How to Actually Use AI Tools in 2026 (Full Guide)

How in the world are you supposed to keep up with all of those AI tools when a new one drops basically every single day? Well, the answer is you don't have to, because almost every single one of those tools fits into just five core categories that haven't changed in years. And in this video, I'll walk you through all five, how to actually use each one while building a tool I genuinely need live, so that by the end, you can do the exact same thing for your own situation. With each category we cover, we'll take our project and upgrade it so that by the end, we've got something genuinely useful, the kind of thing that looks like it could charge 20 bucks a month.

So, let's start with the first category, the one almost everyone ends up running through, your thinking tools: Claude, ChatGPT, Gemini, Grok, and all the rest. This category is obviously used by everyone, and it's also the one almost everyone uses the wrong way, because most people use it like it's Google. They ask a question, get an answer, and that's it. That's the entire relationship. And when that's all you do with it, AI honestly does start to feel a little overhyped. What you should actually be doing is stop asking it questions and start making it build things for you instead of just looking stuff up. And the single best way to do that is something most people rarely try: making the AI write its own prompts. Instead of sitting there trying to engineer the perfect instruction yourself, you just describe the job in plain English and let it write the prompt, because it knows what it needs better than you do.

And that's exactly what I did for this first category. So, let me head over to Claude. I personally use Claude because I think it gives the best quality results. But if you prefer ChatGPT, Gemini, or anything else, those work just as well. I'll go with the desktop version. And in a new chat, I'm going to paste in this prompt. And it lays out the plan, a tool called "Worth It" that helps me make a research-backed decision on whether something's actually worth buying, for exactly those impulsive moments when I want to upgrade my setup. The first thing it does is check that I'm happy with the overall plan. And honestly, it looks spot-on. I get a breakdown of the questions it'll ask, the verdicts it can give, and how it'll look clean and minimal on a white background, which I'm all for. So overall, I'm really happy with it.

And with that, we move straight into the second category, the one that takes that plan and turns it into a real working thing: software and building tools. This is the category that 5 years ago would have taken a developer and a few thousand. Now, you just describe exactly what you want. The AI writes all the code and does all the hard work, and the only thing left for you to do is be creative. By the way, I actually made a free resource down in the description below that will break down this entire process and give you the master prompt, as well as some other secret tips on making this whole process better. You can grab it completely for free in the description below.

First thing worth knowing is that these tools split into two types, and picking the right one saves you a world of pain. On one side, you've got what I call the "describe it and it appears" tools, stuff like Lovable, Bolt, and Base 44, where you just talk to it in plain English and a few minutes later you've got a working web app. Perfect for quick tools, prototypes, and web apps. On the other side, you've got the agentic coding tools, your Claude Code, Cursor, and Codex. These work inside your actual computer, creating files, running things, and building with real depth. And most of the time, using something like Claude Code or Cursor actually works out cheaper than Lovable or Base 44, just because of how the costs work. The "describe it" tools mostly run on credits. You get a set number of builds and edits a month, and since making something good takes a lot of back and forth, those credits run out pretty fast, and you end up buying more. The agentic tools like Claude Code usually run off an AI subscription you might already be paying for. So once you're on the plan, you can keep iterating without a meter ticking down on every change. The moment you're building more than one or two things, the agentic route almost always comes out cheaper.

Then there are three things that separate a clean build from a really bad one. First, before you build anything, you have to make a plan, which is one of the first things we'll do here in Claude Code in a second. That's as easy as selecting plan mode, and anywhere else you just tell it to map the whole thing out before it writes a single line. Because a model that thinks the build through upfront makes a fraction of the mistakes, exactly like we do when we plan before we act. The next thing is always keeping the core of your project's context in a markdown file, usually called Claude.md, Gemini.md, or Agents.md. It's basically a file that always goes in front of your prompt, giving the AI all the knowledge it needs about how your specific project works. So instead of re-explaining your project in chat every single time, you keep the whole plan and all the details in one file the tool can always look at. And you don't even have to write it by hand. You go back to a thinking tool from the first category, have it write the MD for you, and hand it over. Last, but just as important, the agentic tools—Claude Code, Cursor, and Codex—can run real commands and change real files on your machine, which means security is something you actually have to watch. For most of the work, I'd recommend staying in the loop. Glance at what it's doing before you approve it. Don't let things run before you understand them, and never paste in real passwords or API keys, because there's always a chance they leak.

