ChatGPT Agents are FINALLY Here + GPT 5 is Bad!

So this has to be some sort of like above and beyond of what operator was doing. Yeah. Oh wow. Literally I I'm not even kidding. That was just what came off the top of my head. Previously operator and deep research each brought you unique strengths. Okay. So I've thought of some use cases for the agent. So for example, I'm working on a new product called Gold Scraper and I just feel like the landing pages are a little bit stale and static. So, what I want to do is figure out if I can't find uh the most cool looking websites and write an HTML and prompt for my product to, you know, improve upon it. Hi, my name is Demetri Bonichi and I'm a content creator, agency owner, and AI enthusiast. You're listening to the AI agents podcast brought to you by Jot Form and featuring

our very own CEO and founder, Idkin Tank. This is the show where artificial intelligence meets innovation, productivity, and the tools shaping the future of work. Enjoy the show. So, there's been a lot of new updates with Chad GPT recently, and there's a couple different things I want to dive into. One being Chat GPT agents, and the other being Chad GBT. So, I figured it would be a nice overview of everything just to cover it all here. It's crazy to me that we've been in a position where all this stuff is moving so fast and I just want to make sure that we're up to date with everything. So, let's dive right into another episode of the AI Agents podcast and show you all of how it works. So, diving right into ChateBT5 and the agent system on their website, you'll see that it does a

bit of a nice promo video showcasing how this tool has the ability to uh, you know, plan a trip on Expedia, for example. And we're not really going after that type of group of people when it comes to uh you know, you guys are like business owners and SMB and I just don't really see the first example being something that matters. Yeah. So, let's see some of the other examples that it showcases. No, creating slideshows and presentations. That's pretty nice. Some more spreadsheet type stuff. Action agents take action on the web. So, this has to be some sort of like above and beyond of what operator was doing. Yeah. Oh, wow. Literally, I I'm not even kidding. It was just what came off the top of my head. Previously, operator and deep research each brought you unique strengths. Okay, so I've thought of some use

cases uh for uh the agent. So, for example, I'm working on a new product called Gold Scraper, and I just feel like the landing pages are a little bit stale and static. Um, so what I want to HTML and prompt for my lovable.dev uh product to you know improve upon it. Uh so let's go here. And the nice thing is good old audio. Could you using uh some deep research figure out and find some of the most cool looking websites and then from that help me improve my current lovable.dev product which is called gold scraper.io to make a better branded uh experience across all my landing pages. cuz I just feel like it could use some spice and look cooler. And I would love for you to generate a uh prompt to give to lovable.dev. Uh and um yeah, just give me a report on

what makes websites look cooler. So sometimes I like to audio prompt because it's easier. Actually, most of the time I like to do that. So I have agent selected and we're going to go here. And while this is running, I do want to just uh talk about GPT5 and kind of how it works for a second or not plan, sorry. There's a couple different models you can choose from here. Auto, which decides how long to think. Fast, which gives instant answers. Thinking, which thinks longer for better answers, and Pro for rearch research grade intelligence, and they recently just let you use uh GPT40 again. So, you can't really use the older models. And I won't lie to you, I'm actually not too thrilled with it. Oh, wait a second. Um, it's starting to use the browser now, which is really cool. Uh, I'm not too

thrilled with GPT5 in the sense that I want to use older models. I want to use 03 and 04 and stuff like that for specific uses. I do think they were awesome models. Uh, and what's just intriguing to me is that they just launched it all and is kind of forcing you and it it kind of is like forcing you to be put in a position where you use the models they want. And just to be blunt, I do not trust and this this is not only skepticism but truth in the sense of from my testing. It's like if I don't put the search functionality on and I ask a question of it, it's like it gives me information too quickly. So what do I mean by that? I just don't feel like it's doing any sort of like reasoning and the pattern detection just

is like set to a thousand and I don't really like that. So, if I ask a question like, "What are some new AI agent tools?" Oh, now it's actually search the web because I put new. That's good. But okay, cool. I like the new layout, by the way. I do think this is cool. Um, what is the best way to get better at increasing your profits in business? Okay, so it automatically went to fast the other day. what it was doing and maybe they've tweaked this is that it was literally the second I pressed enter it seemed like it was responding and I'm betting what it was telling you or what was telling a lot of people was incorrect information and then they had to change it like when I'm been working in a lot of my equation type things if I don't use the

thinking model and even if I use do use the thinking model I'm getting completely wrong equations for stuff like make.com that previously I would have worked before I just don't find the information as accurate it used to be which is kind impressively dumb. Um, so what's really cool is this is going and checking out the actual websites as you saw in the uh preview it was doing it. So that's something really cool. I'm going to spin up another one here while this is going and I'm going to ask the question, find me the latest AI news articles that have interesting information by browsing Google Trends for news. Yes. Spin up an agent. Okay, so I got this markdown report from the first one. Give me it. Objective: Gold scraper current landing page professionally designed, but the founder feels like dark mode and calm color palettes,

motion scrolling and micro interactions. Okay, seems like it actually went to these websites, which is cool. Okay, so it went to the landing page and did find the information. This is accurate. This is what it says. Get answers, not data. So it did go there. Golden yellow. So it figured out the color scheme. That's good. areas where the landing okay lack of motion and it micros okay recommendations for improving the landing page adopt a rip rich richer color palette with dark mode support I don't want that uh dark mode support but scrolltelling okay so it's telling me to add scrolltyelling so that there's more info when I scroll create a unified design system okay I think this is decent decent advice so this is a ne decent idea I don't mind this let's go to latest AI news so it's opening all the different pages

