247 - How I Built an AI Second Brain That Runs My Business (Step-by-Step) with Elizabeth Knopf

0:00Dude, this is incredible. I have folders set up for analyzing deals. If I'm in the middle of something, I'd be like, hey, use my deal analyzer to tell me whether or not this is a good deal. It's just there. Exactly. I love that. That's incredible. I think if you're at the point where you're like about to tune out because you think like, eh, too complicated, I would just stop being stupid and I would listen to the rest of this.

0:22All right, welcome back to Nikonomics. I've got Liz back with me again. And today we are addressing the two biggest pain points in AI right now. One of them being every time you go to a new chat, you got to re-familiarize the chat with who you are, what you're trying to get. You got to give it the context, right? And number two, have you ever like wanted all your Google Docs and your Dropbox and your whatever other files to kind of just be in one place so you could have sort of your AI brain playing in the background? Well, today we're going to talk about how you can architect the data infrastructure necessary to go and access all those data sources without aggregating them, creating a bunch of duplicate information.

0:56So without further ado, Liz, I'm stoked. Please teach me. I'm ready. First, let's start off with something Claude came out with. Of course, I was working on this AI brain project so that everyone in the audience can basically build out their own AI brain, their personal operating system. And then, of course, Claude came out with a new thing called skills. So what are skills? So you may have heard of agents. and again, just to refresh what agents are, that's basically being able to do tasks for you. They have some autonomy, they have memory, they have context, and they're able to access like third-party tools to do things.

1:35So think of them as like your employees and they're doing tasks. Then they came out with this thing called skills because agents is actually a little bit fusing and sort of how do you make them? How do you build them? And people were just sort of wondering how to do that. And agents were really only to be built in Cloud code, but on the browser or desktop, they weren't really accessible. And so what Cloud did or Anthropic did was basically build out a way to simplify agents into something called skills. And so this is like for very simplistic tasks, which I would almost equate to the instructions and file examples that you provide in a Claude project.

2:19This is where Liam Neeson would come over and say, I have a very particular set of skills. You're listing out very particular things that this thing is doing. Exactly. Exactly. So like, for example, just to give a contrast between what an agent would be versus a skill, like a skill might be write this newsletter and you'll have an example newsletter. You'll have instructions and we'll even go through what a skill file looks likes and it'll be much clearer versus an agent is like your marketing agent that might have multiple skills and not just the newsletter, but they can do posts on social media and they can do other things.

2:54How is this different than projects? Cause projects I give custom instructions, I give context documents, and then I, that's kind of how I use projects right now. Like skills sounds the same to me. It is very, very much similar in the sense that it is now accessible to anything. So you can do it once versus like projects that's really contained to that project versus the scale is now accessible to essentially like your cloud, your desktop, and in cloud code. So like in any chat, I could backslash, I don't know what access is it, but be like, hey, I need you to be my sales agent. And it all of a sudden just draws on all that instead of you having to go to that particular project.

3:35Exactly. That's exactly. Oh, okay. Yeah. So that's you can sort of, I would say that's the simplistic way. Like there are a little bit more capabilities that I'm sort of just discovering and learning. Like this came out only a couple of days, but like that's essentially what it does. And so I've just found it already pretty useful rather than again, having to figure out what project and then change the instructions or context, or then, you know, having a million different projects that proliferate and then have to organize them and figure out what did I name it? All of that fun stuff. You want to share your screen?

4:05Yeah. That's really interesting. The way that Claude is doing this, it's almost like I instantaneously now understand agents where like, okay, it's a project that you made, but that project is restricted to that folder that you've created. Whereas the skill is the same as a project, but now I can access it in any project that I'm in when I need its expertise instead of having to go to that particular project. Okay. So first let me show you if you need documentation, because their documentation is pretty good. You want to go to basically their documentation section under agent skills, they'll have an overview.

