Everyday AI Podcast – An AI and ChatGPT Podcast - Ep 638: Agentic Browser Showdown. ChatGPT Atlas vs. Perplexity Comet

Episode Date: October 23, 2025

(Kinda) Hot take 🔥AI agents kinda stink. (For now.) If you want to get more done with AI, ditch the “general” agents until they catch up. Want gains today? Agentic browsers are the real winners. (Like OpenAI's just released Atlas browser.) Agentic Browsers are powered by the world's smartest models and actually keep your context and finish multi-step jobs all while logged into the services you rely on. So, today, we got head-to-head. Errr... bot v bot. 🤖Agentic Browser Showdown. ChatGPT Atlas vs. Perplexity -- An Everyday AI chat with Jordan Wilson Newsletter: 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:Agentic Browser Showdown: Atlas vs CometOpenAI Atlas Browser Live Demo OverviewPerplexity Comet Browser Live Demo ComparisonMulti-Step Workflow Testing in Agentic BrowsersReal-World Use Case Scenarios for Atlas and CometAgent Mode Capabilities: Atlas vs CometDetailed Feature Comparisons: Atlas and CometBest Use Cases for Agentic BrowsersOutput Quality and Results EvaluationAtlas vs Comet: Pros, Cons, and VerdictTimestamps:00:00 Rise of Agentic AI Browsers03:58 Live Stream Updates & Tools07:12 Rethinking AI for Work10:47 Podcast Strategy with AI Tips14:07 Improving AI Newsletter Insights16:47 Live Demo Challenges Explained22:09 Atlas vs. Perplexity Features25:16 Quick Results and Use Cases26:25 Automating Daily Tasks Efficiently30:20 Updating Content with AI Tools32:55 Generative AI's Mixed Results39:03 Comet: Leading Agentic AI Browser40:39 Challenges With Agentic BrowsingKeywords:ChatGPT Atlas, Perplexity Comet, agentic browsers, AI browser showdown, OpenAI Atlas, Perplexity AI, agent mode, reasoning models, browser AI comparison, multi-step processes, AI workflows, agentic features, browser automation, paid AI tools, free AI browsers, Mac AI apps, Windows AI browsers, personalized AI, multi-tab tasks, background agents, browser memory, ChatGPT integration, Slack connector, Chrome extensions, data enrichment, real-world AI use cases, visual content creation, multi-platform automation, supervised task automation, sensitive data handling, multi-application workflows, browser UX, agentic capabilities, AI contextSend 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. Way before the AI agent wave crashed ashore in the second quarter of this year,

I'm not so boldly stated that 2025 would actually be the year of agentic browsers, not AI agents. Why? Well, everyone wants general AI agents from the big AI labs and they aren't that good right now. Yes, there's great narrow agents like Claude code that excel in niche verticals, but there's no name brand, general use agent that's dominating the agent scene. And that's exactly why I'm so bullish on agenic AI browsers in the interim, because they fix the main problems that general use agents can't properly address right now, working seamlessly with your context and consistently accomplishing multi-step tasks.

So that's why we're going to deep dive today on two of the more impressive agenic browsers out right now. The first to the party and perplexity's comment, or at least the first big AI name, and then open AI's brand new agentic browser Atlas. All right. I'm excited for today's show. I hope you are too. Let's get into it. What's what on y'all? My name is Jordan Wilson.

Welcome to Everyday AI. This is your daily live stream podcast and free daily newsletter helping everyday business. are helping everyday business leaders like you and me keep up with the AI developments and make sense of them and know how we can actually use them to grow our companies and our careers. So if that's what you're trying to do, it starts here, unedited, unscripted, live stream podcast. What could go wrong? Demoing two agentic browsers live?

Yeah. But if you want to take it to the next level, make sure to go to our website at your everyday AI.com. There, we're going to highlight the recap from today's show, the most important information that you need to know, as well. as all of the AI news that you need to be the smartest person in AI at your company. All right. Here's what we're going to do.

