Founder’s Guide to AI: Using AI to Increase Productivity

Hey everyone, I hope you're having an awesome day. My name is Daniel. I'm a part of the Founder Institute marketing team, and we're back again with what is the third session of our AI for Founders workshop series. And like always, we have our incredible speaker here with us today, Trevor Longino. Trevor is the founder of Crowdcamers, an organization that helps startups launch. He's helped a bunch of our startups He is a great friend of FYI, so it's always great to have him here. Before we kind of hand things over to Trevor, please feel free to let me know where you're all joining in from. I see people here in Canada, in the US, Colombia, Austin, awesome, Greece, South Africa. So a very global audience. What's really cool is we've had a ton of interest in these workshop series. So what I believe right now, we are on our third workshop and we've crossed over 3,000 total signups, which is really, really cool. And since we're talking about AI, it makes perfect sense. Folks, if you have a question for Trevor, please feel free to share it in the chat. It should be on the right-hand side of your screen. What I'll do is towards the end of the workshop, I'll get to your questions and Trevor will answer. And also, if you are enjoying what Trevor is saying, please feel free to use the emojis right at the bottom side, bottom of your screen, and show Trevor some love. Uh, Trevor, would you want to kick things off?

I'm ready when you are. Awesome. All right, so I'm going to start by showing you a little bit of a version of what I do to prepare myself every morning for work. And let me actually share my screen here to give you just a bit of a— I'm not sure how I add this to the, to the main feed. But what I do in the mornings, let me go back over here, is start by asking myself, all right, what is it that I need to do every day? And I think a big part of productivity isn't doing more work, although of course AI does enable you to do more work. I would note, Daniel, in our attempt to be clever and make it where it was easy for me to share my screen, we have now broken the screen share. Can you guys find a way to unlock it so it's working? I'll go through my discussion while I'm waiting here. Good productivity isn't doing more work. It's a little bit doing more work, but a lot of it is doing the right work. And I think there's no shortage of tooling that will help you do more tasks, track more tasks, accomplish more things. That part is not super hard to do. What is hard to do is to get it to where it is going to effectively steer what it is you're doing and do the right things. So let's talk about what you can do and the impossible things you can do. Now, what I like to do— but what could I be doing, right? Not just what am I doing, but what could I be doing. And normally I do this through speech-to-text. I'm actually on my treadmill here in the mornings And I'll have my hands in my pockets. I'm going to be looking up and thinking about what did I need to accomplish. Now, speech-to-text doesn't share very well over video. And anyway, you're not going to want to listen to the AI read something to you for the next few moments. So I'm going to show you how this works to give you an idea what happens when you start to use AI to help with your day. There's the first thing here. I have got AI hooked up here. This is running with Claude Code. And running just in Windows in the terminal to pull together and look at throughout my entire calendar, my tasks in Notion, anywhere I have deliverables here. This is going to pull together— let me scroll up a touch here. Like, this is today's data, right? For what I have today, you'll notice, uh, right here, using AI to increase productivity, right? That's this one. And what this will start my day with is pull and say, hey, here are all of the things that you need to do today. And it'll also look at tasks and say, are there deliverables that I should be working on? And I'll note there are a couple things here I have deliberately kind of put in my Notion DB as way past due, although I could of course have things due today as well. It's going to call out and say, hey, here are tasks that you need to finish. And then from there, I can actually have it ask a few questions, right, for a daily note, for a log in Obsidian. What's my intention today? I want to nail my Founder Institute demo on AI productivity. I would note, usually, I'd be talking to the computer to make this happen. I tweak this because again, it takes like 5 minutes to read through all that while I'm walking, which is fine if I'm on a treadmill getting my steps in, but pretty boring for you. Then what system do I work on? I'm going to improve my morning planning with AI to make it more robust and share data with my team. And then what skill I work on? I work on AI coding. Coding, of course, if I type better. Coding and Thoughtfulness. There you go. Now it's taken this, and if I go look at Obsidian here, I'm going to see in my daily journal Let me find my daily notes here. There's going to be, uh, let me find the actual link for where it is pulled together here. Well, file notes. Aha. So I, I will have what we just worked through here published into my daily notes. Let me, uh, this one. There we go. And it's pulled together for me here, everything I need to know what's due. I can then take this and, as I frequently do in my company, I'll grab this and I will share this in Slack, or I'll put this into Notion for my daily scrums, right? This kind of information is shared for other people in my company to understand what I'm doing. Now, this is just the first bit of what you need to do. And there's a lot of pieces moving around here I'm going to talk about in a moment. Because I can publish this to Slack. I'm pulling data from Notion, pulling data from my Google Calendar, putting it all into one place and saving me the time of having to go look all this up. There, this is the beginnings of having AI as basically my VA. And as you build this out, you can begin to do what kind of interactions you would do with maybe somebody on the phone or somebody in person to clarify on these tasks, right? If I look here, some of my overdue tasks, I've got a piece of outbound to write. This actually did— I did it 2 Mondays ago, right? And I could, when I'm running the full program and not this silent demo, It'll ask me, do you have any changes you want to make to your task list for the day? I can say, yeah, that one, it's already done. You can move it to complete. And it'll automatically do it. Or I can say, oh, I actually meant to add a new task. I need to do this. And it will create things. So the kind of back and forth planning through speech to text, using tools like Whisper and tools like OpenAI text to speech, make it where you've got effectively an executive assistant who's going to look at all of the sources of data that you want it to. Right, I've got Notion and my calendar. You could have Salesforce or HubSpot or any data points you want here, throwing this into how you pull your information together. And then also, you you could tell it move things around in my calendar. It's got connection to calendar, move that event to later, or block me a time to do this thing. This used to be something only high-powered execs who spent thousands of bucks a month for EA can have. This is something for $100 a month on cloud code, you can enable right away. So what are we going to talk about here? I think a little more depth on how AI unlocks productivity, because that was only a tiny portion of what you can do once you start to use AI to build better systems in your company. And we demoed kind of quickly what it is you're able to do when you start to use AI and put pieces together. But there's a lot more you can build out from here. I'm going to talk about a bit of these pieces and how they fit together, right? Before I go into all that, as a reminder, who am I? Why you're listening to me, right? Uh, and I've been doing startup marketing for 20 years. I've built a number of businesses up past $3 million a year in revenue, some up past $15 million a year in revenue, and I build and launch marketing engines for startups and new ideas. Uh, Crowdtamers as an agency has a— there's a group of 8 of us, although only 3 of us are really full-time. The rest are part-time working in this while they do the rest of their life stuff. Uh, and we built a number of million-dollar marketing engines ourselves, and I have built and launched AI stuff, two of them, since '22. So that's kind of who I am, why maybe you want to listen to me. Part of what I've been able to do, where I'm now, uh, CEO of Crowdtamers, I'm a co-founder of an AI startup right now, and I'm also a founding member of a giant community-based startup, uh, but seems very overcommitted trying to help run three companies. But it's because when you focus on what you must do, you will find you get more done than if you focus on doing everything on your task list. When you're doing it right, you're not focusing just on accomplishing more, although part of using AI and tooling to be more productive is doing more. You're going to figure out what it is you need to do, and to figure out how to tell people what you're doing. And then you're going to figure out the right way to see, am I doing the right thing? Is there work I can give to other people? How do I hand away the parts that aren't useful to me to either to AI, to another team member, or maybe just defer and forget it entirely? When you focus on doing the right work, you will discover things improve. Now, I'm going to go two caveats here. One of them is One, if you haven't already gotten any of the information from CrowdTamer in our prompts for this, as I go through how things are built, how I'm doing things, sharing specific examples of results, all the prompts, everything are already available for you to download if you go to crowdtamers.com/ai4founders or scan the big QR code. If you've already registered previously, You will get an email right as this workshop is ending with links to all the downloads, so you have all the prompts and everything you need to build this stuff yourself, as well as Workshop 2 and Workshop 1, everything that you want to, uh, that's the like, don't lose your mind trying to copy screen shares or quickly write down things. It's going to be available for you so you can kind of focus on what we're talking about today. So that's like Make sure you know what's going on there. All right. As we start to work on delivery, we have to begin by understanding what proper work and proper planning looks like. And this is something that's pretty easy to get wrong, especially if you let your AI just