Transcription
There are two types of people learning AI. Those who are terrified AI will replace them and those who are using Claude to build entire businesses on their lunch break. You see people on Twitter and Instagram building apps in minutes and somehow making massive amounts of money while you're stuck feeling overwhelmed. You aren't sure which buttons to click or which hacks actually work.
But here's the truth. You don't need to be a tech genius or a coder to win. You just need to master 11 specific productivity hacks. I'm going to walk through today. Actually, I threw one more in. So, now it's 12 specific productivity hacks. And I've analyzed the top 1% of AI power users. The ones who are actually shipping products, actually making money, and actually saving 30 hours or more per week. And they all use AI in a very specific way that 99% of people completely ignore. And the best part, none of this requires a computer science degree.
Before we dive in, my name is Sabrina Rammanov. I teach millions of people AI for free after selling my AI company for millions of dollars. And today, I'm going to show you how to transform AI from a simple chatbot into a productivity multiplier.
Okay. Number one skill is prompt engineering. A lot of people confuse themselves by what this means. They think they need to buy like $5,000 prompt engineering courses. And that could be not be further from the truth. So, I'm just going to break down in very simple terms what prompt engineering is, why it matters, and a really simple template you can use to basically learn 99% of prompt engineering as a shortcut.
Prompt engineering is the difference between someone giving really vague instructions, just squiggly line, and getting a really vague, generic, lazing response. you look at the response and you're like, hm, I don't know, that's not super helpful. And that's because the instruction you gave to chat GPT was vague and not really detailed in the first place. When we talk about prompt engineering, it's usually providing a structured set of inputs into chat GPT so that what you get out of it is way better. So you also get a really nice detailed structured output from chat GPT.
Prompt engineering is improving the quality of your inputs over here on the left side. So instead of a squiggly vague line like this, you're feeding in structured list with expectations and context. And chat GPT knows what to do. And so you get this nice detailed deep dive report afterwards.
Now most people get generic answers from their AI because their literal prompt is make it better or like do this and three words. And that can sometimes work if you just need to get started. But you're often going to find that the answers are lacking and you wish CHA GPT would go deeper.
So, I'm going to give you a really simple prompt template and this is what I personally use in my head and it basically encapsulates 99% of prompt engineering. So, if you just use this template and memorize it, internalize it, you will have learned prompt engineering without paying for an expensive course. It's a really simple template.
So, the first line is you are a top one 0.1% expert in some field. That's it. Let's say you are a top 0.1% expert in venture capital or in tech startups or in copywriting or in launching Clickfunnels like whatever that is just put that at the end. You are a top 0.1% expert in this.
The next line is the task that you actually want it to do. And it can be really simple like I want you to research this and output a table with your research whatever the task is.
The next step is context. So let's say I'm using chat GPT to brainstorm what marketing channels I should invest in for my business. Context would be all the information that I'm willing to share. Like I've tried XYZ marketing channels. Here were the conversion rates. Here was the ROI. This didn't work for me so far. This didn't work for me so far. This is my budget. This is the time constraints that I may have. As much context and information that you're willing to share with AI, that's what you would feed it in next.
The third step is constraints. So for example, if you have specific budget constraints, time constraints, it could be in many different ways. So let's say you're considering making content to build your brand. Maybe you don't want to show your face. So being faceless is a constraint. And the ideas you get, the approaches you get for building a faceless brand will be really different than the approaches for building a face forward personal brand. And that's okay. just specify it as a constraint. So chat GBT knows what kind of information to go out there and gather because like I said like if you're building a faceless brand it's probably going to look at a different set of research and playbooks and best practices for building a faceless brand versus building a face forward personal brand.
Okay. And then the last part of this template is ask me clarifying questions. And the variation that I personally like is ask me clarifying questions one at a time until you're 95% confident you can complete the task successfully. I like it asking me the questions one at a time because it gives me an opportunity to like deeply provide a lot more context for every question that it was confused about. And you'll find this is honestly one of my favorite tips. You guys have seen me talk about this in lots of videos, but adding this at the end helps you think through what it is you're asking AI to do. Oftentimes, I tack this at the end and AI asks me a bunch of smart questions and I realize my initial instruction was super unclear and vague. And so, by going through this process, you also get sharper about what it is you're asking a high for. And so if you can just memorize this, internalize this, practice this a few times, you're going to get way better answers than most people out there.