It's a throwdown, right? This is kind of a continuation from yesterday's podcast where we went over OpenAI's new Atlas browser. So kind of like a Thursday throwdown. Probably not going to make that thing. But there is a ton of feedback from you all. So yes, when I put a poll in our newsletter or the Spotify podcast, I actually go back and look and you've all overwhelmingly said, hey, this new open AI Atlas seems pretty cool,

this new agentic browser, but I want to see it head to head against Comet. I want to know the pros and the cons. So that's what we're going to be going over today. So by the end of today's show, and I'm going to make this a fast one, you're going to understand the major capability differences between perplexity Comet and Open AIs Atlas. you'll see the most ideal use cases for each browser or the best for your scenario. And I'm going to show you some live demos and comparison outputs. All right.

And if you did miss that Atlas specific episode, just scroll back. Click that back button and go listen to episode 637. All right. We're going to start at the end. This is one of those, again, cooking shows here. So apologies for all the back and the back. forth live stream audience. If you could let me know if you can see this, but there's a lot to share here.

And normally I might want to split my screen and yeah, podcast audience, this is definitely one of those. You're probably going to want to go watch the video. I'm going to describe my best what's actually going on here. But there's going to be a lot happening on screen. So I'm going to be showing you both the chat, GPT Atlas browser, which I have on my screen now. and then the to the right for those following along,

perplexity comment. So something to keep in mind, if you want the video version, go to our website, your everyday AI.com. All right, here we go. I'm starting with a prompt,

all right? And I'm going to describe what this prompt is doing, but I wanted to get it going first. All right. And I'm actually going to start another one. I am on, just so everyone knows, I'm on a paid plan for each comment and on a paid plan for chatGPT via the Atlas browser.

So you can use comet for free completely. And you can also use Atlas for free as well. Although a lot of the better features of OpenAI's Atlas are more for paid users, specifically a lot of the more agentic features. So there are some like agent. features inside open AIs Atlas, but the actual agent mode feature, which is what I'm demoing here, that is only if you are on a paid plan. Whereas perplexity comment, completely free. But we're going to go, I'm going to show you kind of the spec sheets here in a second.

But what I'm going to do now is I'm going to do a second prompt. And then in a couple of minutes, we're going to go in and I'm going to read all of these. So let me just go ahead and put my second prompt in. All right. Again, doing this live, y'all, doing it live. Bring it up agent mode here in Atlas, getting it going. And then going into, let's see, I'm going to do it right here, into the assistant mode.

Let me clear this. I did run this before just to make sure things worked and to see how long they went. so I wouldn't keep everyone going for like 50 minutes. All right. So I now have two separate kind of agentic processes running, both in OpenAI's Atlas, which is on the left side of my screen and inside of perplexity's comet. I don't know what the max is.

For perplexity comment, it said the max that you can have running is two at a time. Also, hopefully, My audio or video won't be too laggy. But I'm not on the world's most powerful computer. So hopefully things aren't going to be going too slow here. All right. So I have those running. And what I'll tell you, I'm going to read these prompts here at a second.

But I want you to think about your work. Okay. And I talked about this a little bit in our show yesterday. Because on Wednesdays, we do the putting AI to work series. And today's kind of a continuation of that, but more of this head to head. But I want you to think, what is the work that you do? And unfortunately, almost every single demo or a use case or, you know, YouTube tutorial going over both perplexity comment and going over OpenAI's Atlas.

I think they're all doing it wrong. No one is demoing and using these agenic browsers in the way that humans work. that's what I'm doing. But I wanted to call it out first before I even go into the prompt. I try to show you all and show or, you know, I want you to hear if you're just listening on the podcast. The way that I'm actually using AI, that's one of the most common questions I get, right? People are like Jordan, you've been doing this thing for three years.

You've used all these AI tools. You talk to these smart people. How are you using AI? I'm giving you all my secrets. But don't use it like everyone else is talking about it. Because that's not how humans work, right? Like I said, so many demos, use cases, of perplexity comet, and of Open AIs, Atlas are one shot, one trick, one use case. And that's it.