do its default thing. LLMs are trained on the average of all human knowledge. More or less, and as a result, they're going to tend to a very kind of middling median output. What you have to do to make the results you want to see better is put the work into prompting and put the work into examples up front. So for example, with the output from here's my agenda coming into What do I want to see here in Notion? Sorry, here in Obsidian. This is based off of me giving it templates and examples of what I want to see. If this is set up properly and you explain it clearly, AI will give you way better results. So in order to give it good planning, you have to have it ask you questions to make it make more sense for you. So if I go, for example, I want to have AI tell me— that's the wrong one here— how to build my morning prompt. All right, this is one of the downloads that's available if you go to ProTutors and try to access it. This starts by telling it what I want it to understand. Right? Tasks come in different sizes. Estimate on the task. That's AI coming into the loop. Right? Block out parts of my schedule based off of what I tell you so that I have enough time to do the work I'm supposed to do. Help me think about not just the tasks I'm doing, but the work I want to accomplish. And then weekly, let's review how well everything went. How good estimations were, how much productivity I felt, how I felt was accomplishing, how good I feel about my week. And then begin right now. This here is going to walk through if you use Obsidian for putting daily notes in, use Obsidian for, uh, all of your kind of knowledge-based management. There are other tools you can use here, and you can tweak and swap out parts of this prompt. To suit you. If you're an Apple Notes person, or if you want to keep everything in the cloud with Notion, there are ways to make all of this work. But this is just an example of how I do it, right? So what I've done here is given it examples of like, what is the result I want to see out from you for work delivery and for estimations? And then I can go in and over time continue to provide feedback and tweak this file. Because this file here for how to create all this is saved as another file in Obsidian, I can just make the changes right there, save it, and the next time it runs, it'll have whatever updates I want, right? You put all this together. Claude, by default, unlike OpenAI and some of the other tools, doesn't really have a deep memory of what you do. So for Claude, I use Obsidian as basically the memory storage warehouse where I can tell Claude, go read this, this is the context for this conversation. When you do that, you give it a particular file, go use that as part of the— now the task completion you have to do, your outputs are way better. And because Obsidian is easy to use, it's quite easy for me to make changes and put a feedback loop in place. That's the, like, the simple morning prep, right? From there, I might want to build out, modify, do my own iterations from here. What I'm giving you is a skeleton I encourage you to use to build up on top after you've told AI what you want to do. But what's important, regardless of how you're working with AI, whether you're an Evernote user or an Apple Notes user or you're an Obsidian, like any of these things, there are 3 kinds of work. And you have to use AI to help guide you through all 3 of them. Because if you free up the mental space, because you're not tied into, I've got to look at my calendar, look at all my tasks, look at Slack to see what people have given me from, you know, clients and team members have given me for delivery, and then try to wrestle all of that into the shape of the day. You want to step back and say, okay, one, what are my big goals? And on a weekly, monthly, and annual basis, for me generally, I'm looking at those, right? Weekly is just what do we have to do to unlock whatever the clients and the company need in the next 7 days. That's a kind of a very short-term strategy. But then a bigger one is something like, hey, I know I have to bring in this much revenue every month and I want to grow my company by bringing in this much more revenue every month. That's a strategic need. And then tasks that show up will fit into that monthly. And then finally, annually, I want my company to make this shift and do this thing by the end of this year. And then every month I look at what do I do to contribute towards that? And every week, how do I chip away at those big pieces? As the founder, a lot of your thinking power needs to go into there because the other two are ones your AI can really handle a lot more easily. For you. Planned work is if you say, hey, in 60 days' time, I want to be doing this with my business. You can actually use AI to help you unpack what do I do every day to move toward that 60-day goal. But AI is not going to be great at what the 60-day goal should be. It can make up some stuff, sure. But as a founder, your ability to make the judgment call, to look at things, to understand what you want versus what the average of the entire world knowledge might say is worth doing are going to diverge. And you want to steer towards what you— what matters most for you and your business, as opposed to what the kind of generic advice for you might be. Then there's emergent work, which comes in two forms. One of them is, I'm tackling planned work and I realized I have something to do to make the planned work happen that I didn't know is going to happen. This is really common in, uh, when you're working on the kinds of work that is not very checklist-driven. A good example is most coding, most development work. You'll go into it saying, hey, I want to write this feature, and then you'll go, yes, but to write this feature I have to read these documents to understand how to use this bit of code, and then, oh, now that I've done all that, I've realized that actually we're using old data We have to update the data somewhere, or our software is wrong, or this thing. And so you might do 5 things that were not at all part of the planned work, but they had to be done in order to do that work. And the other emergent data is— emergent work, sorry— is if you're in a service business like I am, I've got my plan for what to do for a client for the week, and the client might go, but I want this. Well, okay, I didn't plan for that, but now I have to either say yes, we're going to do it, in which case it's emergent work. Or I'll go, no, we won't, which case I have to manage the client's expectation versus what I'm delivering. But as you put these types of work together, planned and emergent work— emergent work is very easy to have AI help you scope and figure out and put into your day or put into tomorrow. Planned work, you can use AI to say, hey, I want to do this, what are the pieces that have to go into it? And you can use AI to help you figure out What's important for me to do towards my strategy? And let's talk a bit about how I recommend you tackle that. So in, I believe, the first workshop, I talked a bit about peer scoring. And when I teach Founders Institute cohorts, or when I teach any other growth marketing approach to companies, I talk about peer score and Peer scoring is the idea of you want to take a piece of work and rate it on 4 different numbers: priority, importance, effort, and result, right? You can score at 1 to 5 or 1 to 10, whichever way you want to do it. What you want to look at here is the higher the score is, the more important something is. And if I look at a given task, how does it align with the actual strategic work work that I want to do, if I assign a value to my work, to my hours, to a given task, how valuable is that task? Now, generally, when you're doing the work manually of estimating how big or how important something is, you don't do it on planned or emergent work because it takes too much time for a human to do this. But with AI in here, you can very easily give it a prompt and say, here's my to-do list for the day, stack rank everything with these big strategic goals in mind and this system. With that, it's going to go through and say, well, here's what you should tackle first and why. And if you're feeling particularly like, I don't know, I didn't think Claude gave me a good answer, or I want a second opinion, go Go to OpenAI, go to Gemini, you will probably get slightly different but not tremendously different results. Okay then, if you find consensus among the AI, they go, yes, these are the most important things, then you have to either decide, yes, I'm going to follow the logic that I had this AI go through and tackle these things first, or you can go, I don't want to do that, I'm going to go off doing my own thing, right? You're a human, you get the choice. I would note when you stick to the work that aligns with your strategy and your planned work, your outcomes tend to be better. Now, you need to figure out in your day and in your week how you have time to balance all of these things. If you are the founder who is hands-on, who you are You know, you and a couple co-founders, you are the executors of all the work in your company. Probably two-thirds of your time looks like planned and emergent work because you have to do stuff or else stuff just won't get done. As your company begins to scale, you're going to hand off more of this emergent and planned work so you can focus on the why. Why are you doing the thing you're doing How do you make it better? The whole company's direction, the whole company's steering. But you can't, at least I can't, maybe you guys are better than me, I can't just say, hey, I've got 11 minutes in between phone calls. I'm going to work on big strategy with my 11 minutes. If I have 11 minutes between phone calls, I'm going to get a cup of coffee. I'm going to take a moment to breathe. And I might sit in my chair and spin around for 3 minutes dealing with the existential crisis of life. You have to leave time to actually deal with the big question. You have to leave big time. And so blocking out big chunks in your calendar for asking a big question is part of what using AI to help you manage your productivity will do. Because I find I'm maybe 20, 30, 40 minutes into tackling something big. Before my brain is really up and running on it. If I don't bucket 90 minutes for a big task, I might as well not bother. On days when, like today, I've got maybe half an hour at most between my meetings, give up. Not worth it, right? So move scheduled events around, wake up early, like do what you have to, but leave time for the strategy too. And as you get to where you are less hands-on and more executive, Right, then two-thirds of your time or 90% of your time is going to be thinking about