In short
Google’s newly released Nano Banana (Gemini Flash 2.5 image model) and Google AI Studio demos, including building apps via natural language, “stream” live screen/webcam feedback, and multimedia tools like video analyzer.
Guest
Elizabeth Knopf. Background: builder/early adopter using LLM tools for app creation and debugging; discusses using Claude/Gemini, Google AI Studio, and Cloud Code to replace paid tools and build personal workflows (e.g., RSS aggregation, project management dashboards, agents).
Key claims
Nano Banana improves precision for edits (e.g., changing shirts/products) and adds “visual memory” so generated subjects/logos persist across iterations. Google AI Studio is a “suite” that reduces friction for novices by combining chat, live “stream” feedback, app-building templates, and video/image/audio capabilities.
Notable examples
photoshopped Ronaldo with Coca-Cola jersey; same people in day/night Toronto skyline; step-by-step guides with persistent characters; building a history trivia game in-browser; video analyzer counting light bulbs in a Christmas lights video; generating a lead-magnet-style app from a YouTube how-to video; replacing RSS.app with a custom feed; Cloud Code project dashboard using to-do.md files, cloud.md rules, and scheduled agents.
Written by AI. May contain mistakes. Listen to the episode to check what was said.
Chapters
Tap a time to open that second in VOExploring Nano Banana
0:45 to 1:48
Discussion on Google's new AI tool, Nano Banana, and its capabilities.
“You're not a coder, you're not a developer, but you're going to show us how we can do that ourselves.”
Impressive Features of Nano Banana
1:48 to 4:30
In-depth look at the features of Nano Banana, including visual memory and image persistence.
“picture in front of his house it looks like he's in like nova scotia and then they change his clothes and the lighting and it looks like he's in a flea market in Morocco or something like that.”
Using Google AI Studio
4:30 to 7:19
Overview of how to utilize Google's AI Studio for app development and real-time feedback.
“upload some images and videos getting there.”
Building a Trivia Game
7:19 to 13:00
Live demonstration of building a history trivia game using AI tools and real-time feedback.
“all the interviews to be able to pull out information that maybe I'm not seeing, but there was a problem cleaning the data.”
Innovative Use Cases for Maps
13:00 to 14:00
Exploring unique applications of map integration within AI tools for various projects.
“And then we're going to start the game and we're playing.”
Exploring Google Maps Integration for Kids' Games
14:00 to 16:54
Learn how to leverage Google Maps for educational gaming experiences.
“So if you just go down to like the one for showcase, so just scroll.”
Video Analyzer: Enhancing Content Engagement
16:54 to 21:10
Discover features of the video analyzer tool for understanding video dynamics.
“chart of importance type of custom prompt.”
Creating Lead Magnets from YouTube Content
21:10 to 23:08
Explore how to generate engaging lead magnets from existing video content.
“And even though it's so easy to do this stuff now, like friction is still friction and people just, you know, stop if it gets like a little bit hard.”
Building Custom Apps with Google AI Studio
23:28 to 28:00
Understand how to utilize Google AI Studio for building custom applications.
“and it's aggregating all that information into a feed so that you can basically see everything.”
Building the AI Tool
28:00 to 28:59
Learn about the initial steps and challenges in creating an AI project management tool.
“So basically I have, and let me do this so you can actually see my Cloud Code terminal.”
Show all 20 chapters
Setting Up Project Structures
28:59 to 30:15
Explore how to organize projects and files within the AI tool for efficiency.
“Now, I would say when you're going into this, assume that if you haven't done it before, you're probably going to have to redo the whole project more than once until you sort of refine what you need.”
Using AI Agents for Task Management
30:15 to 31:08
Discover how to utilize AI agents to automate task tracking and management.
“And if you're sort of afraid of the black...”
Creating Effective Templates
31:08 to 32:05
Understand how to create templates that streamline project setup and execution.
“Is there any way you could show me what that looks like?”
Developing Systems for Project Management
32:05 to 33:50
Learn about best practices for developing systems that enhance project management.
“So here's my overall cloud code file structure.”
Context Management in AI
33:50 to 34:56
Explore strategies for managing context effectively when using AI tools.
“Yeah, it's like you're almost building your own little cloud project, right?”
Personal Operating System Insights
34:56 to 35:26
Gain insights into building a personal operating system using AI.
“Because when you close out one of those little black boxes, the journal window, it loses everything of what you just did in the chat.”
Implementing Practical AI Solutions
35:26 to 40:04
Discuss practical applications of AI tools in everyday tasks and projects.
“So I basically am building out all these different things.”
Looking Ahead: Future Topics
40:04 to 41:50
Preview upcoming discussions on data scraping and data cleanup techniques.
“Because I have stuff all over the place and I always am like, did I do that in Claw?”
Discussing Data Aggregation and Analysis Challenges
42:00 to 42:40
Learn about challenges in data formatting and analysis in projects.
“I know that's sort of boring, but it's something we all hate doing.”
Growing the Podcast: How You Can Help
42:40 to 43:26
Understand the importance of audience engagement in podcast growth.
“And if you've made it this far, you're either really committed or you're stuck doing yard work and you can't actually skip on your phone.”
