[00:02] every month, but end up using it to ask questions and get answers, right? Kind of like Google, but smarter. So this is the most expensive mistake for anyone using AI right now, because the same tool, [00:17] for the same price, can do two completely different things. A version will respond to you inside a small box in the chat. The other one opens her computer, analyzes [00:29] files, reads her emails, assembles and finishes a task that would have taken you an entire afternoon, while you can just sit around having a coffee. The difference between these two worlds isn't the tool, it's the key that almost [00:44] nobody knows exists. And in this video I'm going to show you exactly how to turn that key, what the four tools that are really worth it tools that are really worth it today are, and most importantly, the right way [00:58] to give them commands. Because if you provide the command the old way, you're wasting money and time. In the end, you'll leave here knowing how to get an agent to truly work for you, and not just talk to you. [01:13] My name is Tainara Shimite and welcome to the Business in Mind channel. Here you'll learn everything you need to know about artificial intelligence. Before we get down to business, I need to explain something to you. If you skip this [01:26] part, the rest of the video won't make sense. So stay here with me because it will be quick. And look, I'm telling you this because it took me a long time to understand it myself. At first, I used these tools like a normal chat and [01:40] thought it was all the same thing. Well, it turns out I learned the hard way that it isn't. The structure is simple. First, let me explain the difference between a chatbot and an agent. Then I'll show you the four [01:54] tools and in which areas each one is best. And then comes the tricky part. I'll show you the formula to create commands that make the agent deliver exactly what you want. So let's go . Chatbots and humans can use the [02:10] same brain internally. The engine may be the same. What changes is what is connected around that brain. Think of a company. The chatbot is like a very intelligent customer service representative, but one that's stuck behind a counter. He can't [02:26] log into the company system, he ca n't open files, he can't send any emails, he can't access a website, and he certainly can't continue a task after you've left. He talks, but he doesn't act. [02:41] Now the agent is just like that same customer service representative, with the same knowledge. Now you can access your company computer, including systems, files, browser, calendar, and a clear to- [02:58] do list. And then, my friends, everything changes. Instead of just answering how to do something, it can actually do the research, compare information, fill out a spreadsheet, [03:10] organize files, create a report, send a message, test different approaches, and continue testing and trying until the final result is achieved. If you're looking for IAS to do more in less time, then you need to know about [03:26] Monday. Even before artificial intelligence emerged, Mondei was already one of the best productivity tools on the market, used by companies of all sizes, freelancers, or even people who just [03:39] want to be more organized and productive in their daily lives. Now, with its new air conditioning features, it has reached another level. In addition to helping you organize your work and your life, it 's ideal for those who want to bring projects to [03:53] life and achieve their goals much more quickly. And best of all, there's an excellent free plan for you to test all the features. Click the link in the description and try it now. [04:07] For this agent to function, it needs four things. First, of course, there's the brain, which is the artificial intelligence itself , the part that understands, , the part that understands, reasons, and makes decisions. Secondly, there are the [04:21] tools, which are the access points that allow you to act in the real world, such as opening a browser, manipulating files, running commands, or using applications. Third, memory, which functions as a record or a contextual notebook, [04:37] eliminates the need to constantly ask the same questions. And fourth, the objective. A clear task with an expected outcome. Not something vague like "help me," but a mission [04:50] with a beginning, middle, and end. When you put all of that together—brain, tools, memories, goals—and set it all running in a cycle, you end up with more than just [05:02] an ordinary chatbot that simply answers questions. Putting these four things together, running them in a cycle, there you have it, an agent. And this cycle is very simple, it only has three steps. Look how easy it is. She observes what is happening, such as [05:19] which files exist or what has appeared on the screen. Then she thinks about the next step based on what is true at that moment, and then she acts, uses one of the [05:31] tools and changes something, and then repeats everything again until she achieves the repeats everything again until she achieves the goal. Observe, think, and act. This is the driving force behind any agent, anywhere in the world. Keep this [05:45] information in mind because now I'm going to tell you about the tools. But first, take the opportunity to like this video, subscribe to the channel to stay updated, leave a comment below, tell me what you [05:58] think, and especially tell me about a task you would like to use with an agent. Before I tell you about the four tools, I want to show you in practice how significant this is. Come with me here on the screen. Oh, this folder here is a [06:12] here on the screen. Oh, this folder here is a real mess. It has printouts, PDFs, receipts, photos, travel plans, PDFs for studying different languages, strange names, really everything thrown together. And I bet your downloads folder [06:28] looks like that too. Now, look what I'm going to do. I opened Cloud Code, selected the "mess" folder, and here I'm going to give a command written only in regular Portuguese, as if I were asking a person. You'll see it's not [06:42] complicated at all. I asked him to do the following: open and read all the files in that folder. Separate the folders by category, rename each file