← The Growth Podcast
The Growth Podcast
AI Agent Browsers: Should you use one? | ChatGPT Atlas vs Perplexity Comet vs Arc Dia
00:58:07
Loading summary
Transcript84 lines
- [00:00]
Naman Pandey
We are going to be putting the three browsers being Perplexity Atlas and TIA head to head against each other and we'll find out which is the best for most people for most purposes.
- [00:09]
Akash
I've brought in Naman Pandey today. He is dropping all the sauce, the use cases to use the areas they fall over so that you can be more productive as a PM with an AI browser.
- [00:19]
Naman Pandey
So once you select your resume, all you have to do is upload your resume and say fill out this application for me.
- [00:26]
Akash
Wow, that's a dream for job seekers.
- [00:28]
Naman Pandey
These are all extremely expensive operations when it comes to toke optimization. It kind of breaks my brain that how is this stuff free?
- [00:36]
Akash
Do the browsers hallucinate as much as the LLMs?
- [00:39]
Naman Pandey
I have not found many cases of hallucination at all. ChatGPT Atlas to be the winner here. Perplexity is a close second. Just because you could argue that a lot of time is spent doing research which it completely automates really well. And then finally you'll have dia. These are the most insane AI tools I've tried yet. The use cases are literally endless.
- [00:59]
Akash
So we just went through the strengths of all the browsers. What are the weaknesses? What should you not be using these browsers for? If you get any value out of this podcast, do me a huge favor and follow on Spotify and Apple podcasts and subscribe on YouTube. It helps the show tremendously. If you become an annual subscriber to my newsletter, you get access to nine incredible AI products for an entire year. This is an over $3,000 value across tools like Mobin, Linear, Descript, Magic Patterns, Reforge, Build, Relay, Deep Sky, Dovetail and Arise for AI evals. Most of these brands have never done a product package like this, so go take advantage@buildle.akashg.com and now into today's episode. Naman. Welcome to the podcast.
- [01:55]
Naman Pandey
Thanks for having me. Akash, what are we going to do today? We are going to be putting the three browsers being Perplexity Atlas and TIA head to head against each other and we'll find out which is the best for most people for most purposes.
- [02:09]
Akash
All right guys, so watch until the end to see his recommendation for the best. Can you walk us through some examples first?
- [02:14]
Naman Pandey
Sure can. So before we jump in in terms of an overview, I would like to share the broadest overview level use cases for each. So from like a 300 foot view here is that Perplexity Comet is your go to for most research oriented tasks, especially those that need to be Real Time TIA does a really good job at remembering context, especially because it treats all of your open tabs as its operating system itself, which is super unique. And the other two browsers don't really do that. Finally, ChatGPT Atlas is the most agentic one out there. So if you're trying to perform operations on your web browser instead of just aggregating information and such, it will probably be your pick if that's your use case.
- [02:58]
Akash
Okay, awesome. Looking forward to seeing these live.
- [03:00]
Naman Pandey
Yep. So right off the bat to install each of them, all we have to do is for Perplexity, you just navigate to Perplexity AI download Comet. It'll bring you to this page right here. You just click that and it does that for you. For ChatGPT, log in like you would normally, except you'll find Atlas here in the top left corner. If you click that, it'll give you the option to download for DIA. If you just simply, you know, google Dr. It'll bring you to this website. It's called dabrowser.com getting started. And then all you have to do is hit this top right icon and it'll download. If you have a Mac, you'll get the DMG for all three. All you have to do is drag it into your Applications folder and you're good to go.
- [03:38]
Akash
Simple enough.
- [03:39]
Naman Pandey
All right, so now then jumping into the use cases themselves. The first one I had here was we'll start off with actually something that all three can do equally, right? Just to kind of temper expectations around possibilities, for lack of a better word. So say that you're trying to research Nvidia stock, right? So you fire up your this is just what the interface looks like. So if I open a new tab, it just starts off like that, which I always thought was interesting because you know, there is no homepage, so to speak. So that takes a while getting used to. But say you have you're doing your research on Nvidia ticker that updates in real time. You have a different tab that goes over I just wanted it to grab the latest earnings call transcript, which it did for me, along with all the links that it needed. And then any stock news. And then finally just another play on that in terms of what Google's Gemini 3 release the means for, you know, the stock as it pertains. Now if you go to your new tab and all you have to do is generate a succinct and pardon my typing, if I make any typos, one pager for all stock related info that I Just went through, put in any graphics and such as needed and then if you just run that. So firstly, notice how I have a bunch of tabs here that do not pertain to stocks. Right. Except obviously it is smart enough to find out which tabs that I'm talking about. So it group them together. This is really good if you're trying to write stuff, if you're like a creator of any sort and if you have a bunch of tabs open to do research. And then it combined all of the information that was pertinent for my ask into, as I had said, just a one pager. And yeah, the reason I'm starting with this is that all three browsers can do this. This is not something that only Atlas. This is just to kind of get you guys thinking in terms of how to use all of your open tabs and kind of consolidate information in a way that's really useful that all, all three of these browsers really excel at.
