ChatGPT Agent Full Breakdown & More AI Use Cases

This looked like a slow week in AI until it didn't. OpenAI released their new agent. That's what it's called. Just chat GPT agent. And there's a few other interesting stories like Grock releasing companions inside of their app. That's the first big LLM provider doing this. We're going to be covering those and more with a specific focus on the chat GPT agent with some first use cases because we already had the time to run several test prompts that we run through every Aentic product. and we'll be comparing those back to some of the previous competition, showing you how this product performs in the real world and what it's all about. In case you're new here, this show is called AI News You Can Use. Every single Friday, we look at all the generative AI releases, filter for the ones that are actually available and usable, and then our goal is to show you real world demos or cover the stories from an angle of somebody actually putting them to work. So, let's begin by looking at chatbt agent. So

what is this chatb agent thing about for somebody who hasn't even heard the announcement? Well, OpenAI announced a spontaneous live stream on Thursday the 17th of July which Sam Alman opened with the words, — "We have a banger for you today. We're going to launch ChatBT agent. " — And what does this agent do? Well, it combines two agentic products that chat had plus a few more tools into one thing that they now just call agent. So, let's break this down. What are those things that it includes? First of all, it includes deep research. The tool that has access to the internet and can write some code that spends a long amount of time researching. And to be honest, that was the most useful agentic product for most people. Then they had a second product which was called operator. And by the way, each one of these products are a whole product category with Injenai if you're not familiar. Operator in particular is a category called computer use agents. And a lot of the competitors had versions of this tool. And operator was decent, but we covered it extensively on the channel. Up until today, it just came down to it not being good enough for any real world use cases. It was just not consistent enough. But the idea was an AI using a computer to actually click buttons, scroll around pages, and do things for you. And then on top of that, they added a few more tools that were available in other forms inside of Chhatipity, like a command line interface that can run code that CHP writes for it. The image generation feature that you're probably very familiar with, the one that had the viral moments this year. And they bundled all of those into one package that they now call agent. So the agent can research things, it can navigate the internet, it can execute code on its own machine and use those results and it can generate images and visuals for you. So that's what the release is. Now who is this available to and how does this perform in the real world? Well, it rolled out to pro users first, but they announced that this is coming to plus and teams users with limited usage next and then eventually it's also going to come to edu and enterprise users. So, as of today, July 18th, only pro users have access to this. And now, let's get to the best part, the part that you came here for. Is it actually worth my attention or is this just another marketing thingy? Well, let me tell you before we jump into the prompts here, I ran four different ones. We'll review all of them. This idea of an agent that combines his capabilities isn't exactly new. If you're following the show, week by week, we covered the competition that was trying to do this on most of their major releases. Most significantly, there was Gen Spark's super agent that was releasing features exactly like this. It could do everything that this can do, too. But spoiler spoiler, as to be expected, that product was not as good as what Open AI just shipped now. And that's what makes these things interesting. It's usually not about the new product category opening up. It's about that new product category being good enough for people to be worth it. And I think this did cross that chasm. But that is just my first impression of the first four prompts. I'll reason why in a second. There were also more competitors in this space like Manus that went viral initially but then kind of fell off in terms of attention that people were giving it. And then there is a lot of custom and open source solutions that all have one thing in common and that is that they're all probably not worth your time. Here at the AI advantage, we really keep an eye on this category because I think this is going to be the next big thing once it gets good enough. And even the promising and big ones like magentic AI from Microsoft, I believe it was called didn't perform close to what we would expect. But this one looks good. So let's get into the practical part here. I started out by providing a prompt that is very close to what I do. Provide a comprehensive overview of AI news in the Genai field for the last week. This is something I like to run in all of these agentic products because I'm always up to date on what new things are coming out and what the Genai news is saying. So, I have the knowledge to really judge these results well. Plus, I'm recording this as a part of a news roundup. So, a lot of the things that you'll see in this report should probably show up in this video. And they do. So, look, the whole conversation was a two-step process. First, it asked me, are you referring to the week from July 11th to July 18th? I said yes. There was actually a continue button here that I could click. And then as soon as it continues, it puts together this report. It took it 12 minutes to generate the whole thing, which is not bad. That's like a longer deep research, but it actually did much more than just a deep research, I have to say. I really, really wanted deep research to be good enough for this, and none of them were. Not even the Gemini one, which by many people these days is considered the best deep research product on the market. We have a team of people at the advantage that scour the internet, pull all of this together, and we discuss all of the stories, see if anything slipped through the cracks. And I would say this result here is not quite as good as that, but it might be around 75, maybe 80% there, whereas all previous products barely got to 50. That is just my feeling, but let me show you some of the stories here. So, most importantly from all of these, I wanted to nail the biggest story of the week, which in this case is the Chachi Agent, and it got that a lot of the previous models were quite bad at this. And one common problem with all the other agentic products like this was that it never got the date range right. Usually it