the big things, and then the minority of your time is planned and emergent work. Okay, so systems thinking, right? I've talked about a few of these, and, uh, peer scoring is the way I use AI to figure out what is it that I need to do right now based off of the strategy that I have set. I can always tweak my strategy. I can always say You know what, I said this was important, but this week something else is important. I can always do that. And then the peer scores will change as a result. Time blocking is how I manage out my, my calendar here. But I'll be honest, I'm not the kind of guy who would go into Google Calendar and manually make a task time block for every task. Having the AI do that for me and then having reminders in Slack that automatically pop up because one minute before any meeting I have booked, I get the triple knock from Slack and the alert, do this. It forces the context switch, particularly when you're working on a big task that might take several days or weeks or months to churn your way through. It's easy to forget to do the emergent and the planned work because you're grappling with the big thing. And you get that toast notification and you go, ah, I've got to switch work. I've got to do the new thing. At least forces you to do it. Now there's 3 other things I want you to think about and use AI to help you with. So one is, at CrowdTamerz, we have a daily scrum, which I think a lot of agile companies have, but what we don't have is a meeting where we all talk about our work. I don't like meetings in general, but particularly I don't like scrum review meetings. Or your daily agenda meetings for the whole company, because what it ends up being is one person is talking and nobody else is listening, and then the ball passes to the next person. Because if you're the developer, you don't care what the designer is doing or what the email copywriter is doing, and vice versa. So instead, you post to Slack, this is what I will be doing now. With AI and systems and CrowdTamers, what we do here is I look at what the to-do lists are for any given team member, right? What do you say you're going to do today? And then what did you post in scrums you are going to do? Do we find parity? Is there a big gap? Are you maybe not very busy at all? In which case, we can pull something out from the backlog and say, hey, maybe you want to tackle this. You can have a Slack bot that looks at people's daily scrums and just nudges them, prompts them, right? Hey, maybe tackle this now since you've only got 2 things you're doing today. That daily scrum is asynchronous, right? I've got team members in 5 different time zones. With those different time zones, we don't want to all meet at— I mean, if it was the morning, like 3 in the morning for me is I think 10 in the morning in India, right? Flight. That's no good. Or my team members who are in South America, some of them, their morning starts at my 11 AM. How do you span that gap? Making it asynchronous is the way to make it where everyone posts their intentions, and then we can check and make sure, does it match what needs to be delivered? Weekly reviews are where the team actually gets together and says, hey, we did this, and next week we're going to do that, right? That's the tiniest piece of strategy. We know these clients need this work done. So we have to make sure that we have allocated enough human time and AI time to do that work. And then finally, monthly, for every client and for Crowdtamers ourselves, this is what we were supposed to have done for you. Did we do it? If we didn't, how do we fix this so next time we do better? Those systems are all it takes to build a company, to grow it, and to scale it well. And it works. I pulled this dashboard this morning, right? CrowdTamers has 3 full-time employees and then a couple of associate other people. We're not that big a company and we don't make a tremendous amount of money, right? But if I look at January 1st to today, last year where I had more employees and made less money, to now Right? We have more than double the cash on hand and watching this profit. We've spent some of this profit out, but more than double the profit now that we had a year ago because we're using AI to drive more and more and more of how we do stuff. And I would note fewer employees, more clients, and we're able to throughput a lot more work. Now, the work that we do A lot of it looks like what I talked about in the first workshop. A lot of it looks like, let me help you with content creation, right? We have right now, I think, 11 clients somewhere around there who we're doing content for. And what we do is we take an hour of that founder's time, record a video, and we turn that video into all their social posts for the whole week. And we use AI to drive all of this. We use AI to figure out what it is that we need to do with that recording. And we use a couple tools. We're not doing this all with the scratch-built tools in Claude or in, I don't know, ChatGPT or in Claude Code. We are using other AI-based tools for this as well. But we're looking for repeatable bits of work so that we're able to say the things that I used to charge I mean, I charged Standard Chartered Bank 4 years ago now for basically the same thing as the content engine that we give clients right now. I charged them $7,000 a month, or maybe even $8,000 a month to do the work because it was time intensive. And now to do the same amount of work where we take the content, we process it, we make social posts every single day, it's only $1,500 a month to do that. We still have a healthy margin as a company built into this, but we can charge much less because AI lifts so much more. And that's just one little example from what my agency does, right? Everybody out there is able to see similar gains. And what you do is you build the systems that give AI what it needs to be able to take something automated. Now, for content, we use Minvo to automate our video edits. We use Descript to make it really fast to transcribe and edit videos where anybody can do it. Your business doesn't need those necessarily, or it probably does need those, but it also needs other things. And you can teach AI how to do those other things where you need to build in a decision tree. Now, I love Claude Code. Let me go pull open a CloudKit where I can just tell it, I want you to make a new custom repeatable command. Cloud uses, they're called slash commands. I just did a little while ago morning data as a duplicate of how I get data from my calendar in. And so I can say there, right, I've told it what that task is, and it's going to run and go do all of the bits of that task, which are going to involve checking out my calendar, checking out Notion, figuring out what needs to be done today, and then putting it into a JSON file for my computer to read, right? And it will keep running away here while it goes with that. I want to open up a new PowerShell for this is actually command, new command prompt. I'm gonna make a new decision tree, a new repeatable action here. I would note there are some errors firing, but the end, it'll still reach the end goal eventually because Claude navigates around the error threshold. So, okay, I wanna create a new slash command that does a specific thing. And I can ask Claude to prompt me to do exactly that. So I'm going to go, uh, I have put Claude code into its own subdirectory. I would note, uh, last week when we talked, I couldn't show you Claude code because it wasn't working natively in Windows. Monday they fixed that, so the whole presentation I had ready for today changes because Claude code now, instead of just Claude desktop, is available. So I can go here I can run Claude. Now, running Claude instead of Claude Desktop simply because it's more efficient at managing context windows, and you don't— once it starts to fill up the context window, your conversations don't end. When you're in Claude Desktop, if I go to any of my past chats here, or many of them, you will find, uh, this one reaches a point where it is at complete. I can't make more comments, it's full. There are no more conversations I can have here. Claude Code will automatically compress a conversation so that you can continue going. And especially if you're saving files locally and you're like logging things, conversation will continue very transparently. So I can say, great, here I pulled this prompt in, into Claude Code. I just said, great, I want you to walk me through building a new /command. And it's going to ask me all the questions it needs to to understand what's going on. I would note, if you have enabled text-to-speech and speech-to-text, which I have currently disabled for this live event, it can read these to you and you can speak back to it. If you don't like sitting in front of a computer typing, you can make it work that way. Heck, you can record answers on your phone with audio files and then send it as attachments. Like, there's a number of ways to work to make decision trees effective for you. But let's start with the questions. Okay, the first question you want to ask is, um, like, what type of decision are we trying to make here? And my example here was, uh, I want to I think I've got— hey, there we go. A decision-making tree for content strategy. So I'll grab this here. I'll grab this here. And I'll say, I am looking to make a decision tree around content strategy. For clients— I can paste in that bit of the prompt right now giving us some context and it's going to help steer the questions. Now just like previously, I'm not going to go through this entire thing step by step because it makes for pretty boring presentation here but you can see how Claude is working from a to-do list and Claude Code is really good at this. That's another reason why I like Claude Code for non-coding tasks. As well as coding tasks, is it's got this to-do, and until it's completed this first to-do, it's not going to go on to the next one. You may tell it, hey, modify your to-dos, add one in, or do something different, but by default it's going to go through this set because it knows what it wants to go back to. So I can answer all of these, right? And it'll begin to chip away at understanding what the business tree logic should be. What I want in the end is I can tell it something like— I keep hitting Ctrl+Shift+C instead of Ctrl+C because Ctrl+Shift+C is how you copy from PowerShell, not from Windows. Anyway, here. I can say, here's the general logic I want to implement. I would note, I'm not really answering the questions and I probably should be to get the best output here. Does this answer your question, or do you have follow-up? So I'm giving it room to continue here. There we are. Since I already had morningdata.json