Transcription
Cloud Code and Cloud Co-work are now rapidly changing the way that people are getting work done. But with these new tools, we're now bumping our heads against new problems. And there's a free tool that's called Obsidian that really is the answer to most of these problems.
Let's start with the problem that AI practitioners have been struggling with since day one, which is memory. Getting the AI to remember different things about whatever project that you're working on and not confuse those things with other projects. AI companies have tried many different things to fix this problem. But now with the combination of these two tools, we have direct and complete control over what the AI sees anytime we open a new session.
If you don't have the desktop app for Claude, you can grab that at claude.com/download. And when you fire that thing up, you've got access to chat, which is the normal Claude, and then you've got co-work, which is what we're going to focus a lot on today. And also access to code here, which we'll touch upon. But inside co-work, the beauty here is that you can pick a folder that you want to work in. And for today's example, we're going to look into how to build our own open-claw advanced AI agent.
So, I've created a folder on my desktop that just says advanced AI agent. I'm going to open that folder here. Desktop advanced AI agent. Open. Allow. Similarly, in Obsidian, I'm going to go to open a vault. That's the same as just opening a file. And I'm opening that same folder.
Now obviously when we get to building the software we're going to use Claude code but right now for our research phase I'm just saying I'm interested in learning how to create advanced AI agents like open-claw. What's the best process here? And after some substantial back and forth, it has decided to create this architecture document and now we can see that right here in Obsidian.
So since this is a markdown file, it's really hard to open this up in other pieces of software. Your text editor doesn't do it. You know, word documents don't particularly like these types of documents. So, simply using Obsidian as a viewer to these markdown files is just huge. You can pop in here and give a look at everything that it created and making sure that it's all accurate. And this is the part that you really need to work on. This is the manual work I'm convinced that will bring you to the next level of AI. Making sure that everything in here aligns with your vision. Because in my experience, 95% of it is and then 5% is off. And you, if you don't look at these documents [snorts] after a few iterations, you know, your whole project is going to start to drift in a way that's going to be really hard to rein back in.
But now I want to jump into the cheat sheet. I make one of these for every single video that I create. There's now over 176 of these immediately available to anybody who joins my Patreon. There is a link in the description here. This is going to include all the prompts we're going through today and a bunch of extra resources related to using Obsidian along with Claude.
And this is a prompt that I use all the time to summarize the session into a session log. We're going to copy and paste that right out of the cheat sheet into Claude. And this will begin to create our session log, which we'll create each day as we're working and moving forward. And this is really especially important when it comes to compaction. So, I don't know if you've been working with Claude code and ever had it do this compaction process where it squeezes everything you've been talking about into, um, you know, a little summary that's in its knowledge. Now, if that happens at the wrong time, you can lose a lot of work and be, you know, really set back. That is one of the major problems with these new workflows. But keeping these session logs and taking these different snapshots at different times stores all of that accurately into these different files, which you can then double-check here inside of Obsidian.
So, it has now created this session folder and a session log for what we talked about today, and that's in addition to this architecture. So now we're starting to create the little bit of a file structure here inside of our folder. And this is something that coders have understood the importance of since day one. There's no way to create successful code without organized file structure. And now I think knowledge workers and regular non-coders are being forced to kind of learn the value of highly structured folders because we're going to get a lot more out of these AI systems if we're able to give them exactly what they want and exactly what they need at the right times.
Here's a cool prompt in here for cross-project memory. So, reading the last 10 sessions from different projects and comparing and contrasting how you're working with the AI and maybe look for improvements of how you can streamline your sessions.
All right. So now that Claude Co-work is remembering everything accurately, and you can see that clearly by accessing these files through Obsidian, I want to move deeper into that idea of the structure here because many times these AI models will create structures that are not good, and that can lead to so many different problems down the road. If you've ever tried to create a second brain using Notion or something like that, you know how important these structures are, and if they start to get out of whack, that these things start to erode and they become impossible to use. But if you work together, Claude can be really helpful in helping you organize all of this stuff clearly.
So I'm saying, given that this project will require a codebase along with many other support documents outside of the codebase, what might be the best way to structure this project? And Zeph is really just what I am calling my new advanced AI agent. All right. And again, after some back and forth, we've got a pretty solid structure here. So, I'm just saying, "Great, can you create these folders?"
All right. So, now back in Obsidian, we can see clearly all of these different folders and click into them the same way we would access them inside of Notion. So, we can easily spot things that might be off or rearrange these as we will definitely have to do as this project grows and matures.
That brings us right to the next section here. I see people all the time, you know, trying to remember where their prompts are and copying their prompts in. They have, you know, some sort of prompt that worked, or maybe it's a prompt sequence that worked, and they're copying those in, or they're trying to remember them, and there might be a typo there. They're not getting consistent results, and they're most importantly not improving on those processes each time they do it.