4:40And so if you want to read about it, it sort of gives like info. And they also have access to like quick start tutorials, skills, cookbooks, and we'll actually go through some of the ones that they've already like preloaded in so that you're aware. And then we'll actually go ahead and we can build out a skill. All right. So how do you first access to build it? So if you are in just the browser, all you need to do is go down to your settings, which you click where your name is. Then you go to settings and then you go and you see this max plan. Okay. Didn't have to flex that hard. It doesn't help the money.

5:15I mean, it's, yeah. I'm not getting any back. I'm a max plan too. Don't worry. capabilities. And so then you're going to see this section where there's artifacts, AI powered artifacts, which is, I guess how they're like, they're calling that a skill now or capability. You have memory on experimental, and then, then you have skills. And so you're going to have in this one I created. So we'll go in and I'll explain what that is in a moment, but they're going to have preloaded most of these. I think most of these I hadn't built yet. So you'll be able to turn them on or off. So just like to run through a couple of quick examples of what they've provided.

5:53So brand guidelines, pretty straightforward. And let's actually see, I can pull up. You want to see what they look like or like what these documents look like. You can reference them and they have the links again in the Claw docs. So if you want to see their cookbooks, you'll be able to go there and then be able to see what's under the hood of these because they also have this little skill that's a skill creator. So we can basically use one of the skills to create a skill, which is pretty cool. Okay. So I just went in to Claude and there's skills and it says preview and I click upload skill and it's telling me to drag and drop a skill.

6:30Yes. So we do create a file and you'll, unfortunately you can't just like have it automatically load in from when you create it, but we'll go ahead and do it together. And it's just saving it to your desktop and then uploading it. So before we actually go ahead and create a bill, I just want people to see what the files look like. And we can just sort of, for example... And I'm sorry, this is a stupid question. You were just in GitHub? How did you get access to this desk? If you go to the agent skills on their documentation page, there's a link in here and the link is called... Over the last week, I've gotten really upset with Claude Anthropic and it's things like this, like their customer service is, why do they not just have a link in the settings and the capabilities that you're trying to turn on and be like, oh, click here and you can go and see where all the different skills are.

7:21Instead, go to this freaking huge document and find the link. I know. But then we won't have the opportunity to arbitrage the information by showing people stuff. This is true. Thank you, Liz. Thank you. Glass half full. Okay. So it's skills cookbook. And again, I know GitHub is like scary, but it's literally just a file folder, like, or a place to like put your files up to repository repo. That's what it is. So don't be scared. So in the skills here, they have a bunch of examples. And so we'll do... You know, this is just an interesting question. Could you build a skin or a wrapper to make it not look scary like GitHub?

7:57So people can just like, oh, I'll click through the skills. Yeah, I did that for N8N because I hate the UI and there were weird, again, filtering, sorting, simple UI things that should be solved for. But yeah, like I'm surprised like no one has made a better like search engine for GitHub or just something that's for non-technical people to access GitHub. Because literally in if we get to it later in the in the video, one of the steps of my process is to go to GitHub and to look like has someone built this already so I don't have to recreate the wheel? Or can I get inspiration for things that they've done?

8:32But again, it's like sort of scary to, to just navigate this and then figure out, okay, maybe they have okay documentation, but like, what does this look like? How do I load the thing? Right, right. It's very scary. So I think that's a great idea. Wow. Yeah. Just broadly speaking, how do you know whether or not an agent that somebody else has built is good? Like, how do you know that has all the components necessary for it to function well and for it to execute on the tasks that you want it to execute on? But like, that's, there's a skin, there's a wrapper. That'd be a great wrapper. Well, actually, now that you've mentioned it, slide tangent that we'll talk about later in the video as well, plugins.

9:06So they have now made agents or groups of agents or a workflow accessible because what was happening was that people were building these agents. They were building sort of in-cloud code, their different workflows. And even like people have reached out about the personal operating system. Now you can actually create a plugin which enables them to be more shareable and accessible and they have a plugin marketplace. So again, the nomenclature of all this stuff is like, I don't remember what this is called. How is this different than that? But you can just think of a plugin as like something you've built that's agents or your workflow that's now shareable.