- [05:39]
Akash
Yeah. Awesome. So just to confirm, all three browsers can look at your open tabs. And the key, the key here versus regular chatgpt is you don't have to like paste it in or anything like that. It can just go fetch it at end the itself.
- [05:51]
Naman Pandey
Exactly. Yep. Cool. Okay. So now having covered that, we'll be getting into a little bit some of the more nicher use cases because Atlas is open. I'll just start with that. One thing that a lot of people I don't know, maybe I speak for no one, but I struggle with, is filling out forms online. Right. I don't think there's anyone that is like, oh yes, I love online forms. Give me all the forms that you have to fill out. So what I have here is just a made up data engineering job posting. Now if you're like me until like, you know, three years ago when I was trying to apply to jobs, I would just manually, you know, enter all of this information, attach my resume. What if I told you though that I actually have not filled out a single one of these fields? All of this was done for me by this browser. All right, so you're presented with this job opening, right? And you don't, you don't want to fill all of it out. All you have to do is upload your latest resume because obviously it needs some context in order to know what to fill out. So if you just maybe we can clip this part. Should be somewhere here. Okay, there you go. So once you select your resume, you all you have to do is upload your resume and say Fill out this application for me. That's it. Hit enter. And it's kind of spooky slash somewhat creepy to sit and watch this, but you will actually see it. Fill out your information in real time all the way to the end. And a cool add on here is that for any fields where it expects you to enter information like why do you want to work here? What in your experience makes you good for this job? It will, because it's generative AI automatically fill that and fill in those fields as well, even though those were not explicitly mentioned on your resume.
- [07:33]
Akash
Wow, that's a dream for job seekers. I mean, especially if you ever go to like a workday page, you'll upload your resume and then it still makes you fill out all the same fields all over again. Yes, the AI might be be kind of slow, but at least it'll do it for you. And that gets to the point of the skill that people need to develop of managing these AI agents, multiple AI agents. Set it off on a task, then go do something else, then come back and quickly pick up the context again.
- [07:57]
Naman Pandey
Absolutely. That's a wonderful call out the other really big and annoying time sync when it comes to, you know, not just job seekers, but you can be an entrepreneur that's trying to find people to reach out to. You can be, you know, like a product lead at a company that's trying to network with other product leads at a similar company. Pretty much any LinkedIn scraping or for that matter really any website scraping use without you needing to learn how to make bots. Because yeah, it's not the hardest thing in the world to learn how to make scraper bots, but most, you know, non technical people would find that fairly challenging. Atlas solves that completely. You now have one of the world's most powerful scrapers just in your back pocket to do whatever you want with it as you please. So in this use case here, and this is not my LinkedIn, this is just like a burner LinkedIn that I have what I was doing earlier to show Aakash was in my case. So I have a podcast channel, right? A big pain point for me is to find out who to reach out to as potential guests. And I can rerun this query for us here. But essentially what I asked it to do was scrape LinkedIn and find people I can DM for my podcast. Ready, set, do podcast dot com. Keep in mind here that I actually didn't even explain to it what my podcast is. What, what's the point? Like what type of guests do I want? Because it's an agentic browser, it is really good at filling in those ambiguity gaps that exist based on user requests. Like, sure, it would help it if you told it what the podcast was about, but because you gave it a website, it is now smart enough to go to that website, fill in that gap in terms of what type of guests I might be requesting for, and then scrape LinkedIn, as you'll see here, even though it won't like the verbiage of scrape, it'll be like, oh, I can scrape LinkedIn. At least that's what it did the first time. Maybe it's learned its lesson now. But yeah, so what it'll do, and you'll see the cursor move around here is that it'll search through the LinkedIn platform, go on profiles, and then generate that list for you, along with links for the people that you're trying to reach out to. Now this is just the top level, right? Here's where it gets really, really cool. Think of all the pages or all the websites that you go to, where the contact information is hidden behind another button, right? So if I go to my own LinkedIn, almost every LinkedIn profile has a contact info, but you need to click it to make it accessible. Now, because as I was saying, you have the power of Atlas's agent capabilities in your back pocket. You can instruct it to go to whatever it is that you're trying to find, whatever niche of people it might be, click their contact info, copy their emails or phone numbers and generate that in an Excel sheet for you to be ready to really do whatever you want with. I want to stress on really how powerful this is because even if you had the ability to make bots, it is close to impossible to replicate this type of smart click driven behavior as it stands today.
