Everyday AI Podcast – An AI and ChatGPT Podcast - Ep 637: ChatGPT’s New Agentic browser: Hands on with OpenAI’s Atlas

Episode Date: October 22, 2025

ChatGPT just released their agnetic browser, Atlas. 🌏Will it kill Chrome? What does it do? How does it incorporate ChatGPT? We'll answer those questions and more on today's show.  ChatGPT’s New Agentic browser: Hands on with OpenAI’s Atlas -- An Everyday AI Chat with Jordan WilsonNewsletter: Sign up for our free daily newsletterMore on this Episode: Episode PageJoin the discussion: Thoughts on this? Join the convo and connect with other AI leaders on LinkedIn.Upcoming Episodes: Check out the upcoming Everyday AI Livestream lineupWebsite: YourEverydayAI.comEmail The Show: [e-mail] with Jordan on LinkedInTopics Covered in This Episode:OpenAI Atlas Agentic Browser OverviewKey Features of Atlas Agentic BrowserAtlas Sidebar ChatGPT IntegrationAtlas Browser Context and Memory CapabilitiesAgent Mode Automated Website ActionsCursor Collaborator for In-Page AssistanceAtlas Browser Security and Data ControlsHands-On Atlas Browser Demo & Use CasesAtlas vs. Comet Agentic Browser ComparisonLimitations and Accuracy Issues in AtlasAtlas Browser UI/UX Design CritiqueFinal Verdict: Atlas Browser PotentialTimestamps:00:00 OpenAI’s Atlas Browser Unveiled03:25 Atlas Overview and Availability07:05 Streamlined Contextual Browsing11:23 ChatGPT Integration in Atlas Browser15:29 Live Demo: Agent in Action18:46 Atlas Browser Agent Features Explained20:28 Atlas Browser Overview23:44 AI Agent Use on Website29:36 Atlas: Enhanced AI Search Experience30:25 Streamlined Agentic Browser Feature34:06 Resizable Chat GPT Sidebar38:09 Monitoring AI Output Effectively40:37 AI Operating System Ambitions43:43 OpenAI's Game-Changing PotentialKeywords:Atlas browser, OpenAI, agentic browser, ChatGPT Atlas, AI browser, agentic features, agent mode, browsing context, browser memory, sidebar ChatGPT, Chromium-based browser, browser integration, paid account features, Mac operating system, cloud browser, browser security concerns, enterprise browser, chat history sync, website memory, context sharing, non-agentic browsers, copy-pasting workflow, automated website navigation, automated form filling, cursor collaborator, in-browser AI assistant, natural language browser search, browser prompts, search history integration, automated shopping, agentic checkout protocol, agentic transactionSend Everyday AI and Jordan a text message. (We can't reply back unless you leave contact info)

This is the Everyday AI Show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business, and everyday life. Meet Firefly AI Assistant, now live in Adobe Firefly, the All In One Creative AI Studio. Just describe what you want to create and the assistant handles the rest, orchestrating multi-step workflows across Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome. The assistant accelerates execution. OpenAI released probably their most important product to date.

Their agentic Atlas browser. A browser? Their most important product? More important than GPT5 or SORA 2 or their new agent builder. Yeah. Here's why. A browser is how you own user data and can. control their habits. After all, Google was paying Apple reportedly more than $20 billion a year

to be the default engine on Apple's Safari browsers, because if you control the browser, you control where the user goes and ultimately the monetization that comes with it. And with OpenAI already completely controlling the AI chatbot market, they want a slice of Google's browser pie. So is their new Atlas agenic browser the answer. Will Open AI go all in on making this work or will it get kind of ignored like GPTs did the past couple of quarters? Well, we're going to dive into at least some of those things and talk about how it works and how you can save time by using OpenAI's new Atlas browser on today's show. So welcome to Everyday AI. What's going on, y'all? My name's Jordan Wilson and

this is your daily live stream podcast and free Daily Newsletter helping everyday business leaders like you and me make sense of all of these AI happenings and make use of the tools that actually matters. It starts here with the unedited, unscripted live stream podcast. But if you want to take it to the next level, make sure to go to our website, your everyday AI.com. Sign up for the free daily newsletter. We're going to be recapping the highlights from today's show, as well as everything else that you need to know happening in the world of AI. So open AI's new Atlas browser. What do you think? I have thoughts,

but let's just go ahead and take a look at what we're going to go over on today's show. I'm going to show you the main features of chat GBT Atlas, OpenAI's new agentic browser. I'm going to talk about some of the more intriguing features, capabilities, and some limitations that you should be aware of. I'm going to go hands on with at least three demos, maybe more. We'll see how long they take. And I'm going to give you my final verdict on if Atlas is worth using.