9:41And I think they're getting to the point where you might be able to monetize it. But right now it's just - OpenAI did a good job with this with their GPT store. Exactly. Like, again, I don't know why, Claude, the artifacts was like kind of a step in that direction. This seems like they should have something for it. But anyways. Yes. Back to the main assignment. Back to the main assignment. Okay. So basically what I did was I took an image from Canva that I had and I wanted it to basically create brand guidelines based on this image. And so what I did was, we're going to do this together live, is take, uploaded the screenshot.

10:19And then I said, use the skill creator capability. So that's referencing again, the capability itself. Yes. It's representing, referencing itself. The skill creator capability located in the capability section and use the skill creator capability and create a skill called social brand. All right. So, Oh, so you created a skill to help you create other skills. No. So they default and give this to you already. So literally you can go in like when you load, actually, this is a question for you. Do you see skill creator? No, no, I have, there's nothing. It just says upload skill. Oh, okay. That's why.

11:09All right. So, yes. So, so I had done this a day. Yeah. But there's a skill for creating skills. Yes, exactly. So if you literally go into the Anthropic or we can probably leave a link in the description so that people can just download these folders and all you literally will have to do is upload a folder so you don't have to even go to GitHub. And what you'll do is you'll go to this. We'll have a skill and it's called skill creator. And this will literally then help you create other skills so that you don't have to figure out, oh, here's the format of the file. Like, again, you just use AI to create the file format for you.

11:46Then what you'll do is you'll go to upload skill and drag and drop the skill creator, the social brand document in here. And I already, oops, that's not a zip, but you'll upload the skill in there. Okay. So I already have this. So replace the skill that's existing. I'm not going to do that because I already have it there. And then it'll literally show up. So just make sure it's a zip folder. It has to be a folder and it has to have something called skill.md file. You can add examples in there. But again, GitHub scary for somebody like me. So what we're going to do is have a link where Liz is just going to have an easy file set up where there's like whatever, 10 basic folders that you can start with to start building some skills.

12:29Yep. And just upload one by one and name them whatever you want. And then you can preview them, run them. And then if you don't like what you see, say, Hey, can you update it and make an adjustment? So like I took this image and I said for brand guidelines, which we'll have in there as well. and just say like update the brand guidelines to be like this image and it will just do it for you. And so once you do that, so then I, for UI, for instance, I had just to show like an example of the contrast. This is just like the out of the box UI that Claude does, whether it's on Claude code or Claude, this is literally in Claude code, but you always get this weird gradient of purples or blues or like tasteless UI.

13:16that's massive buttons. Exactly. And so I ended up applying my brand guidelines and got something a little bit better. You can't really see, but there's that like background. Yeah. So yeah, so it looks better. I'm not a designer, so feel free to say how bad this is and all of that. But again, it just makes it like more aligned with what you're doing versus again, having to constantly reference that. And so when I'm, when I'm in a chat, how do I access a skill? I just say use XYZ skill or do I like ampersigned backslash? Nope. Literally just say use the skill and you name the file. So they don't have any shortcuts.

13:53I don't think yet. Nope. No, this is incredible. So like I have folders set up for analyzing deals. I have folders set up for different projects that I have. If I'm in the middle of something, I'd be like, Hey, use my deal analyzer to tell me whether or not this is a good deal. I don't have to like click out of that, go to a folder that I've set up a custom GPT. It's just there. Yep, exactly. I love that. That's incredible. And if you have the skill creator, it's like, let's say you've aggregated a bunch of, done a cool analysis rather than like what I do is always like say, okay, now create a framework or again, add this back to my documents, my Google drive, and then add it to a project folder and cloud projects.