Transcript
Automatic transcript. May contain errors.0:00You're saying I could have just had this Talk to Gemini live on and it would have been looking at what I was trying to build and like give me real time feedback? Absolutely. That's insane. If it could generate from this one episode, then I'm going to hang this call up and end early because I'm done for a week. This is incredible. Why the F didn't I know? This is insane. Why the freak wouldn't I have ever done this? This is amazing.
0:23Okay, Liz, I am excited to talk to you today because you are going to show me how to actually use some amazing tools. but the things we're going to talk about today are the things that are currently taking the AI world by storm. The first one is Nano Banana, which is the model that Google just released. And the second one is you're going to actually show us with CloudCode how you replaced one of your apps and created your own app using CloudCode, natural language. You're not a coder, you're not a developer, but you're going to show us how we can do that ourselves. So first question, what the freak is Nano Banana?
0:53I keep seeing this crap all over the internet. I'm like, I don't know what this is. So you know how like a couple of weeks ago we were talking about how Google sucks at marketing. Well, I think they were listening to Herb Hot because someone stepped up their game and they did a great job finally releasing a product called Nano Banana. And all that is, is their Gemini Flash 2.5 image model. So it's just another image model, but it's actually a pretty awesome image model. I found this Reddit thread I'm going to show to you. And these are some examples of Nano Banana, of what they did. So this is, here's a picture of Ronaldo.
1:26He's drinking a water bottle and they photoshopped a coca-cola they also photoshopped a jersey on them the face looks photoshopped to me just because i see them next to each other but pretty cool here's one that i think's nuts like they took this picture on the left it's two guys in the middle of the day with the toronto skyline behind them and then they photoshopped that too at night it's literally the same guys in the same picture you wouldn't have known that incredible editing oh this guy took a picture in front of his house it looks like he's in like nova scotia and then they change his clothes and the lighting and it looks like he's in a flea market in Morocco or something like that.
2:00So really, really good image generation. My question is like, what are people using this for? Is it just like a novelty trick? A couple of key challenges that have come up with a lot of the image generation models. So first off, it's just like replacing things and doing simple edits can be, were a little bit hard to do. And even with like chat GPTs in comparison, you know, you would say like make this, you know, change this one element. And sometimes it would work and then sometimes it wouldn't. And same with Google's previous models. And so now it has just become much more precise with specific edits and iteration of those edits.
2:39So changing the shirt, changing the product, like it actually is much more accurate in terms of how it takes whatever you're prompting it to change and making those changes. And I would say Adobe was still sort of the leader and you're saying Photoshopping, but like literally they were probably still leading the way in some of that editing versus again, going to one of these image models to do your editing. But now Nano Banana really sort of advanced the ball in terms of doing that. The second key feature, and I think is the more exciting and really big deal is what's called visual memory. So this is essentially where you can have from one image to the next, it persists as the same person, same object, same thing.
3:22And that has been the biggest flaw in image generation because you would see these little even ads or short ads and there would be a person, but they'd slightly look different with different bone structure or something in the next cut. And so now persistence has actually been improved so dramatically. And that's because they came up with some interesting techniques while they're building the model, which we can talk about in a second but those are really the oh that's really cool yeah okay so so if i was starting to use these image generating models but that was always a problem right it's like cool i generated this cool logo but how do i get it to stay the same logo across multiple iterations whereas this is like well now you can do that you can you can keep the same person in the same image or the same logo whatever it stays persistent here's my question to you we used to talk about multimodal and maybe this is just me because i'm a caveman but we used to talk about multimodal models where you're like, oh, I can upload an image and then I could upload text and I could upload a video and I could upload, I'm trying to think of what the other mode would be, but those are all three different data types.
4:29And in the past it was only text, but now you can upload some images and videos getting there. However, my perception of multimodal has actually changed a little bit. I think of multimodal now as I have an LLM that I use for synthesizing data. I have an AI image generator that I can use to create whatever, logos, et cetera. I can even get these. They're not LLMs. They're SLMs, small language models that are specific use cases that are maybe they're healthcare, maybe they're tech related, et cetera. Or maybe there's a deep research function. So it's like those are actually different modes of AI.
5:07And I think Google has done a really good job of creating this studio. It kind of incorporates a lot of these different modes. but people don't really use it or know about it. And you were just telling me about Google AI Lab. I can't remember exactly what it's called. What is it? Okay, so since they have so many different, yeah, locations and properties. They're terrible. And it's bad. Aistudio.google.com. So it's Google AI Studio is the name. One of the locations you can access Nano Banana or of course you can go straight to Gemini and access Nano Banana there too. Google AI Studio has now started incorporating in that model into being able to build out little apps and like other tools with this new model.
5:49So that's, I think, pretty exciting. And we'll dive in here in a second just to see what's basically possible. Okay, so this is it. I'm in Google's AI studio. So AI studio.google.com. First of all, I don't know, it must have been two months ago or so, but you showed me a version of this. Yes. And it was overwhelming. This chat interface is clean. I mean, it looks like the OpenAI chat interface. I don't get instant anxiety looking at it. Okay. So on the left side, I have chat, which looks the same as OpenAI, but then I have stream. What is stream? So stream is actually a pretty cool thing. This is where you can basically use your web camera or share your screen.
6:30So like if you are debugging something, if you want to learn something, if you want Google to watch a video with you and give you feedback or see you do something live and talk with you, it is actually doing that and capturing it in real time. So this is like, no way. Yeah, people are sitting on this feature. It is pretty incredible. I use it a lot for just debugging. And when I'm like doing something or like following instructions, I'm like, what am I getting wrong? Or like, look at my screen or so I don't have to take a gazillion screenshot, I'll do that. Or I've tried to make a YouTube video and having it sort of walk through it with me.