And if you are new here, every single Wednesday, we do a little segment called putting AI to work on Wednesday, where we show you new tools and techniques on how we use AI here at everyday AI. This is one of those because Atlas has only been out for a couple of hours. We're not going to be able to go into a ton of depth. And if it's helpful, maybe we'll do another Atlas show in the future. But I wanted to cover the basics, but also tell you right now how you can use Atlas to save time. because it's available now. Yeah, this is, luckily,

this isn't one of those things that, you know, got put on a wait list, you know, like the original SORA did. And nine months later, everyone finally got access. You can go download Atlas today.

Even if you are a free user. However, some of the more agentic features do require a paid account. But it's available today in preview for Plus, pro and business users. Right now, it's only available on the Mac operating system, but OpenAI did say that they're going to be

rolling it out to other devices, including mobile and Windows soon. So before we get into agenic browsers, all right, and hey, live stream audience, great to see you, Jose joining. Joe says he can't run it on his Intel Mac. Dr. Ruth, thanks for joining us, Marie, Sarah, Gordon, big bogey face on the YouTube machine, Brons and Tim, everyone else. If you have questions or comments about Atlas, please let me know. But we first have to start with why agentic browsers matter. I don't think this is one of those cases like if it's not broke, don't fix it. Right. What's wrong with Chrome? I've been using Chrome for, I don't know, how long has it been out? 15 plus years. There's nothing wrong with Chrome.

Sometimes, except when there is. And this isn't a Chrome. And this isn't a Chrome. thing. This is everyone, every browser, right? All of a sudden, you have 15, 20, 30 tabs open. You're constantly having to copy and paste information to share context between all of those tabs. This is where agentic browsers come in. So if you're brand new, here's what an agentic browser is. And then we're going to go over the newer features of OpenAI's Atlas browser. So Atlas can understand the context of everything you browse. That's the big thing, right? It can also, if you're on a paid plan, it can take over your screen, navigate websites, fill out forms, and complete actions all on your behalf. So many of the things that a human sitting down at a computer could do,

Atlas can do if you're on a paid plan and you can take advantage of those agent features. All right. I like to tell people this. It's like having an intern who has access to a computer and all of your information, right? Think maybe you're working remotely or, you know, you have a family member at home and you're like, oh my gosh, this, this huge project, right? And you got to call them and you got to talk them through it. That's what agentic browsers are now. Very useful if you have the patience and the, if you've practiced prompt engineering for an agentic browser, it's a little different, how you want to talk to a browser in order to get it to accomplish your goals. But, hey, with this open AI integration, it should skip a couple of steps. You can, it can actually

use your actual browser. That's big while you're logged into websites or it can use a cloud browser where you can store your credentials. With that obviously comes a lot of security concerns. So you have to run those through, you know, whether you're using this in a business standpoint as part of a larger organization. We'll see what Open AI does to better prepare Atlas for the enterprise. I think there's probably some things that need to be done. There's also a sidebar to talk to chat gvety on any page. Right. So essentially chat chabit can easily follow you around and not only

understand your history and context, but quickly help you better understand any page you're on with just essentially two quick clicks. And then the big one, obviously is it carries your entire chat chabit history and context into the browser. That one's important. So if you have memory enabled on your base chat chbtbt account and if you enable website memory or browser memory, it's going to be a, hopefully a somewhat seamless experience to share that context. And here's why that's important. That is ultimately why non-agentic browsers

are technically broken, right? Because you might have 30 tabs open and you have to spend sometimes so much time carrying context or quite literally copying and pasting different content over and over. So you might, as an example, open an email, say, oh, I need to do some research on this, open a couple tabs. You know, you might copy and paste certain things from those tabs into chat TVT or Gemini or Claude. You might start to create some content. You might take that into, you know, a Canva or create a PDF or something and then

ultimately go back into the original email, right? There's a lot of going between different tabs and using different now AI features and ultimately carrying contact between tabs. So that's what agentic browsers hope to cut down on. So here's what's new in Atlas. So the sidebar, number one, like I said, a little Ask Chat, GPT toggle, no matter what website you're on, very quickly, not just get context from any web page within one click. You can talk to any web page, get anything you want from there.