with its date and type, and finally create a spreadsheet listing everything. The spreadsheet should be in [06:58] XLS format with a professional design, formatted header, adjusted columns, and a summary tab. Do not delete any files. And now all we have to do is watch. Ready. [07:11] Within minutes, he had already told me the structure he used in the folder and created a spreadsheet listing each item and detail himself. And by opening the folder we can see the separation of the categories. And I've opened some [07:26] folders here for you to see what files are inside. And notice that it changed the name following the formatting requested in the prompt, in the command. And what I liked most was definitely the spreadsheet, which turned out beautifully, super [07:41] organized, with categories, new name, original name, type, date, the folder that was added and the description, plus a good summary tab. Notice the difference. [07:55] The chatbot teaches you how to organize this folder, how to create this spreadsheet, all step by step, like a recipe, but for you to follow the manual. The agent simply took the task and [08:11] said, "Okay, I'll do it." And he did. And look, that command I gave wasn't just luck. There's a formula behind it, which is exactly what I'm going to reveal to you later in this video. And when you learn, I'm sure [08:23] you'll never use commands the old way again. Okay, let's talk now about the four agent tools that are worth having. But first, if you want to get ahead and know the latest news from A before everyone else [08:37] , then I recommend you join our free WhatsApp group. good morning messages or people sending stickers there, just news items hand-picked by our team. To enter, simply point [08:53] displayed on the screen or click the link in the pinned comment description. There you can also find our free materials, such as courses, prompts, spreadsheets, as well as discounts on tools that we recommend. So, if I [09:07] were you, I'd take a look. Okay, let's talk about the tools. There are a lot of agent tools out there, but the vast majority are just noise. After testing several of them, four remained that actually deliver good [09:21] results. And the interesting thing is that each one is very good at something different. , because I know how valuable your time is . Also, if you'd like a video dedicated solely to any of them, [09:35] just leave a comment below. The first one is Cloud Code, which is the agent for Antropic, and which I used to show you the example of organizing my folder. It runs like a program that you install on your [09:47] computer. It doesn't matter, you can, it runs on your computer, so you'll download it. It can be Windows or Mac, it works on both. The name has "Code" in it, so many people think it's only for programmers or that it's something very [10:00] complex, but it's not. It can rename files, read PDFs, take screenshots, edit things on your computer— basically anything you can describe in Portuguese. The only limit is your imagination, not the tool. The great [10:17] thing about Cloud Code is that it offers a much wider range of things that it can do. Regarding the price, pay attention here because many people get it wrong, okay? The free version of the cloud does not include the code, [10:32] so you need a paid plan. However, the same subscription covers regular chat, which is what many people use, and also the agent. So it's all together. In addition, the paid option also includes coworking, which is another really cool feature. [10:46] And where does he excel? He lets you see the reasoning. So you see the reasoning. So you follow Iá step by step, and you can interrupt her at any time to correct and redirect her along the [10:58] way to the right path. It's ideal when the task is complex, you want to follow it closely, and you need a brainstorming partner. And why is it good for you to stop and adjust? Because, guys, it consumes tokens, right? It's [11:13] like the gasoline he needs. And then you might exceed your needs. And then you might exceed your daily, weekly, or monthly usage limit. The second tool is Codex, the OpenAI agent, which is the same company [11:25] responsible for GPT chat. So, if you already pay for GPT chat, Codex is already included in the plan. Same account, same login, no new subscription, no new tool to learn, just install and use. And he has a trick that [11:41] others don't have: a cloud-based version. So you hand in a time-consuming task, close your laptop, go about your life, do other things, and when you come back the work is ready for you to review. So it's useful for those who don't want to [11:55] want to leave their computer on while the AI ​​is working. Then you can delegate and come back when you're ready. In short, if you already use GPT chat and want to start with as little friction as possible, Codex is the best choice. And [12:11] by the way, look, I opened this just to show you, the Codex is also available for your phone. Then you can continue your project from where you left off. If you started on your computer, you can continue on your cell phone, [12:23] notifications when it completes a task or when it needs your attention for something. And you can start something new too, start a task on your phone and then continue it here on your computer. It's going by fast, you know? [12:38] Because we've already reached tool three, which is Open Claw, a real-life automation tool. And what does he do that's so special? He doesn't get stuck on one website or program. He lives inside your messaging apps. So this [12:51] is very useful. Where does he live? Ah, Telegram, WhatsApp, and so on. Furthermore, another advantage is that it's free and open source, so you can install it on your own machine. So think about it. You send a message to your [13:05] assistant, just like you would to a friend. The work is done on your home computer, and it responds when it's finished. You can be in line at the supermarket and send a task via your cell phone. So, real-life use cases include [13:18] summarizing emails and creating reminders, assembling documents and organizing ideas, saving video ideas sent via message, and automatically organizing notes . This is the agent that finally delivered [13:32] what they promised us: a real