7:08So there's a lot of different cool things you can do with this feature that again, is like totally slept on. I did all these interviews with people. I wanted to look at all the interviews to be able to pull out information that maybe I'm not seeing, but there was a problem cleaning the data. So I had it all stored in Google and one of the coolest tools google apps script i was writing a script and i was like claude can you help me write this script and so this is me just like going back and forth it's writing the script and then every time it would i'll show it to you so this was the script you see over here and i would copy and paste it into actually google apps script and it would help me create it but every time that it would screw up i had to do this i had to take a screenshot of the execution log and then i would copy it and I would just literally paste it and it was like, oh, this error was because of blah, blah, blah, blah, blah.
8:00And then it gave me a fix. You're saying I could have just had this on? Like I could have just had this talk to Gemini live on and it would have been looking at what I was trying to build and like give me real-time feedback? Absolutely. Yes. That's insane. Yeah, exactly. Yeah. I mean, I will say that I have had in the past where like, again, I love Claude. So sometimes I think Claude is better at feedback because again, they're using the Gemini chat model. So it depends on your opinion there, but Gemini has been getting a lot better, but absolutely, absolutely. Can you do that? Go to that in a second.
8:35Then we'll come back to the apps in a second. So yeah. So like, again, if people haven't been using VO, pretty spectacular for like videos, they have also other sorts of media now you can produce. So whether it's like speech generation, doing audio. Like, again, you can start doing a lot of cool stuff in here. And one of the things also that I think is worth noting on Nano Banana that I also think is cool related to that sort of persistence element is like you could do step-by-step guides. So I haven't yet tried it for software, but I have tried it for like fitness. Like show me a step-by-step guide with images of like how to properly lift weights or do an exercise.
9:14And it will take with the same person going through each of those steps. So you can just, again, get very creative and imaginative on like using those step-by-step guides. Obviously they have a recipe one here. So I think that's a pretty cool feature. And as you can go in here, like they, they have a bunch of cool examples and templates that you can be inspired by different camera angles. That's insane. Yeah. Yeah. Yeah. Okay. So now the cool place that again, people don't even know about. And this is like a helpful way for beginners who, again, are a little bit intimidated to get started in building apps.
9:48So if you go to build, this is similar to some of the other LLMs that will have sort of like little mini apps examples. They used to call them gems, which was just really like confusing. Now they, I don't know, just are calling them apps, I guess, where I don't know if there's even a different terminology, so whatever. But basically they have little starter apps and examples, some of which have incorporated in Nano Banana already. And there's others that have sort of showcased some of the real power of what Google and Gemini can do, which again is really on this sort of like video multimedia components that I would say, you know, the other models each sort of have elements of, but I really think that Google is excelling and doubling down on that capability of both like animation, video, audio, and streaming.
10:41Is this kind of an answer to cloud code? Sort of. Sort of. I would say they're very different in terms of like the workflow, but everybody's trying to build out some kind of coding app because everyone sees like personal software is what's happening. That is what is being sort of cultivated through LLMs as really, I would say, the biggest area of maybe annihilation or augmentation from what's happening. So, you know, everyone's building apps. And so you can do that now with ChatGPT, with Anthropic, with Cloud Code, with Brock. You can all sort of build within the browser. So then still, again, you can build it from here in the browser.
11:24there's still a couple steps to like get it into production. But Google has like that full work stream of getting from a little mini app into production, whereas some of the other guys like don't have the same kind of infrastructure that Google has. It's not as slow. So it's probably more similar to a replet or lovable than it is to cloud code. Sort of, sort of. I would say it's again, it's hard. You're so nice. You're like, yeah, I mean, you're like, yeah, kind of, kind Kind of. But like, again, they all have like, you can start prompting to build something. And then again, to go to that sort of next mile is a little bit more challenging.
11:59But this I would say is like still keeping very simple apps. I would say Cloud Code is still sort of the best in going like, in the same environment interface, you know, from idea to production and building out sort of infrastructure. Like again, I'd say Gemini is doing well. But again, there's something about their interface that I'm just not as like thrilled about. So I think it's personal preference, but you can access a lot of these tools here and test it out and see if you like them. Cause again, I think it's just a preference thing. Yeah. I literally just put this prompt, build me a game.
12:31I can play with my kids. It's a history trivia game with rounds of questions that is tailored to everyone's age. The app should keep score. It's been running for 75 seconds and I can see it real time building this out. So like in this sense, it reminds me of lovable and then I'm like watching it build real time. Boom, there it is. And now I have basically an MVP to test around with. So I could say Nick, age 40. Then I'm going to add another player. We're going to say Liz, age 29. Oh, thank you. And then we're going to start the game and we're playing. It's generating the questions. This is me.
13:11Which Roman emperor famously appointed his favorite horse? incititus to the position of consul or at least threatened to in display of content for the roman senate i have no idea i'm gonna say nero because i just know he was a nutcase okay caligula all right liz here's your question it's generating for you which explorer is often credited with the first circumnavigation oh come on of the earth despite dying halfway through the journey was that magellan christopher no it was magellan he died yeah he died in the philippines i think is where he died. This is freaking cool. I just built this. It's awesome.