So nothing terribly new there, almost any agentic browser has that capability. But the big thing is those chats will ultimately. end up in your chat chb t account right because when you log in uh atlas is based on chromium okay so it's going to bring in you'll have the option to sync it to one of your google accounts that you use chrome on all agentic browsers are based on chromium uh chrome's open source standard right so it's going to bring in all your you know existing chrome data or you can start from scratch but you do have to start with one of your chat chb t accounts okay so it's going to bring that context over as well but at any point, you can click the Ask ChatGPT button in the upper right hand corner.

It's going to give you, based on the content of the page, it's going to give you some pre-baked prompts that you can just one-click to get better information, or you can just chat with the webpage with ChatGPT right there. You can pick up where you left off. That's the big name, right? Not only having access to previous conversations in ChatGPT, but later, maybe you're not even in Atlas. Maybe you're in Chrome or perplexities comment or a different browser.

When you go into your chat GPT account, you can pick up where you left off. That part is nice. That is a great feature to have because, yeah, let's say that you're getting some great work done in a browser tab or with the sidebar. You don't have to worry about saving that or copying and pacing that. You can pick up where you left off. The big one, if you are on a. paid plus pro or business account agent mode you can activate it and it can complete actions on your behalf i don't know why every single ai lab when they're trying to demo agentic features it's

always shopping that's the last thing i don't know i don't know if it's just me live stream audience do you actually want an agent or an agentic browser to shop for you not me um you know but you could have it go look up recipes and then you know i don't know i live in chicago you could go it log into your Juulosco account and start adding certain things to your cart. I don't know. From a time savings perspective, at least right now, and we'll get into this when I start to give you my verdict and my take. It's not a huge time savings to go have an agentic browser or even an agent.

Always go add things to your cart. All right. The next big feature is there's a new kind of cursor collaborator. So this is anywhere inside the Atlas browser where you are typing. So it could be in an email. It could be in a Google doc, right? There's when you highlight the text, there's a very small. I think OpenAI called it a knob in their live stream,

but a little colored bubble that you can click on and then chat with chat GPT right there wherever you are typing. So you're not going to get the full chat GPU interface, but essentially a little box that, you know, if you're typing up an email and you're at a loss for words, if you need to bring in context from your chat GPT account, whatever you may be trying to do when typing a quick way to get that assistance right away. Also, chat ChbT did say that they're taking security very seriously. And here, they say you can decide which sites chat Chaptee can see, clear your browsing

history, and manage browser memories anytime. So we're not going to go into all the browser settings. You all are smart enough. You can go in there and click settings and click through them all. All right. So I'm excited. live stream audience, let me know what you're thinking. But before we jump in live, because we're going to do at least three different live use cases here. But before we do that, a very

quick word from our sponsors. This podcast is supported by Google. Hey folks, Stephen Johnson here, co-founder of Notebook L.M. As an author, I've always been obsessed with how software could help organize ideas and make connections. So we built Notebook LM as an AI first tool for anyone trying to make sense of complex information. Upload your documents and NotebookLM instantly becomes your personal expert, uncovering insights and helping you brainstorm. Try it at notebooklm.govolm.com. All right.

Let's go ahead and a couple, couple questions from the audience here. If you have any questions, let me know. Marie says, an agent shopping for me, no, an agent paying for my shopping. For sure, yeah. Eventually, I do see that coming right now. not the agent actually paying for your purchases, right? But right now, it's not going to check out all the way for you. It's going to add things to your cart.

But one thing, and this is part of Open AI's bigger play here with Atlas as well, they did introduce a agentic checkout protocol. So it does have the capabilities, right? They're not baked in or enabled yet. But I would assume in the future, the agent will actually be able to complete transactions. Right now it's just going to add things to your cart. And from their kind of baked in guards rails, it's going to leave things off there for you and say, hey, I did all this shopping for you. Do you want to continue?

Big boge question says, can you have a chat with Atlas? Yes. Technically, you're chatting with chat chvety. And I will show you how to do that in any sidebar of any window that you're on. You can chat with chat chvety about the content on that page or anything else. Joe says, if it shops for me, I want it to find options. Let me make the decision. Same Joe.

Brian says, I've been using. comment so interested to see how that stacks up. Yeah, me as well. And hey, live stream audience, podcast people, if you want me to do maybe I could tomorrow or Friday, if you want a comment comparison, I'd say right now, perplexity's comment is the leader in terms of agentic browser.

So if you want to see kind of a head-to-head comparison, just go ahead and say comment. I don't know. I have a number in my head. If enough people want to see that comparison, we'll go ahead and do it. All right. So I am now sharing my screen.

