Transcription
This is actually crazy. I just sat back and watched as Chad GPT booked a restaurant reservation for me while I was actually busy filming other videos for this very YouTube channel. And Chad GPT found me a restaurant, checked the availability, and even started the actual booking process while I was doing, as I said, other things. And this new Chad GBT thing I'm talking about is called agent mode. And it's really making me feel like we're living in 2035.
Now, if you are living under a rock and don't know what I'm talking about, Chad GPT agent is basically the newest function from Chad GPT. It combines the best qualities of operator and deep research. And agent mode allows Chad GPT to not only perform extensive online research, but also interact with websites and make basic decisions on your behalf. And this actually marks a really huge leap forward in how we use AI and what AI is reasonably actually able to do.
So, I know you're probably thinking, how do you use this new agent mode? So, today we're going to take an in-depth look at how you can actually use Chad GPT agent and utilize it for yourself. And we'll also look into the pros of it, the cons of it, and I'll give you some really good tips on how to effectively use this Chad GPT agent. And I mean with everything from research to image generation, Chad GPT is actually creating a whole new browsing ecosystem. So let's see how agent mode actually stacks up with this.
So as I already said, agent is the new advanced feature of Chad GPT that uses its own computing environment to perform real-world tasks for you. And this is actually mindblowing, guys. Until now, LLMs required a lot of human intervention to perform any task whatsoever. And getting it to use a web browser was well out of the question. But agent mode finally allows you to prompt Chat GPT with workflows, allowing for a full hands-off experience.
Now, agent mode was launched in July of this year, combining the best of two previous Chad GPT worlds, which was, as I said, operator and deep research. Now, Operator was actually a really short-lived GPT capability that introduced a web browser to the LLM. And deep research is still available today and was created to offer a version of Chad GPT that could answer more complex questions than the generic chat. So, I mean, simply put, it can synthesize answers and provide some basic reasoning, making it great for data analysis or anything more than just a basic summary of your question. But combining these two functions of Chad GPT allows the LLM to handle more complex tasks and react to prompts in a more particular way than ever before. It can now handle really complex workflows, also prioritize information, and carry out actions on your behalf, which is the craziest part. Because whether that is running a competitive sales analysis, for example, or just finding a recipe and buying the ingredients for you, it can actually do all of that. And if you add in connections to popular apps like Gmail or GitHub, you get an AI model that can actually perform all of those daily admin tasks that might pull you away from the more creative side of your business.
Okay, now that I've explained a little bit about what this new Chad GPT agent actually is, let's run through a couple of use cases for Chad GPT agent so you can actually see how to start integrating it into your personal life or your work life as well. So, basically with ChatGpt, if you wanted to plan like a special dinner night, you would type something like, "Find me a highly rated upscale restaurant for a client meeting in London and make a reservation for five people on August 22nd at 7:00 p.m." You know, all that stuff. And uh type something else like, "The restaurant should be in the Hackney neighborhood." And I mean, as you can see, Chad GPT did do its job and has come up with a few options for me. And it basically lists a few restaurants that are very well-rated, talks about the menu options, and even provides a little summary chart of my options for that night. And I mean, this is great. This is awesome. If I'm still not sure what I want to do, I can look at these options. Or if all I wanted were a few unique ideas, this is awesome. I would just settle with the Chad GBT summary and then go with its little plan right here below.
But now let's compare what I'm able to do with Chad GPT agent. So I'll just put the same prompt in here that I put into Chad GPT earlier and walk away. Just leave it there. Just leave it as I typed in the prompt and we'll just see what it does. So I mean, you're seeing my screen right now and Chad GBT has completely taken over in its own browser window and is now finding me a restaurant for a client dinner. So, actually, while this whole process runs, I have a few other important things to do. So, I'll just leave it to do its thing, and I'll be right back in just a few minutes, guys.
[Music]
Okay, well, I'm back and it actually looks like Chad GBT is all finished up. So, let's take a look. So, as you can see, it found me a table for five at a restaurant in Hackney, as I said. So, it actually is very mindblowing to me that it did all this in its own browser window and it was literally that easy. And in the meantime, while it was doing this, I was able to get a couple of other tasks done for more videos that I'm going to be filming today, actually.
So, yeah, basically this is crazy, but we've got all the fun stuff covered, but what about, you know, business admin? No one ever started an online business to do admin tasks all day. I mean, no judgment if you did. To each their own, but especially when there's a long list of other tasks in the way. Admin tasks are not the thing you want to be doing. So, let's just take a look and see if Chad GPT agent can run a basic competitor analysis and put it into a spreadsheet for me. This would be cool. So, let's just say I have a ceramics business. I mean, I don't, but let's say I do, and I don't quite know where to start when it comes to selling online. So, I want Chad GPT agent to run a competitor's analysis. It also maybe generates some SEO keywords for me uh for my online store and create a go-to market presentation so I have some steps to follow as I grow my following. So, uh I wrote this whole prompt down right here as you can see. I'm not going to read it out loud, but once again, I'm just going to hit enter and I'm going to let Chad GPT do its thing. And in the meantime, I'm just going to sit here and watch it.
Okay, so it's done once again. And there it is. Now I have three documents that give me a way better understanding of where I'm at in the ceramics market and a few other actionable steps to get started selling ceramics online. So, as you can see, agent mode can be a pretty effective digital assistant and one that can actually execute a few tasks and run analysis without actually you having to hover over it.
