How I Built My Own AI Second Brain with Claude Code (INSANE)
Chapters22
Describes a “second brain” that runs daily to find, cross-reference, and fix issues in the software, and pushes fixes to development. The speaker promises a detailed live demonstration.
A creator builds an autonomous AI “second brain” with Claude Code to autonomously monitor, diagnose, and fix issues in Harbor, collapsing deployment cycles into a sleep-running workflow.
Summary
Income stream surfers’ Hisham walks through his setup for a Claude-powered second brain, built to continuously scan Harbor for issues, verify them against known problems, and push fixes to development. He credits Cole Medin for the core concept and references a GitHub-based starter pack that powers the project. The system connects to third-party APIs via Python, uses a memory layer called soul.md and memory.mmd, and builds “skills” that automate workflows like product loops and issue triage. Hisham demonstrates real-world results: a deployment visible on Vercel, automatic churn analysis from Stripe, and post-audit fixes flagged by PostHog. He emphasizes that the tool rests on multiple integrations (Linear, Gmail, PostHog, Stripe) and that it operates autonomously, even running while he sleeps. He also acknowledges holes and the need for heavy customization, asserting that you must actively curate your second brain with your own preferences and data. The video blends hands-on setup steps with a candid caveat: AI can be powerful, but it won’t substitute your own vision and decision-making. Finally, he teases Harbor’s future features and founder pricing, inviting viewers to experiment and tailor the system to their own workflows.
Key Takeaways
- The second brain uses Python-based integrations plus MCPS (Mathing/MCPS) to stay connected to live APIs and keep data flowing across tools like Linear, Gmail, GitHub, and Vercel.
- A live deployment check shows Harbor dev v3 updated after autonomous fixes completed by the AI, including a gradient bug removal and early-stage audit findings.
- PostHog and Stripe churn data are scanned and addressed by the system, with auto-generated issues and fixes feeding back into the development loop.
- The setup includes memory layers (soul.md, memory.mmd) and a claw-based agent SDK, enabling the creation of “skills” that run daily and perform end-to-end product management tasks.
- The creator stresses iterative refinement: start with Cole Medin’s framework, then customize project-specific prompts and integrations to realize a practical, sleep-running second brain.
- The system emphasizes safety and governance, showing “never push to main” and “never push to prod” as irreversible controls in the second brain’s policy.
- Harbor’s roadmap hints at client portals, white labeling, and a more capable agency dashboard built on top of the second brain architecture.
Who Is This For?
Essential viewing for developers and product teams who want to automate issue discovery and fixes using Claude-based tooling, especially if you’re already using Linear, PostHog, Stripe, and Vercel.
Notable Quotes
"“As you can see guys, my second brain runs every single day looking for issues inside my software, finds issues, cross references them against known issues, and then fixes them and pushes to dev.”"
—Sets up the core capability of the second brain: autonomous issue finding and remediation.
"“This is next level, guys. I promise you this is absolutely crazy.”"
—Emphasizes how transformative the approach feels to the creator.
"“The cool thing about the second brain here that I really really like is the way that it connects to third party APIs. It actually uses Python files which makes things a lot easier.”"
—Highlights technical design choice enabling persistent connections.
"“I wake up and it's already put into development, right? Not into prod obviously it's in dev.”"
—Illustrates autonomous fixes pushing to development, not production.
"“AI doesn't know what you want. Only you know what you want.”"
—Underscores the need for user-driven customization and curation of the second brain.
Questions This Video Answers
- How do I set up a Claude-powered second brain for automated product improvements?
- What integrations are essential for an autonomous AI working in a SaaS like Harbor?
- Can I safely deploy AI-generated fixes to a dev environment before prod, and how do I enforce that?
- What are the best practices from Cole Medin’s framework for building a second brain with Claude?
- What is PostHog’s role in an AI-driven product management workflow?
Claude CodeSecond BrainAI automationOpenAI/Claude SDKHarbor (SaaS)Product management automationPostHogStripe churnLinearVercel
Full Transcript
As you can see guys, my second brain runs every single day looking for issues inside my software, finds issues, cross references them against known issues, and then finds actual issues and then fixes them and pushes to dev. This is next level, guys. I promise you this is absolutely crazy. I'm going to show you exactly how this is done in this video. Let's jump into things. Now, the first thing to note is I do not take any credit for this method whatsoever. This is all Cole Medin, who actually has been one of my favorite YouTubers for a very long time now.
I really, really like his content. I think he's really, really smart guy and always comes up with new and interesting methods of doing things. So, if you want a video on how to set up your second brain and you don't want to watch this video, guys, go and watch this video from Cole Medin. But basically, in the description, there is a GitHub link. You can also find it in the description of the video that you're watching now on my channel. This is the second brain starter pack. This is how you actually make your second brain.