Again, most people I see are still prompting chat GPT. Like let's say you put a resume in and you're like, "Make it better." You know, it will probably make it better to some extent, but without more specific guidance. Like AI doesn't really know what to do. So just imagine you're talking to somebody like a friend or a co-orker and you're just like, "Make it better." that person would naturally ask you a bunch of clarifying questions. Hey, can you give me more context? Why are we doing this? What does it mean to make it better? How are we measuring that result? Are there any constraints you want me to follow? Etc., etc. So, this is the process we're encapsulating in this prompt template. And it's really easy, honestly, once you do it a few times. It might look like a lot right now, but you don't have to follow it super strictly either. Sometimes I'll just do the first one, the second one, maybe one constraint, and ask me clarifying questions if I don't have a ton of context to feed in, for example. So, if you're still struggling with prompting AI and getting highquality answers, try this prompt engineering template. It is going to teach you basically 99% of prompt engineering without having to pay for a really expensive course.
Okay, so that's number one. Number two, specific productivity hack that will truly transform how you interact with AI and the light bulb moments you get from AI. Instead of using AI as your glazing partner that just makes you feel good, I often ask AI to be my sparring partner and to beat up my ideas. Hey, here's what I'm thinking of doing. Argue with me. Why am I wrong? What are my blind spots? What are the assumptions I'm making that are weak and not backed by data? What data would I need to go collect to convince you that I'm right?
So instead of AI as your glazing partner basically agreeing with you with any crazy idea you throw at it, use AI to make your ideas sharper and better. I'll show you an easy way to do this. You can ask it to identify weak points. Let's say you put together a business plan. What are the weak points that an investor would critique? Like go chat BT, rip my business plan apart, argue with me. These are all things you can append to the end of your prompt. Or if you have an open conversation with chat GPT, you can also just run this prompt based on everything we've talked about. Identify my blind spots, the weak points in my argument, where I'm missing data, my weakest assumptions, etc. These are all just key phrases that I want you to just internalize and throw a chat GPT and you're going to see the nature of your conversation dramatically changes because if you're pretty well versed in AI tools, it's too easy to get AI to say what you want. That's not helpful.
If I'm choosing between five different decisions for my business and they all look appealing, good decisions, how do I choose among them? How would you do it in real life without AI? You probably write everything down, do a pros and cons list. You would talk to your co-founders. You might talk to investors. You might talk to adviserss. And you'd ask them for their candid feedback. Not to just agree with you to make you feel better, but to help you make the actual right decision for your business right now. So instead of treating AI just as a research partner or a faster Google search. So a lot of people beginning with AI will just use it as a faster Google search, which is totally fine as a starting point, but you want to get to the next level of using AI as your intellectual sparring partner so that you can be smarter about the decisions you're making about your thought process. Be smarter about collecting data. This is one of my favorites. Where am I missing data and making a lot of assumptions about what I'm going to do? If I want to try XYZ marketing channel, why chat GB will be like, well, why do you want to try that when this marketing channel over here is already working? Why don't you just double down on that? So, get used to having that back and forth conversations with Chat GBT or Claude or DeepSeek or whichever you're using. It really doesn't matter. This is about how you're interacting with AI tools. So, instead of accepting everything they say at face value, get used to sparring back and forth with them.
Number three, this is for some reason not obvious to a lot of people, but you can use AI as your 24/7 tutor. I say this is not obvious to a lot of people because I will regularly get questions. Hey, I just watched your YouTube and I ran that one thing you said to do and now I don't know what to do. There's this weird thing on the screen. I don't know what to do next. I regularly get questions like this and the thing that I want to encourage people to do and get comfortable doing is to just ask AI. So instead of commenting on my video, open up chatgpt claude, whatever, and go ask AI what to do.
I do this all the time. I'll take multiple screenshots of something I'm confused about and I'll be like, can you help me figure this out? And AI is surprisingly good. It can analyze your screenshots. It'll ask you some clarifying questions, right? if you use step number one that we talked about and just be honest with it. I'm trying to build this. How do I do it? I'm stuck on this. Or I'm following this tutorial. I got to this step. Drop in the screenshot. But now I see this. Drop in another screenshot. What do I do next? A lot of people will wait for the perfect YouTube tutorial or try to find the perfect workflow template. But instead, you could have just taken that time and asked AI instead of trying to wait for the perfect YouTube tutorial or get every answer or get every question you have answered in the comments. The key thing I want people to do here is just to ask AI more questions.
