking annoying. It happened me so many times. It's, oh, I haven't had an error the entire time. All right, well, to be honest guys, as anything does happen on the show, demos just break the second that we're doing demos, but the good news is I don't actually have to do much to pick this. I'm literally just gonna copy and paste this into my club code and it's going to fix itself. So that's the good news. Unfortunately, I didn't get to share the updated dashboard because as we were doing this episode, it broke. But the good news as well on top of that is that this is not the only thing I'm working on right now. I got three different, four different things brewing that you can work on and kind of showcase to the world at the moment I can't, but I think what's really important to note here, that's been mentioned is the value of vibe coding, the value of vibe coding is so, and by working is so high right now, you are able to do so many different things and as a business owner or as an individual, you gotta think about it from this perspective. There are a bunch of different companies you work with or work for, and a lot of different processes that you do on a daily basis. And for me, I'm trying to think of it from this perspective. So if I had to give you one piece of advice to kind of close it out on how does one vibe work and give out the different things you do in your day, think about the different ways it can be automated. It could, or, well, five words. It can be as simple as analyzing your email. It can be as simple as responding to Slack messages. It can be as simple as putting in to-do list items by typing it in a chat. All of these things are things that are vibe work. I know you as a business owner has a lot of different things. I know I have a business owner of a lot of different things that go into our daily lives. Making those things easier is what vibe working can really do right now. And think about it in life departments, and then if you go to work, you can think about it in real department. - I mean, for me, vibe working is about kind of like, you know, at night, like when I'm finishing off working, I like to just give, like, think about like, what kind of, like, information would be useful for me, like, what kind of research would be useful for me? And then just go to, you know, co-work, cloud co-work, or, you know, some other AI agent, and then just ask you to do some work for me. And then, like, come back to next day, and find this useful report. And this is kind of like, in the beginning, it just, it doesn't work. But if you, when you keep doing it, like, you know, like, where it fails, and then you start to actually prompt better. And this actually results in, like, basically, having all these, like, custom research, custom reports, help you with the work. And so, I think it's, like, one of the great things that are happening right now is these, you know, agents that can work in part of us. And that's gonna be, like, that's gonna speed up things so much. And it's, you know, maybe it's not gonna be any more, like, you know, give your agent a task to do, you know, last, to do during the night. But maybe it's gonna be like, okay, I need this. Here's the outcome I want. And give it to me. And then, you know, five minutes later, you've got the job done. And basically, it's just, there's just so much potential here. And the momentum is huge. And so really cool stuff happening. - Absolutely. - It's an article, something that is happening. - Yeah. - Yeah. - I would check that article out, check out everything that's going on over there. And check out everything that we're doing. Make sure to go to check out our YouTube channel if you're listening to this or check out anything audio wise if you're watching this. We're available everywhere. And we just appreciate having you. We're doing full stuff and we want to share it with you. So if you liked this video and want to see more of it, hit the like button and subscribe. Thank you so much for watching. We'll be talking more about the scan of stuff in the future. Give that a thumbs up. Peace. - See you the next time. (upbeat music)

Podcast Summary

Key Points:

  1. AI agents and "vibe working" are transforming how knowledge workers operate, enabling tasks like coding, legal research, and content creation through natural language prompts.
  2. Tools like Claude Code and Cursor are advancing rapidly, allowing teams of AI agents to collaborate on complex projects, though fully reliable software generation from a single prompt remains challenging.
  3. The AI landscape is highly competitive, with major players like OpenAI, Anthropic, and Google heavily subsidizing access to powerful models, creating a temporary opportunity for users to build infrastructure at low cost.
  4. There is a shift from traditional SaaS tools toward AI-enhanced platforms that guide users through entire workflows, not just isolated tasks.
  5. While AI capabilities are accelerating, concerns include "agent stress" (managing idle AI agents), security, integration challenges, and the long-term sustainability of current pricing models.

Summary:

The discussion centers on the rise of "vibe working," where AI agents handle tasks like coding, document generation, and analysis through natural language commands. Tools such as Claude Code enable teams of specialized agents to collaborate, significantly boosting productivity. However, fully replacing software development with AI prompts is still limited by complexities like security and integration.

The AI market is in fierce competition, with companies like Anthropic, OpenAI, and Google offering heavily subsidized access to advanced models, making it an opportune time for businesses and individuals to leverage these tools. The conversation highlights a shift from rigid SaaS products to AI-driven platforms that assist throughout entire workflows. Despite rapid progress, challenges include managing AI agents efficiently and adapting to potential future cost increases as models evolve and subsidies change.

FAQs

Agent stress refers to the feeling of pressure when your AI agents finish a task and are idle, prompting you to constantly find new jobs to keep them busy, even during off-hours like nights or weekends.

Vibe working extends beyond no-code development, allowing knowledge workers to manage entire teams of AI agents using natural language to handle tasks like financial analysis, legal research, and documentation generation.

AI is challenging many SaaS businesses, as tasks that once required software can now be accomplished with prompts. However, robust software that guides users and integrates complex workflows remains valuable and harder to replace.

AI agent teams can work in parallel with specialized roles (e.g., designer, developer, researcher) to complete projects faster and more efficiently, communicating with each other to streamline complex tasks.

Plans like Claude's $200 Max plan offer token usage worth up to $20,000 monthly, as companies subsidize costs to gain market share, gather training data, and encourage user adoption during competitive battles among major AI providers.

While frontier models may become more expensive over time, open-source and local models are improving and becoming cheaper, offering a potential cost-effective alternative for running AI workloads in the future.

Chat with AI

Loading...

Pro features

Go deeper with this episode

Unlock creator-grade tools that turn any transcript into show notes and subtitle files.