So, I'm not going to go too much into like how this works and things because to be honest with you, like Cole explains it much better in his video, but I'm just going to give you a very quick way to actually build this. And also, I'm just going to quickly show you very quickly how it works. So the cool thing about the second brain here that I really really like is the way that it connects to third party APIs. It actually uses Python files which makes things a lot easier. So once it's connected, it's always connected.
So I use either direct integrations using Python layer or I just use MCPS. Both work completely fine. To be honest with you, I'm starting to use MCPs more and more. Previously, I didn't use MCPS anymore, but I'm starting to use them again. So there's direct integrations. There's the memory layer which is the soul.md which if you're familiar with openclaw you'll know exactly what that is. User.md and memory.mmd. This is like open claw but built in a sustainable way and built on top of the claw code and the claude agent SDK instead of using the clawed code oorthth in a way that you're not supposed to use.
Right? So then basically what happens is it creates skills. Right? So if I just go back on my clawed code here and if I write slashsklls you'll see that there are actually a load of skills here that it's made right. So product loop for example product discovery. So if I do slash product loop and obviously this will do this automatically throughout the day as well. If I do slash product loop what it'll do is it will run an entire product management run through of harbor right being my SAS which I've been working on for a few years now.
So you can see here running the product loop. Let me gather all signals to pick up the highest impact item. So what that means is it's actually going to go and fix things in my project which is completely crazy, right? So that's how that works. That's how it actually works. What it does is it creates skills. Skills are like workflows. It runs those workflows every day. It finds things that need to be done depending on what you are trying to do. And this works with everything, right? I also have one running for my um YouTube channel.
I don't really use that one as much. And I have to say, guys, a lot of this stuff is really, really cool and really, really interesting, but there are some holes in this, I have to say. And I I think it's still not quite there yet, right? But you can make one of these yourselves. You can experiment with it. And each one is incredibly unique and works in your way, right? So, however you need it to work, it will work in that way. So, for example, if you if you're on local Windows, it will work on Windows.
If you work on Linux, it will work on Linux. blah blah blah. Right? So, very very interesting stuff, guys. Let's just watch a little bit of this and then we'll jump into actually watching the or going through the actual project. Right? So, I've got the full picture. Here's my privacy assessment picking HA 94. HA 94 is from linear. Right? This was actually added by Rowan last night while I was or this morning while I was sleeping. So, really, really cool. Right. So, my product manager or my business partner, they put things in linear. I wake up and it's already put into development, right?
Not into prod obviously it's in dev. So if I just go on versel and show you what that looks like and also my second brain set this all up for me just so you guys know. If I go to harbor dev v3 right and then go to deployments and then you can see the last deployment was 17 hours ago and you can actually see a lot of these are writer audit findings right very very interesting so these were findings that were found without me right these were not found by me this is so interesting three surgical fixes driven by a fresh brower audit plus 7-day posttop scan what does that mean it does a complete check of the website using Google Chrome inside claude code.
It finds things that are wrong and then it fixes them. And then it also does a 7-day post hog scan. So you I have Post Hog set set up. It can actually find issues inside Post Hog and then fix them. Absolutely crazy. Another one guys and this I didn't set this up myself, right? It set it set it up itself, right? I can basically do slash I think it's churn autopsy. And what that will actually do is it will find someone who has recently churned on Stripe and it will attempt to find why they churned.
That is absolutely crazy guys. If you know anything about product management, this is an absolute game changer, right? So remove duplicate FAQs from homepage. So if I go on the homepage here, I is this in prod? Did I push this? Yeah, I did. Okay. So I already checked this this morning. I made sure it was okay. And look, now there's no more duplicates inside um the homepage, right? The FAQ. So, it did that completely autonomously of me. Again, this was there was basically an issue where this was.5, which looked a bit weird, right? Because um 290 divided by 12 is 24.5, right?
So, it was saying 24.5, but that looks weird. Nobody's like, "Oh, 24.5, that's cool." No, people see 24 and they buy, right? Another thing it did completely autonomously the other day was it looked through why people weren't converting in its opinion and it decided that the the pricing page was just wasn't good enough. Right? So again, it went through the entire thing and fixed it as you can see here and now it looks a hell of a lot better. And then finally from Post Hog, it found that there was an issue um where people were getting an issue with Grammarly and it attempts it has attempted to fix that.
Whether it's fixed that or not, I'm not actually sure. Okay. And now it's fixed the agency gradient bug which was put into linear. Uh linear is just the project management tool that we use uh over here. Uh so there's this was ran basically there's a gradient background behind the agency dashboard pages that's square and takes away from the design substantially acceptance metric whatever. So it will literally it's just gone and fixed that right and now it's pushing it to dev not to prod. It's not going to go into prod. It's just going to go into dev.