- [10:46]
Akash
Pretty cool. Today's episode is brought to you by Jira Product Discovery. If you're like most product managers, you're probably in JIRA tracking tickets and managing the backlog. But what about everything that happens before delivery? JIRA Product Discovery helps you move your discovery, prioritization and even roadmapping work out of spreadsheets and into a purpose built tool designed for product teams. Capture insights, prioritize what matters and create roadmaps you can easily tailor for any audience. And because it's built to work with jira, everything stays connected from idea to delivery. Used by product teams at Canva, Deliveroo and even the Economist. Check out why and try it for free today at atlassian.com product-discovery that's a T L a S S I a n.com product-discovery Jira product discovery build the right thing before we dive deeper, let's talk about something every PM faces. Getting alignment on product decisions. You know that feeling when you're trying to explain a user flow to engineering or justify a design choice to leadership and you're just describing it with your hands? That's where Maubin comes in. Maubin is the world's largest library of real world mobile and web app designs from industry leading apps like Airbnb, Uber and Pinterest. Instead of spending hours taking screenshots or hunting for inspiration, you can instantly find exactly how successful products handle onboarding, paywalls, checkout flows, whatever you're facing. Over 1.7 million product builders use Maubin to benchmark against best in class products and show their teams proven solutions. Whether you need to convince stakeholders there's a better way to handle user activation or research how top apps approach feature discovery, Maubin gives you the visual proof to back up your product decisions. Check out mobin.comakash that's M-O-B-B-I-N.com a k a s h and get 20% off your first year.
- [12:43]
Naman Pandey
Yep. So as I said, it will say that it can't assist with scraping LinkedIn because it doesn't like that term. All you have to be like is use your agent mode to do it and it'll be like oh cool, yeah, I got it. I can do that part pretty easily.
- [13:03]
Akash
Sometimes you just need to remind it of its capabilities.
- [13:06]
Naman Pandey
Exactly. And as I was saying earlier, it gave me the same error and all it was. It told me itself actually that yeah, do you want me to use your browser together? Yes equals I agree. Trigger I yeah, I trigger agent mode. And I just said yes and it was like sure, I'll do it. So they have some, I guess, guardrails around how it looks to the outside world. Right. Because LinkedIn, if I were LinkedIn I'd be pretty pissed off. Right. Because I'd be like what the hell guys, we have measures against this. Which I think which is why to plan against that they do it this way to you know, discourage people from feeling like oh, they gamed the system or they're better than LinkedIn. That's just my POV having spend some time with these tools. But yeah, there's a lot of really interesting under the layer type of stuff which you explore the more you play with them. It's really fascinating. Like I don't know, it's just so much fun to play around with these tools. Another really use case which is slightly different is compared to tools where you're logging in. So I guess what I didn't show on my LinkedIn demo was it'll make you log in. So in agent mode, it'll ask you to enter your email and password and only then can it access any of the information that you were seeing. Now, if you're one of those people that is like, yeah, no, that's not okay with me. There's like privacy issues with that. I don't really want to do that. Totally makes sense. They have custom inbuilt support with more famous tools such as Gmail, in which case you don't actually have to log in. So once you open just like gmail.com on your browser, we'll just go there and then if you hit your little ask GPT here, you'll find that, yeah, so I already connected, so which is why it didn't show up that option for me. But you'll find a custom Gmail connection button here and all you have to do is just authorize that so as to give ChatGPT access to your Gmail. And now that opens the door to some really, really interesting use cases that are built off of your workflows. So I'm even going deeper than just stuff like find out availability for a certain teammate or look through my calendar to find when I can block time for deep work. I'm not even trying to get into that because those are pretty obvious use cases. Something that I found really helpful was, and I struggled with this, just finding out what subscriptions I have enabled, right? Because you'll start discovering subscription. Seven days go by, you never cancel it, and three years later you're still paying that money. Well, guess what? Whether you remember or not, your Gmail remembers because those companies are forced to send your receipts for your subscription to your Gmail. Now all you have to do here is look through all of my emails and I have that written on this earlier tab. Actually look through all of my emails and tell me all of the recurring expenses that I have. And you can even have it make a list and you can even go a step further, further and ask it to link the customer care websites or phone numbers or whatever the case might be for all of your services. And then, you know, as you can imagine, it'll just go through your entire history. So for me, it found Google One, it found YouTube, and then Stuff it wasn't sure about like, Priority Pass Capital One Travel. It's really comprehensive. Like, it'll give you extra information and you'll be like, okay, that's not an expense. Sure, right. But at least we'll have the piece of mind knowing that all of the potential use cases were covered.
