307 | Stop Starting From Scratch: Build AI Projects That Actually Remember with Loren Bartley

21:10And then once I answer them, it's going to then spin everything up for me. So here it's saying, I'm Commodore, here's what I do plainly. I set up work inside Claude so it runs properly. I can stand up brand new projects from scratch. You know, its folder, its instructions, the right knowledge, a tracker and its command center. Or if the project already exists, just brief a command center to run it. So I set it up that way because retrospectively, I had all these projects and I was like, if I just started a project without running it first, I can still use this skill. And then it just explains what it is.

21:39So it's the command center that we're going to be setting up as the first step is one chat that keeps the key documents, holds a single tracker. So you all know, always know where things are up to, decides what to do next and brief separate chats to do the actual building. It coordinates, it does not do the building itself. And that separation is the whole trick. So it's saying we're sitting in your Singapore trick project. So let's start with one choice that shapes it. So it's asking me, is it a new project or that I want to start from scratch or do I just want to brief a command center? So in this case, I'm going to start from scratch and now it'll be able to go off and start.

22:12It knows what it needs to do. So it needs to do all those things that I just mentioned. And, you know, like anything, it takes a little bit of time, but now it's saying, great, now it's going to set it up. It's asking what it is. So is this a personal trip, a work, client work? I'm going to say it's client work. And then, yeah, it'll come back and ask me a few different things. I can also, if I want to, just dump a whole heap of information in there. So because, and this is one of the great things about co-work, is that co-work, you can interrupt it and just keep adding stuff while it's still thinking as well.

22:44So this is saying, quick note, oh, I don't know why that's not working, but let's not worry about that too much. I'm going to put some more information in here. And just to make it easy for everyone else, I kind of typed this before, but what I'm saying is I've basically said, I'm going to Singapore, the dates, the workshops, for. And I've said that I love to do park run and I'm going to be across two weekends. And the weekend, the first weekend, there's actually a special, it's Singapore day. So there's a special park run on. And so I'm hoping to do two park runs in the first weekend and one in the second weekend.

23:15And the other, and then I've told it which park runs I've already done. And I'm saying who I'd like to catch up with while I'm over there. So that's all the instructions that I'm basically going to give it. And then I'll kick that off. So this is more than what it actually asked for. But while this is thinking, what I do want to do is before I said there was the tracker that we have, and I showed you the static tracker that lives inside the chat. The other thing I wanted to show you, and I do have it open. Cool. I'll just make this a bit bigger while that's thinking, is what we can also do by integrating tools.

23:46And so I use Notion a lot. And so you saw that project before, and where was it before? This is the project that we were looking at before, and that was an older version of the tracker. But these are the things that are still going on. So what I get is written into all of my instructions that Commodore puts into the command center. It used to be to spin me up that track, but now I get it to write it to Notion. So this is the great thing. We can use the tools we already use. We don't have to jump. And having one static tracker in each of the chats becomes not as manageable as having them in here.

24:25And I've got the view at the moment that's incomplete, but then we've got, you know, the other views that will also show me what's done, what's ready, and all of this across each of the projects. Now, I will add, if you look at this and go, how do you do this? I have no idea. Claude did it for me. Claude did it for me.

24:41Isar Meitis:So I want to pause for just one second to touch on a few important points that you're mentioning. One is Claude Cowork, and Claude Code for that matter, knows how to use external tools. What are these external tools? any other piece of software that you're using in your company, your ERP or CRM, your marketing platform, your email solution, your accounting set, like whatever it is. It can use it in multiple different ways. It could be through a direct API. It could use third-party connectors like NA10 and Zapier. It can use an MCP if it has an MCP, or it can even use the browser just like you use the browser and do whatever you do in the tool, quote unquote, manually.