So, examples could be like troubleshooting electronics. It could be you're following along a YouTube tutorial, you're stuck, you don't know what to do next. It also could just be brainstorming. I'm brand new to Claude Code. I don't know what it can do. Well, you can ask cloud code. You can open cloud code and be like, "What can you do?" Even better if you apply the prompt template we talked about in the first step. Here's my background. Here's my business. Here are my goals. Here's what I currently do today. Here are some constraints. What can you do as an AI tool to help me? You can go into every single AI tool and ask that. And usually most good AI tools will give you a pretty helpful response like, "Hey, did you know I can do this?" And so instead of looking for a guru that has all the answers or waiting for the perfect YouTube tutorial or asking questions in the YouTube video, just get used to asking AI itself. Do this all the time. New tool comes out. Let's say like Claude remote control came out like last week or something. I just open Claude and I was like, "Hey, can you walk me through how to use remote control?" I followed the instructions and boom, it worked. Like really that simple. It also teaches you to be a bit more like self-sufficient. You realize you don't need to find the answer in someone else. You can actually learn this stuff yourself. You just have this back and forth dialogue with AI. It can show you what to do. You can share what you're seeing, like drop screenshots in there and be like, I'm trying to do this, but I'm stuck.
So, this one's really important because I really see a lot of people are trying to follow along something and it doesn't quite go like they expected. So they'll reach out to the person who made the video asking for help. But at that point, you get used to using AI as your co-pilot, your tutor, and your teacher. This as an exercise for the next 2 weeks. Every question you have, I challenge you to ask AI first before you act upon the impulse to ask a person. Just ask AI first and then you'll realize all the benefits of it. It's 24/7. It's not condescending. It has access to the entire internet at its fingertips. It's fast. It's infinitely patient, right? It's not going to be like, "You're an idiot." If you ask the same question four times, the most important thing is learning to be a little bit more self-sufficient instead of hoping that person has the answer. You realize like, "Oh, you can figure this out." I know there's so much content out there trying to make it seem like other people have the answers. Even my own tutorials, just treat them as a starting point for your educational journey. I fully expect people to to use it as a launchpad and then go deep into the rabbit hole in their own unique way because we're all trying to do slightly different things and so there isn't one person who's going to have the perfect blueprint for you to follow.
Number four is skills and I'm using the word skills within the claude ecosystem. If you're coming from chat GPT you would think of this like a custom GPT whereas in claude ecosystem you would think of it like a skill. So you can think of a skill as a repeatable thing that AI can do almost like a repeatable task, let's say. And claude skills are really powerful because like they can also execute code and you can chain them together. They're really cool. But when you think about a really complex task you have to do like build a product, you can break that down into little skills. So for example, this one might be just design. By the way, if you are using claude code, there is a really popular front-end design skill that was made by the community and it just levels up your design. So, it's like a pre-built pre-loaded skill and it knows all these best practices around designing modern interfaces. So, another one might be the actual development itself, say backend code, front-end code, and then you might have another skill for testing etc etc. And so, when you think about creating skills, think about what is the large task. break it down into smaller concrete tasks and then for each of these skills you want to give it a bunch of context and information so that AI knows how to do this skill well. So like what does it mean to design a mobile app? Well, I'm not the expert, but there's like lots of best practices around how to design the UI and the UX of a mobile app so that it's really modern and feels smooth to users. And same for websites, same for web apps.
This is a technical example, but let's say we want to do a marketing example. So, I'll give you a real example for my YouTube videos, especially my new series where I'm doing education documentary style. So, I have one skill uh just for research. I have another skill that creates an outline for the script. So, it doesn't write the script itself. It just creates the outline. I have a third skill that actually writes it. And then I have a fourth skill which you can call it editor. I call it like a quality check. But these are four distinct skills and they're just they're subsklls in the larger task of like create a YouTube script that's about 20 minutes long based on a trending topic that's like deep journalism. I also have a separate fact check skill which is very loaded. So I should have mentioned that here. So fact check and the really cool thing about skills in the cloud ecosystem is that AI will figure out when to use the skill. So if I give a general prompt to cloud code that's like write a YouTube script, it will actually use these skills in the correct order because like it knows what each skill is meant to do. So you can explicitly trigger a skill by typing slash writer or whatever your skill is named in cloud code. But also you can let AI decide. You can say AI, hey, I'm going to do this. an AI will actually look at your list of available skills and then decide which ones to use to complete your task successfully. And so skills are really really powerful. You don't quite get that composability and flexibility with custom GBTs which is why a lot of people now are moving towards claude and skills and it's just really really powerful.