13:50And so if you want to go back in a second, I'll show you because I think like this is cool. Again, like Cloud Artifacts can do this, like Chachi Boutique can do this, Grot can do this. But I think there's a couple of pieces with the nano banana element and some of the features that they have that's pretty cool. All right. So if you just go down to like the one for showcase, so just scroll. I just wanted to highlight a couple of cool things. So keep going. Okay, so number one, maps. People don't talk about map making or leveraging maps. So map at Explorer or there was like an MCP map thing, whichever one you want to do.
14:23Like you can incorporate in map elements. So I started talking about kids. I started doing like an exploration on the Google Maps for my kids game. Like, you know, something like Carmen Sandiego meets something else. And so Maps is just like, you know, that's embedded and native to Google's stack. And so you can start doing cool things with Maps or, you know, maybe it is more on the business side, logistics and scheduling. Or, you know, there's a lot of different use cases that I just I think Maps is sort of a cool, unique use case that other folks don't have as ease of integration. Of course, you can like integrate Maps into everything else.
15:02But again, it's, you know, Google's sort of core function. Oh, this would be so cool. I could do a total game with my kids. Yeah. Like Gobekli Tepe in Turkey, which they honestly just found recently. I could say, take me somewhere ancient. And I know my oldest son's going to be like, let's go Gobekli Tepe. And then all of a sudden we could just go to YouTube. I could build this out and be like, take me somewhere cool and then give me the top three YouTube videos to link to just because it's in the Google ecosystem. That's awesome. Yep. All right. So let's go to another cool feature they have. Okay.
15:32So this one I haven't got to work perfectly, but it's called video analyzer. And then we'll highlight a couple other. Where is it? To scroll down, down, down, down, down, uh, video, video analyzer. So this is like pretty cool. So we all sort of want to understand, like, you know, you can get a transcript from a video, but like, if we're doing this podcast, we want to say like, where, where's our body language bad? Or I don't know, some, some other thing, or what is happening in this video and you can literally in their little app here add in a video i can drag and drop you can add yeah so this you have to have a local file for i had edited this the other day where you could just put it in a youtube url and it would download the file so you could iterate this little let me see if i can get let me see if i can get a video i made this video a couple of months ago of people hanging christmas lights with some music on it so i'm going to take that and i'm going to add it here and we're going to see what that does.
16:27It's processing the video. Now, do I need to tell it something or is it just processing it? So let's actually see because the way I was doing it was different the last time. So I have no idea. Let me be surprised. I'm so nervous. Explore this video via audiovisual captions, paragraph, key moments, table, haiku, chart, custom. Let's do, what do you think we do? Haiku or chart? Let's do a chart. Let's see. I don't know what it'll do. chart of importance type of custom prompt. I want to see a chart of the total number of light bulbs. I have no idea if that will work. I just thought that'd be interesting to try.
17:12Yeah. Because they're hanging light bulbs. You can start imagining to like, okay, you can actually ingest. Wow. That's pretty cool. count that of course but yeah here's this is the video and you can see this guy's just setting up all the light bulbs i mean i believe it i'm not going to go count it but i believe it how cool is that that's freaking cool it is pretty cool so again like now you can start like ingesting video as a medium and doing more than just you know getting a transcript which I just think is pretty cool. Yeah, that's awesome. All right. So a couple other things to highlight that are just related.
17:57We don't need to do the full demo of them, but just, you know, throwing them out there. So if you just keep on going down, there's like a video learning one. So similar, basically you can put in video to learning app. So you can basically put in a video and it'll generate like an app related to the video. I've tried it a couple of times.
18:19It tutorial type videos. But again, these are just like cool ideas that if you polish up, you can build your own little apps to do cool things with or like use this as a lead magnet. I want to try something. I want to try. Okay. Let's go to my YouTube channel. Oh gosh, that sounds so douchey. I'm sorry. Let's go to my channel. Let's pick a video that would be really cool to turn into a lead magnet. So, okay. So I've had a video that I put out of somebody who walked me through the process of building out a directory. So we've talked about directories before. They're like super high right now because people are like, oh, I could build a directory with AI.
18:58I don't have to do all the manual work, blah, blah, blah. Anyways, that was a really popular episode. People really loved it. And now I'm taking the link. I'm dropping it in here. I'm really curious to see what it generates for me because if it could generate from this one episode of like a how-to video how to build a a directory then i'm gonna hang this call up and end early because we're done i'm done for the week i'm hope it works so that it wasn't working well for me the other day even though i'm like this is one of the apps yeah so we'll see we'll see if it does but we'll see fingers crossed okay well that's that's spinning up we'll see where that goes.
19:38While that's loading, I just feel like Google even still, Hey, I don't know if you remember this, but when we started this podcast, we entered into a social contract. I would spend time, energy, and money producing this podcast, interviewing these individuals and giving you insights into how to build, buy, start, grow your business. And you would like subscribe and leave me five-star review. Now out of that, we both get to talk to really cool people and hear really cool insights. We both get a ton of value, but I just want to help you keep your word. So would you do me a favor? Will you go leave a five-star review for me on Apple or Spotify?