So once that pushes let's just refresh this and there should be a new build and then because this has access to my versel I just said this didn't actually start a build in versel. So now it's checking why that is. Okay so just so you guys know the kind of pinpoint of everything is obsidian. You can read everything in Obsidian. It's really really nice. So, known non-issues, these are basically false positives, which, you know, they're issues, but they're not really issues because either they've already been fixed or, you know, there's something else wrong with them.
There's we got churn autopsy here, which is so cool. Like, you just won't synthesize this information properly yourselves ever. This is so much better. So much better. Okay, so it it did deploy. I don't know what happened there, but let's just go to / agency, which is a feature that will be coming very, very soon to the live website. And there we go. You can see the green uh square at the back has now gone, the green gradient. So absolutely perfect, beautiful. This is just this is next level workflow, right? I wake up and things have been done.
It's absolutely crazy. So let's talk about how this is actually built. Okay, so I did this in a slightly different way to how Cole um lays out in his video. I won't lie to you guys. I did this in a pretty different way to be honest with you. So what you do is you do MK deer second brain and then the project you want to work on. So I'll put Grove because I've already got one for pretty much everything else. And then second brain grove will cd and then what you do is you do claw dash dangerously skip permissions.
I already need a thing for that. Uh actually let's do git clone and then full stop like that and then there we go and then I say I want you to start by reading this entire project and understanding what it is. I often do this with projects instead of just jumping into things and allowing it to read its own memory and blah blah blah. What I often do instead is I let it I I I I do it myself. Right. Just quickly guys before we continue I just want to quickly talk about Harbor. We've got some massive plans for Harbor and honestly now is the best time to sign up because probably the pricing will eventually change.
If you do sign up and if you email me I will give you founder pricing so you get more articles and stuff per month. But the tool is getting absolutely crazy now. We have some really really nice ideas for it including white labeling. We even have a client portal link. So, you know, in the future, if you're running an agency and you want to send a report to your clients, you can send them to the client portal, you can get them, you know, you can publish content here. You can see overall how your clients are doing, etc.
And you can just run like 50 different websites at the same time using this software. Agency is coming soon. We recently increased the quality of the writer. So, now it uses a better model. It uses Sonic 4.6. ICS. You can see that the quality of the writing is much better now. And just overall, I'm really really happy with how this tool is going. Just created a landing page generator as well. A blog rewriter clusters for topical authority where you can basically start to build topical authority to specific keywords that you're trying to sell or trying to push traffic to on your website.
And generally speaking, this is going to be the best tool on the market for the best possible price as well. Even if you don't get founder pricing, guys, it's €29 a month for 25 articles, unlimited sites, and all of this other stuff as well. Or you can go for the growth option, which is the most popular option, and you get 50 articles a month and much more as well. This is the best AI SEO content generator on the market, guys. Many, many people say many, many good things about it. There is a link in the description and in the pin comment.
Go and check it out now. Okay. Okay. So, once that's finished, basically what happens is, so you're supposed to fill this in yourselves, but like I wouldn't even recommend doing that, right? So, just watch for a second. I'll show you exactly how I do this. Let me just find my folder. Okay. So, the second step is I give it the context of what what I want to work on. So, I want to work on SEO Grove. So, inside SEO Grove V5, you'll find my project Grove. This is uh I want you to read it in full and understand it.
text stack and what it is etc. So there's like a two-prong process, right? You first of all you get it to read the second brain creator from Cole Medin and then you get it to read the project that you want to work on and then from there it already knows most of the answers to the questions that it needs to answer to ask. But basically what I do is I say now please ask me all the questions you need to make me a perfect second brain for my project right and then depending on what integrations you want etc.
Let's say you want linear you'd have to give it a linear API key or log in through linear or get it to do the oorthth for linear. Uh let's say you want Gmail you'll have to go on console.cloud.google.com get a JSON from oorthth etc etc. It really depends what you want to integrate, right? Cole Medin, he goes through this in a lot more detail. But yeah, basically this is an example, but I didn't do this, right? I just asked Claude code to make this for me, right? So instead of going through it, I just told it, you know, I'm on GitHub.
Uh I want to use Postto plus Google Analytics. I want to use this basically just saves time, right? because you know it'll take a lot of time to do this yourselves. But if you just ask it to do it for you and ask it to ask all the questions that it needs to know, then it's just going to be much quicker. So watch now. Please ask me all the questions you need to make a perfect second brain for my project. Press enter and then it's going to start asking me the questions that are inside this my second brain requirements.mmd file that we're basically just skipping through because I I can't be bothered.
Right. So name Davidson. Fine. What's your role? What do you do daily? What time zone are you in? Which are these actively used? So, you just answer all of these questions to the best of your ability, right? And then at the very end of the process, you say, "Okay, that's everything." And it will go through and it will create your second brain for you. Okay, guys. So, I'm just going to say, "Please ask me one at a time, not all at once, because otherwise it just gets a little bit complicated. I don't want to write a big message.