That's not how humans work, right? When was the last time you just opened one tab? Nothing else. No email. And you just completed work in one tab. You never had to open up another tab to research. You didn't have to open up an email, a CRM and ERP. doesn't happen. No such thing, right? Usually to get even the simplest task done, you might need

your email open, you might need your project management tool, you might need a PDF open, you know, you might need your CRM. You might, yeah, for the simplest tasks, you're probably going to need at least three, four, ten tabs open. You're probably going to be using three, four, ten different programs for more complex tasks. So that's really what I'm putting Atlas and perplexities comment. through right now. Real world scenarios working through multi-step processes using multiple tools. So let's go ahead here. I'm going to go ahead and read both prompts. So you can see what's going on. I won't make you wait any longer. I need to preface at first. All right. So I'm going to go into I'm going to go into my window here that's working inside of Atlas. All right. Here's what. Here's

I'm having it do. I'm not going to read every single word of this because it's a lot. But I'm saying, go look at my Buzz Sprout podcast stats and find a popular podcast from the first quarter of 2025. Then go to our newsletter website, which is read. Your EverydayAI.com. And then find the corresponding newsletter that goes along with that podcast episode. So essentially I'm saying, go look at my stats.

Find a popular podcast from the first quarter that you think could use a visual aid to accompany it. That's the whole thing, right? So I'm having it both look at quantifiable metrics, but then to also, you know, try to use its reasoning models, right? Because these agenic browsers are powered by reasoning models. So they're supposed to think, plan ahead, sometimes go back and backtrack. So number one, look at my Buzz Sprout podcast stats. Find a podcast that would benefit from having a visual, but a popular one. Then go to our email newsletter website, read what I wrote for that podcast, because for each podcast at the bottom, I always write what's called the leverage AI section.

So I sit here and type on my keyboard and, you know, write about that podcast episode. Then I'm telling each comet and Atlas. to go find more information about this podcast. And I'm not always telling it how, right? But like I'm saying, go find the podcast episode on my website. And I didn't tell it the actual website that the podcast episode is on. I also sometimes put in wrong information on purpose, right? Again, the humans controlling large language models aren't perfect, right?

And even when you're testing your own internal use cases for agenic browsers, you need to factor in both obviously the error for the AI for the agentic browser, but also human error as well. Right. So I intentionally, you know, said something about, you know, find a popular podcast from quarter one, 2025. And then later, I said, we are in quarter four, 2024. We're obviously not.

We're in quarter four, 2025. So we're going to be looking at the summarized chain of thought. here in a minute for both of these. But so, again, the three steps so far, find a popular podcast that could use a visual aid, go to our newsletter website, find the information that I wrote. Number three, think strategically about enriching this data with more up-to-date and relevant information, even though I threw a curveball and gave it the wrong date. And then last but not least, I said, go into my gamma account.

So I've logged it into in each of these browsers because this is something you can do, obviously you need to take your data security, you know, seriously. This is something I'm confident enough in logging into these websites and letting these agenic browsers control it. So last but not least, I'm having it log into my Gamma account and I want it to create a four by three carousel for social media. So this is something in theory that I should be doing, right? We create a lot of, I think, extremely high quality content.

post a lot of it on social media. Maybe we should. And I think visual content is something that could help, right, aside from just the video content and the podcast. So that's the first prompt. This is a real world multi-step scenario. This is a couple elements of data analysis, a couple elements of strategic brainstorming, data enrichment, research, personalization, and then also design, right? Canon AI do all of these things in one run. All right. So that's number one.

Number two is a little more straightforward. Okay. And I'm going to, let's see, there it is. This one's much simpler. So I'll read the whole thing. So for this one, I said, look in the audience and polls section of our Beehive account, which is where we send our email newsletter.

