Transcription
Today, we're covering the single best AI productivity workflow I have ever seen. You are going to get a second brain. We're joined by Matt Wolfe. He's going to break down how to build this system in 15 minutes, and it's going to transform your personal productivity and your work life. Let's get to today's show.
Today on the show, we're going to show you how to build a second brain. I'm joined by Matt Wolfe, and he's going to break down how to take all of the information, your meetings, your notes, to basically build this second brain personal wiki, so that you can always have access to everything you need, and basically be 100 times smarter than you are today. So, Matt, welcome to the show. Thanks so much for being here.
Yeah, thanks for having me again. Love it when you come on. You always have really dope things to show. And what we're showing you today, I would argue is like the most popular non-coding use of something like Claude Code or OpenAI's Codex that exists right now. And it all started with a tweet from Andrej Karpathy, who is one of the early pioneers of artificial intelligence. Maybe like give us a little background of like what the heck we're talking about here.
Yeah, so this tweet here is from Andrej Karpathy, and this sort of set off the whole thing, where everybody was talking about building these LLM wikis. And his tweet here actually kind of gets into the weeds, but you can see 20.8 million views.
That's a lot.
So, this one post here kind of set people down this path. But essentially, what this tweet is saying is that he was experimenting with building like this internal wiki. What he's doing is he's going around and he's using Obsidian, which we'll talk about what Obsidian is in a minute here, and he was using a IDE or coding platform. This actually works with, you know, Claude Co-work, Claude Code, or Codex, or Cursor, or any tool you want to use. It doesn't really matter. I personally use Codex. But, what you do is you use your IDE, and you save these markdown files into Obsidian.
So, the way I've been using it, just to kind of clarify what I'm talking about here, is I watch a lot of YouTube videos, and I read a lot of articles, and I save the transcripts and those articles all into this Obsidian sort of markdown database. And what happens is these just build up and build up and build up. Well, this layer that Andre Karpathy sort of mapped out here is the layer where an LLM looks at all of the stuff that you're throwing inside of Obsidian, all of these markdown files of transcripts and articles and things like that. It looks at all of these and sort of cross-links them. Like when you think of something like Wikipedia, you'll read a Wikipedia article, and you'll see somebody's name mentioned in the Wikipedia article, you click on that person's name, and it links to the Wikipedia article of that person, and you read that article, and maybe it talks about the product that they invented, and you click on that link, and it links to the Wikipedia page of the product that person invented, and everything just sort of becomes this like network of interconnected information. So, that's essentially what it is.
I don't know if that explanation was too in the weeds, but It's like your own little personal curated internet.
Exactly. Exactly. just like you you created all the stuff that you like, and the LLMs are stitching it together to make it its own little mini internet and that serves as a context for any work you want to do, any questions you might have, anything you want to do in your day-to-day life. Is that right?
Yeah, exactly. Exactly. And it finds the interconnections between the things, as well. So, sometimes it'll even find connections between things where maybe you're not making the connection between, right? So, like maybe you're putting in information about, you know, AEO, and you're putting in information about Facebook ads and those are two separate like little silos inside of your document database, but it'll realize that those are both about marketing and growing your business and it'll find those inner connections and sort of link those things together as well. So, it starts to build out this network of interlinked information and ideas and sometimes it'll find interlinkings that you might not even be thinking of.
Which is the magic of having an LM working in the background and do all the things you don't have the time or wouldn't think of in the first place.
Exactly. Exactly. If you're like me and you're watching the show and you're like, I got to go do this now. I want to try to build this. We've got you. You can scan the QR code or you can click the link in the description below. We've got all of the resources, all the instructions you need to build this yourself. So again, scan that QR code, click the link, and get started today.
So we have this tweet that was read by 20 million people. It has inspired thousands, tens of thousands, hundreds of thousands of people like yourself to go and build these systems that you just described. Maybe take us through like you mentioned this Obsidian thing. I'm not a web developer. We're talking to founders and marketers who probably aren't spending all of their day in IDEs and all these technical things. Like give us the like lay of the land like hey, I'm just a guy who wants to be smarter and have more information. What do I got to do?