So a couple more things with skills here that I want to mention. People often look for skills somebody else has made like oh I made a directory of skills over here. There's 30,000 skills and people will be like, "Oh, I'm going to download each skill." I was actually also going to launch a directory of Claude skills, but honestly, they're just a starting point. Most of the skills in these compilations of thousands of skills are really generic and AI generated. And what you want to do is create skills that are really fine-tuned and specific to your process, your preferences, and your constraints. So, how do you do that? So, I'll give you a prompt. I call this self-reflection. It's basically telling AI to reflect on the conversation you've had to create the skill. So you can have a simple prompt like this like based on our convo skills. Really simple counterintuitively very simple. But what Claude will do now is it's going to analyze your conversation. It's going to look for tasks that are repeatable. So maybe things you asked for several times or tasks that had a lot of context like you have a lot of preferences for how to do this particular task. Well, so it's going to look for that. It's going to look for things that are repeatable and it's going to create the skills for you. You don't have to sit there and copy other people's skills. In fact, I really recommend this approach because then you get used to creating and refining your skills over time.
If you want to level it up a notch, use the techniques from our prompt engineering template. So you could say, "Based on our conversation, create three skills, ask me clarifying questions one at a time until you're 95% confident you can complete the task successfully." I actually highly recommend pending that clarifying questions line to this prompt because then Claude will be really clear like, "Oh, should we make this a skill? Is there any additional context you want me to add before I make this skill over here? And I keep bringing up the prompt engineering template because I made this list in order from basic foundational essential to more skills required a little bit more technical, right? But they all layer on top of each. And again, if you don't know what a skill even is, what did we talk about in the previous step? 24/7 tutor. Go to Claude and ask it what is a skill? Why should I make one? Why should I create one? So all of these skills stack on top of another which is the beautiful part.
So number five is memory. I decided to call it memory because it has different terms depending on what AI you're using. But the general concept is like I don't want to repeat myself every single time to AI. I don't want to tell it all of my context every single time cuz that would just be laborious and really tedious for you. So let's talk about memory. So just imagine what memory is. It's basically a big folder that's everything about your business so that you don't have to repeat it again. So if you hired, let's say, five people, imagine having to explain to each one everything about your business every single time. That's what it would be like if you opened chat GPT and you just started new conversations every single time and it just doesn't have that much context, not a lot saved in its memory about you. And so what we want to do is figure out a way to give AI everything only once and then it could reference it in all future conversations.
So in chat GPT you would typically have projects. Claude also has projects. And the cool thing about projects is again you just like load up all of that context into the project and then in all conversations in the project like you have three different conversations in the project. They're all going to know about your business over here because you loaded it into the project. So if you're not already using projects and chatgpt, highly recommend it because it removes a lot of the tedium having to explain your context and constraints every single time to AI.
Now if you're using claude code, the analog for this would be the Claude MD file. And this is my favorite. For those who don't know what it is, it's basically a plain text file that Claude reads automatically at the start of each session or conversation. So in a technical context, you would have your coding standards, architecture decisions, any constraints, what your product does, variable naming conventions, all kinds of things. In a marketing context, your cloud MD file might have like your brand voice, the different types of content you're making, what your brand is about, like your products, services, offers, etc. And the point is you write it once and then cloud code reads it at the start of each new session. And for marketing there you could have different use cases like SEO case studies, short form video scripts. Each one of those could be different projects or they could be different skills but they can each draw upon the Claude MD. So it's a kind of single source of truth for your project so that Claude can always reference it. And it is really worth it to set it up.
And my favorite trick for the claude MD is to continuously update it. So a lot of people do not do this. They make a claude MD once and then they just leave it. So my favorite trick to continuously update this memory is to ask Claude to do it. So I'll put a prompt based on this conversation. What are the most important things you should extract and update in Claude MD? So that way you're not manually sitting there updating memory. Remember, memory is just the folder that's like, here's everything about us that's important. So, we're going to have AI do it automatically based on our conversation. A lot of people honestly don't do this, but it's a massive source of leverage because then you have documentation about your company or specific project or specific thing you're trying to do and it evolves over time and it's highly tailored, highly specific to your company, your goals, your preferences. So, I do this almost every conversation. And so I have a skill called /learn in cloud code. And the skill basically tells claude reflect on this conversation and extract anything important that should be updated in claude MD or that should be made into a separate skill. So it's going to reflect on the conversation and be like hm let me update the memory which is over here or let me create a skill so that it's repeatable in the future. So that's my hack to do this and highly encourage everybody out there to view this. you're going to have these awesome Claude MD files that are highly tailored again to your specific business and goals.