Please find the most popular polls. and give me, and these are the different things I'm asking for. Number one, five very direct ways we can improve the everyday AI podcast newsletter because a lot of times this is what we ask people in polls. What do you want to see more of? How could we improve? Five audience trends I may not be aware of, right? So this is really connecting the dots, hopefully between two maybe related polls that I

forgot about, right, but doing this thing daily for almost three years. Or maybe just connecting dots that I didn't even know could be connected. Number three, saying the five most popular polls, but not based on total respondentist, based on the average number of responses per appearance. Because some polls we used to run like once a week for like a year. So some have been, some of these polls have appeared in our newsletter a lot. And then I said, last but not least, five topic ideas from a popular poll in the last three months that we have not yet covered on everyday AI.

And this is where, again, I didn't even tell it. here's the where the podcast like i didn't say oh i'm jordan here's you know go to your everyday ai dot com nope i said here's my newsletter data and then go find topics we haven't covered yet on everyday i so this one technically it starts out pretty easy and you know most people could do that with enough time right click through you know we've probably done three maybe three or four hundred polls before. But this one, at the end, it gets a little tricky. So those are our two different prompts, right?

For our live stream audience, you're kind of seeing right now, ChadGBT Atlas is running some Python code, right? So they're really getting into it. And then on the other side here, I have comment working as well. So here's what I'm going to do. We're going to take a very quick break in here from our, our sponsors. And then we're going to go and we're going to see the results. This podcast is supported by Google.

Hey folks, Stephen Johnson here, co-founder of Notebook LM. 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 Notebook LM instantly becomes your personal expert, uncovering insights and helping you brain storm. Try it at notebooklm.google.com. All right. So actually, I'm going to let some of these prompts are done. Some of them aren't. So I'm going to continue to let the agentic browsers cook. And then let's just jump straight into a little bit more comparisons between these platforms because we just went straight into

the prompts because I needed to get them in the oven. Right. I knew that they would take a while to get going and, you know, I can see some of them are still at work. I do have, like any cooking show, I have the fully baked cakes ready just in case one of my ovens breaks down, doing live demos as, you know, meta found out when they introduced their new AI glasses. Doing live demos is not for the faint of heart. Luckily, I've done a couple hundred, so I'm okay. So let's look at kind of a little decision cheat sheet that I put together.

Also, this thing is great. There's a ton of content. I put a lot of time putting this together. If you want this entire kind of cheat sheet, just go ahead and repost this episode on LinkedIn. If you're listening on the podcast, check out the show notes. I always leave a direct link to the live stream link for this very episode. Repost this. I will send you this complete guide.

It's actually really, really good. All right. So this is kind of like what are some of the pros and cons, best use cases, et cetera. I'm going to address those in a minute. I do want to go over kind of more of the features. because if you don't know, it's always changing, right? Even Open AI's Atlas that just came out already had, I think, two updates. So who can use it and how?

So I kind of already alluded to this, but right now, perplexity is available to more people because technically it's available for Mac and Windows, whereas right now, OpenAIs Atlas is only available on Mac. They are both available to free users on both sides, although, for the agent mode, which is what I'm using for Open AIs Atlas. You do have to be on a paid plan. Both have said that they're coming out with Android apps. And then Open AIs will be coming out with Windows version of Atlas as well as an iOS app.

So here's kind of the core UX. And they're a little different. And maybe it's even easier to just show this versus showing bullet points on a screen. Right? So with opening eyes Atlas, and I won't spend too much time because I covered this in-depth yesterday, it kind of has this homepage now. So if you open a new tab or a new window, it has this kind of Atlas homepage, what looks

kind of like ChatGBT's homepage, but it's actually completely different. So you can go and, you know, launch different modes. So if you even launch Canvas mode or sorry, not Canvas mode, agent mode, you know, there's options for logged in, not logged in. You know, you can still add tabs and talk with certain tabs. You can talk with your browser memory. So it's, you know, it's a little bit learning and to get used to. And then anytime agent mode is active, you'll kind of see.

So even in my multiple tabs I have open, one of the tabs up there has kind of a bouncing cursor. So that's how you know that that kind of agent mode is running. So it's a fairly straightforward. If you're a chat, ChbPT power user, the Canvas, or sorry, keep saying campus, the Atlas browser is fairly straightforward, although I think they almost made it too much like the chat chabit interface where it can actually be a little confusing. And then on no matter what website you're on, you can always click this little ask chat gbt thing in the side in the upper right hand corner and then talk to that page.