25:20Isar Meitis:So there's multiple ways it can connect. As Lauren said, you do not need to know how to connect it. All you have to do is say, I want to track this in whatever. I do something very similar to what Lauren is doing. I do it in ClickUp because I use ClickUp and not Notion, but you can do this in Monday.com, in Asana, in whatever other Jira, whatever other task management platform you're using, and it will figure out how to connect. It may ask you for some assistance. Give me this key. You can get the key by going to this page, logging onto this, clicking on that. It will tell you exactly what you need to do and then it can set up the screen for you.

25:57The broader idea behind this

26:00Isar Meitis:is human in the loop, right? You don't want the AI just running rogue, doing its thing. You want to know what it's doing and be making the decisions in the main decision points. And you can do this inside each of the chats, but very quickly, and again, you can see Lauren is doing this, I'm doing this, a lot of other people doing this and we did not exchange notes before. It just makes a lot more sense to use the tools you're already using to run your business, to also run your agents and the stuff that's happening inside the agentic environment. And so it becomes as if you're working with other humans, the way you work with other people in your organization.

26:37Isar Meitis:In my universe, and I assume the same thing for Lauren, inside each of those tasks, there could be a chat associated with that task. So when I want to communicate with the AI about a specific task, I don't go back to Claude. I stay in ClickUp and I open the task and I chat inside the task as if I have human people that are watching this and it's going to communicate back through the chat. And again, how did I set it up? Just like Lauren said, I don't have a freaking clue. I asked Claude to set it up and it did. And I can communicate with it just like it's a human. So a few really, really brilliant points here, right?

27:11Isar Meitis:So one is manage the process. Two is you don't have to manage it inside of Claude or ChatGPT or whatever. You can manage it in the tool in which you're managing the entire work that you're doing in the same way you're managing the rest of the work. So you can give Claude access to your existing board and say, this is the way I want to manage your projects as well. And then you will learn exactly how it's set up and you will set it up the same way for that. Let's continue from here. Brilliant so far. Yes. Yeah. And as you can see, and for those that can't see, this basically mimics what was in the spreadsheet that, sorry, the static tracker that I showed before.

27:44So it's got the same traffic light system It's got, you know, what's happening. If I click into these, it's going to give me more details as to, it may or may not actually give me more details as to what's going on. And then there's a whole heap of other fields in terms of data added, all the other stuff that's relevant. And it also tells me which chats this comes from as well. So if I see this task and I, oh yeah, and I want to return to that chat, I can do that really easily. And then if I change the status in here or anything else, as you mentioned, this is where Claude's reading from. So it's not going to come back later and tell me that I still need to do a job, that I didn't tell it in Claude, but I updated the status here to say it's done, it's going to automatically go, okay, I know that's done without me necessarily having to do that, which is super cool.

28:24Okay. So we've come back here and it's asked me a few questions. Is the year -

28:28Isar Meitis:When we're saying come back here again, for those who are listening, we left Notion and we're back inside of Claude setting up a new project. Yeah. And it just asked me a simple question to ask if it's this year or next year. So it's shaping up nicely. And I will also mention, you mentioned before about in the left-hand column, we get these little flashing symbols that go on. So this is a visual representation from those that are watching of just the actual chat going on on the left. We know it's working. And often I'll have multiple of these going on that will then, and as you said, just jumping between them as necessary.

29:00So we can see that it's thinking at the moment. And interestingly, on the left-hand side as well, you'll see I've got one that's got a little jigsaw puzzle instead of the men at work kind of emoji. I said before I used different emojis, This one tells me that I'm working on a skill. So I did update Commodore the other day and I went in there. But you can update your skills in any chat. But I often like to go back to the original chat if I can to keep it. That's just, it works better for me. It doesn't necessarily work better for Claude. But for my logic side of things, it definitely does. And on the left-hand side as well, I've got some of my chats have a little pink brain.

29:34and the pink brain emoji is the emoji that I use for my command center chats. So I know that that's my command center chat that I go to, to assign work to the other chats. And that just works for my logic. And then also what something that I know a lot of people don't realize in the left-hand corner, when you click on any of your chats and click on the three dots, you can rename it, but you can also pin it. And so I pin those chats that I pin my command center chats. And if there's another chat that I'm working on a lot, so you'll see here, take control of your AI. That's a presentation I'm doing on Friday because I'm really active in that at the moment.

