How to turn a Claude Code and Obsidian second brain into a brain for the whole team

The short answer. One person can build a second brain in an afternoon: open Claude Code in a folder of meeting notes, call notes and a CRM export, and have it write clean, linked notes into an Obsidian vault it can search later. A shared folder stops working once the team's Gong calls, Granola notes and Gmail belong in it, and once skills need owners, versions and a way to roll back. The video has Christopher O'Donnell build the solo vault for a fictional company, then connect the folder to Day AI through GTM Brain, Day AI's open-source Claude Code harness, so customer data, agents and skills live in one permissioned graph that RevOps manages from Claude Code.

What the video covers

• 1:31 Claude Code reads a pile of messy files and writes an Obsidian vault: deals linked to reps and buyers, the decisions from internal meetings, and back links that show up in Obsidian's graph view.
• 4:39 Why sharing that folder fails for a team: thousands of calls, meeting notes and inboxes need one secure place with each person's privacy respected, and skills pasted out of a Notion doc can't be updated, versioned or controlled.
• 7:44 With GTM Brain connected to the Day AI workspace, Claude Code turns notes from internal meetings into a rollout plan for a new product: a pilot team, a success threshold, rollback conditions and an owner for each step.
• 9:19 Reps keep the skills and agents they've personalized, while an admin pushes managed skills that a rep can use but can't edit.
• 11:53 Each rollout is saved as a change set, so RevOps can see what changed and when, and roll it back or forward.
• 12:24 Reps hear from their agents in Slack, in email or in the Day AI web app, and give feedback by replying where the message arrived.

Excerpts

[0:00] Right now, everyone's building themselves an AI second brain: a place where all their notes and their processes, everything they know, is in one place that the AI can run around. And they're right to. It's genuinely useful. But almost nobody has taken the leap […]. What about one shared brain across your entire team that every person and every tool feeds into? I'm going to show you how to build both today: the one that you can build in an afternoon using Claude Code and Obsidian, and the one that goes further using what we built at Day AI, […] and lets you and them change selling behavior and automate what everybody's doing without even needing to talk to them. Let me show you.

[0:43] Now, we're going to have some fun with this. Across everything I'm going to show you, we're going to use a fictional example that feels very real. We spent a bunch of time and effort setting up this fake company, Ramble. They make theme park management software. They have a new product for water park management. So I'm playing the role of Althea, who is the director of RevOps here at Ramble. And what I have in front of me is what most people at work have, especially folks in operations. We have all kinds of interesting things here. Now, this needs to be baked into a format, using Claude Code but stored in Obsidian, that's going to be way more useful for the AI.

[1:23] Okay, so we have internal meeting notes. There are a bunch of decisions and action items. We have calls that happened. Here are some notes from a sales rep, Otto. We have Bert's pricing scratchpad, which was pretty good. He Slacked it over to me. We have our CRM deal export here with a bunch of open deals. Okay, great. So let's boil this down and condense it into something actually useful, and there's a pretty cool visual pop to it, too, when we get there.

[1:49] So what I'm going to do is open up Claude Code. I could run it from the terminal. I could do it from the Claude app. I'm going to make sure I'm in my raw material folder, whatever folder I have, and I'm going to prompt it (and we'll share this prompt with you) to go ahead and boil this down into a knowledge graph. Now, why is it so useful to have this in a better, cleaner format? These documents that we're going to be working off of, in any kind of LLM wiki, any kind of knowledge base, any kind of second brain: we want to have our raw inputs and we want to save those, but we want them to be saved in a form that is clean and interconnected, so that future Claude can run around it and get all of the information that it needs to create whatever we want or answer any question that we have.

[2:41] So I go ahead, I punch this prompt in. It's going to grab these raw files, it's going to look at a blank Obsidian vault that I have created there, and it's going to just go ahead and get to business. So we can see here what Claude is doing. It's going through all of my messy raw files, and it's starting to pull out all of the structured information: the deal amounts, the fact that there's a deal over here that's related to this rep and this buyer. It's making sense of everything for me, which is really, really nice. And it's building the vault.