The great thing about Atlas is it works with your entire context of your chat GBT account because you have to hook a chat chbtee account up to open a i's atlas which a huge huge benefit all right so let's talk a little bit um sorry got to jump around uh let's talk a little bit about uh perplexity's comments a little different it doesn't necessarily have a um a traditional home screen in the same way that's um open a i's atlas does it's more of just a slimmed down version of the perplexity home page uh similar you can go through some of the main features of perplexity on this kind of home screen. But the big, you know, what most people are going to be using is launching the assistant

in the upper right hand corner. So similarly, it works in a lot of the same ways in fashion as Open AI's Atlas. It automatically, if you click that assistant in the upper right hand corner, it takes in the context of that page. So some other automation things to look at. So perplexity has foreground assistance as well as background assistance on their max plan. That's the $100 or $200 a month and you can have much more, much more robust agentic flows going in more than two. So I believe on the base paid plan inside perplexity using comment, I think you're going to have two kind of agentic capabilities going on at the same time. For OpenAI's Atlas, similarly, you can open tabs, you can click to complete tasks,

you can also at any point, you know, run the agent mode. A couple things to note, explicit guardrails, you can't run code in the browser, download files, or install extensions. Yeah, I'm not going to go through all this, right? If you want to know all the bullet points, comparison of security, privacy, all that stuff, the different models, you can just go repose post this on on LinkedIn and I'll share this great guide with you. A couple things that I want to talk about is some of the integrations and extras, because this is where I think both perplexity and Atlas can separate themselves from each other. So one thing that I like about perplexity's comment is it does have the,

a Slack connector. It actually includes extensions when you sync this because both of these platforms are built on chromium, which is the open. open source kind of web language from Google's Chrome. So unfortunately, Open AI's Atlas, even though they say that they have support built in sinking for Chrome extensions, it didn't work. So you have to add them one by one.

So that kind of stakes. But everything else, you know, you can kind of import from a Chrome profile. So you would sign in with your Chrome profile. And you can choose what you want to import versus what you don't. You know, big feature. an extra of Open AI is being able to import and talk with your entire history. That's huge.

Both your browser history, if you choose to import it, as well as obviously the browser memory from there on forward and your chat GPT memory and context that brings over with it, right? That's huge. Also, perplexity does have a voice mode built into their browser. I don't use it a ton, although it is. is fairly helpful. Pricing, we already went over this.

It's free for each, although you do have to have a paid plan for OpenAI's Atlas in order to use the actual agent mode. So you do get agentic features, right, essentially stringing together multiple tasks and carrying content over, you know, from different tabs per se, but to actually have something like I'm showing you inside Open AIs Atlas, where it's actually clicking things, it's typing things, right?

That's the big benefit of using agenic browsers. It uses a computer like a human would, right? It can fill out forms. It can fill out job applications. If you give it the access and the information that it needs, it can use drop-down menus, right? It can go inside other large language models, which is something I showed in my show yesterday.

So go check that one out. Right, anything with a computer that would require you to click, drag type you can do in the agent mode, but you can't always do that in the free mode in Open AIs Atlas. All right. So that's enough of the kind of bullet point comparisons. But let me just give you a couple use cases here. So I'll save for a perplexity comment. And then we're going to rat this here up.

I told you it was going to be a fast one today. We're going to go check the results. And if you guys want to, maybe I'll share the actual end. results in the newsletter. I normally don't do that because sometimes it's like, you know, oh, some, you know, specific stats from our own internal, right? It's never anyone's personal information, but I'm not always like, oh, here's all our stats and right, but if you want it, I don't know. Say, what should you say in the comments? I always have a number that I put out there and if enough people say it, I'm like, yeah, I'll release it. Say, say, say gimme. All right,