So, you've got two main tools here. You've got Obsidian. This is Obsidian on my screen right here. Now, Obsidian is just a markdown file organizer. So a markdown file is essentially just a text file with some additional formatting on it. That's really all it is. This is like a markdown file here. This is my index markdown file and you can actually see everything that's saved into my wiki here. This is like a just a giant table of contents of everything. Yep. So, you can see down here, here's different concepts that are saved into it. But, again, this is just a text file with some extra formatting on it. But, all of these inner link to these concepts, right? So, I've got topics and entities here. You can see I've got answer engine optimization as one of them. There's seven sources inside of my wiki for answer engine optimization. If I click into this page, you can see it sort of is inner linked into this answer engine optimization. And then down here, you can see here's the seven sources of content that I have about this specific topic. And all of this stuff kind of cross links between each other. I didn't do any of that cross linking. That's the large language model reading these text files and going, "This text file kind of looks like this text file. Let's connect the two." So, that's what Obsidian is. It's just a a markdown file organizer. So, if I click into concepts here, each one of these concepts is essentially just another text file that links to other text files.
And so, Obsidian is a free application that you download.
Yes. How do you get stuff into it?
Okay. So, Obsidian also has a Chrome extension. So, if I open up my Chrome here, we've got this Obsidian Chrome extension up here. And I literally just click this and it will pull it in. So, right now I'm looking at a tweet, and you can see here's the entire tweet. I just click add to Obsidian and it's now inside of my knowledge base.
Okay. So, what we have here is this free Obsidian app that is essentially a knowledge base personal wiki. And the primary way to get stuff in there, I assume you can manually copy and paste stuff in there, but the primary way is through like a Chrome extension where you're just like, "Hey, I'm consuming this thing." I assume it does like transcripts of YouTube videos and all that kind of stuff and you're just popping them right in.
Yep, exactly. Exactly. Like, in fact, here's the Futurepedia um YouTube channel here. I actually didn't add this video about AO here cuz I wanted to have one that was an actual example here. So, if I come up to the Obsidian web clipper, it takes a second, but look, it's got the entire transcript.
Nice. I see.
already here. So, I just press this one button and it saves it into my Obsidian vault. So, yeah, now we've got this entire transcript from this video with Kevin in it all here inside of my Obsidian vault. Now, this one that I just saved in here, it's not processed yet. See, it's in this folder called raw and this is just anything I dump in here. This is my inbox. This is where the stuff gets saved. Now, you need to have the LLM sort of review all of this, find the interconnections, file it into a concepts, break it out into, you know, what YouTube channel it came from and do all of that kind of stuff. That's where we start playing around with Codex or, you know, it could be Claude code, it could be Claude.
somebody wants to use in this case. We're only talking about Codex today cuz you're a Codex fan.
Yes, so I use Codex for this and what I do in Codex is I would literally come here and just say uh process all the files in the raw folder. That's all I need to do and it knows exactly how to process the files in the raw folder. And the way it processes the files in the raw folder is inside of your Obsidian database, you've got this little file here called agents.md. This literally tells Codex or whatever IDE you're using, it tells it how to handle specific prompts.