- [16:13]
Akash
This reminds me of how Demis Hassibis, the CEO of Google DeepMind, said, One of the key things about AI is that a human is really good at going and looking at, let's say, the first 10 search results, but AI is really good at going and looking at a thousand things. And this is a very cool use case because you can set AI free on your thousands of emails that you haven't been able to check.
- [16:32]
Naman Pandey
I mean, it also kind of opens the door to slightly more dystopian Black mirror ish ideas that I may or may not have had. Like, I don't know, say you have a person that is no longer with us. Right. Except you can point this tool to their email if you have access to that, and build a bot that can forever talk to them. You know, I don't know if you want to keep that, but it's like, yeah, like your imagination is where the capabilities end. It's just unbelievable. The type of things that you can do with these agents as they are today, and this is the worst they'll ever be, is another thing that I have to remind myself of. Like, they're only going to get better from here.
- [17:09]
Akash
Awesome. How else can we become productive with this tool?
- [17:12]
Naman Pandey
So when it came to all of the agentic type use cases, I believe that should be enough to get at least some juices flowing in our viewers in terms of, you know, potential use cases. If it's okay with you, I'd love to pivot to Perplexity and do kind of a deeper dive on some of what it is, you know, better at and the use cases that make more sense for it.
- [17:32]
Akash
Yeah, I want to get to comparing these tools. Let's do it.
- [17:34]
Naman Pandey
All right, so next up, we have Perplexity Comet to like, kick it off. From what I was saying in our overview here, it's really good if you want to consolidate and do research on a bunch of tabs that you have open now, because we are very close to Thanksgiving, I figured it would make sense to involve a use case that is, you know, more Thanksgiving related. So say you have Amazon open, right? And I want to call out here that you actually don't even need to open Amazon. I just did that so that it does a really, really good job. And we can do into the go into the comparison where I just do it on the homepage instead of going into Amazon. But really my idea here is I have a nephew, he's 10 years old, I want to gift him something for Thanksgiving except I have no idea what to give him because, you know, it's just like there's so many things and I also don't have the time really to be honest. So I just want to delegate this task to AI right, not use my brain. So I'm trying to be the worst prompter here because I find that that is the best way to actually assess how good an AI tool is. So all I'll say is find gift ideas for my 10 year old nephew. Honestly, that's all I'm going to give it. And I'll also tell it compare prices outside of Amazon as well and we'll just let it go and see what it finds. So while it does that, what it's going to do actually is not only will it look through all of the options that it has on Amazon concurrently, it will find if that same product is available on, you know, the seller's direct website and such, it'll compare the prices and it'll give you the link for whatever price. Like look at this first example. Even though we're open on Amazon, I'm sure it found Barnes and Noble site to be cheaper. And we can confirm this here by actually searching that, which is why the link it sent me was from Barnes and Noble and not Amazon. Now obviously this is a pretty like a loose example, right? For lack of a better word. But think of all the possibilities here. Say you're working on a case analysis at work, right? You have like six different Excel tabs open a lot of context here to work with. You can simply open Perplexity and ask it to consolidate. Say you've been doing some rough work or scratch work and you have like a Google Doc open that lists just all of your ideas. Now you can ask it to go through all of the ideas that you have and then find concrete examples of whether or not they're accurate and it'll do that for you. So in this case, if I just open Google Sheets, for instance, we can just open a blank sheet here and have that just be open as a tab and then here I'll say compile the list of all gifts that you found on my open Excel Sheet tab. And when I did this last time, it was smart enough to figure out exactly what I was looking for and it had done that I'm not sure why some of that verbiage was. Alright, so it said appears to be empty. There are no gifts listed. So I'm going to say yes, I want you to populate the sheet for me, which again goes back to my intentional bad prompting skills, which. Right. So it wants the URL, which will give it that and I think that should be good enough for it. We'll see here. All right, there you go.
- [21:05]
Akash
Oh, there we go.
- [21:06]
Naman Pandey
So yeah, while it does that going back, think research, think lifetime, it's probably going to really excel at those type of works. It didn't list link here for some reason, but we can just ask it to include the link as well. Or maybe it did actually. Yeah, so I guess it didn't. We'll say be more specific with the price because it has a range right now and list links to all presents. And I think that should have this update. And I don't know if you caught that, but it actually opened a new Amazon tab and a new Untitled spreadsheet because I think it's trying to go through like look, you can find it scrape all of this information. It searched for the Lego Technic racing car automatically. That's where it found $55. A really good use case for this also is you can ask it to find the biggest discounts that are available right now and track them against discounts that were available throughout the year. So because it has history as well of what prices used to be, it will find you the biggest discount that exists for these products and you can even use that to optimize for what gift you want to buy because you know, you might want to be saving $80 instead of just $50 just. Just because of the fact that there was a bigger discount. So all of these time sensitive historical data related research, it just absolutely sheds at.
