Full transcript of “Automating ANY Process: 5 Levels of AI Automation (Full Guide)”. Click a timestamp to open the video at that moment. Need another video? Use the YouTube transcript generator.
0:00 I hate seeing people waste months and thousands of dollars on failed AI automation systems. Sadly, most of those systems were doomed from the start. I want to make sure this doesn't happen to you. So, I tested every level of AI automation so you don't have to. From simple prompts to complex vibe coding projects that are changing the world to the sad truth about some of these magical AI agents that everybody's hyping, this framework separates what flies from what flops. Let me save you months of frustration and thousands of dollars chasing shiny objects like this one. This is an Eola file, a steam powered device from ancient Greece, often considered the very first automation. It's crazy to think that something as simple as steam powered the industrial revolution and really brought us into this modern age. I think about
0:46 simplicity a lot. It's a major theme of this channel as I always tend to over complicate things. When I started AI coaching about two years ago, I would spend days coming up with complex and convoluted prompts for my clients only to have them point out that their seven-word prompt performed much better. Simple prompts in aggregate can achieve some really complex things. Here's a tool called GPT for sheets. It's pretty easy to install. Let me show you how it works. In this sheet, I'm trying to look through a bunch of keywords to find out if they're relevant for a car wash. In this first column, I'm just going to enter a simple prompt. Is this keyword good for a car wash? In the second column, I'm going to enter all of my different keywords here and drag this prompt all the way down. And here is the formula that we use to combine these. So I do equals to GPT open brackets prompt
1:34 space keyword. And this loads up and tells us if this keyword is relevant. You can then just drag this down and have it analyze all of your different keywords. And there you have it. This would have saved me so many hours when I was an SEO consultant. But you can do so many different things with this tool. Here's another example where I've loaded in a bunch of my favorite songs and then this prompt, hey, can you give me the chords to these songs and this GPT for sheets has returned all of the chords to my favorite songs. So remember, these large language models are a huge database of knowledge that you can pull from with these simple prompts. I use this at least once a week when I'm loading up a new resource for my Patreons. I have this spreadsheet that has all of the different resources
2:19 available to my patreons and I use GPT for sheets to fill out the description, category. I use it to do my tagging, etc. Speaking of the resources that I create for my Patreons, I'm jumping into the cheat sheet here. I create a cheat sheet like this for every single video that I make. This one includes over a hundred use cases for this GPT4 sheets. So, if this is something you're interested in, there's a link in the description. There's also some coaching options in there as well. So, remember, never underestimate the power of a simple prompt, but simple prompts won't do everything. And that gets us to our next level of AI automation, the big ass prompt, or BAP as they are called. I find it's best to use a massive prompt when you need more control and more
3:05 specificity over what you're creating. Typically, this includes providing examples of what you're trying to create, at least three. This is called few shot training and this is how I create my newsletter. Every week I use this B which is basically still a simple prompt here but it just includes examples of my favorite emails that I've ever sent out. I just copy and paste this in into chat GPT. After I copy and paste that in, I paste in the entire guide that that particular newsletter is about. And I separate that guide with these three little quotation marks. This is a helpful way to tell the large language model what part is the prompt and what part is the context that they should be focusing on. So I just do a select all that it that grabs the entire 20 pages of this guide and I copy and paste that right into the prompt as
3:52 well. I use the 01 pro mode for this and it returns some amazing results that need very little editing. Bam. Newsletter is done. Let's see how many words this prompt is. Oh mama, that prompt is over 10,000 words. If you really want it to follow your tone and voices is a great way to do it. But these long prompts can be valuable in a lot of other ways as well, especially if you're trying to follow a similar format for legal documentation, market analysis, product descriptions, technical documentation, project briefs, and even in and throughout the interview process. There's another type of extensive prompt that I think can be incredibly valuable when it comes to process automation that deserves a category all on its own. So, here we are at level three. This is a sequence of prompts and I've created a
4:38 lot of videos about this. I'm going to link to some of those in the description, but this is one of my favorite ways to use these custom GPTs, claude projects, and perplexity projects. Let's say you're trying to automate a process that requires a bunch of different deliverables, a bunch of different things to be created. This sequence of prompts is the best way I've ever found to do that. The general overview is you want to work with the large language model to first document your process. Second, you want to convert that process document into a series of prompts that help you work towards your goal and automate your process. Third, you want to take that list of prompts, that sequence of prompts and convert it into instructions that a custom GPT or a cloud project can use. Then you simply load those custom instructions right into that custom GPT
5:25 or that cloud project and let it walk you through those steps creating all the different pieces. Let me show you how this works. Let's say we're trying to plan an event. I just entered this quick prompt right into GPT40 asking for a step-by-step process for planning a professional meetup. And now I'm grabbing this prompt to convert that process into a prompt sequence. Drop that right in here. Okay, that's looking good. And now comes the tricky part of converting this into instructions. This took me a lot of trial and error. So this custom GPT that I've built comes in really handy. Open that puppy up right here. It's my automate any process. And I'm just going to drop in that uh prompt sequence right in here and have it generate the instructions for us. Awesome. Now we can just select all of