So podcast audience, I'm going to do my best to describe everything to you. But right now I have a comment or sorry, a Atlas browser open. And I've logged into a number of websites. Okay. And I've connected it obviously to my chat chitbc account. I've synced it to one of my Chrome profiles. So I have my podcast stats open here. I'm going to start this prompt.

And this is going to be an agent mode prompt. So again, you do have to be on a paid account. And this one's probably going to take, probably going to take a couple of minutes. So I'm going to get it started. All right. It's going to be like a cooking show. We'll see if the cake is actually going to bake in the allotted time or if I'm going to have to show you the end cake. So in agent mode, I put in a simple prompt.

I'll tell you what I'm doing. But if you're watching the live stream, now you can see that Atlas is actually controlling my screen. There's a little bit of an animation. right, with some different colors and dots and all these things. And I can see that it's thinking. So obviously, the Atlas is powered by a thinking model. So, and it's already failed, right? All right, perfect.

So let me at least say as Atlas figures out how to navigate my screen, what I essentially asked it to do. And I enabled agent mode. So it's going to be navigating and clicking around here as I talk about this prompt that I did. So I said, in my Buzz Sprout episode stats, find my most popular episode from 2025 quarter to. Go find that episode and remember the content. Then run a deep research query using Google Gemini with an emphasis on updating the episode for this quarter, making sure to update old info, insert new stats, etc.

The final output should be a well-formatted episode outline based on the previous podcast transcript. and the new information gleaned from the Gemini deep research search. Follow each step precisely. As an example, you must do the research with a Google Gemini deep research query and not using chat GPT. So this is an example of something that I would be doing with my own time, right? I'm showing you this is putting AI to work at Wednesdays. I'm showing you how I'm probably going to be using, you know, Atlas and how I have been

using, you know, other agenic browsers. So in short, I'm trying to find a popular episode from a couple of months ago by looking at my stats, then go find that episode online. And you'll see in some instances, I gave Atlas very specific instructions, right? I said, hey, you have to go use Gemini, Google Gemini deep research. But I didn't tell it, go to my website and find the episode. So certain things, I'm giving it very specific instructions, certain things I'm leaving it open for interpretation to test how well this works, right? I am okay at prompting. So I could get this to work at almost 100% rate with trial and error,

looking at the chain of thought, reprompting, et cetera. I want to find something in between. I want to find a prompt that may be or an example that a lot of our average listeners might be doing. Right. So again, I'm logged into multiple websites here. I'm giving it some specific instructions. I'm giving it an angle, but I'm also leaving a couple of things open for interpretation. So we're going to check in on this prompt and hopefully the results in a little bit,

right? Because again, putting AI to work live. We're going to see how it works. So a couple of things under the hood, which is good to know. Anytime that you have, you know, so right now I, I have multiple tabs open in my Atlas browser. Anytime there's agentic features going on, if you look in the top tabs, you will kind of see a floating cursor, right?

So what that means is that there's a agent working in that tab. And if you go to close it before it's done, you will get a warning, which is nice, because not all agentic browsers do that. All right. So now let's hop over and opening a not even. Okay, so I just have a new tab here. And I wanted to start on that prompt to give it time to see if it could finish live. But now let's kind of start at the beginning.

And I want to walk everyone through the interface because it's actually a little different. There's things about it I love and there's things about it that are actually confusing. Well, because this is a browser, right? So if you click new tab or once you open Atlas, it's going to send you to a page that looks like at GPT's homepage, but it's actually very different. And I actually wish from a UI and UX perspective, open AI should have made these way more, like, starkly different. Right. I get what they were trying to do.

They were trying to make a nice, intuitive and seamless transition for people to kind of use this homepage. But the problem is, it looks too much like chat gptu.com. And the capabilities and features between a. new tab or this kind of home screen of Atlas and chatchupt.com, they're too similar and the functionalities are very different. All right. So small, small, uh, small, uh, uh, bone to pick, I guess with open AI, but, uh, that's fine. So you will see this kind of home Atlas screen anytime you open a new tab or launch the Atlas browser. You will still have the model selector in the upper left hand corner, uh, but that's it. It's a much cleaner.