written, it's now asking if it can overwrite the previous file. I'm just letting it know, yes, it can, so it can make that file. Cool. So now This window has come back and it is, uh, giving me answers into what I have asked it. I would also note something else about Claude Code that you can't do with Claude Desktop. I have Claude Code open in two windows simultaneously, and I can have four or six or eight windows open if I want to go full sicko mode, so that while something is happening in one window, I can progress on a different task in the window. Right? So I can build out my decision tree here. And I can say, this is the logic I want to go through. And I want to say, hey, this is the information from my client. What's the content strategy? I could even— I think we've talked about this last workshop. In insight.crowdtinger.com. I could even go use a tool like my Insight AI tooling. That's not going to be correct. Let me open up a different browser window where I've got passwords saved. Hey, there we go. And I do, I'm a site user too. There you go. I think I'm logged in. Hey, there we are. I could use a tool like this. And I could say, hey, This is my client. Let me grab this information about them and about their competition and feed this in as part of my strategic messaging decision tree, right? I can start to use other tools to give me input for these tools. However I want to make it work, what it comes down to is I need to build a repeatable step where if I say, hey, I'm going to go here, I'm going to grab information from my internal tooling. I'm going to pass it over to Claude Code. Maybe I'll put it into a PDF or into a markdown file in Obsidian and tell Obsidian to read the file, whatever, whatever way I want to have it work. Then do these things, go through these steps, and this becomes repeatable where not just you can do it, but anybody in the company can do it. And eventually you can automate it almost entirely. Using AI and have the AI write some automations and some scripts for you to make this all go. So there's a lot you can do with this beyond just what I have shown. All right, I saw today, this morning, a gentleman I follow, Adam Robinson, was talking about exactly this as I was reviewing the presentation to go in today. I thought, you know what, This is a pretty good example of two things. One is most jobs I don't think are at threat with AI. Some of them for surely are. And as Adam mentioned here, right, he replaced two full-time employees with an AI clone. So he fired two people, which definitely means these people lost their job. But if you can be replaced with a $99 a month AI clone with no loss in quality, you're probably, the value you're adding isn't as much as the pay you are getting as a person right now, because what we need to be is more able to guide AI, more able to make AI's outcomes better than just doing repeatable steps of work. But for a lot of businesses, like Adam is doing $5 million a year. I'm not doing $5 million a year. I don't have any full-time sales employees. At my current run rate of about $400-something K a year, I will never have any full-time employees who do sales because I can't afford them. But I can afford a $99 a month AI clone that is almost as good as salespeople are to grow my business. And this is really where most of what AI's disruption in the short term is going to be doing. Because 70% of people work in a small company. Most of those small companies don't have enough capacity to fill all of the work they need done, but they also don't have enough profit margin to hire people to do those things. Project management is a great example. Like, I don't have a full-time PM at CrowdTamers. You kind of need really good PM to be a good service agency, because once you've passed 5 clients, you can't keep all the work work in your head. The better we get at using AI to run the project in my company, the better quality service provider we are for all of our clients. So I'm able to unlock happier clients, more productivity, better team member collaboration, all of these things with AI that would never have been possible before. And it's not just sales, it's not just content marketing, You're the key part of this loop, right? You being in here is required to make this work effectively. And your goal as a founder over the next year, as you build out your company and your business and your operations, is how do you add these strategic layer on top here, because a lot of the execution bits, you can pay, you know, Cloud Max $100 a month right now. I would note, I keep talking about Cloud Max as a great value. Writing this software that I demoed for how I plan my mornings, I started doing this about 2 weeks ago. Took me almost probably 2 days start to finish. To write the whole setup. And if I had been spending per API credit, would have cost me almost $500 for just those 2 days worth of coding. But I pay Claude a flat rate every month, so only $100. Never mind what it would have cost to have a developer write this code for me a year ago, right? That this would be a, I don't know, 3-month-long, 4-developer project costing $120,000 to build, right? You as the human in the loop here need to figure out what can I do that is not the bits that AI can do? How do I optimize me there? How do I optimize AI for the bits that I could do, but I shouldn't have to anymore? And that does nicely tie into kind of what we're doing. That's the wrong slide. We're talking about next week. This week's AI productivity. Next week, we're talking about AI agentic systems. So I've talked a lot about how you can use Claude and Notion and Google Calendar and all of these pieces you put together to unlock these productivity games. But going into how to set up MCP servers, how to make sure that you are using, uh, installing and setting up and running all of these other technologies effectively, is a full hour and 15 minutes all by itself. And that is what next workshop is going to be, is about how you can unlock— I mean, everyone's talking about this, right? I've got thousands of leads a month coming from, uh, my content posting, or McKinsey is sunk, or here's my AISR, or do QA testing. Like, all of these things are possible with basically 3 ideas. One is you, the human, considering and creating a system that you want to have your AI automate, because computers are only as smart as you tell them to be. And AI and LLMs are only as smart as you tell them to be. You need some AI tooling, and I strongly encourage whether it is Claude Code or whether it is Claude Desktop, you need to have— I just forgot what the last text here is. Um, we have Claude Desktop running here, like either of these, if they're able to work on your local computer, and they're able to work on the cloud is when you really unlock the productivity gains that 99% of people who heard about the AI revolution are missing out on. And when you put all these pieces together is when you're going to be able to accomplish way more than you think you are going to be able to accomplish. And this matters. This matters because all of you building right now are building into a headwind. We are at the lowest point in, I don't know, what is this, a decade, more than a decade of growth for scale-oriented startups, much less the new ones. And the growth has slowed down for two reasons. One is the economy is weird. We've all seen it. We all know it. Things are kind of woolly out there. It is hard to predict what's going to happen next month. So how much do you want to invest in something expensive. But the other thing is, everybody kind of vaguely knows that AI can do a lot of stuff that we used to pay some SaaS vendor to do for you. And so that slows down your deployment of buying a $10,000 a month or $100,000 a month or even $100 a month product because, you know, well, but If I need it enough, I could probably build something myself that makes this work. This means raising funds is harder. This means landing clients is harder. This means you need to run a leaner operation, and you need to get the right things done as fast as possible and not worry about the things that don't matter for your business. The more you focus on the right kind of work, the more you will be able to make headwind against this kind of traction. I would note, as crowd tamers, the agency has focused on more clear definition of the work we do, focused on bringing people in to do their content marketing for them, and then expand those contracts. We've improved the quality of our average client base, improved the retention for how long they stick around, and we've been growing This is my 5th year running Crowdtamers. This is the first year where July and— well, June and July were both growth months for me. Summer has always been slow for me. As we focus on better messaging and as we focus on better delivery and being able to outperform other people out in the market, we're able to grow despite the fact that everybody who I work with, all of my clients, are the ones who are seeing growth slow. Because that efficiency and that focus makes a difference. So if you haven't, if you're here now and you're like, I haven't signed up for the next one, which Trevor's going to talk about how I make MCPs work on Cloud Code or on Cloud Desktop, I'm going to do a live install and demo for you of setting up the 3 most important MCP servers to me, one of which is you connect to your calendar, the other is you connect to wherever you store your tasks, And the third is you connect and store something local on your hard drive. I like Obsidian, Apple Notes, Evernote, just a bunch of text files in a folder, whatever you want. If you've got those three, you'll be amazed at what you can accomplish. Register now. Make sure that you are going to show up for this last one of the series. This has been, I believe, and Daniel can validate if once we get back on the discussion in a few moments, the most, uh, signed up for workshop boot camp for Founder Institute. I hope people have found this useful. If you have questions, if for some reason the, uh, sign-up link for the downloads didn't work— I saw at least one person say it didn't work, people have had problems putting in their phone number and getting it to work— if it's not working for some reason, email me. I will forward you the email with everything so you don't have to muck around with trying to make the form work. It seems like about 5 or 10% of people can't make it work for some reason, and I haven't been able to figure out the technology problem that's causing it. Regardless, I want you to get these prompts. I want you to find this useful. I hope you found this discussion useful today, and, uh, thanks very much. I will try to go through some questions.