6:12 these instructions, copy those. To create a custom GPT, just go to explore GPTs, create, go to this configure tab, and drop those instructions right in here. Give it a little name. You can give it a little logo if you like, and click create. And that should walk you through these 10 steps of defining the purpose, identifying the target audience, setting the budget, choosing the time and date, finding the venue, planning the agenda, inviting the speakers or guests, promoting, managing the RSVPs, and day of preparation. This is my four-step automate anything process. I've got a lot of details about that in the cheat sheet, including that link to the custom GPT that creates the instructions. These custom GPTs and cloud projects can do a ton, but there's
6:57 still some limitations as they're not interacting with the outside world at all. That's where the next level of AI automation comes in. And that's where things get a little fuzzy for me. This level four of AI automation includes tool use and agents. And when I first started to see these tools come on the scene, I thought we had found the holy grail. But after many weeks and months of experimenting with these tools, I find that they fall down a lot more than they work. Unfortunately, maybe your experience is different. If so, please let me know in the comments. These tools can be useful for simple things like scraping websites or moving one piece of information from one part of your tech stack to another, maybe like from a Mailchimp to a CRM or something like that. But when it comes to this idea of
7:43 agent swarms and agents controlling agents autonomously, I found that my experimentation has really fallen flat to create anything that is practical and that can be used over and over again. So if you're struggling with that, do not feel bad at all. I honestly think the technology is not quite there yet. If you've got something working like that, something that you're using in a practical way that's actually working consistently, let me know about it because I really don't think that that is possible right now. And if you think about it, even these major AI labs are just starting to build agents that truly work. Some evidence of this is the deep research that Chat GPT put out. That is a legit agent that is working well. Also, their operator tool is working
8:30 better than expected. But this is the part if you're trying to automate a system that does a lot of this thinking for you, you may need to pump the brakes and wait a little bit. One exciting glimmer of hope in this area is anthropics MCPS. Now, this is a whole concept that we don't have time to get into, but this is basically a way for these large language models to control these tools in a uniform way. So, I'm really excited to see how that progresses. I think that could unlock some major tool use and agentic uh use cases in the future. After wasting a lot of time with some of these automation tools, I felt like I was at the end of what was possible with AI automation, at least for us mere mortals who don't work at one of these big AI labs or something similar to that. Then I started to play
9:15 around with another set of tools that really got me excited. And what I'm about to show you is really what separates the world from those who are just, you know, talking about AI, using it here and there to the people that are really transforming the world with this technology. This is level five. Code with AI integration. I'm sure that you know that you can use AI to code, but what a lot of people haven't understood is that what you code can also have AI at its heart and its brain. Just take a look at what this guy Peter Levels is working on. and he's building an advanced flight simulator that people can actually play as he's building it. He's also built and heavily monetized an AI photographer app and he's doing a lot of this using vibe coding which is basically just using the English
10:01 language with very little understanding of code in and of itself. If you've often said, "Hey, coding is not for me." I'd really encourage you to reconsider that now that these tools have become so much easier to use. You may find these easier than some of the automation tools that I just talked about, especially if you're just building a piece of software to automate a process for yourself and not planning on selling it to the world. The basic process for this comes down to using a large language model to decide on what features your software should have and then having a discussion with it about what technology you should use. This could include Python packages, etc. From there, you get the core functionality built, then start adding features. There's a ton more in the cheat sheet about this process as well as looking into Peter Levelvel's take on
10:46 how he builds software. If this is something that's interesting to you, you might want to look into Cursor. This is probably the most well-known AI tool out there. My personal favorite is Replet because it makes it so easy to launch and host whatever you create. Windsurf is a competitor to Cursor that a lot of people are loving. There's also tools like Versel and so many others that can help you build custom code to automate your different processes. But the crazy thing is is that this all builds upon the simple principles we went over in step one with those simple prompts. When you're coding, sometimes a simple prompt is the best way to go. The cheat sheet is absolutely packed with all kinds of resources that will help you improve upon what I've just gone over. It includes detailed notes on starting your first AI code project, which I think any serious AI automator needs to take a
11:32 look at. Like I said, there's a lot more about Peter Levelvel's project in there as well. It's over 20 pages long and there's over 120 other cheat sheets just like it, all immediately available to anybody who supports this channel on Patreon. There's a link in the description to that. There's also some coaching options there as well. Next, I want to congratulate you on investing in yourself and taking the time to learn all this stuff. Like I mentioned earlier, that multi-step prompt sequence is one of my all-time favorite techniques. I've got a ton of use out of that. So, here's a video that dives into that with one of my all-time favorite tools, Perplexity. This is the next step in your journey to mastering AI. I'll see you over there. Make your dreams come true.