G-I-M-M-E. Say give me. And I'll send those. resources today out in our newsletter so you can see the actual end outputs from the different um uh authentic browsers so best use cases for perplexity comment uh anything complex research and sourcing i think that's a great use case another one is automating tedious chores so as an example you know go through my inbox and unsubscribe me from all marketing emails or go read my lincoln messages and flag any of them that sound important or you know go through my slack my calendar and my hub spot you know if you can have a prompt each day start with that and triage and tell me the most important things I need to get to right those things that you would normally do every single day

I think are great for perplexity comment also multi-step background tasks especially if you have the higher tiered plan there's no shortage of things that you can do that I think can really save time running those background agents. So as an example, maybe you're trying to book some business travel, you know, saying find three nonstop flights to San Francisco for next Tuesday, book the best one under $400 and add it to my calendar. All right. Open AI's Atlas. A couple of things, a couple different use cases that I think are worth paying attention to. So number one is high volume communications, you know, drafting 20 personalized followups in a CRM as an example. Number two, sensitive data handling. This is something you can go read all of open AIs specs on their data

handling and privacy data security. Again, this is a hot topic. I'm not going to get too much into it. Everyone always wants me to, you know, rip apart. You know, I'm a small business owner. I understand the risks and the benefits and I am a responsible AI user, right? You may not see it because you're like, all right, it doesn't seem responsible to do a bunch of live demos with all this stuff connected to your data. I do them first beforehand. I always read the chain of thought. I understand it.

I always reiterate, right? All these things that I teach you guys to do. So I have a decent control over I know what the results for the most part are going to be with a little bit of range. But anything with sensitive data handling, I think is good in opening eyes Atlas. And then supervised task automation, right? So as an example, you know, log into our partner portal, pull this. Q3 report and save the URL.

So just some example use cases that I think are great for each platform. In this guide, we have a lot more here at the end of it as well. All right. Let's get back. Did anyone fail? Did anyone succeed? Let's take a very quick look. And well, let's also see if they finished.

All right. So let's see what we got here. This is the right one. No, yes it is. Okay. So here's the results from the first one. So it worked for 15 minutes. So this is Open AIs Atlas.

The first prompt I gave it, remember, it was going through the Buzz Sprout stats, finding a popular podcast that could benefit from having a visual kind of companion, a visual to supplement or complement the podcast. it was supposed to go find the podcast episode on my website without me telling it, you know, what that was. And then go do more research, bring it up to date. I gave it the wrong date. Okay. So yeah, I'm kind of looking, I'm kind of looking through here to see how each of them did. I'll have to do that later. And then the end deliverable was it was supposed to take all of this. And this is a project I would have, an intern or, you know, a contractor or, you know, a contractor.

part-time person on my team probably work on, right? Go find one of our more popular podcasts. You know, that's older. Let's create a visual out of it. Let's put some more up-to-date and accurate information, right? Because we're just creating a ton of content and then it sits there and dies. All right. So this is a project I would actually give to someone. And then the end goal was to go inside of Gamma. So it had access to all of our accounts. And the end goal was it was supposed to create a visual based on this older, popular podcast, but to make it more up to date with new information. So I'm looking at the results here. And from Open AI's Atlas, it looks like it found, which I don't think it did a great job here thinking about this. Because one of the things about

creating content that's old and try to make it a little more evergreen is it can be based around a hard release, right? Because you can't talk about a hard release a year later or three months later. So that's what it did. So it went through, uh, it's, it found that one of my more popular episodes from that quarter was going over Microsoft co-pilot's new agents. But those agents, you can't like, you can't say like, oh, GBT4, right? And then update it a year later. So, uh, don't really like what, um, Atlas did, uh, in this instance in the agent mode. It put together a very non-visual deck, like a slide. I wouldn't share this.

And I did in the instructions, I believe I said a four by three carousel, which I don't think that's correct. I told it to use the agent mode. Yeah, getting real crazy now. Tell it agent modes to use other agent mode. So I told it to use the agent mode inside Gamma to make it more visual, which it looks like it did not. So I'd say probably like a C minus here.

from Atlas in the first prompt. It completed it, which is not easy. It was stringing together a lot of information. So it looks like factually, kind of got mostly everything right, although I don't agree with its decision-making process and the end goal is not that good.