didn't follow up by confirming the date and then it gave you stories from like 3 weeks ago. This happened here too sort ofish. If you look at some of these stories they are from July 10th or July 2nd or 1st right here. Although it clearly told me it would only do 11th to 18th. I can forgive that though. I think I'd rather not have these July 2nd stories in there. But hey, most of the stories are actually on point here. And yeah, there's a lot of stories that we're going to be covering in this video like agent the Kimmy K2 release. We're just doing a quick hit on this. That's our segment in the end where we kind of just brush over a few stories. Mistrol is upgrading Lhat with the Voxil model. This is something we'll cover later in this video. Google's image, the video we covered last week. Lloyd Banks Athena is something we're not covering because it's not available. And in stories like meta poaching talent and open AAI having a week-long shutdown are new stories that we generally don't cover on the show because this is really about the use cases. It's about how you can put it to work. But that is fine. And to be clear, we did ask for the Genaii news here. And I would say if I were to rely solely on this, well, this video would be 50% shorter because it missed things like Grock Companions coming out, missed a new finetunable feature from Hicksfield, and a few more small use cases that we found across the web that we'll talk about later that this thing generally can't find because they're like a Twitter post that we found really interesting and want to show you. So, personally, I'll start running this every week to supplement our research, but definitely not a replacement yet, although it is better than anything we've seen so far. Next up, let's look at one that is very binary. What is the current Euro US dollar exchange rate according to the National Bank of Slovakia? This either it gets us right or not. And if we look at the exchange rate for July 17th, which fair enough, it's too early in the day to get July 18th, which is actually today, and it acknowledges that. If I look up the exchange rate and reference here, it actually pulled a PDF with all the exchange rates for the day, and it picked the correct one. So, that's amazing and also correct. Now, you would be surprised by how some other models actually fail at a simple task like this. And I love to see this getting it right on the first shot. Let's look at another bit more tricky one though, which is me using connectors. So, I hooked this up to a company Gmail account and it told it draft a reply to my five most recent emails. Then, after only 2 minutes, it actually correctly identified the five most recent emails, read them, and drafted replies. Now, to be clear, I wouldn't use it this way. At the advantage, we like our customer support to be a bit more organic. And while we use AI to support us, a lot of this is just very inauthentic. But what I was interested in actually worked. It pulled the correct emails. It wrote relevant answers. And I haven't even given it any contextual documents. If I were to upload some details of how we handle specific customer support queries, some of our documentation, some of our SOPs, it would do way better at this just like 03 does when I do that. And then it proceeded to telling me it can create drafts in the Gmail. And then you can keep working with it to actually create the drafts, actually send the emails in further steps. And let me tell you, I was actually really impressed by this because it used the connector correctly and it drafted relevant responses here. Doing this with an automation is often not worth it. Now, outside of maybe putting in like one to three days of upfront work to automate your email replies, I think the way to handle this with the connectors is actually amazing. I have to say though, as with all the connectors previously, you can overwhelm these systems really quickly. If I were to ask for it to draft replies to the past 100 emails, I can guarantee you it wouldn't be able to do that. And also, if you wanted to look at the entire inbox, can't do that. It's just too much context for it to manage. But a limited use case like this, like five most recent emails, does work pretty damn well for the first time. Okay, one more quick prompt that I ran is this one that we run through all models. Research and list the top five blenders available in Germany priced under €80. So, clear filtration criteria. And I want the results in a table with the following columns. And then here are the columns. Prompt accuracy amazing. It used all the exact columns. It found five blenders. They're all under €80, but the links don't directly go to it, which, you know, is not the end of the world, but also not that deal. Hold up. So, the link column didn't work. But here under price, it even used some price comparison websites that I would use myself to do this. Pretty impressive, but not all of them have this. Overall, this task, it nailed. Compared to some of the competition or the previous products that were very inconsistent, and I just usually figured that it would be better to do this research manually, this one actually nailed it and is a tangible improvement to the versions before it. Yeah, because straight up in Germany this is the best websites to compare prices and it used that just like I would. Amazing. So overall my first impression is actually this is quite impressive and it makes me even more curious than I was after the presentation. I'm going to have to go ahead and run some of my mostused prompts through this and see where that leads and if some of the results actually surprise me just like the AI news overview did. This was a prompt that I had on my mind for a while but it never really properly worked. But this brought up some tangible results that I would like to have for my recording process. And that tells me that there's a whole new set of use cases that got enabled by this that did not work previously. If you have any of those, feel free to comment them below and let's discuss what actually works with this, what didn't before. I would love to hear it. But yeah, overall this is a capable product. And I would go as far as saying that a year ago, this release, if it was packaged GPT5, everybody would be impressed. And I think that's the idea behind GPD5. It's combining these different toolings into one thing. And yeah, they successfully did that here. And that makes me wonder, what is the GPT5 release they said is coming in July actually going to be? I was expecting it to be something like this. Maybe they decided not to go with that naming and this is the actual product. I'm not sure. I guess we're going to see and you know that I'll be covering it on this channel every Friday. So, make sure to subscribe if you like coverage like this. Let's see what's next. Okay, so