So, one of the files, probably the most important file that you can have in any of your different documents here, is a process file that tells Claude exactly how you want to execute something. So, like I said, when I'm doing research for these videos, there is a process file that lives inside of the folder that in clear English states all the different steps and then clearly points to different software scripts that are needed at each step. And so then I can just open that folder inside of Claude and say, "Let's work through this topic generation process." So that's helpful to just get that going. But most importantly is I can then continuously improve that. So each time it might be doing kind of the same thing, making the same error. I can say, "Okay, we've done this before. Let's update those instructions to make sure that we never hit this error again."
And this is really where it's important to jump in and edit those different files inside of Obsidian so that they match exactly what you're trying to say and they're not overcomplicated by the AI being verbose or a little bit out of sync. Of course, you can go back and forth with the AI to make sure that it's correct, but for little things, I found it is so much easier to jump in with Obsidian and make just tiny little edits [snorts] or just little additions than trying to get the AI to just put it exactly perfectly. So that's the thing where the AI is getting you 95% of the way to these process documents, and then inside of Obsidian, bringing it that next 5%, you're going to start to get not only a great powerful document but a process for continuously improving that each time I'm running these processes, getting a little bit better, a little bit better, and my results are beginning to compound.
And now on to the next problem that people are still running into, and I hate to see it, is they have to go to their different pieces of software and grab things and pull them in. And you don't need to do that anymore for most things. You can start to build and explore these different slash commands and embed them into your different process documents. So when you hit slash, you can get in here and start exploring these different plugins or more importantly, creating your own plugins here by building them out in Claude code using just simple language saying, "Hey, I want to, you know, scrape Twitter for the latest AI thing. What do I need to do to be able to do that?" And from there, you can start to get really fancy with these where you're not only going through these different prompt sequences, but accessing all different pieces of data in a truly agentic format, getting them to control your computer and go to different browsers and look at your Salesforce and all these different things.
This cheat sheet has four different operational prompts that you might want to take a look at. For example, there's one called the dashboard builder. There's one called the client pulse if you're doing any sort of client work. There's the experiment tracker. You know, it's a big part of AI is experimenting, but we want to make sure we're keeping track and experimenting in the right direction there with that experiment tracker. Weekly ops review. These are all copy-and-paste ready to just begin to fuel your workflows in any direction. You can obviously customize these to whatever you're working on.
One of the big things you might be asking yourself, if you use Notion, can't we just use Notion for all this? And I am a heavy Notion user, and I still prefer Obsidian for my knowledge base for Claude. And that's because it lives right on the desktop, and it's actually Claude is in control creating and editing these files right there on the desktop that I can access and view inside of Obsidian. When it has to ping and update Notion, there is just quite a lot of friction to updating those files. It never quite works the right way.
Okay. So, in order to simplify that and rapidly speed up your workflow, Obsidian just living right here on your computer viewing those files that are on your computer, you can then, you know, control those. So, if you decide, you know what, I want to try to use Notebook LM to create a video around this, you can just easily grab your files, paste those right into Notebook LM, and create a video from all that. You have complete control of the files, which is a brand new thing because previously using these models, we would, you know, not really have direct access to updating them in Claude projects or in custom GPTs. You know, they were kind of baked in to the back end of it, and it was, it was kind of a hassle to update these different process files and knowledge files. But now we have the ability to control them directly right here on our desktop.
That's a big thing related to the fifth problem here is when all that stuff was stored in all these different areas. Our work did not compound, and you know, we were really, there was a lot of friction moving that stuff around. But with this workflow, we have the ability to continuously progress our knowledge, to keep it clean, to keep it structured, and to spend some time on that because that is the job now where AI is doing all of the writing and the creating and the researching. We need to make sure that it has what it needs at the right time to continuously move as fast as possible.
I even go so far as to create one main projects folder, and then inside that folder at the very root level is a Claude MD file that has exactly how I want all of the work to be done across all of my different projects so that I don't have to run into these same problems over and over again. We can just continuously update exactly how I want to work, and you know, it can continuously understand a little bit more and more about, you know, these global preferences that way. And this has become a game-changer because I don't have to tell that to each different project, let alone each different chat, every time that I fire it up.
This cheat sheet goes way more in-depth into all of this stuff, including pointing Claude at maybe other places where you have this context and pulling that in to update this system that you now own that you can continuously grow and use over time. So, definitely check this cheat sheet out. There's a link in the description. I also have some great group coaching options in there where you can jump on calls with me, ask me questions directly, collaborate on whatever you're working on. There's a bunch of smart people in that group. So, I love that. Highly recommend it.
And the next video I want you to check out is all about Opus 4.6. So, Opus and Sonnet, these new models have started to unleash some really new capabilities that weren't possible, you know, just a few weeks ago. So, that's the next step on your AI journey. I'll see you over there.