You're Not Behind (Yet): How to Learn AI 11 Minutes

0:00 I spent the last 2 years obsessing over every single AI tool that came out, but one thing that I learned from this is that the whole year behind idea is complete nonsense. Because once I stopped chasing every new release and looked at how some of the best people were actually getting good with AI, it

0:14 came down to three very clear phases that nobody really talks about. That's why in this video, I'll lay out all three of them so you'll know exactly which phase you're at and how you [music] can reach the third one which almost nobody is at. And to understand why nobody really reaches that third

0:28 phase, you first have to understand the concept we're actually aiming at. It's called AI fluency. It works just like being fluent in a language where you're not translating inside your head, you're just speaking. And being fluent with AI is the gap between the people who feel stuck and don't really know what to do

0:41 with it and the ones who are using it in ways that are completely changing their lives. And almost everyone in that first group assumes the people in the second group just learned a couple of secret prompts, but they didn't. They just went through a process the first group never did. That process is a staircase with

0:54 three phases and you can only climb them in order. Phase one is understanding how AI actually works. Phase two is building systems with it. And the final phase, phase three, is the one almost nobody reaches, which is complete self-sufficiency. The reason most people feel permanently stuck is that they

1:07 don't even know the other two steps exist, so they spend years perfecting the first one. But now you do, which means we can start building those foundations beginning with the first step of the staircase, understanding how AI actually works. Because without that foundation, you'll never be able to

1:21 climb to any of the next steps. You just lack the fundamental knowledge of how the whole thing works. Now, by the way, I actually made a free resource breaking down every single one of the phases and exactly how you break out of it. And you can grab that completely for free in the

1:32 description below. At its core, AI does one thing. It predicts. When you give it context, it works out the most likely next word, adds it on, and then runs that loop again and again until the entire answer is built. Those words, or chunks of words, are called tokens. They're basically just pieces of words.

1:48 And there's a little randomness in which one the AI picks, which is why the same question can come back slightly different each and every time. You've probably noticed it yourself. And it happens because it's not actually looking anything up. It's just guessing very, very well. And if the model is

2:01 just guessing, you can actually start building on top of that and become scarily good. Because the AI has two completely different kinds of knowledge, and they behave nothing alike. The first is what it learned through training, when it read a massive chunk of the internet and soaked up [music] the

2:13 patterns. That knowledge is now built into the model, but it's blurry. It's more like a vague memory of everything it ever read than an exact copy. And it's frozen on the day training stopped. [music] So, anything recent or anything really specific is something it only half remembers. The second is a kind of

2:27 sharp memory that lives inside the context window. You can picture it as a box with a fixed size. [music] And whatever you put in that box, whether it's your question, a document, or the conversation so far, is what the AI sees clearly. The catch, though, is that the box is pretty small and temporary. And

2:41 if you fill it up in a long chat, the older the context is, the more likely it is to completely fall out, which is why you might notice the AI seems to forget stuff. The oldest information gets buried so deep [music] that it basically gets skipped over in favor of whatever is most recent. And if you close that

2:55 chat, the box closes with it, meaning every conversation starts new. Now, where it really starts to get interesting is that the box isn't actually sealed shut. The AI can be handed tools to reach outside it, like when you ask it to search the web. It can also pull files from other apps, and

3:08 whatever it finds gets dropped straight back into the box as that same sharp live knowledge. [music] That, by the way, is all an agent really is. The model in a loop, reaching out, pulling things into the box, and acting on them. But, we'll cover that in a bit. Now, every frustrating thing AI does falls

3:21 right out of this loop. It makes things up when it's running on that blurry frozen memory with nothing good in the box to pull from. And it sounds generic when the box is empty, because the safest guess is almost always the most average one. And the average is what it falls back on when it has nothing else

3:34 to guide it. And so, to get past this step, the one thing that'll make the biggest difference is giving it the right context, because your output is only ever as good as what you actually put inside that box. And that completely [music] changes how you should think about AI, because you should actually

3:46 stop describing what you want and start showing it the real thing. Don't tell it you're writing is casual or formal, hand it three things you've actually written. Don't explain your business, give it the real documents and let it figure them out itself. And once you see AI as blurry memory plus a box you can fill,

4:00 you can now start building around it. And that's exactly what we're going to do in the second step. And the single most important thing you can do to actually learn AI is to use it, to take the foundations and build real solutions to your real problems. Because building systems with AI comes down to one big

4:14 shift. Instead of using it one task at a time in a chat window, going back and forth, you build a system. So, rather than doing the task yourself, you build something that does the task for you. And then you walk away completely or just step in when you actually need to. And once you do that, it keeps paying

4:27 you back over and over without you ever touching it again. And spotting what's worth turning into a system is incredibly simple. Look for the things you do over and over in pretty much the same way. All the repetitive, predictable, or slightly boring jobs can be handed off. And honestly, they should

4:40 be. The judgment calls and the creative stuff, you keep. You're not automating your whole life, you're deleting the busy work that quietly eats your entire week. So, there are basically three kinds of systems that are worth building with AI. And each one hands the AI more of the work than the last. The first

4:54 type of system is the assistant that already knows your entire world and can help you accordingly. So, let's go ahead and start with that one. For this, I'm using Claude, but any model with the same features works and I'd really recommend the desktop version. In the top left, click over to co-work, then

5:07 click projects. At the top right, [music] hit create new project and start from scratch. I'm naming this one the assistant. Then you give it a brief instruction of what you want it to do. And here's the most important part. You're not just dropping a few files in, you can give it full access to specific