But here's another thing. How do you actually make sure that the suggestions that are popping up are what you actually want? So, prompting for standard chat and agent mode are slightly different. So, we have a few tips and tricks to make sure you have effective prompts for the right chat mode. So, as I just said, Chat GPT agent mode is different from the standard chat mode. And so you should adjust your prompting strategy accordingly and think of it less as a knowledge base and more as a virtual assistant, one that needs all the instructions up front. So your prompts should be really directive and designed to trigger workflows and actual actions that combine research and analysis with logic and real decision-making. An agent mode is best at handling complex tasks with multiple steps and clear outcomes. So instead of a simple question-answer structure, use prompts with an action-output structure instead. Because standard chat is more about getting information from a knowledge base, while agent mode is pushing this to the next level by actually asking Chat GPT to act on the information it gets.
Now, tip number two is to provide agent mode with context and permissions. I mean, in agent mode, you give Chat GPT the permission to browse the internet on your behalf. And I mean, this is really great news if you like hands-off browsing, but it does mean you need to be extra careful about what you include in a query. So, include full context about the prompt and make sure you specify what Chat GPT can or cannot do. Also, make sure to give the correct file permissions and set up boundaries around what actions Chad GPT can take before actually asking it to do something. So, to keep Chad GPT from running into a strange loop, you must really, really make it clear.
Now, tip number three is using conditional logic, aka if and then statements. Now, these can really help increase efficiency in your prompts and make Chad GPT agent act on its own reasoning without you actually needing to come in to reprompt it. Now, with your original prompt, you can instruct Chad GPT agent to take different paths based on what it finds. And this really saves time and makes the tool truly hands-free, something that can't happen in standard chat.
Now, we're here with tip number four, which is automating. Now, with the sources menu, you can sign into your accounts on other websites like Google, Canva, or GitHub so that Chad GBT can pull data from those sources directly. And this is especially helpful for things like creating presentations or spreadsheets or analyzing code and organizing your schedule or just organizing your email inbox. So rather than just uploading documents, you can ask agent mode to look into them for you directly from Chad GBT. And that way, it can really handle all the boring stuff for you. And to take this automation just one step further, you can connect Chat GBT to an automation client like NADN. Once you do that, you can start running pre-scheduled automations, which can really save you a lot of time. And once you set up a trigger, you've basically got a personal assistant right at your fingertips, which is really, really cool. And if you need any NAD ideas, look no further than the upper right-hand corner because we have a previous video all about NAD. So, make sure to check it out.
And we're going to tip number five, which is cross-reference research. Now, with the combination of a browser and deep research, agent mode can more easily pull from multiple sources in a single workflow. That means you can use agent mode to analyze uploaded documents and cross-reference them with live data or just pull in research from the wider web. And that creates an outcome that is more than just a summary of data, but has a deeper insight or analysis that can really help you decide on your next steps.
But here's the thing, agent mode is asking for more than just a standard prompt. It wants a workflow that clearly defines the trigger, also the data, the sources, and the action steps, and also the most important part, the output. So, I mean, just keep this formula in mind as you're adapting your prompts from standard chat to agent, because they really are different.
Okay, guys, but now let's talk about the elephant in the room. Now, a Chat GPT agent, as cool as it is, is still very, very new. So, it, as you may have guessed, has a few kinks to work out. In some ways, it's way more powerful and cooler than the standard chat, but that comes with a few limitations to really think about as you start experimenting with this tool. I mean, at the end of the day, Chat GPT agent is still a bot that is browsing the internet. That means it's going to get stuck on the barriers we've put up against bots browsing the internet. So if you come across a Cloudflare or CAPTCHA verification, agent mode can't get past it in any way, even if you take over control of the browser environment.
Another thing with access comes exposure, and we still don't know what can be exploited with agent mode. For one, it's really vulnerable to prompt injection attacks and hallucinations. And also, we don't really know what kind of memory agent mode has. I mean, OpenAI is clear that agent mode doesn't see your sensitive data, but that doesn't mean you can kick back and relax because there's no actual way to really restrict its permissions. So, it's really important to be clearing your saved logins and cookies throughout your Chat GPT settings really regularly. And it's also worth it to think twice before entering any sensitive data within the browsing environment.
Another con is that it's actually pretty slow. I mean, in testing out agent mode for this video, we found that some of the tasks we asked it to complete were just, I mean, very, very slow. It's kind of like letting your grandma use the computer, even if she says she can do it herself. I mean, you still have to sit there and watch just in case she gets confused or something. I mean, we've probably all been there. And sometimes she misses, you know, a button or two or three or whatever. Well, with some tasks, it's still faster to just do it yourself.
But I mean, with all these cons I just listed, keep in mind that agent mode is barely a month old, and OpenAI is undoubtedly working on these little kinks. But it's still worth asking whether or not you need agent mode, or maybe if a combination of standard chat and human intervention is faster and more accurate. And something else that's worth noting is that the tool is currently only available for paid plans and is restricted to certain locales. But I mean, still, even with all of these limitations, agent mode is a fascinating update and one we're really excited to keep our eye on.
So, what do you guys think about agent mode? Let me know in the comments. And also, let me know if you have tried it yet or not. Also, let us know some of your workflows in the comments down below. I would really love to hear about it as well. And make sure to check out this prompting with AI video that we have on our channel as well. It also offers some really good prompting tips.
Now, thank you so much for watching, guys.