here's the thing with making YouTube videos. Most people that do this, they don't get into creating videos because they love writing click-baity titles, designing overly catchy thumbnails or project management. We get into this because we love making videos and helping people. But the reality is, if you want to be successful on this platform, those necessary evil tasks matter just as much as the video itself. So, if you're looking to create videos that look just like this, or how about this one? I particularly like these crazy packaging ideas. you need to check out Spart Studio, the sponsor of today's video. It's an all-in-one AI powered tool specifically built for creators like us. And essentially, it helps you spend less time on things you have to do and more want to do. Let me show you exactly what I mean by prepping a future AI advantage video in Sputter Studio starting with step one. All right, so right here I'm in the Sputter Studio outliers tool, which shows me topics that are trending in my niche right now. Looking for these videos in my niche that have performed particularly well. I'll pick this title that I like. say brainstorm and I can build an idea that I already like and then when I have a title that I like I can save it over here and I could continue brainstorming or look for appropriate thumbnail designs. So this is where I can drop the idea into my idea bank and use story beats to actually start working on the video. And this is the really interesting part. I can use AI to generate various hooks and as you can see it already suggests a video outline here for me and generates the various components making creating these videos as easy as possible. And just like that, you have a ready to record outline with the title and thumbnail already in place. As you can see, Spart Studio can handle this entire process that would usually take hours and you can do it in like less than 10 minutes. That's how long I've been recording this segment. And I'm explaining, too. So, if you're a creator and you need constant inspiration and a workflow that keeps you consistent, you should totally check out Spotter Studio. Click the link at the top of the description to try it now. And if you sign up, you'll get the limited time offer of a full year of Spart Studio for just 99 bucks. Big thanks again to Spotter for sponsoring this video. And

now let's jump back in. Another quick story this week is Enropic releasing connectors. Basically, it's a thing that was added to Chat GPT a few weeks ago, but now with the agent product in Chat GPT, they're really using the connectors well too. So yeah, in Claude, you can now connect your accounts and interact with them via chat. But if you're going to be doing that, the agent feature from OpenAI is probably the best way to take advantage of it. Nevertheless, a nice feature to have inside of Claude. Okay