Mhm. So, essentially think of them as sub prompts. These are just prompts that if I tell it to go and process something, well, I've got down here the ingest. When the user adds a source and asks the LLM to process it, do these steps. And this is just a prompt, so it's just taking this specific prompt here and throwing it into Codex and now Codex is running this entire prompt for me. Read the source from the raw. Before processing, validate that the same source URL hasn't already been processed. A lot of times I'll throw the same YouTube video in there, not realizing I've already saved it once, so it has a little double-check on it. Create or update generated sources, create or update topic pages in the wiki/topics, create or update the entity overview, synthesis, or comparison pages. So, usually it takes like 2 minutes to process. But, one other thing that I like about Codex is they've got these automations here. And I can actually set these automations to run on a schedule. So, if I click in here, this is just a prompt that's going to run on a schedule. If there are any unprocessed files in the raw directory, please process them now. And this one actually is set to run at 12:50 a.m. pretty much every single day. So, this just happens at night every single night. I'll just be kind of dumping stuff in throughout the day all day long, and then at night when I'm asleep, it processes everything that was in the raw folder. I just manually told it to process this now, so we can kind of see what happens, but this actually does happen in the background without me actually needing to do anything on a normal basis.
And so, what's happening is like you're going out, you're clipping stuff, it's going in that raw folder, then the agent's going in, reading it all, interconnecting it, and making it easily accessible for the future. Cuz you might park it there for days, weeks, until you're ready. In this case, we you did an AEO article, you're saying, "Hey, I'm working on an AEO project, and I've got these specific questions." And then it's going to go access that and all the other content really quickly to get you the best answer possible.
Exactly. Exactly. In fact, I actually gave it this prompt before we jumped on.
Based on what I saved into the wiki, tell me the best strategy to get my website Future Tools to show up within chat GPT and Claude responses. We can see this should optimize less like a directory and more like a setable recommendation authority. The highest leverage strategy. You can see exactly what it referenced from within my wiki here. Keep the gates open. Your robots.text currently allows all, which is good. OpenAI says OAI search bot chat GPT search inclusion." And Anthropic says, "Claude search bot." Etc. here. Keep those allowed, monitor logs, and treat training bots separately from search retrieval bots. And then it shows the sources. And these are various sources inside of my wiki.
It's pretty remarkable that you can get just a super high-quality and also concise response, I would say. It's not like this like long typical LLM of like, "Let me generate a 30-page report that doesn't say much of anything for you, right?" Um, and it can do that because when you've given it good instructions and you've given it all the context. Matt, we are doing this in Codex. And for folks who aren't familiar with Codex, maybe just explain what the heck Codex is and how do you set it up?
Yeah, so Codex is sort of chat GPT's version of Claude code, right? It's their app for essentially writing code. And you mentioned that that last response was fairly concise. Well, that's one of the things that Codex is known for. When you think of like chat GPT versus Codex, there a sort of buzzword now called harnesses, right? They're both using the same underlying large language model, but the system prompt around each one is different. Codex is actually tuned to be very concise, get to the point, just write the code that's necessary, and don't sort of ramble on. I can actually tell Codex, you know, treat this like a chat GPT response, don't be brief, ramble as much as you want. And it will actually give me a long, sort of drawn-out message. Same with chat GPT. Chat GPT tends to be optimized to give you longer, more detailed messages, but if you gave it a prompt, be as concise as possible, tell me as little as I possibly need to know, it will do that as well. But Codex is just like their sort of app for doing code. So, if you've ever used Visual Studio Code, if you've ever used Claude Code, um, if you've ever used cursor or whatever, this is kind of chat GPT or OpenAI's version of that. It is their platform for writing code. And if you've ever used any of those other ones that are sort of based on Visual Studio Code, this one looks quite a bit different. But it's designed to just be very simple and give you just what you need to know and what you need to do and not a whole lot else.
So, you get this by just going to OpenAI on OpenAI's website. I believe they just have like a link to download It's a standalone app on your computer. And what I like about it, Matt, is it's a little less intimidating than Claude code. Like there's some actual UI and structure. You're not just in a terminal window, which I think can be intimidating for people who aren't used to being in a terminal window. But you're just going to go and and download Codex like you're showing here. And it's going to use just whatever credits you have in your chat GPT subscription, whether it's free or max or anywhere in between. Like that's it's just going to consume whatever you have access to in your regular chat GPT account.