30:16And I probably don't need to pin it because of the fact that I am so active and it sits at the top of my chats usually anyway, but I just pin it there as a reminder to me, Hey Lauren, that's what you should be working on. So that's just a few other little hacks that can be really good to help organize yourself. And then when I clicked on those three dots alongside a chat, the other thing it does allow you to do is to change the project. So some people, sometimes I'll be working on something and then go, actually, that really should be in a different project. And you can move it, which becomes handy because you can move things backwards and forwards as well.

30:48So that becomes handy. The conference one I was talking about previously, I was building out a brochure to use at the stand. And I was doing it all within the conference chat. And then when I finished it, I thought, you know what, because it's done now. And I thought, but this would be really handy context to be inside the branding that, because I'm doing a rebrand at the moment with my AR Mega Success brand. And so I took that because it was basically an AR Mega Success brochure that I created and I moved it across into the ARMegaSuccess rebrand project because I wanted that context saved in there for later.

31:27So this is some of the flexibility that you've got with this and quite often. So I always go to a project first and then start the chat within the project. But sometimes you just go straight up here on the left-hand side, you click that new task thing and then you start working and then you go, oh, actually this should have been inside a project. So I just wanted to give people the heads up that you haven't broken anything, you can just move it when it suits you.

31:51Isar Meitis:Fantastic. One last quick question. How do you pick which files to connect? Do you use AI to help you pick the files or define what needs to be imported? Or do you have a system for that? Because again, so far we spoke about creating the actual instructions, which I assume that's what it's doing right now when it's running the thing after asking you all the questions. While it's doing its thing and we're waiting for the instructions, let's talk about the files themselves. Like which files do you upload based on what? Do you have a system for that as well? Yeah. So I have a folder which I call Impactivate Knowledge.

32:28Impactivate's my company brand. And so I have in there my brand voice, my ideal customer profile, like a whole heap of things that are relevant to that. That gets put into most of my personal, so my business work, but not all of it. So for example, in the client's side of things, I'll put their brand voice documents and those types of things in there. But then I have folders specific to each project where I put in the relevant information. So for this one, so I have a travel folder, as you saw, I've got other folders, like I have a branding, the branding folder. So they really become specific to the project.

33:02And that again, goes back to that context window because we don't want to overstuff it with too much stuff that might be irrelevant or contradictory information as well. So that's kind of how I work on that. But I can see this has come back now and it's given me quite a few documents. So the first thing that it's given me is my project instructions. And if I click on that in the center panel, it pops up in the right hand side with what those instructions are. And as you can see, it's quite a bit considering I didn't give it very much at all. Like two to three pages

33:31Isar Meitis:of instruction. Yeah. And the reason for that is because I've created a template and said, these are the things that kind of needs to go into the template, knowing what's good. So basically, we get a project snapshot. We get this is, and then it'll say what type of work. So in this case, this is client work, the knowledge files and resources of truth. So it's telling me which folders and files need to go into that, the confirmed facts, because that's what I gave it already and how this project works. Then it's got the save location. And I've set this up. And this is an important thing is that I have this other rule that I do, which is versioning.

34:08So what happens is whenever my chats spin up any form of document, I ask for it to be in markdown format. And then I also ask it to save it into the project folder. But then each project folder gets an archive subfolder. And when it makes a new version of it, it moves the previous one into the archive version. So the main folder for the project only ever has the most recent version, but I don't ditch the old versions. I don't override the old versions because sometimes we go, actually, I'm pretty sure like we used to do it differently. And I can go back through those or get Claude to go back through them, find the bit that I want to, that we might've, you know, deleted or changed or whatever, and then reinstated if I want to later.