So number six is repetition. So this is a surprising one. I'm actually most surprised by it. So I think it was Google who just came out with a research study that basically said repeating the same instructions to AI whether it's chat GPT cloud or whatever actually results in better answers. And intuitively this makes sense. Like if you are interacting with a team and giving them instructions, you generally want to repeat the most important thing so that it stays top of mind, right? So one way I do this, for example, in cloud code is I actually have a shortcut. All of my shortcuts have Q at the beginning if they're just casually defined shortcuts in my Claude MD file. But anyway, this shortcut I run it at the start of every Claude session and it tells Claude to go reread my Claude MD file. So I know that every session Claude is supposed to read the Claude MD file. It does do that. But I type this shortcut to force it to read it again. And I get better results that way. And now this research study has come out basically validating that approach. Repeating instructions, the most important instructions will give you better answers. So, what you might want to do for this is just repeat the most important things, which could be your goals, it could be certain key pieces of context, or it could be certain constraints. The key here is don't repeat a massive amount of information cuz that's not helpful either. You just want to repeat what are the most important things.
So, let's say you have a long conversation with chat GPT and it's pretty long. People have noticed with super long conversations, they tend to get worse. And that's because of that context filling up, you have so much information, AI doesn't know what to prioritize. So even in this long conversation though, here in the middle of it, you could say, "Hey, just a reminder, these are the things that are most important to me." Maybe it's like making income while you're still employed. Like that's the most important subtext or sub goal underlying your conversation. Maybe it's something else. Building a brand, maybe it's something else. whatever it is like remind chat GBT or whatever AI tool you're using and don't be afraid to repeat yourself multiple times.
One of my most used skills for development. I have a shortcut fault for it. It's called Qmin and I repeat it all the time. I basically spam it. So it asks Claude, review the code and make sure that it introduces minimal changes to the codebase. This is really simple. I'm essentially saying reuse existing parts of the codebase and don't add a lot of stuff. Don't import libraries. Reuse existing components and fe functions where you can. I don't want a lot of mess and slop added to my codebase. In a typical coding session where I'm implementing a feature, I will run this cumin shortcut at least two times, maybe three times before I let Claude implement the feature. I spend a lot of the time in planning and especially for development now that I have an existing complex codebase. My biggest concern is AI slop code that breaks a bunch of other stuff or code that I don't understand so that when I have to debug stuff or refactor stuff later I don't know what's going on. Those are my two biggest concerns. So I literally spam this again and again and within that conversation Claude eventually learns like this is really important to Sabrina. she really cares about uh making sure all changes are minimal and reuses existing components etc.
So number seven is planning and specifically I meant it in the cloud code context but it can apply to any AI you're using. So planning is like imagine a fork in the road and your choices are to plan or just go ahead and build it say in a development context. I personally have noticed far better results if I spend about 80 to 90% of my time in plan mode. So in cloud code there's actually a mode called plan mode. In my opinion it didn't work very well like even 6 months ago. So I'm just in a planning mode not literally using plan mode now. I think plan mode works fine. So I am literally in plan mode in cloud code and I spend 90% of the time there. This is where I use the techniques from number one and number two. So prompt engineering and sparring. So when I say plan, we are going back and forth trying to understand is this the right way to approach it. What if I do this over here? What about this over here? Did you consider this? In the context of building product, it might be well what percentage of users are clicking this button. What support tickets do you have come in that suggests we should change the feature in this way or change the copy of a button in this way? Are people confused and not clicking the button? But this is where I'm using these skills number one and two while in plan mode and I spend 90% of my time here. Only when I'm really happy with the plan will I switch to what's called the edits automatically mode in cloud code. And that basically means okay cloud will just we'll follow the plan and go implement everything in the plan.