Yeah, exactly. The app itself is free to use. The bigger the OpenAI account you have, right? You've got like the $8 a month go plan all the way up to the $200 a month pro plan. And depending on which plan you have, you get a different amount of credits or tokens to use inside. Even free plans, even if you're on a chat GPT free plan, you do get a certain amount of usage. But the usage is, you know, it grows based on how big of a plan you've got. But you can absolutely do this with the free plan. Um, you you don't actually need to have a Codex or a chat GPT paid plan to do what we're showing off here.
Yeah, and so when we go back into Codex, what we're really saying is like we now have this kind of easy-to-use way to access this personal wiki that you build over in Obsidian, right? It's like you got this little personal internet wiki database. Codex has access to everything. The question here is like how do Codex get access to it before we go and talk about some of the cool use cases here.
So, when you set up Obsidian for the first time, it's going to say what folder do you want to install Obsidian on? You install it in whatever folder you want on your hard drive. And then when you come into Codex here, you would just click on uh create a new project, use an existing folder. When you select this use an existing folder, you would just select whatever folder you would installed your Obsidian vault on. And then it can just read everything inside of that folder from here on out. That's as simple as it is. So, that's how it does that. But, you can actually go even further with Codex and give it access to a whole bunch of other stuff, right? So, you can give it access to Chrome. This is actually brand new.
Oh, that's cool.
Uh they just gave access to Chrome like this week that we're recording this. You can give it access to spreadsheets, GitHub, Slack, Notion. Mine has access to my Gmail, my Google Calendar, my Google Drive. So, it can actually cross-reference stuff that I'm asking it with my calendar. It can actually read my emails. One of the things that I often do inside of Codex now is say skim my entire inbox and look for emails that you can respond to based on what's on my calendar and what's in my wiki. So, it will actually use that information to help respond to emails as well. So, you can do some really, really cool stuff once you start adding in some of these plugins as well.
Well, and I think what's really interesting about it, Matt, is that we were talking before the show. We're like, "You know, this is really awesome." Like, there is so much you can do here. And it's also like you can do so much and accomplish nothing. Right?
[laughter] Because there's so much to build, you can just keep piling information and information into a system like this. And you're like, "Wait, I'm like a week in and I have this really cool second brain, but like I'm not using it for that much." And so, what I wanted to talk about before we close out was just like what are the ways if I was trying to grow for business, if I was a marketer, if I was an entrepreneur, that I would use a system like this to be really powerful.
For me, you know, I I'm constantly in consume mode, but I feel like a lot of the stuff that I consume just sort of goes nowhere. I consume it and then it's just kind of like, all right, I move on with my life. This is a way to sort of resurface it again in the future. So, one of the things you can do is you can actually create an automation here and let's say like every day at 9:00 a.m., send me a message with some recommendations of what I can do in my business based on some of the stuff I've recently saved into my wiki. I actually have a morning brief synced up to my Slack where every morning in Slack, I get a message at 9:00 a.m. that says like based on the stuff that you've saved inside of your wiki over the last couple weeks, here's what I recommend you try next. And it just proactively
That's very cool.
brings me things. That's one of the directions AI is moving into right now that I'm sort of most excited about is AI being more proactive versus you needing to go and prompt it all the time. We're getting to a point now where AI is starting to uh run on either like cron jobs or like these automations inside of something like Codex where it can actually start to like come to you and say, "Hey, here's what I think you should be doing next." So, I really really love that element of it. But, some of the other things that I'm doing with it is I created a sort of personal networking CRM inside of it as well. So, whenever I meet somebody at a conference, I say like, "Here's the name, here's where I met them." It automatically knows the date already just from the date I'm entering it. Here's some of the things we talked about. And then later on, if I bump into that person at another event, I can kind of, you know, pull up in my phone and be like, "Where did I meet this person again?" And it'll say, "Oh, yeah, you met this person at this place. They have a dog named this. You guys talked about um you know, baseball for a little bit or whatever, right?" And I can quickly remember what conversation I had with that person 6 months ago. And I've actively used that and it works really really well. So I've I've been doing that kind of stuff and it saves the the the contact information of the people inside of the wiki as well.