14:34And so it's like, you're managing all of this stuff. Now you can just have it as a skill and then reference it. So anytime, just think of it like a framework or a template that you're using, that's sort of a skill. Okay. Just like knee jerk, when would I use skills? And when would I use projects now? Projects, I would say, is like isolated. So you could have multiple skills referencing projects. Like projects, I would say, is more when it's stagnant information versus like a skill. Like you might be, again, brand guidelines apply to more than like maybe you're doing a newsletter and you need to apply brand guidelines.

15:05So maybe you've used all your context in your instructions for the actual examples or like you're putting in some new news story. And it's like, OK, now I have to add in the brand guidelines. And it's like you have the guidelines plus the content plus something. So now you can reference the skill brand guidelines to then visualize the content that you have in the folder. So it's like being able to say, like, I have my copywriter who's doing the copy. Now that's in the project folder and they can really focus on that. And then now I can bring in the designer to do that rather than having that within the project folder.

15:38Hey, 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? We go leave a five-star review for me on Apple or Spotify. It would really help.

16:11And if you want, even share this with a friend. So the skill itself doesn't have a ton of context. So if I create a project, for example, I've got instructions. I've got a, this is how I structure my projects instructions. Obviously I have a context doc that is like, Hey, here's how I want you to act. And then I have a bunch of context documents of examples of how I want them to act. So for the deal analyzer, for example, the context doc would be like, Hey, you're whatever this analyst, or you have this much experience. And this is, these are the multiple ranges that I like. This is how I like EBITDA to be, et cetera.

16:42Here are examples of good businesses that I've seen. And then, then I have a bunch of context documents that I can go and pull the skills, not doing that. Like it doesn't have all the other context documents. It's just kind of that either instruction or context document. Exactly. You can think of it like that. Like I would say from what you just shared, like all the dynamic stuff that you might have new deals that you're populating in there, new information. If you have, I want it structured and formatted that this certain way, like you're probably gonna have that on a bunch of different projects that are maybe similar.

17:11And so you can do that. Like, again, it's a folder, like even if you have one deal in one project, you're probably not having multiple deals in the same project. And so you might have different folders. And so rather than, again, take up that context by saying, I like my formatting this way, you can sort of invoke the skill into each of those with them having, without having to use up that context with the skill or that file. Or you could have it be like, let's say that you, your text files in a certain format for when you're building out your databases, or if you want a certain Excel spreadsheet populated a certain way, you could have, Hey, this is the skill.

17:50It's the data extractor skill or whatever. And then it, it automatically knows, okay, this is what I'm looking for in this data document. I'm going to go ahead and help it repopulate this. So yeah, it's focusing on skills or tasks. Interesting. Or like even here's an example, like you just had a meeting and again, from the meeting, you're going to need a summary. You're going to need to email that out. You probably have a format in which you do at that email. So you could say you have an email formatting skill. You have a meeting notes skill again, to keep everything standardized. And so you can say, you know, here's the context from the meeting that just came into my, my Google doc.

18:26Here's some other information about the project or things weren't related. That's all in the project cloud project folder. And then you can say, here's, you know, use the skill for my email formatter to create an email, use the skill for doing meeting notes to create the meeting notes, so on and so forth. So again, you don't have to constantly be adding that in since it's like universal to every time. Very cool. I think that anybody who has messed with AI even a little bit has gotten to the point where they said, man, it'd be nice if they just remembered, if there's persistent memory. and chat GPT does a pretty good job of having persistent memory across chats.

19:01However, you've still got to input information into chat GPT in order for it to have that persistent memory. And there are some ways where you can tie in your Google drive. You can tie in Dropbox and these other file saving formats. However, there you haven't exactly built your own LLM. It's not like it's, Oh, this is my second brain. It knows everything about me. It knows my whole schedule knows. And it's simple. at least that's what I thought. I don't, I didn't think there was a, you know, a way to do it or even a simple way to do it. And Liz and I were talking last week or the week before, and she started showing me and talking to me about her second brain and sort of the way that she's got the data organized in order for it to populate.