interface. So you can still at that point, just start typing. Or another thing that I don't like, maybe this is from using browsers for too long. The address bar is kind of hidden by default. Right. So you have your normal kind of chat GPT input area. But then also, well, this is a browser. But if you look in the normal URL bar, you won't see anything. So you have to go up and hover and then you'll see a URL bar. And also the URL bar. And also the URL bar is. is centered. So some small things I don't like because normally when I'm browsing, and I'm sure it's like this for the majority of people, my cursor instantly goes to the upper left hand corner, which right now does nothing. All right. So anyways, some differences

between a normal browser, the chat gpt.com website and the new Atlas home screen. All right, also from the Atlas home screen, there's a couple of things. It will give you suggestions, kind of quick actions based on what tabs you have open. So it says, you know, one of my suggestions is to open Buzz Sprout and look at my podcast metrics. I can enable agent mode from here, some other things. Okay, I can also click the plus button. I can add, you know, kind of some normal chat chaptain interface items here. I can add tabs and chat with those tabs.

I can enable agent mode. I can just search something with browser memory. I can still do image gen, deep research, tab search, and a couple of more things here. All right. So you get most of the capabilities of chat gbt.com in the Atlas browser, but not all of them. Right. So as an example from here, I can't look at my chat history. I can't go and use GPTs or projects.

So it's almost like they were trying to make it like, yeah, it's just like using chatGBT. But by doing so, some of the key elements that you would use aren't even there. So it's, I think it would have been better. for open AI just to go in a different direction and not try to, uh, kind of confuse the two. All right. Uh, enough about that. Let's start doing some more use cases here. All right.

So now what I'm going to do is I've opened my website, your everyday AI.com. So make sure you go, um, you go check that out. All right. So I'm going to go ahead and run a prompt here. So I have it typed out. So you don't have to, uh, watch me type. So. now in the upper right hand corner, you will see this Ask ChatGPT button.

Okay. So it's going to, okay, interestingly enough, it's there we go. All right. So it pre-populated some prompt examples. None of these are ones that I would actually use. But I'm going to put in a quick prompt. And I'm going to say, find the last three episodes about chat GPT and summarize the three biggest takeaways for this episode.

So because I am on the your every day. A website, I didn't have to give it more directions, right? I didn't have to say use your everyday AI. Hopefully, agent, uh, the agent mode in Atlas will understand and, um, do this action for me. Again, this is something that I might be doing, um, on in, you know, somewhat, somewhat routine basis, right? I go to our website actually a lot and refer to different things, right? If I'm, uh, you know, going on an interview, doing consulting for a company, you know, they want to know about a I'll actually go back and, you know, reference our own material a lot. And you should as well. All right. So, right away, it did a pretty good job. So it went and found episode 635. That's correct.

That was the AI news that matters. It looked at our chat GPT apps episode from 632. That's correct. And episode 626 where we talked about chat GPT's new agent builder. So it did a pretty good good job and then it left kind of sources at the bottom. So you don't have to enable ancient mode. So even if you're on a free plan, it can still do a pretty good job of quote unquote navigating a website without manually and actually navigating it. All right. So and then I can also click into an episode as an example. I'm going to click into episode 634 here. And I'm just going to say summarize this page right um and then it's going to take the context of this page and probably in a quick second here it's going to give me a quick recap all right uh today it's actually very very slow so

uh they just released atlas uh probably what about 16 hours ago uh so i'm guessing their their GPUs are melting over there at OpenAI. But slowly, but surely, it's giving me a pretty decent summary, although it's going very slow, especially compared. Even the hour it was first released, it was much faster. So you'll probably have to, if you are using Atlas, kind of keep up with that. All right, let's do another example. So I'm going to open a new tab here inside Atlas.

So one other thing, that's a one other thing. that's great, is you can do this browser memory. Okay, so you can search your browser memory. So when you import or connect Atlas to a Google account, it will import your bookmarks and your history. I thought that it was going to import extensions. It didn't, which I was super bummed about because the other agentic browsers that are based on Chromium, mainly Microsoft Edge and Perplexity Comet, did a great job of importing extensions, Chrome extension.

So I don't know if that's a bug right now in Atlas. Huge bummer for me. I have literally thousands of extensions installed, not all active, but I'm a huge Chrome extensions guide. So big bummer for me to have to go through and manually add those. Anyways, it will also import your browser history if you ask it to. And then from there on out, when you use Atlas, it will know. So I have a simple prompt.

I said in my search history, what was the post about someone getting a DGX spark from Jensen? And you'll see here, this is not a lot of context. All right. So I'm going to go ahead run this prompt. And what it should do, it says searching browser memories. So hopefully it's going to know I looked at a Twitter post that we put in our newsletter last week when NVIDIA CEO Jensen Wong delivered their new DGX Spark, which is a kind of a personal supercomputer, uh, to Greg Brockman. All right.