The first time I ran it, it gave me, I'm showing it here to everyone else. It gave me a much better visual, and it actually took the exact same podcast. So it did a better job on the visual the first time than the second time, but that is Gen A.I. All right.

Let's look at, let's see, where are we? Where are we? Let's see. Oh, interesting. So it looks like this time, comment followed up with some questions, which the first time, it did not. So luckily, I finished this once already.

So let's go and find the first version that actually finished. There we go. So in this case, interesting, interestingly enough, right? And this is why generative AI, it's like a roll of the dice sometimes. So the first time, I'll show you the results ready to go, comment, went through and found an episode on how AI is making us more productive and dumber at the same time. So this was based on an older interview. It looks like episode 487. So it went through in the actual kind of slides that it put together.

It's pretty good, right? I do know, because I looked at this earlier, it did properly go out and find updated data, updated stats that were relevant. It took quotes from the episode, actual quotes, and it put together in an actual good-looking visual, right? Our live stream audience can see this. So Comet did get tripped up on the, I told it it was the quarter one of 2024, even though it wasn't. So on the first run through, it did get mixed up on that. On the second run through, it didn't, but it didn't complete the full run. So it's hard to grade this one.

So I'll say the second grade for Comet, the one we did live, was incomplete. But the first one, I'll say was probably a B pretty good, right? Again, this is not an easy task. All right. So let's look at the second prompt. And this was, again, looking at all these different polls. There's hundreds of polls. I asked for some specific information, right? Five direct ways we can improve everyday AI. Five audience trends I may not be aware of. The five most popular polls by the average number of responses per appearance. And then five topic ideas from a popular poll in the last three months that we have not covered yet on everyday AI. So pretty tricky. The last one, a little tricky. So let's go over to opening eyes Atlas and see how it did. One thing I like about Atlas is I can drag kind of the ask chatypti side note over there. So this one went on for nine minutes. And again, you should always, I clicked on the kind of chain of thought.

You should always read it, check, accuracy, verify all those things. This is the quote unquote human in the loop. All right. So first, it gave me the direct. ways to improve the Everyday AI podcast newsletter. Okay. And it looks like it's pulling out some specifics. Pretty good.

Give me the audience trends I may not have noticed. So let's see. It says the most click poll options focus on hands-on productivity tools. Excel, Agent Mode, AI and sheets, connectors, listeners want content that helps them apply AI immediately. That's something I knew, but it's nice to have comment, reiterate that.

Also, one of the audience trends I didn't know is people are curious about HGI timelines. Then it found me the most popular polls by average responses per appearance. That's great. Save me a little bit of math that I didn't want to do exporting, sorting. All right. I didn't want to do that. But it looks like a lot of people participate in the polls that I ask people for AI work on Wednesday's topic ideas. Then it got me topic ideas from recent polls.

So this is correct. I did not yet do agent mode in Excel. It actually tied for a poll one day, so I didn't do that. AI functions in Google Sheets. I kind of already did a show on that. Chad GPT polls didn't do that. SORA 2 didn't really do a hands-on tutorial and then Gemini deep think mode. All right.

Interesting. So gave me some ideas there. All right, let's now quickly check in on comment. So comment. I'm trying to remember if I can pull this over. It's hard. Okay, there we go. So let's see if comet passed.

So it gave me five very direct ways to improve the everyday AI newsletter podcast. What I see immediately is it gave me a much more analytical and detailed breakdown. Comet did. So there we go. It gave me five audience trends you may not be aware of. So it said polls about AI ethics. trust and social impact. Generate consistent 30 to 40 response rates

indicating your audience cares deeply. Okay, that's something I didn't know. Maybe I should be asking more polls on those kind of softer topics. Five most popular polls, there we go, did a good job. And then it gave me five topic ideas. Let's see if it's things I've covered. So I've kind of covered some of these things.

Top AI agent platforms, real world business applications. Oh, no, wait. Actually, perplexity comet was the only one that did this right. So I said, give me five topic ideas. And for each of those topic ideas, there should be five supporting bullet points. Oh, no. Open AIs Atlas did it as well. So pretty good job.