19:40Cause that's, that's kind of the hardest thing is in order for these models to know how you would think or act or to know about your last week or, Hey, what, what items do I have coming up? What did I get done last week? What are the most important projects, et cetera, it's got to have access to your calendar. It's got to have access to your Google drive and it's got to have access to your Gmail and it's got to have access to everything else that you're kind of working in on a daily basis. And so what we're going to do right now is going to show you how do you access those things without necessarily building a brand new database and a whole new LLM.

20:09Absolutely. The problem I was trying to solve for is I have amassed and hoarded a bunch of files and digital information in my Google workspace. And probably like many of you, whether you're using Dropbox, Box, it doesn't matter. You've probably accumulated stuff and you have it in these folders, already set up a system. And now you're like, oh my gosh, now there's this LLM that I would love it to just go through and figure things out. Now, of course, in Google, they've started to iterate this within like the Google workflow and Microsoft within their suite, where they can access things within those documents.

20:43But then if you want to go and build tools, whether it's in NAN, if it's cloud code, or really any system, codex, whatever, you have to then figure out how do I utilize that set of documents for reference? Do I have to put this into this thing called a vector database, which is, you know, specialized for LLM so that they make all these connections and understand each other? And so do I have to now create a whole new database and then have two and then start proliferating and it costs and like all of these new challenges. And so the things I wanted to try to solve for as like step one was how do I take my Google Drive and just make it accessible for cloud code and for all of my projects so that I don't have to duplicate it and I can get at least a ton of the, like, you know, I won't have it as a vector database, meaning it's not going to be perfectly accessible as you would if you put it into an actual vector database that is creating all these cool connections.

21:39but actually I was able to create good enough so that I can reference documents much more easily and then incorporate things in for my workflow because that's sort of the starting point is my database. Any questions before we dive in on how to do that? And it's actually not that hard or not as hard as you think. No, I think if you're at the point where you're like about to tune out because you think like too complicated, I would just stop being stupid. And I would listen to the rest of this because in order for these models to actually provide the most value in your life, they've got to have access to all the data that you have.

22:14If you want to give it to them, right? If you don't want to give it to them, whatever, but they've got to have access to the data. And just because the UI isn't pretty, the UI being pretty doesn't solve for the problem that eventually you're going to have to organize your data. Like it's just going to take time to figure out how do you map all these things together. So either do it now or you're going to do it later. I would do it now so that when the cool UI comes out, it's already done and you can actually take advantage of it right away while other people are trying to figure it out. So shut up, sit down, watch the rest of this.

22:39Cause I think this is insanely valuable. Keep going. Yeah. Okay. So I'm going to start from the beginning. So just again, give people a quick refresher on like how to even access this. So you're going to open this terminal. If you don't know where this is, like, again, search on your desktop for it, terminal will come up. And then all you got to do, I'm going to just get to my place where I have my folders. And I would just say like know where you want to put your stuff in your local files because um it's spoken to people who will just put clod right away and then they don't realize that all the files being created are just sort of like all over the place so you do sort of want to think through a file structure and so as you can see i'll just pull up my file structure so people can see where i'm going i actually had clod help me with that i was like my my files are a mess can you make it organized I was like, sure.

23:26And we like worked through it. It was cool. Exactly. So CD that just says change directory to document slash cloud code files. I probably should shorten that name. And now I'm going to invoke Claude. All I've typed in was Claude. And if you don't know how to install Claude, we had an earlier video a couple episodes ago. You can go check out to see how to get that set up. Okay. So now Claude is running as you can see this little guy. And so I'm going to again, switch over to a different folder. But again, just so you guys can see on the screen what files I have here. I have systems, which is really where all my agents are.

24:01This is sort of a UI I created again, which we just showed that will basically show all the files I index. So how did we get there? Okay. So again, went through a couple steps to get to this little terminal that's black, scary. And just so you know, again, where we're located within my file structure. So we're in documents, cloud code files. Okay. So we have systems, which is basically where all my agents are located. I have active projects that I have going on here. So as you can see, I have a ton of projects going to finish at some point. And then I have like archive of stuff like I no longer need and staging.