YT YouTube2Text

YouTube2Text → Transcripts → The 4 Components of Every Successful AI Second Brain (Build These Now)

The 4 Components of Every Successful AI Second Brain (Build These Now) — Transcript

by Cole Medin · 4,615 words · 641 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00I use my second brain to run my entire
0:02business. Not that I'm automating
0:04everything, but no matter what part of
0:06the business I'm working in, I'm using
0:08my second brain to help in big ways.
0:10Researching the latest in AI news,
0:12helping me prepare my content, creating
0:13diagrams and presentations, helping me
0:16improve my agentic systems that I teach
0:18and bring to companies. No matter where
0:20I'm working, I'm talking to my second
0:22brain and using it constantly. And I
0:24truly believe now that you need a second
0:26brain, no matter what your role is, what
0:28you're doing in your life. It's become a
0:29necessity. And it's also so incredibly
0:31easy to build and incorporate these
0:33kinds of systems now. So, I've covered
0:36my second brain a lot on my channel
0:38already, but I've never made a single
0:40video where I cover the fundamentals,
0:42right? Like, what's the main framework
0:44to make a second brain just click for
0:46you? And so, that's what I want to cover
0:47in this video to help you get started
0:49building this kind of system if you
0:51haven't yet. But even if you already
0:52have a second brain that you've been
0:54evolving and using over time, I want to
0:56show you the fundamentals here so you
0:58can use that to sort of audit your
1:00second brain and just make sure that
1:01it's working as well as it can for you.
1:04Because here's the thing, I have seen so
1:05many people completely overcomplicate
1:08their second brain, either as they're
1:09evolving over time or building it
1:11initially. That is the number one thing
1:13I want to help you avoid in this video.
1:15Because we just have to think, like,
1:16what's the main objective for our second
1:18brain? Well, it should know our business
1:20off the top of its head. That's why we
1:22call it a second brain. It's really an
1:23extension of us. And there's not that
1:26many components we have to build to make
1:27that work really well. So, there's
1:29really only four components that I'm
1:31going to cover in this video with you.
1:32So, you build that as a foundation if
1:34you're creating your second brain
1:35initially, and you're auditing against
1:37that if you already have your second
1:39brain. And I'll cover both here. And I
1:41also want to mention quickly that
1:43everything I'm covering here applies no
1:44matter how you built your second brain.
1:46Maybe you built it up from scratch with
1:47Claude code like I did, or you're using
1:50Grok bots or Hermes or Open Claude. It
1:51literally doesn't matter. The four core
1:53fundamental components that I'm talking
1:55about here are your responsibility to
1:57build up, audit, and improve over time,
1:59no matter what your system is built on.
2:02Now, one thing you'll notice as we go
2:03through this together is my second brain
2:05is just a few tools and a folder of
2:08markdown files. Even though I've added a
2:10ton of capabilities over time to make my
2:12second brain literally save me 25 to 30
2:15hours every single week, I still haven't
2:18had to create this super sprawling
2:20system with a ton of tools I piled on.
2:22And it's really because coding agents
2:24have gotten that good. So, Claude Code
2:26runs my second brain under the hood, and
2:28I don't have to pay for a service for
2:31video editing for it, or for creating
2:32diagrams or PowerPoint presentations.
2:34Just running commands directly on my
2:36computer, my second brain is able to do
2:38all of that. And so, it's really the
2:39power of the LLM that allows us to keep
2:42it that simple. And that's really the
2:44biggest reason why it's so important to
2:46build a second brain now, because the
2:48ROI is insane. It doesn't take a lot of
2:50time to build these systems like it used
2:52to, and it takes even less time to
2:53improve them, especially because you can
2:56use your second brain to improve itself
2:58once you've built the foundation. And
3:00so, speaking of the foundation, this
3:02brings us to the four C's. This is
3:05everything that I want to cover for, you
3:06know, what makes your second brain
3:07really click. If you implement each one
3:09of these C's well, you're set up for
3:11success. And so, the framework of the
3:14four C's it actually come from Nate
3:15Hurk. He's another fellow AI educator on
3:18YouTube. I have a lot of respect for
3:19him. And I spoke at his virtual
3:22conference a couple of weeks ago giving
3:24a very similar presentation. And so,
3:26that's why I'm kind of bringing in his
3:28framework of the four C's cuz it's also
3:30just a really solid framework to
3:32understand what does our second brain
3:33really need. And so, this is what you're
3:35going to build and audit against if you
3:37already have a second brain. And so,
3:40we'll get into that in a little bit, but
3:41first I just want to talk about the
3:43tools that I use for my system. Like I
3:45said, I keep it incredibly simple. The
3:47agent that runs everything under the
3:48hood is Claude Code. And at one point I
3:51was using Codex for everything as well.