20:12It would really help. And if you want, even share this with a friend. What we just went through, the AI studio is incredible. Why do more people not talk about that? Are we just spoiled? I don't know. I don't know. Cause it's like, I think the thing about cloud code versus like Google's suite and even everybody else's like coding suite, there's still friction in terms of like when you start to how you get to production versus I think with clod code because it's in the terminal. It's like even though you're still dealing with this weird black box interface, you can go through the full process and then when you want to visualize it, it just sort of like pops up on your browser versus the other ones.
20:55It's like you have to then figure out how to get it to view it. Sometimes there's a previewer, then sometimes you have to like connect it with like either Vercel or go with like Vertex or there's like still steps. And at every point, every point in something you have to change or do, you lose people. And even though it's so easy to do this stuff now, like friction is still friction and people just, you know, stop if it gets like a little bit hard. Understandably so. Yeah. Okay. So that, that learning code app, I got an error. You can see the vision and I think with some iteration, you can get there.
21:30I just think with this AI studio that Google has, man, for like novices like me, I can have the video running while it's watching me go through stuff. It can tell me in real time what things I'm messing up. I can create apps. It can show me the apps. It takes the good of lovable replet and it takes Claude, which helps me debug things. And then it has a good image generator. Oh, and then I can integrate my YouTube videos. I can integrate my Google Sheets. Like that's one thing that bugs the crap out of me. And maybe you can help me fix it. When I ask Claude to parse data, and then it's like, sure, go ahead, just copy this and paste it.
22:07I'm like, no, just give me the CSV. Such a pain in the butt. Whereas Google can obviously just do that with Google Sheets, but money on the table. Yeah. No, and it's interesting because right now people that are watching and paying attention, Like this is a great arbitrage opportunity for any of this stuff. Like, again, people don't know about Google AI Studios. So you can literally take that code, find a good URL that has some good traffic already and just put it up like minimal changes. Obviously, you need to do a couple of things. But, you know, if people get ad revenue or whatever for a free tool.
22:40So I'm working on a couple of those. So we'll see how it goes. But I mean, right now is a good little arbitrage until everybody's doing it. That is such a freaking good point. It's an amazing arbitrage because people, it used to be the software development was only for a select few who had gone through a ton of school, understood how to code. Now we can do it with natural language and something like Google's AI studio. It's a built out suite. It has all the tools that you might ever possibly need. You have no coding experience and you just told me you built an app that replaced one of your apps recently.
23:10What was that? Okay. So let me share my screen and show you what I've been working on. So I used to pay for an RSS feed tool and that basically aggregates. Tell me what that is. Oh, sorry. Okay. So that basically is listening to websites, keywords, different social media sources, and it's aggregating all that information into a feed so that you can basically see everything. It had organizational collections and groups and just different ways to parse information so that you could just see it in one spot versus going to a gazillion different sites or having to read through a million newsletters or whatever.
23:48So I basically was like, I'm sick of paying for this. It was not that expensive. So probably not the best use of my time for something to build. But at the same time, it was a good starting tool as I'm learning cloud code and sort of sharpening my skills. And so can you show me what, like, what did it look like before? Sure. Let me see if I still have access to. Okay. Let me see. Sorry. I'm just like, I'm trying to understand. okay so this is like the tool i used to be paying so this is rss.app yeah and it would bring in all the things that you were interested in yeah so it's not really showing how it used to show just because i've canceled this on the free version here but like let's do the opening night so basically it would be a feed so like it's still better than what i built but like for my purposes this is great so i was paying 20 a month for this and what was it what was it going out and scraping was it going to like twitter was it going yeah so all of these youtube or like how was it yep so it was scraping specific sites so like xai news i was going to oh okay i inputted all these i had my own you know twitter feed here just ai sites like different peoples who twitter accounts i had like this is incredible yeah why the f didn't i know this is insane well the cheaper one that is worth doing and it's actually free now it used to you have to pay a little bit or i use api I use APIs, so that's why I end up paying for everything.
25:14Seedly is a little bit more inexpensive. This is like$10 a month for more feeds. But yeah, you can put in different sources and just have your own little feed here. I mean, I go to Twitter because it's like, I've kind of curated my feed so that I see the stuff or I'll go to their search and see what's trending, et cetera. Why the freak wouldn't I have ever done this? This is amazing. Exactly. Of course. Because there's specific information I want. I want to know the AI updates. I want to know the SMB updates and I want to know the tech updates. That's pretty much it. I don't really care about the politics anymore.
25:46And I have to freaking filter through Elon posting, you know, half naked 10 year old girls on the feed. Like Elon, settle the freak down. Anyways, this would be an amazing way to cultivate my like my own. Sorry. OK. Yeah. So so you had an app for that for your own feed and you decided I'm going to build my own. Yeah. And again, this is just something to cut my teeth on because I'm just like, you know, this is like annoying that I'm paying 10,$20 a month. Like, again, I'm paying a lot more in other subscriptions, but I was just like, easy tool to build. Let me see if I can do it. Ingesting information is useful for a lot of different use cases and apps that I have on my backlog here.
Read the full transcript
26:22This is insane. Okay. All right. Go back to your screen. You can also say that in some ways, this is a little bit moot because like you have perplexity, like it sort of does it for you or any of the cool things. But anyway, so I just decided, let me see if I can build this myself, see what would happen. So I built out, oops, dashboard, don't worry. Okay. So I built out just again, like to get different content ideas. So I put in a bunch of different sources. How do you stay organized with all the things you're building? Because I've only built a handful of things and I'm like, okay, which folder did I put that in?