I just want to Okay, so what's your name? Hish and CEO." And I do uh linear checks every day, bug checks, and look for stuff using v using various product management techniques to try and find ways to make this tool better. There we go. Then it'll ask me the second question. What time zone you in? I'm in Ireland. Okay. What's next? Got it. Which platforms do you use every day? So linear, Gmail. Okay. Uh, so let's say Gmail. So it's just going to ask me one question at a time. Calendar. Let's say Google calendar. Sure.
Google calendar. Let's see what else there is. Task management. Linear. GitHub. No, let's just say linear. Okay. Chat messaging. Don't need any chat. Want to use Obsidian. Okay. Okay. Uh code hosting GitHub plus versel CRM or custom tracking. Do you uh post hog plus stripe? So it's just going to keep asking these questions. Just go through these one after another after another. No, that's fine. And then eventually at the very end of this, it will have all of the information it needs to create you the perfect tool. Right. So summarize linear overdue tasks. Monitor GitHub PRs and issues.
Track post hog metrics and flag anomalies. Monitor stripe revenue churn. Draft email replies. Keep track of product decisions while you made them. All of them are important. You can also add to this list, right? Of course. Uh I want you to be a partner. I would recommend partner guys, honestly. So ask only for irreversible actions. It always says bold choice. What should your second brain never do? Uh it should never push to main. It should never push to prod convex. It should never um delete anything. Right? So there that that'll put it in its brain.
So it will never do those things specifically. What types of knowledge matter works for you? Check all that apply. Uh all of them. Yes, I'm on MacOSS. Local only. Should have answered that, but local only. Great. Last question. of all your platforms, all of them. So, it always does this. I want all of them. I don't just want a few. Right. Fair enough. But the P. Okay. Um. Right. Sure. So, linear, obsidian, post hog, stripe. Let's say that's four, but I'll make it work. Shut the hell up, you bloody clanker, and just build it for me.
Does this all look right? If so, I'll fill out the requirements. Yes. Right. So now what it's going to do is it's going to edit this uh second brain requirements thing inside this folder right so not inside the second brain harbor inside the second brain grove so first of all it reads the skill so it understands and then it basically will just fill in this so we can actually see this. So if I go here you'll see that it will have now filled this in. Uh it's harder to see because it's not on obsidian but you can still see the little X here.
Uh, Obsidian. I like Obsidian, but honestly, personally, I can read Markdown anyway. I don't really necessarily need Obsidian personally, but okay. So, now it's just going to go through everything. It's just going to create the entire second brain. You can come back in like half an hour to an hour, and then you've got pretty much everything done. The only thing that would then be missing is integrations, right? You basically just need to integrate things one by one, and then you're done. Literally. Um, what I would really recommend doing after this is finished though is saying something like, "Okay, that's great, but now let's really make this way better and more proactive." And also, for example, um, have a product management template which finds actual issues and creates those issues in linear proactively.
So, this is what I would do personally after you've finished the first prompt, right? It normally does a pretty good prompt. It normally does a pretty good job, but I have found that it wasn't that good the first time round. You do also have to give it time, like it needs like a week or two to be honest with you, to actually start understanding what you want every single day. But overall, this is how you build these second brain guys. And they are effective. They can be a little bit more, you know, they might be lacking in substance a little bit if you don't really dive deep into it, but I'm incredibly happy with my Harbor one, right?
My Harbor one is genuinely useful. I implemented the entire agency dashboard in under 24 hours using this system. So, there is definitely something to this. I just think it needs a little bit more oomph and a little bit more from you because at the end of the day, these things have to come from you. AI doesn't know what you want. Only you know what you want. So, you have to sit here and make sure that this is your second brain in the sense that it has everything it needs to be your second brain. It can't be your second brain if it only has half of the information.
But yeah, overall guys, I think I'll probably leave the video there. I mean, after this, it's just literally it just does research. It creates a plan. You execute the plan. You come back, you say, "I need more. I need more. Do this. Do that." Then you put all the integrations inside it. And then you have this incredibly powerful tool that every single day runs for me while I'm sleeping or whatever and finds issues with Harbor and even potentially sometimes fixes those issues, which for me is absolutely crazy. This is like Open Claw, but actually good because it's using Opus 4.6 six and it uses the Claude or the Claude agent SDK which is also really really good.
I'm going to leave the video there guys. Thank you so much for watching. This is my take on Cole Medan's second brain. Shout out to Cole Medin. I'll leave a link to his video in the description and also tag him because I think he's a legend as well. Thank you so much for watching guys if you are watching all the way to the end of the video. You're an absolute legend and I will see you very very soon with some more content. Peace out.
More from Income stream surfers
Get daily recaps from
Income stream surfers
AI-powered summaries delivered to your inbox. Save hours every week while staying fully informed.