3:52It's incredibly easy to switch between
3:54the harness and LLMs whenever you feel
3:56the need to. You shouldn't do it often,
3:58but it's definitely possible. And then
4:00for the memory, the whole vault of
4:03knowledge for my agent, I use Obsidian.
4:06You can use something like Notion if you
4:07want as well. I just like Obsidian
4:09because then my entire knowledge base is
4:11just markdown documents on my computer
4:13that I sync between my devices. And so I
4:15use a sort of Carpathi LLM Wiki style
4:18knowledge base, which I've covered in a
4:20video that I'll link to right here. It's
4:22a great setup just to keep things
4:23simple, but still allow our agent to
4:25very efficiently search through
4:26everything, all the core memories and
4:28past conversations about our business.
4:30And then for all of the diagrams that I
4:32create, the visualizations that I show
4:34on my channel a lot, I use the
4:35Excalidraw plugin directly in Obsidian.
4:38And it's nice because then all of my
4:40diagrams are also just markdown
4:42documents. It is wonderful. And then I
4:44use Open Router for all my images and
4:46other models. So when I'm benchmarking
4:48different models and generating images
4:50like thumbnails, for example, I'm doing
4:51that all through Open Router. And then
4:53of course I have my second brain
4:54connected to literally every single
4:56place I work, like my Google Calendar
4:58and Gmail, my Asana where I do my task
5:00management, my GitHub repositories, uh
5:04my Google Drive for all my files, Slack
5:06where I talk to it remotely on my phone
5:07and then also helping me manage
5:08communication with my team. It is
5:10ingrained everywhere, and that's one of
5:12the C's we'll get into in a little bit,
5:14being able to work where we actually
5:16work.
5:17And then for where I deploy things, cuz
5:18like I said, I talk to it remotely on my
5:20phone even when my computer is off, I
5:22just have a simple VPS just hosted in
5:24the cloud. I pay like $20 a month for it
5:26where I have my entire system running
5:28with the whole memory vault syncs to my
5:30local devices. And so I use Digital
5:33Ocean. Hostinger is also really good
5:35recommendation if you want something
5:36cheap. It really doesn't matter. Just
5:38keep it simple. You don't even have to
5:40have it hosted in the cloud if you don't
5:41want. And definitely when you're
5:42starting and just building it locally,
5:44you won't even worry about it. But I
5:45just wanted to be very comprehensive cuz
5:47like literally what I'm showing you here
5:49is my entire stack. I I told you it's
5:51simple. It fits on a single slide. And
5:54so now going into the four C's, our
5:56fundamentals here with capabilities,
5:58connection, cadence, and context. The
6:00most important thing to understand is if
6:03you get this right, it doesn't even
6:05matter the underlying large language
6:07model that you use for your system.
6:08You're going to be set up for success.
6:10I've run my entire system with Claude
6:12code using even Haiku for some things
6:15and a lot of Sonnet, a lot of Opus. I've
6:16made my entire second brain work with
6:18Codex like I mentioned earlier. I've
6:20moved the entire thing to Kimmy just for
6:22the sake of testing. Like we have the
6:24new Kimmy K3.1 model now. And every
6:27single time, my second brain just works
6:29wonderfully cuz it really isn't up to
6:31the model. It's really the context that
6:34how well your second brain works, how
6:36much it knows your business off the top
6:38of its head, and that is our core
6:40objective. And so getting into the first
6:42C with context now. The sponsor of
6:44today's video is Arcade, the MCP runtime
6:47for AI agents. These days it's so
6:49incredibly easy to connect your agents
6:51to the platforms you work in like your
6:53email and ticket management where the
6:55real challenge is the permissions.