34:50So that's something else that I do there. It gives instructions as to how the command center is to work. And then it puts some project guard rails. And then if there's any overrides and these overrides are usually things that might override, let's say the system wide instructions or anything like that, it'll list them there as well. So now that I've got this, I can just basically copy these instructions and it up here in this middle section, it's telling me what I need to do. So it's telling me that I need to rename the chat and that's how I get my little brain emoji in there. So it's saying rename the chat to brain emoji Singapore trip V1.

35:22And I'll tell you why V1 is important in a sec. And then it's saying, put in place in the app, create a project called Singapore. I've already done that. So that's fine. I did that at the beginning, but it's reminding me in case I didn't. And then I've got an issue for some reason, Notion's not tracking properly. It could be because I upgraded the account last night. And then it's asking me what I need to do. It's telling me what I need to do next. So interestingly, it didn't tell me, it usually also tells me to go and update the instructions, but I'm going to go and do that now. So basically, because I'm working in a chat that's inside the project, I just have to click on the name of the project, which is in the top left of the middle panel.

35:58And I click on that, which says Singapore trip. I click on the right-hand side where it says instructions, and I just paste in those instructions and I go save. So that's basically saved the instructions for me. I can go back into that chat that I was working on, and now I can do what it's asking me to do, which is to rename this chat into my command center. And the reason it says V1 is because this is the version one of my command center chat. When I feel that it is getting to the end of its lifespan, I usually just say, hey, we've been chatting for a while because it doesn't tell you in co-work whether or not you've hit your context window.

36:35There is a way you can do that in code, but not in co-work, not that I've found anyway. And then I just say, do you think if we're getting towards at the end of your context window, it'll usually honestly say, yeah, I think I have. It's been great working with you. And I'll say, can you spin up a new brief? And it will do a new brief that I then put into a new chat. And then I basically take the emoji off the beginning of the name of this one. And then I rename the next one Singapore trip V2. And then I work out of V2 moving forward, but none of the other stuff is lost. I've just working with someone who's had a nap or had their coffee instead.

37:09So that's, so I'm not quite finished yet because there's a couple more things. So I've done the instructions. So they were the instructions. It spun me up the tracker straight away because that's what it's instructed to do. If Notion was working, and for some reason, I'm not sure why it's not at the moment, but it would have then gone and created the Notion version as well. And then it's given me the command center briefing, which is here. Now what I do, and it may or may not be necessary to do this, but if I wanted to do a different chat, I would take this into the different chat and say, okay, give it this brief and then say, you're now the command center.

37:42But I just do this. I just basically copy that because I'm going to use this chat as the command center. And then I'm just going to paste that into here. But before I do that, let's just look at the other documents that they've got. So it's given me the command center briefing. It's given me also my parkrun schedule. It's just gone and done that straight away as to what I could potentially be doing for parkrun. And for those that aren't aware, parkrun is a five kilometer run that happens Saturday all around the world. And I'm a bit obsessed with it. And even though I'm not a runner and I do it without fail, I love the tourism side of it.

38:10And then it's given me a command center checklist as well. So all of this has happened. This is now my command center. So I'm just going to basically brief it to say you're now the command center. It's going to know what it needs to do. And then the next thing I can do in here is I can, so it's basically telling what's going on here. So it's saying the command center's active. These are the open decisions that I need to make. It's ready to run now. And okay. So what I can do now, it's basically telling me where we're at and then it's saying flights. So I can now say, can you create me a brief that I can put into another chat to organise my flights?

38:48And then I'll have a second brief that can then go and research some accommodation. I'd like to be somewhere around Orchard Road. And then also, if you can draft me up an email, because I want to let Vicky, Koucha, Melissa and Jodie know that I'm going to be in town. And if you can write them as drafts and just have them sitting in my draft folder, that would be great. That's in a separate brief. So now that's me just verbal diarering this into Claude. And Claude's now going to go and create those multiple briefs for me. And then I'm basically going to go and cut and paste them into other chats within this project and then let them all work together to do the things that they need to do.