26:52Where is it saved under? What's the latest version of that thing? This is my project management dashboard that I started building in Cloud Code that I've already iterated. Of course you did. Of course you built a content management, your project management dashboard. How did you build this? Is it automatically pulling in new projects? Do you have to populate it every time? Like, what am I looking at here? No, no, no. Okay. So what you're looking at right now, as I'm going to load my other one. Okay. So this is like a project management dashboard, because of course, as you start building projects and me being like a obsessive multitasker, especially with cloud code that I had started having too many projects.
27:27And I was like, I need to keep track of what's what I started sort of learning to that. You want to build out to do list. You want to have certain rules that I started wanting templates. and then I started having like agents or different frameworks that I was starting to leverage and I needed to keep all this organized. So of course, what do you do is project management tool. So I first went to GitHub to see like someone might have already solved this. So let me see if that's better. I'm still testing out other people's tools, but this one was my own because again, like you sort of have to decide what should you build for yourself versus like, what should you build for show me how it works okay so this is so freaking cool is not fully functional yet and i had to redo a number of different features on this and so it's still like a mega work in progress so right now like how is it deciding it looks like there's a priority associated with it highest high medium low how does it decide priority do you tell it that so I had put in basically a set of rules in a cloud.md file that's sort of like overarching in my personal operating system.
28:37And let me see if I can. Did you upload that to Cloud Code? Yes. So basically I have, and let me do this so you can actually see my Cloud Code terminal. So, or I guess I will talk to it. In Cloud Code, all you need to do is like start with planning, tell it what you want to achieve, and it will help sort of think through what that is. Now, I would say when you're going into this, assume that if you haven't done it before, you're probably going to have to redo the whole project more than once until you sort of refine what you need. Because sometimes I just think it's starting from scratch is better versus trying to iterate and fix because it might have to like redo a bunch of stuff.
29:15So this is probably my third iteration of this tool, and I'm going to probably redo it again. But essentially what I have right now is my local file folder that has all the different files and folders for each of these projects. In the upper level sort of file folder structure, so I have everything in something called cloud code files. Then under that, within that, I have some files. And then under that, I have a bunch of like categories of projects. I have active and then I have like archived and then I have a couple of different stages. And in the active ones, I have then these projects. And from these, it's ones I'm actively working on.
29:57And then I've built out some agents, which again, all I did was tell Claude, I wanted to have agents that achieve these sort of things. So I have like a systems folder that's thinking about my system of project management. it. So I created a, and again, I'm just prompting Claude and telling it what I want to achieve. So like what I'm saying literally to the camera right now, there's nothing more magical than what I was doing besides just like interacting with Claude and figuring out how to do that and having Claude tell me how I should approach it in planning. And if you're sort of afraid of the black...
30:30So like what was that agent doing? What was that agent doing? Yes. I have a couple of different agents. I have a security agent that will help just make sure I have security best practices going. I have an agent that does pulling my to-do lists on a daily basis, or I can manually have it basically refresh. And so it's pulling in my to-dos because I have a to-do.md file in every project so that it basically tracks all of the tasks that I'm doing. And I have set in the cloud.md file the rules of what I wanted for that project. And in every one, I have it say create a to-do file, that to-do.md, and break down every part of the project into bite-sized tasks.
31:17And then up - Can you show me that? Is there any way you could show me what that looks like? Let me see. What's an mt? I know what an md is, the markdown, but what's an mt? No, md. Sorry, it was md. Oh, md. Okay, sorry. I just misheard it. And so every time you go into Cloud Code, you're like, okay, here, this is like my template of things. I'm going to upload. Yes. Basically the context. Exactly. So I'm having it build templates as I'm learning so that I can say like, oh, I've done a data cleaning project. So I know here are the rules or things that come up so that I don't have to like recreate the wheel.
31:49And then also when I'm just like doing projects, like I want it to break it down into bite size pieces, always have a to-do list, always have certain security things. So let me see here. So these are like my different little projects I have going on right now. Let me just make sure there's no APIs showing. Okay. Okay. So here's my overall cloud code file structure. So I have my active ones, I have staging, and then I have my systems. So these are different little systems that I'm telling it to do. So I have a to-do aggregation log. What does that mean? I'm so sorry. Systems? What do you mean systems?
32:22So I just decided, again, for my agents that are doing things or have a system for doing project management. So like think of your management systems, Like when you're managing things, like how you would have a team and how you manage a team. You do on a daily basis a check-in. You want to have what are all the two-do's for the day, your prioritization. Would it be fair to say like this is an SOP folder? Exactly. It's like this is how you do things? It is more than it is both an SOP plus the execution. So I have actual agents in there, which again, all I did was tell Claude, make an agent that can do X, Y, and Z things.
33:01An agent that will get all of my to-dos from every project on a daily basis and aggregate them and add them to that dashboard. So that is literally - And that runs independent of you. You don't have to like push a button to have it do it. Correct. Because I said - It's a certain time that it goes and runs. Exactly. I said, here's the schedule, you do it. And so it will go ahead and update it. I didn't do it in real time because that takes a lot of CPUs and just will kill your system. But, you know, I have a way to do that if I wanted. So basically I'm thinking through, okay, this is what I want it to do.