[3:12] Okay, great. The vault is actually ready and I can look at it now. So this is terrific. I told it to document all the decisions we made in internal meetings, and it did a great job of that. Coolest thing here is the graph visualization. Since I told it to identify all of the connections, the back links between these things, it's super easy for me to just hop around from this capacity model that the sales team has. And look, it's beautifully written up, it's all totally polished, totally nice. And then I can see all of the reps here that are connected to it. I can go into this pilot file, I can just keep hopping around. This is like the canonical set of rules that we've decided on at Ramble, all documented really beautifully. People love the fact that you can just see it in this fun graph. But now I want to do something with it. Now I want to bring it to life.

[4:05] So I want to have some way of developing skills here, you know what I mean? I want to be able to think through: if we have the call transcripts and recordings, we can probably come up with a really good way of prepping people for their calls. And wouldn't it be great if that could have a little bit of a flavor of, "Jeez, if you can introduce this new product into the conversation, go ahead and do it," or, "Don't bother on this one, they don't have a water park." That's one problem with multiplayer. The other problem with multiplayer is the data. Here I can see all of the accounts. It's not hard at all to document any information if I'm the only one who's ever going to have to see it.

[4:42] But what about thousands and thousands of Gong calls and Granola meeting notes and everybody's Gmail? My gosh, how are we going to put that in some place? We're not going to put that in a folder and share that folder around between people. That's completely ridiculous. So we want to be able to do a couple of things. We want to solve the data privacy problem, put everybody's data in one place securely, and then we want stuff to happen automatically. We want AI agents to wake up in the middle of the night and start looking for connections between deals.

[5:12] It's easy enough for me to use Claude to just come up with some skill definitions for Claude. But then what do I do? I could put them in a Notion doc and tell everybody to copy and paste them into Claude. That doesn't work super well, and it doesn't give me any kind of control over who has what and what the plan is. And I don't have any way of updating those either. Also, I don't have any control over which of those stay frozen and are controlled entirely by me and are actually controlled by the owner of that agent that is running that skill.

[5:45] If you want the team version of this […], that's what we do at Day AI. It's the shared company brain I'm describing. It's a secure cloud graph your whole team and every tool feeds into. The link's in the description if you want to see it on your own data. And the payoff is where this gets ridiculous. It's the part I still can't quite believe we can do. Stay with me.

[6:07] So you have all of this automation, and it's pretty easy for you to come up with how you want this to work. How do you press deploy and send this all out to the people on your team? Day AI makes this really easy. We have this resource called GTM Brain, and it's just something that you can point your Claude Code to and it will figure out the rest. I can literally just take this URL and go back to my second brain folder in Claude Code and say, "Take a look at this and set me up to be able to deploy agents and skills to my sales reps. Set it all up and leave me with a rig where I can do real change management. You can see that we want to push attach rates of our new product Ripple. Let's make that happen. You tell me how, and let's get it done."

[7:00] So that's literally all I need to do. It can connect the dots between whatever I have started with and what Day AI has in terms of binding in. It will just figure out the right way to blend these things together to take whatever initiatives I have for my sales team and be able to push those out to the sales reps. Get their data securely in one place and just start ripping. So this is what it looks like once Claude has tied in all of that GTM Brain logic, connected to my Day AI workspace, and run around understanding everybody at the company, what they're up to, who they're talking to, and then the change that I want to push out. And the conclusion lives in an initiative file.

[7:44] So here's this initiative. I want to drive Ripple attach rates. And here is the plan for doing that. We're going to pilot it with Jed's team. We are going to declare it a success if Ripple is getting mentioned in greater than 50% of deals. We have rollback conditions. We have responsibility very clear in here for everybody. This is great. This is just all coming out of Granola notes from internal meetings. So now that Day AI is connected into my local repo, my local folder, all of the customer data lives in the cloud. The agents and skills are going to live and run in the cloud. It's all going to be super secure, and I can control it all from Claude Code right here.

[8:27] All I have to do here to kick it off at any point, even brand new, I can just run the start command. So now it's seeing what we have set up in Day AI already in terms of skills and agents. It understands, from reading the documents in the Obsidian vault, what we're trying to do, what our standards are, the way that our sales pipeline runs. Everything about the company is now clear to Claude, and it has a way to reach out into the cloud to look at what skills and agents are running for whom at our company, and potentially make some changes, right? So let's get in there.