So, with that in mind, let's get into the second phase of the prompt. I head back to the Claude chat, tell it to move to the next phase, and it gives us a master build prompt we can paste into Claude Code or any other tool, Lovable, whatever you're using, and it'll give out a really decent looking website right off the bat. So, I copy that over, and because I'm using Claude, I can just click over to the code icon on the left, switch into Claude Code, make sure this chat is connected to the correct folder I created for this project, paste the prompt, and ensure plan mode is selected, and click generate. This runs the whole prompt through a planning phase first, so the AI figures everything out before it writes anything and presents us with a plan. And there we go. We've got our plan. Normally, I'd review the whole thing and make sure it makes sense. But for this example, it all looks really good. So, I'll let it proceed. I'll sit here for a couple of minutes while it runs through everything, approving steps as they come up. And at the end, we get something that looks like this.

Now, one of the best ways to get great results fast with these tools is to turn on the audio recorder in the bottom left here. And as you go, just talk through anything that's not working or you change. And in the end, we get something that looks like this, which already looks great, but it's still lacking something. That something is a brain. Because even though our tool technically already works, right now, it doesn't actually know a single real thing. Ask it about a product and it would just be guessing or it wouldn't work at all. Which brings us to the third category, and the one I personally love the most: research and knowledge tools. This category fixes one of the single biggest weaknesses in AI, because a normal chatbot, as smart as it sounds, is really just optimized to be decent at everything. It's a generalist. And a generalist will never beat a specialist. A tool built specifically to read real current sources and show you exactly where every answer comes from.

There are three actually worth using. The first is Perplexity, a search engine that actually answers your question and hands you the sources in the easiest way possible. The second is Deep Research, the mode most large language models have rolled out now. It doesn't work quite as well as Perplexity, but you give it a huge question and it spends a few minutes reading dozens of sources and comes back with a full written report. That's not really what we need here, though. And honestly, I rarely reach for it, especially since the third one exists. And that's Notebook LM, where instead of searching the whole internet, you point it at your own documents, your notes, your sources, a stack of reports, and it only answers from those. So, it physically can't make things up. Now, the rule for all three is dead simple: if the answer actually has to be true, current, or specific, you should be using one of them. And we're going to build that right into our website.

So, I head back to our master prompt, tell it to continue, and it drops me into the third phase. This phase also includes some instructions for you to follow. But what I found works best is to just copy everything over, get into Claude Code, paste it in, and tell it to handle most of it itself. This is honestly one of the best perks of using Claude Code, Cursor, or Codex over the other tools. You get way more customizability without losing the ease of use, because 90% of the time the tool either hands you the exact step-by-step instructions for anything it needs or just does it for you. So, I hand the prompt to the tool and come back once it's done. After a quick stop to grab a Perplexity API key from their platform, we're met with this result. It's a pretty cool workflow. It walks me through the exact product I'm considering, what I need it for, then runs it through Perplexity to see what prices are actually out there right now, flags any discount, and finally gives me a verdict on whether to buy it or wait for a better deal. This honestly took me no time at all. So far, we've spent just a couple of minutes going back and forth and letting the AI run in the background.

But there's still one big problem we haven't solved yet, and that problem is the design. Because right now, Worth It is a very plain and boring looking tool. Which brings us to the fourth category: image creation tools. This category is one of the most slept on, not because people don't know it exists, but because they think AI image tools are just for messing around, making weird memes, and sticking your dog in a Renaissance painting. And sure, they're great for that, don't get me wrong. But what they actually are is a free design department. Logos, icons, product mockups, thumbnails, social graphics—the stuff you'd normally pay a designer a lot of money for, you can now make in seconds. And to be completely honest, this won't fully replace designers. But for something quick like the example I'm about to show you, it's 100% worth it.