5:20 folders on your computer to work from. For me, I want help writing my email list, so I'm pointing it at a folder full of my old emails and the ones I'm working on now, so it can suggest fresh ideas, flag what I haven't covered, and call out what I've overused. Once that's set, I click create. Inside the project,

5:33 I've got all my past conversations and everything it can see, and I can edit the instructions anytime. So, let's run a prompt. I'll ask what the most covered topics on my email list are. And look at that, it's not guessing. It read my actual archive and told me straight up which topics I lean on too hard and

5:47 where the gaps are. But, notice it still only moves when you ask. Now, the second system is the worker. This is an agent, where the model runs the loop itself. The idea is you set it up once, and then it runs completely on its own. You hand it a repeatable job, and it just does it

5:59 on a schedule in the background. You can think of it as your own personal virtual assistant. Since we want it running without any input from us after setup, I'll click into the scheduled window on the left. Here, I can either have Claude create a task for me, or set one up manually, and I'll do this one manually.

6:13 I'll give it a name, the researcher. For the description, I'll paste this in. For the actual prompt, you want to keep it pretty slim, because the models are good enough at plain language now that they fill in most of the blanks themselves. So, I'll use this, and in the bracket, I'll drop a random topic. Let's say new

6:26 dishes from different countries. It'll go pull real sources, write a one-page brief with the key facts, and save it in whatever folder I choose. I'll make a new folder called food, set the frequency to weekly, save it, and now it's live. Here's an example of what it gave back, and that's the part that gets

6:40 me. I did nothing after setup. It goes out every week on its own, does the research, and leaves a finished brief sitting in my folder. Now, this final one is probably my favorite. The first two systems were about using AI to handle stuff you already do day-to-day. The builder makes AI create things you'd

6:54 never be able to make yourself, things that never existed before, actual tools built for your exact life, just from you describing them. For this one, we switch over to Claude code, which sounds intimidating, but honestly, the only code you'll see is the code the AI writes and edits for you on screen, and

7:07 the payoff is massive. One of the coolest things you can do here is build custom tools for your own life. For a few weeks, I'd wanted something simple to track my finances and save a bit of money just by being more aware of where it goes. So, I'll paste in this prompt. But, before I hit enter, let me show you

7:20 one of the best parts of Claude code. You can pick the mode it runs in. For any build that needs real thinking up front, you want plan mode. It's like how we work. If you think a problem all the way through before you start, you're way more likely to get it right, and it's the exact same with AI. In plan mode, it

7:34 maps out the whole build before it writes a single line, so the result comes faster and with fewer problems. So, I'll switch to plan mode, answer the couple of questions it asks, and hand it my spending for the month. Now, of course, I'm just using an example here, but as you can imagine, that would not

7:46 affect the result. And it builds me something like this, and I made that just by describing it once. Once you can build systems, AI stops being something you open and use, and becomes something that quietly works for you in the background. But every system runs on tools that exist right now, and these

7:59 tools change every single week. So, the real question is whether you can keep building when everything around you changes. You can, but not the way most people try to, because keeping up was never about chasing every new tool. It's all about where you choose to stand. Because everything in AI really sits in

8:13 two layers. There's the surface, where all the tools, models, and apps that change every single week. And underneath all of that, the foundation. How the thing actually works, and how you think about using it, which honestly barely moves at all. People feel behind because they're standing on the surface,

8:26 measuring themselves against a layer that's built to outrun them. Because there's always a newer model, and always one more thing you still haven't tried. But the people who never fall behind live on the foundation, and from there, every new tool is just a fresh take on something they already understand. And

8:39 there's a dead simple way to always stay on that foundation. One that almost nobody actually does. You build your own test. Pick five things you genuinely do with AI. Real tasks. Maybe it's a piece of writing, a workflow you repeat every single day that you'd love an agent to take over, a part of your job, or even

8:53 an assistant for all your school work. Save them all in one place. Then every time a new thing gets released, and everyone loses their mind over it, you don't have to interpret the hype, or trust the benchmarks, which are wrong about half the time anyway. You just run it against your four tasks. Actually use

9:07 it, and judge whether the results are worth it. Takes about 20 minutes. And from that point, you'll know whether it actually makes your life better, or just sounds like it does. That one habit takes you out of the audience, and puts you in the seat of the person actually deciding. And just like that, you're not

9:20 chasing releases anymore. You're testing them. And once you're testing instead of chasing, every new release gets a lot easier to handle. You just check one thing, whether it actually changes how your specific system works. And almost every single time, the honest answer is no, it doesn't. The foundation stays

9:35 exactly where it was. So, you take whatever's genuinely better, fold it into what you already do, and keep moving. But, there's still one part I can't solve for you in this video. Because everything I just walked you through, all three phases, that's the map. You can see the whole staircase

9:48 now, the foundation, the systems, and the self-sufficiency. And understanding the map already puts you ahead of most people out there. Actually building the foundation, building the systems, and getting enough reps that your judgment turns sharp, none of it happens from watching. It happens from doing, day

10:02 after day. And almost everyone who tries to do that alone never actually pulls it off. Not because it's hard, but because there's no path in front of them and no one doing it with them. And sure, you could piece all of this together on your own. It's completely possible. But, doing it that way takes so long that by

10:16 the time you finally get there, you're behind anyway, which is the exact thing you were trying to avoid, which is exactly why I built AI Fluency. It's a school where each of those three phases becomes a month, around 30 daily steps each, all part of one 84-day road map you build alongside a whole community

10:30 doing the exact same thing. [music] So, if you actually want to become squarely good with AI in the next 12 weeks, get hours of your week back, and completely change how you work with AI, go ahead and sign up using the first link down in the description below. Thanks for watching, and I'll see you in the next

10:43 one.