33:33So I don't have to manually do this. I'm allergic to doing manual tasks and anything administrative. So I just say, I don't want to do this. So can you do this for me? So let me even just see if I can open up just so that you can see what this looks like. Move that over. Yeah, it's like you're almost building your own little cloud project, right? Because the cloud projects have context where it's like, this is how I want you to act. I want you to always give me to-dos. I want you to make sure that you're attacking problems this way. And in cloud code, you don't have that context. You need to recreate it again, which is why you're building the systems folder.
34:12Totally. And I've had to rebuild this. I've had to say, okay, scratch everything. Let's start again. So like as you're building and you're trying to do this for yourself, don't be afraid to do that because you'll figure something out and be like, that was stupid. Or you'll build one thing and then you forgot you build it and you build it again and then they're conflicting and it creates problems. So just like – Well, I wonder – Yeah. I wonder too if I could get to the end of – maybe the end of creating something and just say, please write me my operating system based on the way that I interacted with you.
34:41I don't even know what the real prompt would be, but some type of markdown that helps you remember what I like and what I don't. Exactly. Exactly. You can absolutely do that. And another thing I found is helpful is it doesn't have great context management or sort of ways to remember your whole chat. Because when you close out one of those little black boxes, the journal window, it loses everything of what you just did in the chat. So what I've been doing now regularly is saying like create an MD file of the whole chat and create a file that basically summarizes it. So when I start this again, I have context I can reference.
35:13And so that's really helpful to have those things. So then you can manage your context because in those chats, if you're not clearing it out regularly, it'll actually like get worse over time because you're overloading it. Holy crap. This is insane. Personal operating system. And then it has all the files. Yeah. So it has a file. Yeah. So like I have stuff. Oh, look at that. Family management. Yeah. Financial tools. Dude, that is amazing. Yeah. That's amazing. So I basically am building out all these different things. Obviously, I have to iterate and update this. And I've had to redo this a few times, so I'm sure things are missing.
35:49So what is that? Do you tell it to when you're creating something new? You just say, hey, please reference my personal OS architecture in this folder? Yep. You have to be in sort of the right place within the folders for it to know how to reference things, which is a little tricky sometimes if you're early on in the process. But as long as you're sort of at the, you know, my sort of high level folder is the cloud code files folder. So if I'm in there, I basically can say, go reference my operating system and do X, Y, and Z thing or do this. Yeah. So I basically will have it do that. If I'm in a lower file structure, it'll only sort of see what's in that lower level file.
36:30What are the different categories and frameworks that you have? So we have, yeah, go up to the top again. so systems architecture you've got the smart context manager basically you're telling it to say you're saying be token efficient yeah modular design okay so you're talking about the way that you're scaling your security framework and your master launcher and then you've got the enhanced project management like basically you're just telling it the things that you like yeah right exactly or like again when i said always make the chat log or always make the context or whatever those will be rules in this folder.
37:07I'm still sort of iterating through for like when I create a new project, always inherit these sets of rules and create the cloud.md file and always create the to-do list file. So that's sort of like, again, I've iterated this a few times, so I don't even know what's right now. I have to figure out what's what at the moment. Those are like the rules and agents I'm building out, or sorry, those are the rules I'm building out. And then from there, I'll have like little agents that'll go and say, okay, on a daily basis, go and get those to do files, go and get, you know, check the security, go and, you know, see, you know, debug something, go like you could have like.
37:44So let me ask you this. I get, I get that the agents are running at the end of the day and they're gathering that information. They're putting it in the dashboard. If you're in the middle of a project though, what do you have to do in order for it to generate the data so that the agent will find it and take it, right? Because you have one agent that's like, go find all the to-dos. Cool. If the to-do is not generated, it's not going to be able to find anything, right? So how do you make sure that at the end of the day, do you have a mental checklist or do you also prompt it to say, at the end of this, make sure you do this?
38:16Yep. So you can do it either way, like both work. Right now, I have been doing it more on the mental side of things, but I've just started adding these rules and tasks. So let me even see if I can find in one of my other folders rules that I can show you. I mean, theoretically, you could create an agent that's like, go at the end of the day, look at every new folder that's created or new project, whatever. I don't know exactly how you define it. Pull out all of the to-dos. And if there are any to-dos that are missing, put that on the dashboard so that when you go to the dashboard, you're like, oh crap, I didn't put the to-dos.
38:48Exactly. Right. Like it's like, oh, there's a yellow flag there for me. Exactly. So here's one of my rules folders for one of my projects, this one. And let me see if one more that we can pull up. So this is like my to-do collector agent that's doing that. Okay. So I'm right now trying to build out a little, what's this the same one? Yeah. Okay, cool. A vendor quote collector. So basically like we have to get our windows is redone in our house. So I call gazillion places. So I want it to basically create a little way that will email out all that basically scrape data of a bunch of vendors, get their contact info, and then start cursing.
39:27So anyway, so I started this project. So I put together these rules for what it needs to do. Project overview, technical considerations, development phases, context management, and status stuff. So here's where it says like create the to-do file, analyze requirements, build out the architecture on the design. And then in my to-dos, it basically built out a checklist of stuff it needs to do. And then I haven't really been working on this, but it'll go through and check off hypothetically each of these little things as it gets it done. I know this just comes easy and natural to you, but I think there's a huge opportunity to start businesses just around this, around helping people, companies get organized with their data infrastructure with respect to AI.