assistant that lives inside the apps you already use. In fact, Rodrigo has already made a Open Claw. I'll leave the video playing here in the cards, and it will also [13:47] fourth tool is Antigravity, Google's agent for visual things. It's Google's agent platform built on top of Gemini. It's currently in open beta testing, so it's free for any [14:03] individual without needing a credit card. I even confirmed this before recording this you're watching, that may have changed, but for now it's free. And what are the advantages? Jemini can actually see what he's [14:18] doing. So he looks at a printout, compares two designs, and generates images with real visual context. So, without a doubt, it's the best for creating screens, making layout adjustments, and working with images. Ideal for those who work in [14:33] design and marketing. So, if your work involves design, marketing, or anything visual, this is your agent. It shines where others remain in the dark. Do you realize how you can use more than one, how they [14:47] complement each other? That's not a problem either, but I'll talk about it in a moment. So, now a quick summary for you. Which one to choose? Choose the one that best suits your daily routine and go all in on just one thing . You don't need to learn all four at [15:00] once. This is a great place to start getting familiar with agents; you can add others later. Therefore, cloud code is best for you to closely follow the reasoning and use in complex tasks. The codex is for those who [15:13] already pay for GPT chat, which is the easiest way . Open Claw is for those who want to automate real-life tasks using a message window. And antigravity, visual work, design and marketing. Each tool has its strengths. [15:27] Identify what best fits into your routine and start with that. Depth, too, folks, comes with use. Now, pay close attention here, because nobody's telling you this. You can choose the best tool in the [15:41] world, but if you give it the wrong command, it will deliver something terrible. And what most undermines results with the agent is precisely that: the way he gives commands. For chatbots, the command is a description of what you want. But [15:56] for the agent, the command has to be almost like a contract, simplified, of course, but you get the analogy. A briefing outlining what he is required to do cannot be short and vague. Why? Because the agent truly has autonomy in their hands [16:11] . With a vague phrase, he'll start making things up. They'll choose a color you didn't ask for, create a session you don't want, and 10 minutes later deliver something that wasn't what you imagined. And all of this involves consuming resources, [16:25] spending tokens, for example. The solution is not to write a longer command, but to write a structured command. And it only has four parts, so it's quite easy. It's even easy for you to take notes, which I highly recommend. So, memorize [16:39] these four words, because every command you give to an agent for the rest of your life will use these four things: objective, limits, format, and output. Let's go one by one. But [16:53] ask you to like this video, subscribe to the channel, talk to me in the comments, and then let's see part one, starting with the objective. The objective, folks, is different from the action, okay? So, the key question [17:08] define your goal. What does the finished product look like? Again, the goal is not the action, it's the end result. He will define the finish line for the agent. So, what would be a bad example? Create a page. [17:23] This is a stock. How do we turn this into a result, right? For you to add to the objective section. Create a single-column page with an email capture form at the top, ready for me to review before publishing. The second [17:36] sentence shows the agent how he knows when he has finished the task. And if you don't provide a finish line, the agent will invent one. And again, you probably won't like her. So, remember, the objective answers the [17:49] question: What does the finished work look like? Part two, limits. And here are the safety rails. Limits are everything that the agent cannot do. This part here prevents disasters. Here 's a very simple analogy to help you [18:04] understand better. Imagine a child playing near the stairs. You put up a gate not to stop her from playing, but to prevent her from falling. So the limit is this grid. Here are some examples of limits: don't touch any files outside of this [18:19] folder. Don't publish anything. Just let me revise it. Do not copy text from the revise it. Do not copy text from the competitor's website or delete files from that folder, which is what I used in the example I showed you. Your list [18:33] of limits is your learning scar. It just keeps growing, and each one saves you a future headache. So, you've set your boundaries, and you're talking to your agent, he's creating the [18:47] task, and then you see that he's done something wrong again. That will become a limit, and you'll keep updating your list of limits. Part three is the format. The format is exactly how you want to receive the result. This is where [19:01] most commands completely fall apart . Either we do the work right, but the delivery is in a format you can't use. So tell us exactly how you want to receive it, for example, in bullet points, in a table, in a single file, [19:16] in a folder with a specific name. When in charge, you need to remember to be specific. The work may be correct and still be useless to you if, for example, it is in the wrong format. So, if you can't describe [19:29] your desired format in one sentence, you still don't know what you want. So, you 're not ready to press enter yet. You need to think a little about how you want this delivery to be. And you can use the regular chat to help you [19:42] choose the best format for your task. This way, you also don't waste tokens us, you have to come with everything ready . And finally, our fourth part is the exit strategy, which is what he should do [19:55] when he gets stuck. Almost no one writes this part, but it's the one that saves the most tokens and time, okay? It defines what we should do when things get stuck or when information is lacking. Without instructions, the agent [20:10] information is