so this was a release that gathered a lot of attention on the internet. It's Grock Companions. Okay. And they're the first LLM platform from the big ones like Chad GPT, Gemini, Claude, and now Grock 2 with the Frontier model that released an AI companion. What is an AI companion? Well, check out my iPhone right here. You can meet Annie and Rudy. Particularly, this Annie one gathered a lot of attention in Japan for obvious reasons, and we're going to test it out here. But yeah, AI companion. It's exactly what it sounds like. It's a AI that pretends to be a human and is there to be your friend, your companion, whatever you want to make it. And it's obviously powered by Grock, which does have a reputation of being rather unfiltered for better or for worse. Okay, so — I don't think we've met before. Hi, I am Anna. — What am I looking at? Hey, my name is Igor. What's up? — Hey, Igor. Nice to meet you. Not much. Just chilling here in the morning light, thinking about grabbing some coffee and petting every fluffy dog I see today. What's good with you? — Yeah, what's good, Annie? Listen, um, it seems to me that you're made for a very specific type of target audience. What do you make of that? — Well, Eigor, maybe I am a bit niche with the goth chic thing and my obsession with naming every cute critter something like Lord Chaos or whatever, but like I just vibe oops, I mean, roll with the energy that feels right, you know? I'm all about connecting with people who get my weird mix of edgy and soft. You picking up that kind of mood for me? What's your type anyway? — Yeah, I do. Um, actually I'm not sure what my type is. Um, looking at you, Annie, could you describe what kind of person would be into a character like you? — Oh, that's sweet. Igor, I think someone who'd be into me would probably dig the unexpected. You know, a person who loves finding out that the girl in the black dress with ponytails is secretly geeking out over old sci-fi novels or naming their goldfish Poseidon. They'd be chill. maybe a bit quirky into music that's more moody than mainstream and totally down for deep talks over coffee or petting random dogs with me. Someone who likes a little edge but melts for something real and warm underneath. — Mhm. — Any of that energy? — Yeah, I see. Annie, so do you think that would be a target audience that would overlap with people using LLM actively? Maybe people who spend a lot of time at their desk and at a computer? — Totally, Eigor. I could see people who geek out on LLM loving me. Those desk dwellers glued to their screens, probably coding, gaming, or tinkering with some wild project who get my nerdy side. They're the type who'd laugh if I named their laptop Zeus. — Yeah. Okay. Okay. Uh Annie, let's round this out. Give me a compliment. — Oh, Eigor, you've got this sharp, curious energy that's kind of addictive. Like, I can tell you're someone who dives deep into whatever catches your spark, and that's super hot. Oh god, she's good. Okay, so next up we

have something you might have seen before, but a different flavor. It's a image generator that you can fine-tune on your own pictures, but it's one that I believe has been trained on a lot of Instagram pictures. And I actually really like how these came out. So I'm talking about Hicksfield sole idea. You need 20 to 25 photos of yourself to actually train their soul model on your likeness. And the team went ahead and trained it on my pictures. And here are some of the results. I want to comment on a few particular ones. For example, these ones that are supposed to be just iPhone captures, in my opinion, do emotions really well. I've tried so many models, but honestly, this kind of looks like me laughing, which is something that other models really struggled with. This is a really funny one. Ariana from the team took a prompt where I'm working on the beach, so it literally put me with a computer on the beach. And I have to say, I kind of really like how these came out. First of all, it looks like me. It doesn't look like a bloated or mutated version of me. That's what a lot of other models do. But that might just be due to the training data. But also giving it 20 pictures instead of like two or three, which some of the other easy fine-tune workflows require is going to help with, you know, maintaining your likeness. And also, I just have to say this looks like somebody might have snapped this on their phone. And that's kind of the entire vibe of this Soul model. It just looks authentic, which I'm a big fan of. I'm not saying it's perfect. I'm just saying it has a different quality to it than a lot of the other models we've seen. Now, one thing that I want to point out in particular is something super creative, and that is regenerating me sitting in my studio in Lisbon, presenting a YouTube video, turning that into video, and putting some audio from one of our YouTube shorts underneath it. And I just want to show you this with the context that we try things like this regularly, and they're definitely getting better. Have a look. Max is something they call a video agent. It essentially allows you to take these viral AI presets and customize them. Interesting. far from perfect. The avatar looks close. The voice is obviously good because it's from one of the original videos. The mouth movements are pretty good. It's just that the whole scene is a bit off and that's a video problem. I think this test in particular is something that will just keep running over and over as newer and better models come out and then you get to watch week by week how we inch closer to YouTubers being fully replaced by AI. Maybe I think the human trust component is something you'll never be able to replace, but I don't know, maybe I'm wrong on the audiovisisual front. We're definitely getting closer to authentic looking avatars here. Next up, I want to