I would say this whole category talk about is like kind of proactive personal productivity where AI and agents are helping you be more productive but not like you having to remember to ask them all these things but like they're starting to insert themselves into your kind of core workflow and daily schedule which is super cool. Because email, Slack, and calendar access lets them do that and the ability to to schedule tasks and everything kind of makes all of that possible which is really really cool. I think there's some other really cool marketing use cases here. One is like everybody's always interested in what their competitors are doing. You can easily see making a competitive intelligence database here, right? Where you have agents that are looking for changes on websites, looking for new products, looking for new news articles, all of those things and not just giving you summaries but giving you proactive recommendations on if you need to do something based on this information, right?
Absolutely. Like I I've got one automation set up. So a lot of sites these days don't use RSS feeds anymore which kind of bumps me out because it's really easy to set up an It's really easy to set up an automation where whenever OpenAI or Anthropic writes a new blog post the RSS feed like alerts you, "Hey, there's a new post on this site." But those two specific sites don't have RSS feeds. So I actually set up an automation that looks at the website's site map and alerts me whenever there's a new page added to the website's site map and so I get like almost instant notifications whenever Anthropic, OpenAI, Google, DeepMind, Meta's AI blog, any of these websites actually add a new page to their site map. Now sometimes it's like they updated their terms of service or something like that, right? But I would say 95% of the time it's a new blog post that I'm getting almost like instantly that the moment it goes up. So, that's uh you know, one of the secret weapons that I use to stay on top of the news as it's happening is like create a little agent that's constantly watching the site maps of these websites for a new page added to the site map.
Well, and when you were showing the plugins a minute ago, you like you you you talked about that Chrome became a new plugin. And it's not just like getting data from Chrome, it's that Codex can use Chrome to browse, right? And if you're a marketer, that's really, really powerful because then you can use Codex for AO audits. You can use it for ad testing and optimization. You can use it for website audits, for product marketing collateral audits. All of these things where you can basically just build a section of your wiki that, you know, product marketing knowledge, best practices, skills, and then have Codex go and deploy it on whatever part of your site or a competitor's site that you're working on that day. Honestly, that setup is going to save you so much time because you then you're able to just like speed run all of the work and optimization and really like go through all the iterations and options you might have as you're going trying to rethink that work, which I think probably makes Codex for marketers the most interesting option to run this system in because the computer use is I think the best currently in Codex.
Yeah, yeah, it works really well. I will admit the computer use and the Chrome use, they're um almost agonizingly slow if you try to sit there and watch it go and do the stuff. You're just kind of like, all right, you need to go and walk away.
Yeah, yeah, you need to walk away. Like I almost want to start moving the mouse for it and be like, all right, let let me nudge you along here.
Uh but they
[laughter]
they do work really, really well. One other use case that I I really love that sort of ties back to the wiki and this is something I've really gotten the habit of doing is I'm a big journaler myself. So, at the end of every single day, I love to just sort of journal on my day. I love to say like here's the stuff that I accomplished throughout the day. Here's the things that I build. Here's the conversations I had as well as, you know, struggles. Here's something that I'm still struggling with. I needed idea for a video for Tuesday, but I don't have an idea yet or, you know, whatever's on my mind, I just sort of brain dump it all into a journal and I journal now directly into Codex. And I don't know if you noticed, but when I was showing you my Obsidian vault here, there is actually a journal sort of, you know, code here. And what it actually does is it cross-references my journal with everything that's in my wiki. So, I save a lot of mindset stuff in my wiki. I save a lot of like YouTube growth strategy stuff in my wiki, things like that. And it will actually read my journal entry and then say, "Based on everything you saved over time, here's what I found that's relevant to what you just journaled on." And for me, that's probably been the absolute most useful, helpful thing that I've done with all of this is just journal on everything. Just brain dump everything that's in my head and let it use the wiki. And it also will pull the knowledge from like the the LLM's knowledge that it's just, you know, trained on anyway, but it will sort of ground itself with the wiki first and it'll say, "Hey, this video from, you know, Marques Brownlee said you should be thinking about this. This video, you know, from Futurepedia said you should try this." Right? So, it will actually pull in a lot of the stuff from the wiki and combine that with the sort of general knowledge that's in the LLM. And I don't know if you've ever like journaled into ChatGPT, but it's usually pretty helpful and has some pretty good advice. But now the advice is very, very, very tailored to the type of content you're looking and actively consuming as well.