[9:02] Okay. Criterion: every seller has a configured agent. Okay, great. Mason does not have an agent yet. Good to know. All right, call prep v1.6. Is everybody using it? Okay, Mason, nah, Mason's in bad shape. Now, let's say Mason's new, right? We haven't deployed this stuff to Mason yet. And it flags that for me immediately. All I've done is run the start command.

[9:22] Because we're working with salespeople who are responsible for their numbers, they have to be able to update the logic, the definition, the behavior of these agents alongside ops. This is a really tricky problem that's just not hard when we do it this way. We're going to keep intact all of the personalization that the reps have added in, custom skills, new agents, whatever it is, but we're going to be able to push new stuff and also make some tweaks along the way. And as I mentioned before, I'm an admin, so I can push out managed skills that are not editable by that seller, by that end user. They're just going to work the way that I've defined, and that's the way it is. I can update them from here, but it is what it is.

[10:03] So right here again, I've only run the start command. It sees our rollout plan, and it says, "Look, why don't we audit this? Why don't we implement it?" I'm feeling a little bit spicy, so why don't we just implement this? And it's really important that both me as the admin and Claude, frankly, following its own instructions in the repo here, in the folder, that we're very thoughtful about this. We want to be able to do big rollouts and big version upgrades to the rig across the company. We want to go through it step by step and develop a lot of confidence that we're doing exactly the right thing.

[10:38] I'm going to go ahead and allow this. What it's doing here is checking everybody's behavior, everybody's ops, matching that up against the instructions that I have in my Obsidian vault, doing a really nice last review before pushing out these new skills and changes to existing skills. Part of this plan we're rolling out is to give Mason an agent, and it's not just going to create some generic agent. It's going to look at everything that's working across the team and build this virtual worker to be just the ideal sidekick to this seller who's new to the team. So it's going to push these out to the cloud, and then there you go. Mason is going to have an agent running, doing all this stuff perfectly. The other agents are going to have little subtle behavior changes that are going to get our new product Ripple introduced into the conversation at the right moments.

[11:27] How do we build those skills? Well, of course, by looking at historical deals and calls and every bit of customer conversation. That's where the Day AI context graph comes in. It makes it super easy and super secure from a data privacy perspective to grab all of the actual primary source material to come up with: what do we even want the AI to do? Be very, very thoughtful about how we take action, roll that out to everybody, and then evaluate it over time. That's all super easy for me to do from Claude Code, just like this.

[11:59] And now we are living large. We have everybody running the current version of the call prep. That's going to include, in the right places, where to mention Ripple and how. We have an agent for Mason, and that's going to be running absolutely state of the art. We're in really, really good shape, and it recorded a change set. This is a really critical thing, and Andrej Karpathy talks about this a lot. You want to log all the changes automatically here, so that you know for posterity exactly what you changed when. You can roll things back, you can roll things forward, and there's never any mystery as to why anything is happening in a particular way.

[12:35] Now, let's take a look at what this looks like for the actual seller. They're going to hear from their agents in Slack, and they can just respond right there: give feedback, ask follow-up questions, move things along however they want. Or over email, or in the actual web UI for Day AI. Our web app becomes a really clean, elegant work surface for the rep to just go about their business that day, get a ton done. It's really clean, it's elegant […].

[13:09] So now you see what happens when you take a second brain, and you bring it to the whole team and elevate it to the level of being a go-to-market brain, a sales brain, a company brain. You're able to drive this behavior off of way more data. You're able to make much better decisions. You're able to save a ton of time and energy from the people that you're paying to actually sell your product. You're learning and compounding every single day that anything happens. The whole system is getting better, and you're able to modify and mold the behavior of your team and how it is bringing your product to market in ways that are very helpful and unobtrusive to the people doing this hard work day in, day out, every day, face-to-face with customers.

[13:49] If you want to know more, we have free courses, and you can also get on the phone, get a demo from us, whatever you want. We're here for you. Let us know in the comments what you think. Give a like and subscribe if you're so inclined, and we'll see you at the next one.