6:57Arcade sits between your agents and the
6:59systems where they act. And so each
7:01person signs in with their own account
7:03so the agent acts as them and never goes
7:05past its scope. The credentials never
7:07reach the model and your policies are
7:09enforced outside the model where a
7:11prompt can't undo them. Arcade also just
7:14launched tool recommendation, which is a
7:15big deal because Arcade itself has over
7:188,000 tools your agents can use because
7:20if there's a platform you want to
7:22connect to, Arcade has you covered. But
7:24then it also means if you try to connect
7:26your agent to all these tools and load
7:28them in up front, it would cost your
7:29agent 1.75 million tokens before it even
7:33reads a single message. And so we need a
7:35way to dynamically figure out the tools
7:37that we want our agent to have access
7:39to, and that is what tool recommendation
7:40does. And so here within my Claude code
7:42session, I'm connected to Arcade through
7:44the MCP server. All I have to do is
7:47describe the task I want my agent to
7:49perform, and it's going to call Arcade
7:51with a description which is going to
7:53give us tool recommendations. And so
7:55instead of loading them all up front, it
7:57gives the agent the tools it needs and
7:59then it calls those. And so I get the
8:00final answer, and this is perfect but I
8:03also don't have to dump all the tools up
8:05front. Arcade is free to get started and
8:07no credit card is required. I'll have a
8:09link to them in the description. Now,
8:11there are a million different ways to
8:12get the context right for your second
8:14brain and I'll show my strategy at a
8:16high level in a little bit. But the main
8:18thing is you need your second brain to
8:21feel like it knows your business off the
8:23top of its head. So without having to
8:25dig really deep, it should be able to
8:27know what you're working on and your
8:28main objectives, things like that. And
8:30so the audit for yourself, if you really
8:32want to test your second brain or you're
8:34building it for the first time, is to
8:36open up a fresh session and ask what
8:38your business does and who works there.
8:41It should answer like a teammate. It
8:42shouldn't have to dig really deep
8:44through the memory systems or the
8:45platforms where you work. It should just
8:47know. And if it doesn't, if you realize
8:50like, "Oh man, I'm not really getting
8:52the context right for my second brain."
8:54The simple solution is literally just to
8:56tell your second brain that is the case.
8:58Like, "Hey, it doesn't really feel like
8:59you know my business off the top of your
9:01head right now. What can we do to make
9:03that better?" You ideate and solution
9:05with your second brain. Even from the
9:07get-go, you can essentially use your
9:08second brain to make it better over
9:10time. So if you don't, you know, pass
9:12any of the audits that I cover in this
9:13video, you just tell that to your second
9:15brain and work on a solution together.
9:17And I know that sounds kind of cheesy,
9:19almost like it's too easy, but let me
9:20tell you, it just simply works. It's how
9:22I got my second brain to saving me about
9:2410 hours a week when I first built it to
9:26now upwards of 30, which sounds crazy to
9:29even say out loud. But I have a whole
9:30system for measuring this that I don't
9:32have time to talk about in this video,
9:33but I will cover in future content. But
9:36a big part of my day now is talking to
9:38my second brain saying, "This is working
9:39well, this isn't. What can we improve in
9:42our context system or the skills that we
9:44have?" Like I'll cover in a little bit.
9:46Constantly improving, not by itself
9:48fully, but it I definitely lean on it to
9:51make improvements to itself every single
9:53day. And so the context system that I've
9:55gotten to at this point consists of two
9:58layers. I have my core markdown
10:00documents that are loaded into every
10:02single session with my second brain.