And then another key thing here and although not strictly related to planning is I actually watch what cloud code does especially for development for marketing I don't to be to be honest I have all these skills for helping write the YouTube videos and I only look at the output however for coding especially if you have a complex existing product you can't afford to just push a feature you don't know how it was implemented down the road it breaks three other things and you don't even know why you just can't afford that situ situation. So, I actually look very closely and monitor what AI is doing when it's in edit automatically mode. The reason for this is because I want to abort if it goes down the wrong rabbit hole. So, AI is far from perfect. Claude and I will have this amazing plan to implement stuff and still for some reason it just decides to do something else over here for part of the plan. That's when you really want to watch it and make sure you are able to abort and stop before it goes down the wrong rabbit hole. then you have to revert all the changes and do everything again. So I pay very close attention to in a coding context. For other types of projects, for example, YouTube videos, then I don't pay that much attention to what it's doing. I just review its final output. But for anything really high sensitive, high stakes, you do want to watch Claude Code while it's thinking and make sure you can stop it before it pursues the wrong rabbit hole. Even if you've come up with the most amazing plan, sometimes AI just doesn't follow it for whatever reason.
Number eight is MCP. This stands for model context protocol. So the way to think about it is right now when you use chat GPT for example, you open up on your computer, it's thinking, you're chatting with it, you're brainstorming ideas. But MCP is what allows AI to actually do realworld work. So it can act on your real tools, your accounts, your data. Without MCP, you know, your conversation will chat will tell you what to do. Do this, open this app, steps at 1, 2, 3. But with MCP, AI can actually do those steps for you. So imagine claude or whatever AI you use over here. And because of MCP, it can use tools like your Google Drive or Air Table. It can use tools, let's say Slack. It can hook up to your CRM. It can do Google Sheets, like all kinds of tools that it can hook up to. And that's what allows AI to actually do work for you. The gap between these two things is the gap between like a consultant who's just telling you the things you should do versus an employee who's actually executing the tasks and saving you a bunch of time. You can connect Stripe, you can connect email. There's so many things that now you can connect to AI through MCP. For example, pull last week's sales from Stripe, summarize the support tickets from Intercom or Slack and draft a Monday morning report and put it in Notion or Gmail or wherever, right? It can actually do all of those things without you even leaving the chat. You basically hook up your tools once and then you can tell AI what you want to do with those tools.
So, I actually built my own MCP server recently for potato and it's pretty cool. So what it allows Cloud Code to do or anything you integrate with MCP is you can basically scrape Tik Tok, YouTube, podcasts, PDFs and websites. You can extract that content. You can generate visuals. So it's hooked up to nano banana and VO. So you can generate videos, images, infographics that go along with your social media posts. And then you can post or schedule everything to social media. And this is all within the interface of AI. You can just chat. In fact, I have a skill. I have lots of skills, but I have like one basic skills called slashpost and I can type anything slashpost to LinkedIn, Twitter threads. I have a live stream coming up like it's going to make a post about it. Generate an infographic because it knows from my memory my Claude MD file that I usually like an infographic with my live stream announcement. And then it's going to go post it. Claude will actually probably ask me to clarify, do you want to post it now or do you want to post it in the future? And then I'll say post it now. But this is just one example of the power of hooking up AI tools to your daily productivity tools that you actually use. That's the difference between using chat GPT. It tells you what to do like a consultant versus an employee or partner who actually then goes ahead and does the thing.
I'll give you a prompt. So, I know if you're not technical, you have heard of MC, we have no idea how to set it up. It might be intimidating, but I promise once you set it up, it's really mind-blowing because of all the possibilities it opens up in terms of automating work. So, one prompt here is here are the apps I use and then just list some apps you use. So, that's the first part of the prompt. And then the next part is which ones have MCP and then you can just say walk me through it. So, here what we're doing is we're telling AI here are the apps you use. So let's say it's notion, Gmail, Air Table. Now you're saying which ones have MCP servers and then help me set them up or walk me through how to set them up. And Claude will literally walk you step by step through setting up each of these MCP servers. And you just have to follow those directions. And then once you set up those MTP servers, you can do all kinds of things like, hey, read my emails and surface any urgent issues. This is one I use. So read my support tickets and tell me what is urgent. Are there any urgent bugs? Are there any clusters of feature requests, etc. In fact, I I actually have a GitHub action now that just does this daily. But MCP is the first step in figuring out all these different ways you can further automate your flows.
Number nine is what I call stacking. And I'm specifically referring to stacking skills like Claude skills with MCP. So now we're combining these two things together. And it's really, really, really powerful. So the way to think about this is if AI is a robot here we have a robot skills and MCP the way to think about it is skills tell your robot what to do and MCP is that how it actually does it cuz you can talk to AI and it'll tell you what to do but how do you actually use these different tools? That's what MCP is for. And then when you combine it with skills, skills tell AI here's exactly how to use these different MCP servers and tools to accomplish this goal. So when you're able to combine MCP plus skills, it's the closest approximation to an AI employee that I've ever experienced because then you can explain what you want to do and then you can give AI the tools it needs to actually get that work done because it knows what to do. It knows all your preferences and everything. That's the skill. It knows what good looks like. That's part of the skill. And then by giving it access to tools via MCP, it can go ahead and actually do those things.