lacking. Without instructions, the agent chooses the worst possible course of action, okay? I swear, it 's like this: he keeps trying, he keeps going in cycles, in circles, and he keeps wasting resources, trying, I'll try this way, that way, [20:22] that way. When he could have simply asked you, right? So, one sentence will do. You can add it. If any important information is missing, stop and ask me before continuing. Have you defined how he will handle a [20:36] question? So, with the output defined, the agent stops and asks for help. Without that, he tries to guess and gets caught in this infinite loop, wasting his tokens in the process. Either you define the solution, or we'll define it for you. And he usually decides on the [20:50] most expensive way, right? So, what does this complete formula look like? This is our contract with the agent, outlining the objective, limit, format, and exit strategy. For example, the objective would be all of this here, folks, it would be one thing, I would put it all together in a [21:03] single prompt, okay? I can add the objective and then write it next to it in the same prompt, right? And then skip a line , then format, and then output, or without the titles I can also just send these white texts, okay? All at once [21:17] . So, the prompt would look like this: create a one-column page for my product launch focused on capturing the visitor's email right at the top, ready for me to review before publishing. Use only one file, don't modify [21:29] anything outside the page folder, and don't publish it anywhere. The page folder should contain the page file plus a short summary listing each section that was created. If you don't know who the product is for or what its unique selling point is, stop and [21:45] ask me a question instead of guessing. So this is a complete prompt. You define the outcome, the timeline, the delivery method, and what to do when information is missing. One more thing, what would I do in this [21:59] real-life scenario? Obviously, I would feed the agent information about my product, right? So, a document about the product, with its unique selling points, the marketing terms we're using, and everything else. Try this test [22:13] on your commands before pressing enter. Write the four parts. The "Wow, this is extra work that seems unnecessary." In the third one, you'll be wondering how you could send commands in [22:28] another way. So remember, for a chatbot, you'll use a description; for an agent, you 'll use a contract. Now, let me tell you where most people give up on using agents. They ask for one thing, the agent messes up, they [22:44] correct it, then they ask for something else and he repeats the same mistake again. And again , and again. And then the person thinks the agent is broke, that he's bad, that he's not worth it, and they just give up. But it's not broken. You're the one who [22:57] forgot to remind him of that information. Remember when I mentioned the notebook for contextual memories? That's it. Every serious tool has a small text file that the agent reads before starting any [23:12] task. Everything written there becomes a rule that he always follows. You place this file in your project folder and, voila, you've just given your agent long-term memory. So you need to ask him to create [23:26] this little memory book. You need to ask him to remember this, unfold, there will be more things he'll need to remember. I'll give you a simple example. Imagine you 're creating something serious and [23:39] professional, and the agent keeps putting emojis in every single text; you remove them, and the next time they put them back. It's tiring, isn't it? At least I think so. But then you say it just once: "Create a rules file and write it down. Never use [23:55] emojis in this project. It's professional. Remember this in all our future conversations about this project." Look how simple, look how wonderful, everyone. It doesn't take a single minute and that's it, it will create it, [24:07] save it, and that problem will never come back. Whether in that project or if you want it to be a general rule, it won't come back in any project. You can excited today, so I'm going to give you another trick. You can ask the agent [24:21] trick. You can ask the agent to update the memory automatically. You say: "Every time I correct you, you must write down that correction as a new rule at the end of the file." Again. Pretty simple, right? Then, look what happens. [24:34] The first time you have one rule, but by the fifth time you already have about 20 rules written down. And a few days later, there comes a point where the agent almost never makes mistakes anymore, because you've created a perfect customization for your [24:48] case, almost as if it could read your mind. It's like a new employee who is learning. The way that company works is something you'll never forget. And so you don't leave here with just theory, here's an action plan for [25:03] today. You're going to choose one of the four tools, select a boring task—a truly boring task , not a playful test, okay? Something boring that you do every week. You're going to write the command in [25:17] contract format with those four elements that I hope you've noted down. So, if you wrote them down, they're easily accessible. Otherwise, you can just go back to this video, ask it to create a memory file, and then add [25:29] at least three rules and run it from beginning to end. And then, if you feel comfortable, come back here and tell me in the comments how your experience was. Because the truth is : you can watch 1000 videos [25:43] about A-agents. And I love that you 're here watching my video, but the turning point only happens when you open the tool and execute. You only learn when you actually do it in practice. So stop reading and watching [25:56] videos and start executing. And then you'll keep doing it. In parallel. Oh, but of course, folks, you'll improve, you'll watch more videos, but remember to always practice as well. And if you choose to start with Cloud [26:08] Code, there are seven secret commands that are easy for you to apply and that will help you a lot, including saving tokens. And I showed them to you in this video tokens. And I showed them to you in this video that's appearing here.