39:36And as you would have seen, one of the tasks I gave it was to write emails to certain people. And then, and because I have my Gmail account connected, it's going to write them and put them in as drafts, which is kind of super cool.

39:50Isar Meitis:Awesome. I want to do a quick recap for those who are somewhat lost because you covered a lot. So it all starts with a skill. A skill is a set of instructions that tell AI how to do something specific. In this case, it does several different things. But one of the things that it does is it knows how to set up a new project. So when Lauren said, hey, run Commodore, that's the skill. The skill is called Commodore. And Claude says, oh, what do you want to do? Because there's several different things it can do. It can either update an existing project or create a brand new project. Let's create a brand new project.

40:30Isar Meitis:So the skill itself contains a whole set of instructions on what needs to happen in order to set up a new project properly. And that includes the task setup and the command center and the components that needs to go into it and ask a bunch of questions to understand what is this project about. And it does all these things on its own. So this is step one. One of the outputs that it creates after it asks her all the questions is additional prompts. She calls them briefs, right? But it's just another prompt that you can copy and paste into a new chat inside the project to achieve different things.

41:08Isar Meitis:From that moment on, it's a snowball effect because every one of these briefs can create more options that can create more briefs, which again are just new prompts that they will do more things. But the cool thing is from like the second conversation and on, because the project is already set in a proper way with the right context and the right instructions and the right reference files and so on, she can just open the microphone and say whatever she wants to say. And it will happen because the The AI already has the scaffolding and the context in order to do all these things that needs to happen.

41:49Isar Meitis:The other cool thing that you saw, and these are two things that I do very much the same. One is stop typing. Like all these tools are really, really, really good with voice and you can speak way faster than you can type, even if you type very fast. For Lauren and me, it's very obvious we're very good speakers because that's what we do for a living. We speak on stages and we have podcasts and stuff like that. But for any person, it's easier to explain yourself in just speaking out loud. And you don't have to be exact and you don't have to be perfect. And the sentences could be horrible from a grammar perspective.

42:25Isar Meitis:And the AI will be perfectly fine in figuring it out. So this is one thing. The other thing that you saw is that you can give it multiple tasks in one prompt. And because the scaffolding in there and the instructions is there and the reference information is there, it will figure it out. So a great example was the email. She didn't say who these people are or what their email addresses are. She gave names. However, the project knows she's going to Singapore. It knows it's connected to a specific client project because that's how we got started. She said it's a client project. It understands who the clients are.

42:59Isar Meitis:So it understands who the people are in context to these clients. This cannot happen in the open chat. If you just go and click on chat and say, hey, literally the same exact prompt. I won't just anywhere else, so on, so on, so on. And like, it will not know what to do because it don't know who these people are. It doesn't know they're in Singapore. It doesn't know she's going to Singapore. Like all of that information that it gets from the project and from the skill setting of the project, it doesn't have. And this is why this is so magical. Because in 10 minutes, and the reality is most of those 10 minutes was Lauren explaining what she's doing.

43:31Isar Meitis:She can probably do it in three minutes. she has a setup of a new project and from that moment on she can speak on very broad terms and the ai will still figure out what to do and will go and execute multiple tasks in parallel again in this particular case it figured out the flights it figured out the accommodation i mean it didn't figure it out but it came up with options for flights options for accommodation and a draft email by her saying well it hasn't actually done that it hasn't done that it's done the brief for that. Oh, correct, correct, correct, correct. Yes. It creates the prompts. Yes.

44:02Yes. So can you give me one sec to just now show what happens here? So basically I can go and say show in Finder if I wanted to, and you can see now that it's popped it into this folder that I've got here. And I mentioned before I have an archive. So this is the archive stuff that would have been put across into here. But now what I can actually do is I can go back to my Singapore project now, and then I can just take that this here, this fold, this thing that it's done, which is the Singapore trip catch up emails. And I can just say run brief. And then that one's kicked off. Then I can go back to my command center and I know which one it is because it's got the little face there.