So I'm going to say on this one, Open AIs Atlas, yeah, I'm going to give it a B minus. And I'm going to give Comet actually a B plus here. Much better job. So in these two scenarios, kind of a mixed bag, we got some incomplete, but hey, such as doing live demos on generative AI. Let's go ahead and wrap this up. So here's what I want to end with. My final take or verdict. Which one should you use?

Are they good? Are they worth your time? Well, I'll say this. I say there's no real downside to using an agentic browser, right? This is coming from someone. that I've been using Google Chrome, I don't know, at least 15 years. I think it's been out for like 17 years. I'm sure I've been using it since essentially it came out.

It was hard for me at first, but given these are based on chromium, they import, they do properly import your bookmarks, your passwords, etc. Comet properly imports your bookmarks, or sorry, your extensions. Atlas does not. So I'll say this. Comet was the first big AI player to the market in agentic browsers. I say big market, you know, big player. We had Dia, we had some others, right?

Arc. And there's been agentic features in Microsoft Edge and obviously in Google Chrome. But Comet is the first big AI player to go fully agentic browser. And they've had months of updates. So even in terms of output in deliverables, even in our head-to-head showdown, I think Comet did a little better. But that's also because Atlas has been out like not even like 48 hours, right? I would assume it's going to continue to approve. So, Comet is fairly stable, and it's very diverse in terms of its utility.

It's seemingly quicker for most everyday tasks versus Chatscheebt in their Atlas, especially if you're using the agent mode. I think right now, it has a higher floor than Atlas, but a much lower ceiling. Atlas, on the other hand, from OpenAI, it feels a little more premium. Right? It feels a little more polished. Even though there's some things on the UI-U-X I don't like, it does have a very Apple-esque feel, where I think sometimes comment is a little clunky.

Atlas, to me, just makes sense. Even though there's a lot of features and bells and whistles, it's easy to use. However, it is painfully slow, especially in agent mode. And it's not good at scrolling, right? That's a small thing, but it makes a big difference. I think if you really want to take advantage of what these agentic browsers have to offer, it is stringing together, like I talked about in the beginning of the show,

it's stringing together, multitask, multi-platform, multi-tab, multi-software application, strings, because that's how humans work. And so much of that is scrolling, right? So it's a small thing that Open AI and Atlas will hopefully get figured out, but it's a huge pain right now. In agent mode in particular is not that great. If you're using just the kind of built-in agentic features inside Atlas, really good.

The actual agent mode where it's clicking, it's scrolling, it's using the graphical interface, it's typing, it's a little slower than common. I do think that'll go better and go faster and improve over time. But I do think that the upside of Atlas is very high, especially if Open AI continues to update it and strengthens and improves and integrate some of the core chat chatsypte integrations into atlas i do think atlas is open a i's biggest play and biggest gamble if they want to be the

a i operating system of the future they have to get atlas right but at least right away i got i got to say there's a huge benefit that i wasn't really testing here uh for atlas which is bringing in all of your chat chp t context so i'll end with this if you are a chat chabit t power user like myself, I think Atlas is worth it for that reason. If you're not a Chad ChaptiPT power user, I think Comet, at least for now, is the agentic browser for you. All right. I hope this was helpful.

If so, go repost this show. All right. And I will send you that complete Comet versus Atlas deep dive guide. I didn't have time to get to all of it in today's show. We already went longer than I wanted to. But go repose this. I will share this. You know, we spend so much time putting this together, try to give you a no BS look and to give you a cheat code to be the smartest person in AI in your company or in your department.

Sometimes we just ask, hey, go, go share the love. Don't keep this thing to yourself. So please go repost this. Give me a day or a week. You know, sometimes I miss people if they repost it and I will send that to you. All right. So make sure you go do that. Tell someone about it.

And then when you're done, make sure you go to your every day. DayAI.com. Sign up for the free daily newsletter. Thank you for tuning in. Hope to see 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. and 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.