which is I feel like the preview window. It's weird that it's not showing it properly like it was. Now it may be that I'm not in the web search possibility or sorry that I'm not on the web version that it's not showing it right. Ah, so I'm not on the web version. It's not showing it right. That's kind of silly. I don't like that whatsoever. I don't like that it's forcing me to go to a browser. I want to download an app. So this shows it better. And I apologize for not showing it through the worst means, but apparently this is what it was designed for. So, can't access the previous source due to cross-sight restrictions. Okay, nice. So, we can rewind though, which is pretty cool. Let's go back to uh the website design. Can I rewind? Oh, I can rewind here. This is

very cool. Okay, so it's showing that it's scraped stuff and scraped websites. Okay. Well, now, okay, it checked out my website, which is cool. I want to ask another thing. Can you go on some of the websites and scan them to enhance the prompt? Also, I really don't want a dark mode. Light mode is fine. Give concrete suggestions on what micro interactions would look like and scrolling. I do not like that I'm having to go here into like what's the point of it having a desktop version if it doesn't have all the components. How hard would have it how hard would it be? I don't understand how how hard would it have been to implement this into the desktop version. Would have maybe it's so hard that it obviously it's not doable. I don't know like what what I like how it's talking through what

it's doing. I'll scroll down to the micro interactions and scroll it telling effects. I don't feel like it's the best designer on the planet, though, because this website's very okay. But I do like some of the components it's finding. I just don't feel like this is like some new sleek modern website at all. All right, this is the more sleek website. Uh, more commentary on this, by the way, from GPT5 is this isn't just my opinion. We're seeing when I looked up GPT5 reviews, this is the top result. GBT's long promises faster speeds and more time spent thinking, but the new generative AI model has turned off some users and with a tone shift away from its casual conversational style. Um, apparently it's been in the work for months. GPT5 itself is really a couple of different models. There's fast a straightforward LLM. Users quickly

rebel to GPT5. Many said was it's more stodgy and had less personality. So, uh, I use GPT. You could just use GPT40 if you have the API uh, connected to automations, by the way. So, don't worry about that. Um, on the code writing side, it seems like it's pretty solid. It has some tweaks to make things safer. That real that makes sense. It can connect to your mail and calendar and it's a significant step along the path to AGI. However, maybe it's the AGI of a bad associate because I don't think it really got much better. And you'll notice a lot of people like if I look up reviews, all of it says underwhelming. So, I don't think it's just me. The Reddit GBT5 is horrible. GBT5 is awful. Let's check out this Reddit thread. If it gets something wrong, no worries. Just ask again.

Seems like the inaccuracy thing I mentioned earlier is consistent with what they're seeing. Yeah, some people are saying it improves their business model, not our experience. And that's the immediate skepticism I had. Like it was doing immediate Oh, this is actually a really cool website. Let me scroll through what it's been looking through. Yeah, okay. It's going through better websites, so that's good. Um, but an immediate problem that I, like I said, found was I don't understand why they would make it so quick responding. And all that can mean if it is so quick responding is just that it's quite literally utilizing information that it's like cached and then pattern recognizing an answer. But that means it's just not going to directly answer my question. It's going to answer something related to my question. Um, we're not at that point in time, right? Like and

even if we get to insanely fast cloud compute speeds like it's going to be another couple decades before the speed I was seeing like it was literally press enter response was a paragraph done in 04 seconds and the difference between what we're seeing what you just saw was it was like 2 seconds and I know that might not seem like a lot but that means that it at least thought about it and that's what I was concerned about. So, but as you can see, this fully can go through different websites, and I just really want to get an answer to the question. That's good. So, let's see. Latest AI news, what it did. Went through the different articles, gave me a breakdown, has weird spacing. I do not like the spacing at all. Why is it to the right like that? It's not centered. Somebody

at OpenAI is going to get fired. No, just kidding. Uh, summer of these trends. Open eyes open GPOSs. Open also reverse course by restoring GBT40 for paying customers. Yeah. So that wasn't available a couple days ago. So um like literally you couldn't you couldn't see this. Oh, sorry. Not temporary. You couldn't see this which I find it so funny they only gave us 40 back. So overall pretty good giving me these news notes for me. And now it's delivering the final message for me. Ooh, enhancing Gold Scrapers aesthetic. It's giving me some more insights. Revise prompt. Oh, I like this. This is a pretty good prompt. Base the site on a light neutral gradient with our gold accent. All right. So, let's see what this actually does to the website. I want to show you a practical outcome. Let's see what it does to the

homepage, which you can see here is decent, but it's clearly like AI generated. And I'm trying to make it less AI generated. So, with agents, I can give it some insights, you know. Now, should I hire a designer and have them make a Figma file? Probably. But here we are. And you know, I could add this to the website and it would probably make some improvements. Um, that's something for me to try on my own though. Don't want to break my site. But I will say I think it's really cool how it managed to go through and analyze different things and you know cuz when I say to AI and this is very common, right? Do XYZ thing, right? Make it look better. That's cool and all, but you got to tell the AI something that is different. And something to make it different would

be to have it look at things. And having it look at websites is going to be different than it being remembering whatever's in the the language model or on blogs. Uh saying what is a good uh website. So that that would be the key here is whenever you're doing this type of stuff, use it so that it has a visual reference. I think that's going to be cool agentically. Uh but for all intents and purposes, make sure that you're doing this the right way. Make sure you're QAing everything. I wouldn't go and just pop this in my website like right away. I'd have to make sure it actually works, right? So, please, please, please check everything out in here and check out everything that we have on the podcast. We really appreciate it that we've grown so much recently, but you guys are the reason

we do our work. We are so grateful to make the AI agents podcast for you. Look out for all the episodes with myself and the guests out there, and we'll see you in the next one. Peace.