44:40And again, I can go show and find her. I could have just scrolled through and found the next one, but I'm just going to do this to make sure I've got the right one happening. Pull that back up. Oops, where have I gone? I'm going back to my Singapore trip again. In a new chat again, Then I'm going to pull that one in. And again, I'm going to say, I'm going to say run brief. And then it's going to run that brief. And then I go back to my command center again. You can see two of them are working. And I find that third one, which is the catch-up emails. I'm going to go back to here into my main project.

45:10And I'm going to put the catch-up emails in there again. And I'll say run brief. And then, oh, run brave, it says. This is Whisper Flow I'm using in case anyone's wondering. And now in the left-hand side, we can see I've got three things happening at the one time. Now, I didn't even read those briefs because I was pretty confident now that it's going to do the right thing. I could have read them. I could have edited them. And I could have come back now and finessed it further. I could have given feedback to the command center and the command center could have rewritten the briefs if need be. But most of the time, I don't because I've got so much confidence in it now that it's going to do it right.

45:45And then in this chat, if it goes a bit haywire, then that's when I'll just re-brief it as part of the chat and say, oh, no, actually this and give it more context. And you can see this one's popped up as blue, the Singapore catch-up brief. And then it's telling me that those emails are already sitting in my outbox ready to catch up with those people. So that's one of the really cool things about this. Now I can go into my drafts and there'll be a draft there saying, hey, it's Lauren here. We're coming to Singapore. Let's catch up. And then I can finesse a little bit. Obviously, if I'd given them more context at the beginning, you know, I could have said, you know, tell Jodie that I want to meet her at Parkrun on Saturday, like any of those things, and it would have done a better job.

46:21But that was just for demonstration purposes, showing you at the really high level what it can do rather than the quality of what it can do. So that really kind of does the full circle on how to use this. And if anyone's asking, like, how do you set up a Commodore? Like, that was just me taking, like with any skill I create, taking what I've been, once I'd done the thing and I'd set it up the way that I liked it. And then just basically saying to Claude, okay, can we package that as a skill now so that I can repeat this over and over? That's the magic.

46:57Isar Meitis:Absolutely brilliant. so again this is a whole different level of creating and using projects and what you see is that the concepts are the following one is you don't do the work claude does the work two you manage the work you feedback the work you sanction the work so there are steps in which you as the human is heavily involved to push it in the direction you want, but the actual work is done by Claude. The last thing is you can do many of these in parallel, and as long as you're watching them and seeing what they're doing, then you can do that as well. And all of that is managed by skills.

47:43Isar Meitis:And skills are just the same kind of thing as a project. It's a bunch of instructions with some context that it knows how to repeatedly do the same thing time and time again. In this particular case, the skills knows how to set up a new project and how to set up the command center so it can spit out briefs that can do more or less anything else you want inside that project. Lauren, this was absolutely fantastic. If people want to follow you, learn from you, connect with you, meet you for a park run somewhere around the world, the best ways to do that. Oh, that park run one would be the best one, but failing that, If they go to AIMegasuccess.com, that's probably where they can find out and connect with me best.

48:25And then if anyone actually wants, I've created a generic version of Commodore. So if they go to AIMegasuccess.com forward slash Commodore, then they can download Commodore as a skill. It is generic. It'll set up the trackers. It won't connect you to Notion or Asana or ClickUp or any of those things, but you can build on it. So you can use that as your base and then say, okay, and now let's have this connect to monday.com. and it will do that. Or I prefer this in my instructions and just keep updating that skill until it works right for you. And then you're off to the races. So that'll be a good starting point rather than necessarily starting from scratch if it was a little bit overwhelming for anyone.

49:02Isar Meitis:Awesome, awesome. So thank you so much. Again, absolutely fantastic. You're already on the 7th of July, which is the very last day to use your Fable 5 credits. Anthropic takes them away, so I will not take more of your time so you can go and do that. I really appreciate you. I appreciate your time. And thank you so much for sharing with us. Thanks. It's been a pleasure.