Awesome, awesome, Trevor. I think you can see the emojis. A whole lot of hearts, a whole lot of love, folks. I absolutely enjoyed that. There was a lot of, you know, technical bits on how to do stuff, but there was also a lot about thinking. And when I kind of think about how people use AI, what I really like is it's kind of given us the ability to really think about the bigger picture a lot more. And I think that's really important. And as we go into questions, I'm gonna jump right into it because we don't have a ton of time. We have this really interesting question here from Simon. How do you intentionally use AI to augment and elevate your own thinking quality rather than just automate or outsource routine mental work?

That's an excellent question. And so there are two things I have found that AI is really good at surfacing. That's easy for me to miss. I'll give you an actual, like, a story of it in action, and then a theoretical discussion. Sorry, there's an ad. The story of this in action is, uh, for a client, we're doing content marketing for them, we're producing content for them every week. And one of the things I strongly recommend you do is you put a small ad budget behind your content, because I am not patient enough to wait for organic visibility. I can pay for visibility and put $100 a week, and I will get thousands of views instantly. But why? And for one client, he went, I've spent $400 in the last month. What did it get me? Because I have no proof anything worthwhile came from it. And I went, well, I mean, view count, but that's a vanity metric. Why do we care? Right? So I go to Claude, my, my AI co-founder here, and I say, hey, let me connect you to this client's Google Analytics. Let me connect you to this client's Search Console. And let me download the reporting from YouTube Studio since I can't really connect to it directly with the AI and just get all the raw data I can and dump it into the AI and say, what happened? Explain to me if there was any value to putting $400 behind promoting this content. And what it finds out instantly is, so you're running these ads not to get clicks. They are reach ads. They're designed to just get views. So no one clicks on the ad. So you go, wow, it's a waste of money. But the branded search terms for this company's domain, right? So company that— not the domain itself, but searching for company name— went up 300%. And because of that boost in company name searches, they got 27 signups in the course of, uh, 2 weeks as we were promoting this on, you know, their, their brand term. And as a result, those 27 signups came in at— I mean, I'll make the math a little simpler, right? 30 signups, $400. What is that, $14 something per signup? Actually a pretty good CPA for a B2B company trying to acquire a, you know, a business that's going to be worth hundreds or thousands of bucks a month want their site on board. So directly you look at YouTube, there's no correlation. And if you're clever enough to look at analytics, you might see, oh, there was an increase in traffic while we were running these ads. But the increase wasn't in ads coming from YouTube or searches coming from YouTube. The increase was in Google search traffic. So you'd have to look pretty deeply to find that on your own. Instead, the AI looks and goes, oh, This is an opportunity. And so then I've systematized this because of course I have. Now as part of reporting for every single client, we ingest all this data and we just ask the AI, look at the report that we've prepared for the client and find any interesting trends or information from this dataset that's not in the report that we might want to include. Now I'm uncovering information that helps me steer business decisions by asking the AI, look at all these things and figure it out for me because I'm not going to. It takes too much time. This is the kind of thing that used to be you pay McKinsey $40,000 to go spend the 3 months it takes to learn this, or I can spend $400 and have Claude analyze the ad spend and instantly say this is the result you get. There's an example like exactly how it can surface unknown information. But then the trick is, it doesn't have the strategic next step, right? It might say, oh, keep spending on YouTube. That's cool. It's clearly worth it. But there's a strategic question you should ask, which is branded spend with no actual click goal was positive, super positive on YouTube. Where else could I invest in branded spent. Could I— these are random ideas, I'm not saying do this— could I buy Spotify ads and say, hey, uh, I know that my persona listens to 99% Invisible, a UX podcast with I think a million subscribers, right? I know that they listen to that. I want to advertise there on Spotify, not because people will click the banner, which I don't think I've ever done in Spotify despite having listened on there for a decade, but because the brand's in my head now and I might search for it later, right? Is that where my branded spend should go? Or do I go on to, I don't know, Taboola, run in the bottom of all those news websites with something about this product solves bunions or whatever you do? I don't know, right? Like, you ask the next question and then you can use AI to validate some of what to do from there. But I have still found AI will do the literal thing of, oh, you should increase your YouTube ad spend, but not the strategic thing of, but what does this mean?