So I'll give you a real example. I recently refactored a lot of how I make content. So I have a skill called slashcross. So I'll film my 20 Tik Tok video drafts. I have other skills to help me brainstorm ideas, refine the hooks, etc. But this scale is basically when those 20 Tik Tok drafts are ready to go, what do I do next? This will actually search my local Google Drive, it'll find all these latest Tik Tok videos that are ready to go. It will transcribe each one so that I have the transcript for reference. This is really helpful in the future if I want to remix this post into another type of post. Then it's going to write platform specific captions for each video. And then it's going to for some videos generate visuals or transform them into visuals. So what I mean by that is you could take a video and transform it into an infographic. So it just depends on the platform and then it will use potato mcp to actually schedule everything out for the week. So in this one skill I've told it what to do. The skill is you're going to do xyz. You're going to do 1 2 3 4 5. Like this is the process. But how it's going to do that is through these MCP connections. So Plot MCP I also have it hooked up to Google Drive so it has permissions. So now we have essentially like a mini employee that knows what to do and has the means to do it. So we can go ahead and do it. Oh, and the last step is it logs everything in Air Table. So it updates Air Table and this is also an MCP connection. This is something even just a few years ago everyone was doing basically manually very messy as well. It's like hand off this, hand off that, update the systems of record, etc. And yes, you could build this in nadn or make.com like a workflow automation tool. But the flexibility of the skill is that I can add any additional context. So I could say /cross, but let's say I want to make a carousel for the prompt videos. You can type a full prompt along with the skill so that Claude has context on any unique things you want to do. Maybe you want to process 18 out of 20 videos this exact way and schedule it in your next free slot, but maybe two out of these 20 videos you want to publish right now. You can just say that in this skill and it will adapt accordingly. That is not something that like traditionally a workflow automation like any or make.com would be able to do and that's why it's so powerful. That's why this is the closest approximation to a mini employee because it's not just so rigid. You can be like use this skill but also take into account these exceptions and then go ahead and do it. Or let's say you want to do it just one at a time and get your per permission and approval before proceeding with each one. You can also just type it into the prompt and it will do that. So, it's really powerful if you can stack skills with MCP. It's kind of your mini employee that now knows what to do. It knows the playbook and it has the skills or it has the tools through MCP to be able to like actually carry out the tasks that are in your playbook.
Number 10 is using the mobile apps. highly highly underrated but obviously every major AI tool has an accompanying mobile app. So Chachi BT also has a mobile app. Claude has a nice mobile app. So the cool thing about the mobile app is you can basically sync your phone. So this is supposed to be a phone with your laptop. This is supposed to be a laptop and they can actually stay in sync which is really cool. So you can start a conversation on your laptop and then if you're on the go or you're on the in the bathroom or whatever, you can continue that same session on the phone. And so that's really cool because you don't have to start from scratch. You can continue that same conversation. And with Claude Code, they just released a tool called Remote Control, which basically allows you to run Claude Code from your phone. The powerful thing about this is Claude Code has access to your local computer, your local files. So on the phone, you can search for that photo I took yesterday and post it on social media. It'll actually go to your photos, find it, confirm it with you, and then it can use the MCP tool to post it to social media. And the nice thing is you don't have to dangerously skip permissions, right? In Cloud Code, if you're like stepping away, you can just use remote control and still get all the same permission notifications. That way you still know what Claude is doing and it's not running rogue on your system or something just because you want to get work done because you're stepping away. Here with remote control you can approve every single thing that it's doing. You don't have to enable dangerously skip permissions and you can just continue the session. And it's really powerful because you can say, "Hey, open up like that CSV I downloaded yesterday. Scrape this information and make a new CSV and put it in this folder over here." It can do all of that on your computer even though you're telling it from the phone. And so I love this new feature. So yeah, definitely check that out if you haven't already.