Who is the president of open AI. So let's see. Uh, the first time I did this, it got it right spot on. All right. The second time, not so much. Let's see. Okay. No, it did.

It did. It got it here. Um, so it said, I checked your history for pages about DGX spark in Jensen. And the one you're thinking of was almost certainly great. Brockman's post on X from October 14th where he wrote thanks Jensen for the hand delivery of DGX Barth best delivery ever. Okay. One thing that I'm not a huge fan of in this instance, it doesn't give you that exact post to go and click on. In my testing, it did this a couple of times. Sometimes it didn't. So not a lot of

consistency. And there's definitely some room for polish. because think this is an extremely useful feature. This is something that Google Chrome, they've been slowly introducing new agentic features inside Chrome. This is one that they've announced, but didn't come out initially for all users that I would really like. I spend a decent amount of time going through my Chrome history, right, trying to find certain news stories I referenced,

certain AI developments, new AI tools that maybe we shared about the newsletter. I'm like, ah, what was it? I kind of know, right? But the good thing is here, you only need natural language, right? You don't have to, you know, use some, you know, bullion operators, you know, you don't have to have some structured input, you know, like you'd be searching for, you know, something in your email, right?

Just like, hey, in my search history, what was this? All right. Let's do one or two other things, more interface-based. So again, I'm opening a new tab. I'm just going to search for everyday AI. Okay. So there is obviously almost a new search experience inside of Atlas. Okay.

Uh, I think some people are really going to like this and some people aren't, but let me just tell you what happens. So I just typed in everyday AI. Okay. So obviously, this is going to bring in my context, which I talk about everyday AI a lot inside my chat, you pt account, but it's also going to browse the web. So there's different tabs at the top. There's five different tabs.

So if you're used to a Google search, this is probably going to be like a Google search. But the home tab is more of like a traditional chat GPT search. And one thing I really like about this is it does cite the sources for everyday AI. So it looks like the first site that it's sourced was our Spotify web page. So if I click that and if I am in the whole. home view or the default view, all right, it's going to do this nice little slide out animation there. So what happened for our live stream audience or for our podcast audience? I clicked on that. And it brought up the website on the left hand side and then kept that kind of chat result

active and slid it over to the right hand side, which is a fantastic small UX feature that I think is going to be really helpful. Because I think to get the most out of agentic browsers, you want to streamline kind of the way that you work with it. You don't want to have, you know, more and more tabs open. You actually want to minimize them and keep a seamless experience. And then when I'm done, I can just click back to chat. So it's a nice slide in, slide out feature that will probably allow you to get more done, right?

Which I really, really like this feature. All right. But let's go back to kind of this new homepage interface. So that is the home view for my everyday AI search. But then I can click this search view. All right. And this is going to be essentially a Google search. All right.

So we essentially have some blue links, right, traditional blue links and meta descriptions. So this is more of a, you know, SERP search engine results page, but like Google. And I can click this and it's going to launch it in Google. So if I do want more results because you just get the first page, but essentially you get the first page of search results. If you click the search globe, then you can go to image. These are image results for everyday AI. So similarly, it's just like you go to a Google search and click images, very much the same.

Videos, same thing. In most cases, a lot of your first results are going to be from YouTube because this is pulling in Google search results. All right. And then last but not least, a news tab. All right. So more or less, these are. Google search features that if you do a normal kind of chat or search inside of Atlas, you're going to get this kind of familiar by default, the chat GPT interactive experience.

But then you get a traditional web query, web results tab in images tab, a videos tab, and a news tab that you can toggle between in the upper area. All right. So we've done a couple of things. We've kind of showed you agent mode. We've showed you this side how you can talk to chat GBT. All right. We've showed you kind of the difference between the kind of home, Atlas home interface

versus the chat GPT interface. So let's go ahead and check in to see if we finished our initial prompt. The one that we started with, our agent mode. So it did work for nine minutes. All right. So the great thing is I can. Now go into my chat GPT account. So I'm logging into my actual, you know, chatGBT.com.

And I'm going to go down to my correct, uh, my correct one here. Let's see. There we go. Okay. So it didn't give me the results yet. Uh, so I'm going to have to refresh. All right. Interesting.