To answer your question, yeah, it does, uh, and in a ton of detail as well. So one of the things that you mentioned was, uh, it's important for AI to, like, for example, to know stuff about your business, right? Like, I think one of the things that I noticed as well is when I'm trying to come up with those missing pieces in strategies or whatever I'm working on, I feel like sometimes the quality of information I'm providing AI is not up to the mark. That's why I'm not getting what I want. Do you have any tips or tricks on how do you make sure AI has good quality data on what you're trying to do?

Sure. So I use Obsidian a lot, right? I keep talking about it. It's free. I'm not vlogging something you have to pay for. And it has— it's inherently just a bunch of folders with Markdown files and CSV files in there. And that simplicity and portability means that any device, like any device I've got, This is my Raspberry Pi. I can load up Obsidian on this and read things here, or on my phone, or on my laptop. I could probably, if I was sufficiently motivated, get this loading and running on a smart refrigerator, right? The portability and the extensibility means two things. One is it works anywhere. Two is 20 years from now, that data will still be readable because it's all just text files. So there's no vendor lock, right? And then, as I do almost any work with AI, I tell AI to log this information to the right folder into Obsidian so that it's storing information. Then I can tell Obsidian, go look in this folder. So for example, I've got— I'll share my screen here real quick. I think you've probably seen this already though. I've got in Notion— I'm sorry, in Obsidian, I have everything for AI for Founders. Look at here, Founder Institute, AI for Founders. There's my hub with all of my notes and everything listed. And then there's outlines for all my workshops, right? All my downloadables, they're created here. So when you're like, man, Trevor, you're PDF downloadable isn't very fancy. Yes, that's because it's exported from here. That's why there's no branding on it. But then for other things, right? For, I mean, CrowdTamer, I've got my podcast or my clients or my content marketing or whatever, like all of this sits here. So I can tell Claude, and specifically Claude Code, I like better for this because as I mentioned, it's more efficient at managing the context. I will tell it, go look in this folder. This has the information you need. It has all the information I have, right? The other place I will go is because I can't synchronize my other huge source of data, which is phone calls. I can't synchronize it to Obsidian without a bunch of extra steps. I end up taking the transcripts for every single call and putting them in Notion. And so they're all in one Notion space. And I can tell Claude, so in Notion, which you're connected to, look for either this call or this client, or like there's search for this thing. You will find the call transcript from yesterday's call or last Thursday's call or whenever. Look at that before you start to tackle this project. And so it also has for anything that's a client-facing bit of work, or I've had a call with a team member bit of work, I've got the actual full context of what we talked about, plus whatever's in my file. And this combined together to give you— think of it as like if I was a researcher trying to write, you know, the biography of this task 400 years from now, I've got all the primary sources of data available. And so you then can draw from first principles what conclusions you need to to make. I find if you preprocess the data, it can help some if there's a huge dataset that you have to narrow down so AI can figure it out. But Claude's context window is 200,000 tokens, which is close to 80,000 words. That's usually enough, right? You can't have long conversations necessarily once you've loaded a bunch of context, but you can have enough to begin to get things started and to begin to say, for what I need right now, what are the most relevant facts. So you can then write that to a separate file in Obsidian, and we can move forward from there. If I've used up my full context window— awesome.