10:04That's what really makes it feel like it
10:05knows my business off the top of its
10:07head. So I have the soul.md. That's the
10:10behavior I have for it and how I want it
10:13to act. I have the user.md, so it knows
10:15me and my business. I document all the
10:17core things there. And then I have the
10:19memory.md. These are all the most
10:21important memories that I bubble up from
10:23the underlying knowledge base, things
10:25like my main objectives for my business
10:27that I want it to know immediately
10:28without having to search for. So, that's
10:30the high-level layer of the second brain
10:32that it knows about all of the time. And
10:34then I have an underlying Carpathia
10:36LLM-style wiki that manages the
10:38knowledge base that sprawls over
10:40everything in my business. And so, every
10:42single conversation I have with my
10:44second brain, a summary is created that
10:46is ingested into the daily logs for my
10:50knowledge base. And so, this is what
10:52hydrates everything that I'm working on
10:54every single day. And so, I essentially
10:56have an entity markdown document for
10:59everything in my business, every video
11:01I'm working on, every company I'm
11:03working with, every person that I'm
11:05working with on my team for open-source
11:06projects and beyond. All of that is
11:09stored and maintained over time.
11:11And the way that I load the top-level
11:13knowledge base into my second brain is
11:15using a hook, a Claude code hook. And
11:17so, I start the conversation. It reads,
11:19of course, my global rules like my
11:20Claude.md, but then it also brings in
11:22the soul.md, user.md, and memory.md. And
11:26of course, these files speak to the
11:27structure of my underlying knowledge
11:29base, so it knows how to very
11:30efficiently search through that as well.
11:32And so, that is everything for context.
11:35The main audit here is, can you have a
11:36conversation asking about the core of
11:38your business and not have it search
11:40super deep to get you the right answers.
11:43And so, now that brings us to
11:44connections. Now, I would argue that
11:46context is the most important part of
11:48our second brain. But if we only have
11:50that, it's really not useful at all
11:52because it's not able to help us where
11:54we actually work. It's just a
11:56ultra-smart chatbot. But if it can't
11:58reach our task management to help us
12:00there, helping us draft replies to our
12:01emails, uploading things to Google
12:03Drive, then it's not reaching its full
12:05potential at all. And so, the audit here
12:08for connections is you ask for a Monday
12:10morning catch-up. Right, just asking
12:12about your priorities, things you have
12:14to reply to, people that need your
12:15attention. If it can't find all of that
12:17very quickly, then you're not passing
12:19this audit, and that's another
12:21opportunity for you to improve your
12:22second brain. It should be able to pull
12:25from literally every single platform
12:27where you work. And so, I mentioned my
12:29list earlier, but like my email, my task
12:31management, my GitHub repositories, my
12:33communication platforms like Slack, I
12:35have it connected everywhere. And
12:36generally, there's going to be some
12:38simple API or MCP server that I can
12:41bring into my second brain, so it
12:42doesn't take a lot of effort to
12:44incorporate each platform. And for
12:45anything that doesn't really have an
12:47easy MCP server to connect, I'm just
12:49going to build my own simple little CLI
12:51tool, and of course, using my second
12:53brain to help me build that, so I don't
12:55have to have some separate coding agent
12:57set up my second brain. It sets itself
12:59up. And so, what I've actually done is
13:01I've created a small little wrapper
13:03around all of the APIs and MCP servers
13:05that I use to connect to my platforms.
13:07And the reason I did that is because
13:09then I can determine the exact tools
13:11that my second brain has per
13:12integration. So, for example, I want my
13:14second brain to be able to draft emails
13:16on my behalf, but not send them, because
13:18I want to review before I send. So,
13:20instead of just bringing in the entire
13:22MCP server, and giving it access to
13:24every Gmail tool, for example, I have a
13:26lot of control. And then, also allows me
13:28to add more determinism in these
13:30scripts. I don't really want to get into
13:32the details of that right now, but yeah,
13:34it's just a lot of customization that
13:36I'm enabled to have when I have this
13:38sort of wrapper around my integrations.
13:40All right, so that's connections. Now,
13:42let's talk about the third C,
13:43capabilities. So, we have the core
13:45knowledge of our second brain, our
13:47context. We have it reaching out to our
13:49platforms, the connections, but right
13:51now, it doesn't know how we actually
13:53work, and that is our capabilities. This
13:55is where we define our workflows, our
13:57skills for our second brain, so we
13:59codify how we do certain tasks, so we
14:02can then delegate it to our second
14:04brain. Because here's the thing, we
14:06don't want our second brain to reinvent
14:08the wheel every single time for a task
14:10that we give it. We don't want it to
14:11figure out how to edit videos or create
14:13diagrams every single time, we want to
14:16figure that out with our second brain
14:17once and then turn that into a skill.