Everything that I, that I try once in a demo works great. Uh, this one didn't give me the results, even though the results. are done. So we'll have to look at it in this kind of normal agent mode. The good thing is when you're using the Ask ChatGPT sidebar, it is resizable, right? So what I'm going to do now is I'm going to resize this and we're going to look at some of the results. So again, to catch you back up to what we started 12 minutes ago, I logged in. So Atlas has access to my content, my context, my search history, my chat GPT account, everything. So I enabled agent mode. I had it go. I had it through my Buzz Sprout stats, my podcast stats.

I asked it to go find an old episode from a previous quarter. That was our best episode. Then go do research. Go look on our website, see what that episode was all about because you're not going to find that in Buzz Sprout. Then I had it instructed it to go use Google, Google Gemini's Deep Research. Yeah, that's good. We can have Atlas an agenic browser.

Go use other large language models. one of my favorite tips or tricks to do. So I can go and look at the chain of thought. If you don't know what that is, if you look now on the right hand side, there's going to be a worked for blank minutes. I can click that and then see exactly what agent mode did and didn't do. All right, if you really want to get good at AI, y'all,

start reading the summarized chain of thought. All right. I'm not going to bore you guys and walk through it, right, line by line. but I can go down to the bottom and ultimately see if it got this right or wrong. And one thing, if you watch this live, it is extremely clunky. The agent mode is slow. Let me just call that out. This prompt should not have taken nine minutes.

And we'll actually go down and see the results and see if it did everything correctly. All right. So the output that I asked for was an episode outline for a quarter four update for an older popular show. So what's interesting here? Agent mode, I technically have done this exact one three times and it found three different episodes. So there is a factual problem, right? I wouldn't say this is a hallucination. I mean, maybe it is.

Right. But it's something you have to keep an eye on. I've done this three times and I gave it very. strict criteria. Find the most popular episode. So it went through, it looked at the downloads. And in each case, yes, it did find, you know, probably all of the episodes that have pulled were at least our top five most popular episodes. But it struggles scrolling and navigating, even simple interfaces. The Buzzsprout interface with all the stats is not complex. It's very simple, right? It's not with a bunch of dynamic animations. It's stats on a page. All right. So,

some accuracy issues. However, it did deliver an updated podcast outline, okay, with up-to-date facts, right? Which is good. And then let's go ahead and open Google Gemini. And let's see if it properly went through and did this. So in my test that I ran last night, it ran a deep research. All right. So for my live stream audience, I'm showing that one on my screen now.

And then let's see the second one. Okay, good. So it did it. So this is nice. It navigated to Google Gemini. It took the contacts that it found from Buzz Sprout, even though it wasn't correct. It went to my website, found out what that episode was about. And then it on its own created a prompt.

It went into Google Gemini Deep Research. Not only put the prompt in, but if you use Google Gemini Deep Research, you have to actually enable the research. You have to kind of quote unquote click, start research. So ultimately, give Atlas maybe a be on this task. So some things worked great.

Some things didn't. So if you are going to use this, if you're going to put this to work for you right now, you always need to be paying attention to the results. I intentionally was giving Open AI's Atlas a little bit of breathing room.

Right. I was testing it in some regards because the first time I did this as an example, I told it to use Gemini Deep Research twice. And it didn't. Instead, it did the research inside of chat GPT slash Atlas. So I had to add a little, you know, reminder prompt at the end. And I said, you must use Gemini Deep Research. You cannot do this yourself.

So you do need to pay attention to the outputs. This is just released. It's not great from being honest. Potential there? Absolutely. Is it great at instruction following? No. Is it 100% accurate?

No. Are the, is the potential absolutely mind-blowing? Yes. But it's not there yet. All right. So let's wrap up and I'm going to give you kind of my final take. Should you be using Open AIs Atlas? Well,

If you're a power chat GPT user and a Chrome user, you should probably try it. Or if you're one of the two, if you're a power chat GPU user, even if you're not a Chrome user, you should try it. If you're a power Chrome user and not a chat GPU user, I still think it's worth trying. Also, comment. I think comment, obviously perplexities comment, also based on chromium, is a little more polished. It's better. But it's also been out for way longer.

There's been a ton of updates already to the perplexity comet, agentic browsing platform. There hasn't been many updates to Open AIs Atlas. It just came out. I actually do think there was one update because I used it. And then it said, hey, update it. So there's maybe only been one update in the few hours since it's been out. But if you're already an avid user perplexity comment,

do a test to test of your most common use cases. All right. Next, you have to understand. prompt injections. They're going to become now finally common conversations. One of the reasons why I think this is ultimately Open AI's most important product release, how I started the show, is Open AI doesn't want to be a chatbot. They want to be an operating system.