Uh, yeah, so on, uh, notes and trying to, you know, extract information, one of the things that I do is there's a tool called Granola. It basically listens to your meetings in the background, so it also has this option to export a transcript. So what I do is after a meeting, I'll export the transcript, put it into GPT-03, and then kind of get deeper insights. I've found that to be super helpful. But for more questions from our audience, we have a question here from Jay. Shout out to Jay because Jay's been here since day one. There— are there ways that AI can be used as an effective tool to deal with ad hoc emails, instant messages, tasks that can surface during the day? I know you answered this a little bit, but if you could maybe like touch on like the emails and instant messages bit.

And that is definitely something I want to talk about next week, but I'll also discuss it briefly here. I have Claude hooked up to my inbox, and what I have it do is it knows, and it pulls down basically from, uh, from Slack, it's pulling down who my clients, I'm at, the clients right? Based off of what the Slack channels that I'm sharing with somebody are. It knows who's active. So it knows if you're with an @CrowdTamer's email, or with any of my clients' domains emails, I actually care about your email. So every morning, I will tell it to make me a digest, what do I need to respond to? I am terrible at email, by the by. I only do this once a day. If I was like, really into it, I would build an automation that maybe every hour hits me up on Slack. But when I'm like Slack, people message me on Slack, I know that either they pay me or I pay them. So I will respond quickly on Slack. Email is a, it's a war zone. I will check once a day, but in general, even that is way better than what I used to do, which is every Friday, go to my unread tab, hit Ctrl+A and hit mark as read and go, week's done. Right? Like, I'm not, I'm not a big email user. I don't use AI to process Slack because my Slack channels don't tend to be too noisy. If you've got really noisy Slack channels, I would almost suggest, especially if you're the person in charge, this is AI for founders after all, I would start to talk to my team about AI etiquette. Sorry, Slack etiquette, not AI etiquette. Slack etiquette on when do you tag me? Right? Don't just tag me to be like, hey, by the by, this happened. I don't care. Right? If you are able to solve a problem on your own, go solve it. I love you for it. If you need my feedback on a thing, tag me, then I will show up and try to help. Like, that etiquette means I probably only get 8, maybe 12 Slack messages that I must respond to a day. That's not hard to manage, right? With I think 15, 16 active clients, and with 19 members, 18 members reading my feedback any given day, right? That's quite efficient. Email is the noisy channel that I have to manage, and so I use AI once a day to go through kind of all that and figure it out. And I also do things like, um, there are some emails that I know I want to reply to, but they're, they're kind of low-value replies. Heyro is a good example. Helpful reporter out. It is a request by journalists to get experts to answer their questions. I have areas of expertise. I have at 6 AM, noon, and 3 PM, I have a script that fires off that looks for the hero email that's going to show up in the next 5 or 10 minutes, whenever it shows up, scan it. If there's something it can answer, writes an answer and sends it. Because the fastest reply tends to be the one that gets quoted. So I have those going off. It looks at my background, my processes. It knows the kind of questions I want to answer and the kind of answers I want to write. And it will frequently quote me from my own blog posts in the replies it sends the journalist. But it goes really quickly. Like, I want a reply to be done there, but I don't want to write that reply. It's too much time. So I've entirely given those away. And I would note maybe 1 in 10 or 1 in 20 gets quoted. Not a very high accept rate. That's why I don't bother writing them, because even when I do the work, I still don't get quoted. So like, you can automate away things that are low enough value to get rid of, right? Emergent work. That's something that showed up in my inbox I don't really want to tackle. I can automate the whole thing away. Cool. It goes away. Strategically, I've got a big marketing campaign I want to plan for. It's way better to spend half an hour working on that than it is answering hey-ro queries that probably won't get answered or quoted anyway.

So because, uh, you know, you're a founder, we have a bunch of people here who are founders. They have teams of all sizes. So this might be, uh, kind of like off the rails a little bit, but how do you get your team to start adopting like an AI workflow framework when it comes to productivity? Or like, what are some tips and tricks of like how do you onboard your team?