40:15Because I have stuff all over the place and I always am like, did I do that in Claw? Did I do that in ChatCBT? Did I do that in Perplexity? Whatever it is. I would bet I could build agents that are like, at the end of the day, go to these chats or at least figure out some way to do that to stay organized. You always blow my mind, Liz. You know what, Agent, we really need to build, though, or maybe I'll start working on this, is basically all the stuff we talk about from the pod that can just then go and build off those little apps for it. Yeah. Which is possible. It's possible. The other thing I was going to say is I know the RSS feed got us over to here.
40:52Yeah. That is an amazing idea. You're curating information for people. And yes, I can go to perplexity. However, I'm still at the mercy of perplexity and what they're giving to me. I think most people would still like to have a human who's like cheer rating the news. That's why in like local newsletters, yes, you can go and gather a bunch of information, but the editor's insanely important because that editor knows the local area so much better than any LLM at the moment. So which news stories do you pick? What do you put in there? What do you not put in there? And so having this RSS speed built out, you could have custom, like I could, it's almost like building a newsletter.
41:31I could build a custom RSS feed that just goes out every single day of like, hey, Eagle Idaho, here's what happened yesterday. And I spend five minutes a day. Yeah. Just curating that information. Yep. Absolutely. Man, that is so freaking cool. I'm going to end it here. Liz, as always, this was fantastic. What do we have to look forward to next week? Oh gosh, next week. Well, there's been a couple of things I feel like we've talked about. I have to show you how to do some, I think data scraping and data cleanup with CloudCode, I think would be interesting. I know that's sort of boring, but it's something we all hate doing.
42:04And I know you've had use cases and I just finished a big project around that that saved me a very long time. That would be awesome. I'm working on something right now where it's like I've aggregated all the data. I have good data, but now how do I put it in the format where I can then analyze it? And I'm, spoiler alert, I'm not doing well. Well, that's where Cloud Code will be out. And again, everyone wants to do scraping, but you know, there's still some friction there and doing scrapers. So it makes that easy too. So we can do all that fun. Okay. Next week. It's a date. Sounds good. All right.
42:40Hopefully you liked that episode. And if you've made it this far, you're either really committed or you're stuck doing yard work and you can't actually skip on your phone. So while I have you, the show is growing, but I have a favor to ask of you. Will you please help me grow the show? I want to reach more people. there's a couple things that you can do like and subscribe is the simplest thing obviously you want to get notifications for when the next episode is coming out but if you go the next step will you leave me a review five star on spotify or apple what that does is it tells the algorithm that oh hey this is a high value podcast because more people are leaving reviews for it and it then pushes it out to more people so that's why when people are like will you like and subscribe and put the five star rating it's not just to make themselves feel better it's actually to get more exposure for the show.
43:25So if you do that for me, I would greatly appreciate it. And I'll see you next time.
From the publisher
🚨MY NEWSLETTER https://nikolas-newsletter-241a64.beehiiv.com/subscribe 🚨
Join me, Nik (https://x.com/CoFoundersNik), as I interview Elizabeth Knopf (https://x.com/leveragedupside) for an incredible dive into AI app building and no-code development tools transforming entrepreneurship. We explore Google's Nano Banana AI, their revolutionary Gemini Flash 2.5 image model with game-changing visual memory technology for precise editing and AI image generation.
Elizabeth demonstrates Google AI Studio live, showcasing the powerful Stream feature for real-time debugging and AI automation. She reveals how she used Claude Code tutorial methods to build custom RSS feed automation, replacing paid subscriptions, plus her personal AI assistant dashboard with AI agents for project management and task automation.
Questions This Episode Answers:
What is Nano Banana AI, and why is visual memory revolutionary for AI image editing?
How does Google AI Studio Stream enable real-time debugging during custom app development?
How can entrepreneurs use Claude Code and AI automation tools to build apps without coding?
What are the benefits of AI-powered project management with personal AI assistant agents?
Where are the arbitrage opportunities in AI app building for non-technical entrepreneurs?
Tools Covered:
Google AI Studio complete walkthrough
Claude Code tutorial for beginners
Nano Banana (Gemini Flash 2.5) AI image generation
Custom dashboard creation with AI agents
RSS feed automation and app replacement strategies
Enjoy the conversation!
__________________________
Love it or hate it, I'd love your feedback.
Please fill out this brief survey with your opinion or email me at nik@cofounders.com with your thoughts.
__________________________
MY NEWSLETTER: https://nikolas-newsletter-241a64.beehiiv.com/subscribe
Spotify: https://tinyurl.com/5avyu98y
Apple: https://tinyurl.com/bdxbr284
YouTube: https://tinyurl.com/nikonomicsYT
__________________________
This week we covered:
00:00 Introduction to AI Tools and Excitement for Innovation
03:09 Exploring NanoBanana: Google's New Image Model
05:50 Visual Memory and Image Generation Improvements
09:01 Google AI Studio: A New Interface for Building Apps
11:53 Real-Time App Development with Claude Code
15:04 Innovative Features: Maps and Video Analysis
17:53 Creating Engaging Content with AI Tools
21:47 Harnessing AI for Arbitrage Opportunities
23:13 Building Custom Tools with Google AI Studio
24:36 Creating a Personal RSS Feed Tool
26:27 Project Management and Organization Strategies
28:39 Developing a Personal Operating System
32:14 Automating Tasks with AI Agents
35:25 Curating Information and Building Newsletters