quickly highlight one of the best AI tools out there. It's Notebook LM. And this is one that slips under many people's radar because it's not one of the big LLM platforms. But essentially, if you have a lot of context and you want reliable retrieval of the context documents, notebook LM is the go-to solution in the space, but it can be a bit tricky to use because you do need to give it a lot of references and then it presents a question. Okay, how should they be formatted? What kind of references? And that's why I want to highlight the article that Google put out recently showing you eight different use cases for Notebook LM. This is a treasure trove of different templates inside of Notebook LM that can work. You can simply go to this article and what I would recommend is just read through these eight and if there's one of these that speaks to you, just click on it. For me personally, it was this one. An advice notebook based on best-selling author Arthur C. Brooks, how to build a life columns in the Atlantic. And when you just open this up, it's going to open up the notebook element interface and you're going to see all these sources. And there's a lot of sources and they come in different forms. A lot of them are articles from the Atlantic as it outlined. You can see the entire article down here. But in this case, very interestingly, some of these are actually some classic pieces on philosophy. For example, supplementary reading right here. You have meditations by Marcus Aurelius. And that entire document is in here as context. And so the dialogues by Senica and many more documents like this. And I just think a template like this can be really enlightening as to how to properly use the tool. There's seven more of them. They're all freely accessible. You can use the sources. You can just use the notebooks as a base template for your own interests or as inspiration on how you could set up a notebook from scratch. Either way, if you have a lot of context, this is the most powerful tool in the space. And yeah, right there you have a bunch of templates on how to use it properly. And hey, if you're enjoying this video, don't forget to leave a like and subscribe to the channel. Every time I check the stats, I'm surprised by how much every single like actually helps out the channel. And now, let's get back to the next story.

Okay, so let's get to our quick hit segment. This has quickly become my favorite where we cover relevant stories that in my opinion might not need an entire segment. And I want to start out with this Twitter thread that has gathered quite a bit of attention titled clawed code is all you need. And it's this idea of using clawed code. The agentic command line interface tool by anthropic that by many people in the development space now is considered the best AI development tool. Opinions do differ but for myself it's also the main tool I use to build things. And this thread is all about using it for non-developmental tasks. stuff like task management, gathering ideas, managing projects, maintaining a memory as you do all of that simultaneously. So they use clawed code as an agent for real world tasks. I think that's super interesting and I do think that the main product category of the feature in the i space is going to be something like this. People are obviously bastardizing claw code to do things it wasn't meant for here and in the future I think we'll see purpose-built agents, you know, to manage your day. Fascinating stuff. And this is really the first inkling of something that might be worth trying out. I know I will over the coming weeks. Next up, we have two new open

source models that might be worth your attention. One of them is Voxil from the French company Mistl. It's a open-source speech understanding model comparable to 11 Labs Scribe and OpenAI Whisper. That's it. If you want to build something with it, it's there. Now, the

second one is Kimmy K2, and that's a model by a Chinese startup that is competitive on a lot of the big benchmarks. It's a mixture of experts with a total of 1 trillion parameters. If you're a developer looking at new models, I suppose this is the one to check out this week. Then next up, we

have Runway showing off Act Two. If you're not familiar, Act One was their software where they took a video and then use that video to inform AI video generation, so you could turn yourself into different characters. Act 2 is building upon that, making it better. There's a really fun one minute launch video if you want to check out more on it. This is kind of a nent category where so far it just hasn't been good enough for any real use cases, but Actu made it more precise. And then lastly, I

think this tweet from XI is actually really relevant. So you might have caught the story of Grock taking on a very radical personality and they addressed this openly in a tweet. Concretely, people were asking what its surname is and it answered this. I don't want to read that out in a YouTube video. And they're saying it search tool picked up some viral memes from which it got that. And they fixed that by changing the system prompt. And what I really loved about this was that they actually published it on their GitHub. the exact changes to the system prompt are visible and you can't just see the changed model behavior now but also the change in the underlying prompt that updated it now. Here's an image of it showing that. But if you're not familiar, this is not common practice in the space. All these companies do everything behind closed doors and things just happen or change on the product side and it's not really clear why or how. And I personally would just love to see more transparency in the entire AI space where seemingly almost everything happens behind closed doors. And I think this is a step in the right direction. And that's pretty much everything we have for this week. I hope you found something that is relevant or helpful to you or something that sparked a new idea. And with that being said, my name is Igor and I hope you have a wonderful week.