So it is a mind shift. I've got a developer we're onboarding right now who is 10 years of experience writing code. I've mentioned I'm co-founding an AI app right now, uh, and we are making some of our code a little more effective, a little more scalable, a little less prone to failing at 2 in the morning and telling no one, whatever the reason is. But we have our developer and he's used to writing code in VS Code the way a sensible human would. I said, look, the way we hire at CrowdTamer is that there's always a paid pilot project. For a month, tackle something in your free time, we'll pay you for it. If we like each other enough to keep going forward, cool. If not, no worries, you got paid, we got work. Everyone still did okay. Half of your pilot project is not the code you write. Half of your pilot project is learning how to write code the way you write code in 2025. I showed an example. I've since closed the window, so it's gone, but I had set up for a code refactor of actually Insight AI, the tool I showed earlier. It's a mess. It was vibe-coded the worst way because I didn't know how to do AI coding when I did this. It has like 80,000 characters worth of code across 37 different files. It's, it's, it's just not right. And so I said, hey, here's how we need to refactor it. I had AI tell me how to refactor it. And then I said to AI, great, I want you to make 6 different agents because Claude Code can make subagents which have their own capacity to go solve a problem, do everything Claude does, finish their to-do list, and come back and tell you it's finished. So I basically had 8 different instances of Claude all writing code on the same codebase simultaneously. Which is insane. When you try and watch it, it looks like a seizure happening on your screen. Let me— I think I have a video I recorded of it to send to him. Uh, let me do this in real quick. Hey, oh, this. All right, um, I will— yeah, there we go. I'll share my screen to show you this real quick because as it plays You're like, that's not, that's not a very useful thing. And in truth, the way it runs is a bit overwhelming, right? So this is, uh, this is it trying to show me 8 different screens updating simultaneously. So it's very kind of insane looking. But what's happening in the background is as each of these 8 subagents does anything, it is then saying, great, I'm pushing to the screen that this thing has been done. In the end, where we got to, I think if I do— where did my refactor report go to? It's going to be here, this one. Then basically I had each of these files created and it ran them all on its own. Running the code this way, learning to tell AI, go do all of this, here's what the tests should look like, make sure it's working, is completely different than how code writing was one year ago, much less a decade ago. And you have to either yourself be enough of an expert to tell your team, I know you can do the work, but I want you to learn how to do the work better and model the behavior. Or YouTube has everything that any human in the world knows how to do. And if you go search YouTube, you will find other people who are smarter than me, who have documented how they do the same thing, or whatever thing you want your team to get better at, and have your team self-train into the new system, and then importantly, share what they've learned on how it works with the rest of the team. Because if one person in your company suddenly 5x's their productivity, most of the rest of the company is going to go, I want more of that. Even people who are not super motivated by productivity tend to be motivated by, I don't have to do the stupid parts of my job that I don't like. And being able to give those away to AI is a huge win for almost anybody.

Awesome. Thank you for your super detailed answer. We don't have a ton of time left and I do want to leave some space for networking. We'll just go into the last question and I'll expand on it a little bit. I know you've answered this in previous workshops, but just because we have a bunch of people who are new, Matthew asked, what is the reason for preference for Cloud Code over V0? For ChatGPT, deep research, all that.

So what you really need to unlock what I consider the, the useful powers of agentic AI is something that runs on your computer. You can set up Cursor or WinSurf or a bunch of these softwares, and there are also just CLI tools you can use that you install them on yourself. All of them to use ChatGPT or Gemini or any of them. But I haven't found one that's as easy to set up and install as Claude Desktop. And once you have Claude Desktop set up and installed, you can give it Desktop Commander MCP as its functionality. And again, next week we'll talk about how you do that. Once you've given it Desktop Commander MCP and filesystem MCP access to write to your computer, then you can have cloud desktops that have cloud code for you. Once you have cloud code, then you've got the most powerful tool I know of. It is a command line interface that does everything you want to. You hook up MCP servers to it, connect it to your calendar, to your email, to Slack, like to everywhere. It's the easiest bootstrap in for a non-technical founder. If you really like Gemini, go use it. I actually have a Gemini command line interface that I wrote myself for me before Google bothered to ship their own. And I was using it for 2 months before Google finally rolled out their own CLI. And it doesn't work as well as Cloud Code. And it's more of a pain to set up. So I'm not flogging my own tooling. If you go check out my GitHub profile, which I think I've linked to in the very first email set of prompts you get from from the first workshop, you'll find the Gemini CLI is there. It's called Twinkly, twin for Gemini CLI. Whatever one you want, it's not super important. Two things you need: one is a good context window, and Claude is number two right now, right? Gemini is number one, but I find Gemini quality a little bit lower. And you need easy setup if you're not yourself a technical founder. So I like Claude Code because it runs locally, you connect MCP servers to it, and it does everything you want from there. If you can make that happen without Claude Code, go. But also, the pricing for Claude Code Max is among the best. If you're paying straight API credits, I would not use Claude. It's the most expensive API. But if you're using someone's all-you-can-eat plan— ChatGPT, Gemini, Claude— I like Claude's outputs better. I like their pricing. I like their tool. So that's why. Awesome, awesome.

And folks, you heard it, uh, next week in what is our finale for our workshop series, Trevor will be going through everything in, uh, detail, right? How to set up Cloud and all that good stuff. Trevor, thank you so much for doing this. Uh, as always, it's a pleasure having you on, and it's also a pleasure having all of you folks here who've been, uh, you know, following us, uh, along, uh, in this workshop series.

And we—

what we'll be doing is we'll be sending you a replay of this workshop in the follow-up email along with Trevor's link to Crowdtamers. And we'll also have all of the replays up on the page where you signed up for this workshop. So don't miss that out, folks.

If you don't—

if you do want to stick around and network with one another, you know, meet founders from across across the world, feel free to do that as well. Uh, Trevor, thank you so much for joining us. Any words before we close out?

Uh, I think just the skill of using AI to help you focus on what's right and help you delegate mostly to AI what you can give away is the most important thing for you as a founder to learn over the next 90 days. Last time we talked about building an MVP using Claude Code and Lovable and other tools to make the technical shipping easier, right? That's, that's kind of a one-time problem in that once you've built the MVP, it is done. And hopefully you can earn some money with it. But helping you figure out on what you need to do, what your most important focus for work is, and how to unblock you to be your best self is a perpetual problem. So learning that skill and learning how to use and trust AI to help you be your best founder is the thing I would say to founders. You must unlock this capacity to be competitive in 2025 and 2026. Thanks, everybody. Awesome.

Have a great day, guys. And remember, do sign up for the last, uh, workshop. We'll be sharing the link in the chat as well.

Have a great day, Ed. Take care.