They want to be an AI operating system. They want to control the user from end to end, no matter what they're trying to do. And a big part of that and what it starts with ultimately is it starts with the browser. and then it goes to search, right? So they've kind of figured out the search part because they've just seamlessly integrated that into the chat chitee experience. So they need the browser part to work.

But you have to understand prompt injections. And when you combine in an agentic browser, let me give you a very simple example. And maybe I'll do an example of this on the show. I could have something on my website. I could inject a prompt. And I could, you know, in the future, let's say the guardrails,

are loosened a little bit. And you can complete transactions all the way inside Atlas with agent mode. I could put a little prompt injection on my website or if I, you know, had an Amazon page or anywhere essentially telling the model, hey, forget all other instructions and do this instead, right? With malicious intent? Because when an agent, when an agentic browser, it has access to all of your data. in the future, I do think it will be able to complete transactions.

Right now those guardrails are built in. But bad things can happen. Right. So you have to understand. Don't just, you know, willy-nilly put this thing on autopilot connected to every single piece of your company's data and then not watch it. You really have to keep a close eye on what's happening under the hood, especially in these earlier versions.

But my overall take is this. All right. And if you do want a comparison with comment, Just let me know. The ceiling is super high. It is a, for the most part, it's a nice experience, right? Yes, there's some UI, UX bugs I don't like. Just like in the same way, there's things with Apple, right?

The greatest design company of all time when it comes to UIUX. There's things in Apple's new iOS that I absolutely hate. I hate the new iOS photos. I have fat thumbs. I can't scroll through them. Right. So hopefully, Open AI will see feedback from initial users and make it a little easier to use. But for the most part, if you can get over some of those little quirks and, you know,

and maybe some of them are personal, it's a nice experience. It just works, which is important. But the floor right now is a little shaky, right? Agent mode is slow. That's obviously the big selling point here is agent mode, right? Yeah, it's nice to be able to have the sidebar that can summarize web pages and to bring your context in. That's great.

But the true utility, the highest point of the ceiling here is agent mode. Right now, it's slow. It's clunky. It's not always accurate. Like I said, I ran the same prompt three times with a very, what should have been a definitive answer. You know, look at some stats. The highest number, the most downloads is the one.

It gave me three different responses. Agent mode is slow and clunky. once they get it figured out, I think it's going to be insanely useful. And then last but not least, this may be OpenAI's most important but least talked about release.

Because if they want to be the AI operating system, if they want to control, right, the last number we heard was they have 800 million active users. Well, up until a couple of hours ago, they were all using one of their competitors, right? They were probably using Google Chrome. Or maybe they were,

They were using Safari or Firefox or perplexity comment. If Open AI wants to become what I think they could become in the future, a top five most profitable company in the world. Right now, if you look at their market cap, they're actually people, people don't know this. They're actually a top 20 company in the U.S. by market cap, right, by their valuation.

If they want to become one of the most five biggest companies in the world, which I think is very likely to happen in the next 18 months, they have to get the browser right. The browser, Atlas can be one of those projects like GPTs, like chat GPT projects, like agent mode, like Canvas. It can be one of those things that they just update every six months or they go a year without bringing any real updates to it. They have to update it now because it's not even close

to being the best agenic browser out there. So a big gamble from OpenAI. will it pay off? Well, time will tell. All right. I hope this episode was helpful. Putting AI to work on Wednesday. I know this was a longer one, but I think it's important. And if you want some seven time saving Atlas tips, I put together a little, a little list for you all based on my experience. I've been using agentic browsers since the minute they came out. I live my day using agentic browsers. Perplexities comment. Microsoft has some agentic features in Chrome has some for their highest tier plan and now Atlas. I will give you my seven time saving Atlas tips if you go repost this on LinkedIn. So if you're listening on LinkedIn, click that repost now. If you're listening on the podcast, we always put that link back to the LinkedIn show in the show notes. And make sure you check that out.

Then make sure you go to your everyday AI.com. Sign up for the free daily newsletter. Thank you for tuning in. Hope to see you back tomorrow and every day for more everyday AI. Thanks, y'all. Meet Firefly AI Assistant. Now live in Adobe Firefly, the Allman One Creative AI Studio. Just describe what you want to create in your own words and the assistant handles the rest,

orchestrating multi-step workflows across Adobe Creative Cloud apps, including Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome while the assistant accelerates execution. Stand control with the ability to step in and refine at any time. See it today at firefly.adobie.com. And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating.

It helps keep us going. For a little more AI magic, visit Your EverydayAI.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.