Number 11 is GitHub. Especially if you're using Claude Code, it can get like messy how to organize everything and save everything. So developers already know about this tool, but it's basically like the ability to save your work, especially for anything involving code. But if you use cloud code, I really recommend hooking it up to GitHub because that way you can just save all of your work. So the typical workflow is like you'll have what's called like a main branch. Okay? And then let's say I want to improve a project. For example, I have a project called content and that's what contains like my YouTube writing skills, my crossosting, my posting skills, etc. And let's say I want to improve it to make it better. So in GitHub, you would create what's called a branch. And it's basically a nice area for you to experiment with stuff like you can add stuff, you can delete stuff without affecting your main copy over here, right? So you can mess stuff up as much as you want in this branch. Even if you mess it up, that's okay. You can just go back to your main copy over here. But let's you do make a lot of progress on this branch. You add a bunch of stuff. You're super excited about it. You can basically merge or replace this main copy. So this will be your new main copy over here after including all the changes from your branch. It's not just like saving it at a point in time. It allows you to have a playground to test a bunch of stuff. If you don't like it, you can just go back to the main branch. If you do like it, then this basically will update your main branch to incorporate the changes that you just made. So you can just think of it as like a really fancy floppy disc, I guess. And yeah, I know
It can seem intimidating, but this is when you use AI as your 24/7 tutor, like we talked about. Just like literally ask AI, "What is GitHub and is it useful for me?" If you're using Cloud Code, you should definitely use GitHub. If you're still using ChatGPT or Claude in the web, I don't think you need it yet.
So this is really for people who are already progressing to Cloud Code, but you don't have a traditional developer background. So you don't realize like there are tools like this that exist that that could massively help you. There are even people that use GitHub to manage their different cloud settings cuz you can have a global user cloud setting for your user and then you can have custom settings for each project. And it's all nice cuz it allows you to just save all of that.
So I started this off by saying like, "What are the 12 productivity hacks that really will separate you from 99% of the population that are still prompting ChatGPT like, 'Make it better'?" And we talked about prompt engineering, using AI as your sparring partner and your 24/7 tutor, things like skills and memory and repetition as ways to to become more productive, planning MCP, stacking skills plus MCP to have a mini-employee. This is the step everybody should want to get to because it's truly mind-blowing what becomes possible when you reach this step. So I encourage everyone to at least make it to number nine. Number 10, mobile apps, Cloud Code, remote controls. Probably my favorite feature they've released in the past probably six months, hub to save your work and be able to safely experiment.
And the last one is putting in the reps. So, I know that sounds very guru and motivational, but a lot of people consume my content, but they don't actually do it. So, it would be like watching a movie of people working out and just like never actually working out yourself. Would you seriously expect results from doing that? No, of course not. That doesn't make any sense. It's the same with learning AI stuff and learning AI tools. You have to actually put in the work. So, watching a video honestly doesn't count. I appreciate everyone who's on this live stream, but unless you're literally following along, you haven't started. You have to put in the reps. It's going to suck. You're going to fail. You're going to mess up. You're going to not know what to do. You're going to get stressed. You're going to accidentally delete stuff. You're going to have to restart projects about five times at least. But that's all part of the work. That is the work of learning something.
So, I just really want to emphasize this. The people who really, really, really succeed in AI today are just putting in reps at a way crazy volume and pace compared to you. Seriously, I've been using Cloud Code for the past year for hours a day. That's what I mean by reps. I just sit there and use it for hours and hours every day. Watching a tutorial without following along doesn't count. Go back to the gym analogy. If you were just watching people work out without actually working out, you would not expect to have any results, right? So, just take that same mindset when it comes to learning this AI stuff. The real way you're going to learn it is by trying all of these things that I've talked about today. And this is why I focus as well on the high level. There's about a billion tutorials for each of these things on YouTube. If you don't sit there and actually do the work in the tutorial, it doesn't matter that you passively consumed it because you're just going to forget. Actually doing the work helps you not only remember, retain the information, but it helps you like have an intuitive understanding of how to interact better with these AI tools. And so don't forget to do the reps. That's number 12.
Just wanted to leave on that note. All of this is amazing. I'm glad everybody stayed and listened, but you got to do the work and follow along the YouTube tutorial. If you get stuck, use AI as your 24/7 tutor. As you build confidence, start creating some skills. Hey, we're having a lot of conversations, Claude. What are some skills you can make from this conversation together? You can have Claude create your entire CloudMD file for memory. MCP is going to unlock like a huge, incredible layer of productivity. And then if you can stack MCP with skills, you will be able to literally have mini-employees. So yeah, just, just do that. All right. Yeah, thanks everybody for watching.