14:20And so, the audit here is to give it a
14:22short request for a job you do every
14:24single week. Something real should come
14:26out the other end and it shouldn't have
14:28to clarify a ton of things with you or
14:30reinvent the process. You need a skill
14:33if it does either of those because you
14:35teach it how you work and then you
14:36should just be able to delegate and
14:38trust that your second brain, especially
14:40more over time when you've iterated on
14:42the skill a bit. So, when I talk about
14:44improving my second brain over time so
14:46it saves me more and more time, I'm
14:48mainly referring to adding and adjusting
14:50the capabilities, the skills. And so,
14:53every single day when I'm telling my
14:54second brain this is working well, this
14:56isn't, it's a lot of times making
14:58adjustments to those skills and it's
15:00just prompting in these markdown
15:01documents. So, it's incredibly easy for
15:03me to review things and change them as
15:05needed. Like we've been covering here,
15:07most of our second brain really just
15:09does consist of markdown documents,
15:11which is part of what makes it so simple
15:13for us to mold over time. My rule of
15:15thumb here is if you find yourself
15:17explaining any kind of process, no
15:19matter how small to your second brain
15:21more than three times, you should
15:22definitely turn it into a skill. So,
15:25instead of saying I want you to do this
15:26process, do this, this, this, this and
15:28give me this output format, you just
15:30turn it into the skill. So, you do
15:31{slash} edit video or {slash} create
15:33diagram and then you just supply the
15:35extra context around what it's doing
15:37with that process. Like what's the video
15:38we're editing? What are we supposed to
15:40be visualizing with the diagram? And an
15:43insane collection of skills in my second
15:45brain at this point. Skills to edit
15:47videos, create diagrams, create
15:48PowerPoints, do web research, filtering
15:51all of the latest AI news for me. I have
15:53skills to help me create scripts for my
15:55content. I have skills to help me manage
15:57my inbox. Like a lot of skills that also
15:59help my second brain understand how to
16:01use the connections that I have. And so,
16:03all of these C's, they really do relate
16:05together because context helps you
16:07understand when we have to use
16:08connections. Capabilities will rope in
16:10the connections. Like, okay, part of
16:12this skill we're going to go reach out
16:13to my Dynamis community integration. And
16:16so, we're bringing all these things
16:17together, but the capabilities is really
16:19every single process we have for our
16:21second brain. All right. So, the last
16:23thing with capabilities that I want to
16:24cover here is how I use them to audit my
16:27second brain over time. This is the most
16:29important part of the entire video. If
16:31you already have your second brain
16:32because this is how I make sure that as
16:34I'm changing it, I'm not breaking things
16:36or making it worse in different ways.