But before that, let's talk about why most people don't get good results with AI imagery. Because it comes down to two things. The first is they just don't know how to prompt. Most people describe what they want as "a logo for my app" or "a thumbnail for a YouTube video about history." But that gives the AI almost nothing. It doesn't actually know what you mean. So, it falls back on the most average, generic version of what it thinks you might like. A better way is to be as descriptive as possible while staying completely on topic. Something like "a flat minimalist logo, one accent color, lots of negative space, no text." I'm not adding anything unnecessary. I'm just giving it way more detail while staying laser focused on the logo itself. Specificity is the whole game. And a quick trick: hand your prompt to one of the chat bots from category 1 and have it sharpen the prompt right before you paste it into the image generator. The second is that you don't have to generate from nothing anymore. You can hand it a reference, an image of a style you love, or a photo of a real product and tell it to work from that. And once it gives you something close, you don't start over. You just edit it in plain English.

So, let's give Worth It a graphic update. For image generation, there's a ton of choice, but one of the easiest ways to get high-quality images is to head over to Gemini and, up in the top left here, select Image. That lets you create images completely free, even if it's a little limited. A tier above that is AI Studio, which gives you Imagen 3, one of the best image models out right now, but this one's paid, so you link an API key and pay as you go. Fiddly to set up once, easy after. For this example, though, I'm going with the free option, so any one of you, no matter your budget, can follow along. And because we already made a really good master prompt, this part is easy. I head to phase four, copy each image prompt into the generator, and after running through all of them, I've got clean visuals. Then I drop those images into the project folder, the same folder Claude Code is working out of, and tell it exactly how I want each one used. I hit send, wait a couple of minutes, and there we go—the updated website. We kept the updates pretty simple here, but even this gives you a picture of how far you can take the design if you want to go all out.

Still, great images are one thing. There's one more step that can massively elevate any design work you do, which of course is adding some actual motion in the background, because that makes a huge difference in how professional your tool looks. And so video is the next category. There are just a couple of things to keep in mind to make these as good as possible. The first is to always start from an image whenever you can, instead of generating video from nothing. That way, you get way more control over the final result because you're starting from an image you've already confirmed looks good. It also keeps things consistent. If you're generating people, it lessens the chance of their face drifting and changing over the clip. The second is if you want high quality results, keep the motion small and short. That tends to look far more realistic, whereas forcing the AI into huge movements is exactly where it's most likely to fall apart.

Back inside the master prompt, we've got one final phase left, the one that gives us the image-to-video prompt. So, I hit go. And here's the prompt, which I just copy over. Then I switch back to Gemini because it also does video. Now, unfortunately, this one isn't free. You'll need a subscription. And that's true for Gemini's video specifically and honestly most video tools out there, just because AI video is so expensive for the companies to run. With that said, I paste in our prompt and pick the image I want to animate. For this one, I'm going with our header image, which is going to loop quietly in the background to create that professional feel. I select the image, make sure I've got the landscape option so we get a clean 16x9, pick the best model, which here is 3.1 Pro, and set the thinking level to extended for the best possible result. Click send. And here we have it.

It's a simple, clean animation, perfect for a background. There's some sound on it, but don't worry about that. We won't be using it in the final result. So, I move that video into our project folder. Jump back into Claude Code and just tell it we've uploaded a video into the folder and want to replace the header image with it. I hit send, and let's see what it does. And there we go. You can see the animation up top giving this really nice looping effect. And our tool is now fully functional, every button clickable, ready to use.

But the biggest takeaway here isn't the tool or even that you could recreate any other tool the exact same way. Because this was just one example, one tiny combination of those five categories. The categories themselves are the point. Because once you actually understand how to use them, you can build pretty much anything. And that's exactly what we do inside AI Fluency. It's where you go from just knowing the five categories exist to being genuinely fluent in every one. It's a 90-day step-by-step map you can follow a little each day to become completely fluent with AI. So, you can build things like this on the spot whenever you need them and genuinely change the way you work and live with AI. The links in the description below. Thanks for watching. I'll see you in the next one.