That's pretty sweet. Okay. So, my closing thought of where I think this is all going is that AI is still very single-player mode. Right? Like if you look at what we talked about today, it's like this is all just math stuff.
Mhm. The files actually sit on your computer, right? They're not in the cloud. Um all of this stuff resides on your actual computer and uh I think this is going to be transformative for teams as this technology continues to evolve. And imagine you're on a small team and all of you you have this collaborative wiki of context is going to be really valuable and I think going to be a game changer. The other like just pet rock I have Matt and we keep talking about this um Kiernan I offline is that like I can't wait for there to be a book format that's just a markdown file. Like when you pay a lot of money for just a markdown file of some of the books you love to be able to like inject them into these types of systems.
Yeah, I would love that. I mean right now I think the bottleneck is probably more context windows from these models than anything else. But yeah, I mean that would
You still can't go and get the raw text file of the book anywhere.
Yeah, I mean if you look hard enough there's definitely places, but they're not there there's some ethical questions there. Meta and Anthropic have already gone those routes. I that's fair. I'm just saying your your mortal like me. I'd love to pay like 30 bucks and have the context of this book in my AI systems. I think Kiernan I just finished the Loop Marketing book and I think we're going to probably have a special markdown file version for people so they can do that which which will be pretty cool. But yeah.
wouldn't be surprised if you know authors start doing that as a sort of like value add thing. That would be kind of cool to see. I mean I can also see why they absolutely would not want to do that, but it would be kind of cool. I can see some marketer brains doing that in the future.
Well, this marketer brain is going to do it cuz I want the ideas to spread. And this is how you get the ideas to spread. If you're optimizing for idea spread, I think it's a very smart smart thing to do. So I don't think this is a flash in the pan use case or or product to me. This seems like we're very early on in what is going to be just a new way that a lot of individuals and in the future teams work. And I think we just went over some really helpful use cases for both personal productivity and for marketing that can take this system in action. And my last closing question for you, Matt. People probably wondering wow, it looks like Matt spent a bunch of time to do all of this. Like how long do you think it takes to get like a good basic version of this up and running?
Um probably 15 minutes or so. And I say that because Andre Karpathy, he actually put up a page on GitHub with this method. So, you can literally open up your Codex, open up your Obsidian vault, copy and paste his GitHub page in, and say build this for me. And then just let AI sort of build the initial bones for you. So, just doing it that way is very very very quick. You know, and then obviously the amount of time it takes to just like inject new stuff into it. That's that's a sort of ongoing forever process.
a day kind of thing.
Yeah, but you can set it up really really quickly because you can just point AI at the place that teaches it how to go and build this for you. The beauty of AI is that you as long as you get the right instructions, it can do stuff while you're doing other things, which is fantastic. Matt, as always, it has been awesome to have you on the show. I've learned a ton. I've been fascinated with the second brain concept. I've yet to build mine. Now I'm going to go get in Codex and get this going. So, thank you so much, and we'll see everybody really soon on Marketing Against the Grain. Thank you.
This data is wrong every freaking time. Have you heard of HubSpot? HubSpot is a CRM platform where everything is fully integrated. Woah, I can see the client's whole history, calls, support tickets, emails, and here's a task from 3 days ago I totally missed. HubSpot, grow better.