16:38Because as I'm changing big things like
16:40how I delegate work to my second brain
16:42or the LLM that I'm using, I keep a
16:45running list of tasks that I have my
16:48second brain do with all my capabilities
16:50like video editing, diagrams, helping
16:52with my scripting, the news digest that
16:53I have it do daily. And so, every time I
16:55make a big change, I just run all of
16:57these as sort of my eval suite just to
16:59make sure my second brain is still
17:00working well or hopefully even better
17:03with the big changes that I'm making. I
17:04would highly encourage you to do
17:06something like this no matter what
17:08you're using your second brain for. For
17:10example, every single time I change the
17:11LLM in my second brain like going from
17:13Opus 5 to Opus 5.5, I use my edit video
17:16skill to just edit like a basic 3-minute
17:18video that I recorded a while ago. So, I
17:20do that each time along with all these
17:22other tasks just to make sure I'm set up
17:24for success. Now, I'm going into the
17:26rest of my week using the second brain
17:27cuz I don't want to encounter all these
17:28issues as I'm trying to do real work
17:30just because I made a bigger change to
17:32my system. So, that is everything for
17:34capabilities. Now, on to the last C here
17:37with cadence. This is all about having
17:39your coding agent, having your second
17:41brain do work while you're not asking it
17:43to do so. So, proactive nature. Now,
17:45even as you evolve your second brain,
17:47still most of the work you do with it is
17:49going to be you interfacing with it
17:50directly delegating work. But as you
17:52trust it to do more and more, you want
17:54to automate some of these things. So, it
17:56just does it proactively and you don't
17:58even have to ask it to do so. So, for
18:00example, in my second brain, the
18:01cadence, I call it my heartbeat. I have
18:03it run every 30 minutes and so, it just
18:05runs on that VPS that I have deployed in
18:07the cloud and it's going to look at my
18:09platforms like Asana and Gmail and
18:12figure out things I can just take off my
18:13plate autonomously. Like one of the
18:15biggest things that I have it do is
18:17draft replies to all of the emails in my
18:19inbox which I get a lot. That alone
18:21saves me like 30 to minutes to an hour
18:23every single day. And I will say the
18:25cadence is the last C for a reason.
18:28You're only going to be automating
18:29things once you trust your second brain
18:30to do it well. So, you build the
18:32capability initially. You work with the
18:34second brain one-on-one to use that
18:36capability. And then once you trust it
18:38to do a good job, that is when you
18:40automate it. So, the test here is to put
18:42one workflow on a schedule once it works
18:45well when you run it by a hand. And it
18:47should report back to you without you
18:48asking. You can't just have it do the
18:50work silently. It needs to come back to
18:52you with at least a summary of what it
18:53did. Like, for example, with my
18:55heartbeat, when it does these things
18:56proactively every 30 minutes, it gives
18:58me a Slack update just saying what it
19:00did very briefly and anything that needs
19:02my attention. So, those are the four C's
19:05at a high level. And the most important
19:06thing for you to understand here when
19:08you're just getting started building
19:09your second brain is to keep it
19:11incredibly simple. Even my more evolved
19:14version is still simple. But, like when
19:16you get started with your context, maybe
19:18even just start with some of those core
19:19files, like the soul.md and user.md. And
19:22then start building a sort of Carpathia
19:24LLM wiki. For the connections, just
19:27bring in a few MCP servers for those
19:28tools you first want to work with. Start
19:30with just one skill or a few skills for
19:32the most repetitive things you want to
19:34delegate. And for the cadence, that
19:36comes later once you trust things, build
19:38a heartbeat, just something that runs on
19:40a schedule basis to maybe even just
19:41start by drafting replies to your
19:43emails. You don't have to build all the
19:45skills at once. You don't have to make
19:46the proactive nature of the agent super
19:48advanced right away. Just do something
19:51to save you a few hours a week to start.
19:53And then it'll start to build on itself.
19:54As I said, we'll use our second brain to
19:57improve our second brain over time
19:59because it knows our priorities and it
20:01knows the things that we want to
20:02delegate to it more over time. And if
20:04you want to build the exact same system
20:06I use every single day, I have a GitHub
20:07repo I'll link to in the description
20:09with a skill to help you build the same
20:11thing I use. I'm giving it away for
20:13free. I even have a video where I cover
20:15this that I'll link to right here. Not
20:17that you have to build my exact system.
20:20Even if you want to take just the
20:21inspiration from the ideas I'm
20:23presenting here, I would encourage you
20:24to do that as well. It's just important
20:27to build that strong foundation and
20:29audit things over time as you evolve
20:31your second brain. That's why everything
20:32here applies no matter if you're getting
20:33started or you already have a system.
20:36And so, there you go. That is everything
20:37I wanted to cover today. If you
20:39appreciated this video, you're looking
20:40forward to more things on building
20:42second brains and genetic engineering, I
20:44would really appreciate a like and a
20:45subscribe. And with that, I will see you
20:47in the next video.

About this transcript

This page contains the full transcript of The 4 Components of Every Successful AI Second Brain (Build These Now) by Cole Medin, generated from the public captions YouTube serves with the video. The transcript has 4,615 words across 641 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.