- [22:29]
Akash
And why does it, how does it have historical price data from Amazon? Because I imagine that would be like a huge amount of data for it to store.
- [22:35]
Naman Pandey
I'm sure you've heard of tools or extensions like Honey, right, that do basically this exact thing. Honey or Capital One Shopping. So what it does is it ties up with those extensions which I have given it access to earlier. I just set that up once. That's how it's able to access all of that data. So you're right in calling out that. Perplexity itself doesn't have that, but it kind of has the bridge to where that data is stored. So that's why it can go back and forth and fetch information based off
- [23:01]
Akash
of that, ooh, okay, there's the power of tool use for your agentic browser. If you have the right tools available,
- [23:09]
Naman Pandey
then it can go do cool things, especially extensions, right? And this is I think something that's super underrated and I haven't found a lot of even YouTube content around this. But once you can get extensions that are already so useful to sing along with these tools, that's when there's like a really cool unlock of absolutely amazing things that you know, it's like some of this tool is much greater than those of its individual parts. Like it just becomes something much, much more valuable than individually extensions and individual browser. Like it just gets super powerful like that. Only thing is, it does take a while sometimes. And I also want to call out that I used to have Perplexity Pro. I no longer have it because I just don't use Comet as much anymore. But this would all be much faster if you do have the Pro. But like I said, you don't need it, right? It will still get you pretty far off the free account as well. And keep in mind these are all extremely expensive operations when it comes to token optimization. Now I don't know how it's doing that behind the scenes, but all of this that you're seeing is extremely token intensive. So I mean, I just want to call out that, you know, Perplexity Comet does a really good job of really being efficient with its token use because I have no idea how they're allowing non pro users to be doing such in depth research on their tool just for free. But again, I don't have behind the scenes insight into this. Maybe you have a perspective, Aakash. I don't know, but it kind of like breaks my brain that how is this stuff free right now?
- [24:35]
Akash
Yeah, no, it breaks my brain too. How is this free? This is like this is alpha for you guys out there. You know, for you guys who are watching your budgets carefully, you are getting a pro product that should be paid for free right now.
- [24:48]
Naman Pandey
Yeah. So anyway, you'll find it painstakingly attach all the links. Clearly it's doing what we ask it to notice how to the scent accurate the prices are. Because if you remember, we saw 55.24 on that Amazon page, which is exactly what it has here.
- [25:03]
Akash
Do the browsers hallucinate as much as the LLMs?
- [25:06]
Naman Pandey
I have not found many cases of hallucination at all actually. That's such a good question. Yeah, it didn't even occur to me until just now, but all of the work that I'VE done in terms of the form filling. What else did we cover? I guess this type of research around tools and such like this will be done soon. I promise. None of these links will be broken at all. Like they will work exactly perfectly like to the dot. So maybe they hallucinate a little bit. But in my use, in my usage of them, I have not found a lot of hallucination. Also notice how because I have the same tab open in two different tabs, it also shows my name here. It just went away. But I mean, yeah, let's quick, quickly test some of these links if they work. So that says 2399 and it was minus 40% that that's, you know exactly what it has here. There's some notes as well. I never asked it for notes. I guess it just attached metadata that it found. But yeah, so I. The reason I flagged sheets here specifically is because I know for most people, right? For most PMs as well as really just normal people, Google sheets is kind of their bread and butter when it comes to. It can be life expense tracking, right. It can be just trip planning, whatever the case might be. Keep in mind that, and this is funny because even Gemini has started showing their own like stuff here. I found this to be much less helpful, I won't lie. So I just like to close it. But remember anything that you're doing on sheets, if you can imagine it and if you can find a way to word it, what you, whatever the operation is that you're trying to do, Gemini will do a wonderful job of doing that thing. So like I can look up here, sample expense sheet, right? And actually let's do that. And then if you ask it to find where there might be gaps where you can be spending or saving more money, it look through your data and make personalized recommendations based on what it found on your specific sheet. So I think this is a good way for most people to think about or reconcile use cases when it comes to perplexity research sheets and anything that involves comparison. For me, the first thing that comes to mind is let's test out Comet for those purposes.
- [27:06]
Akash
Awesome. So that's Comet. What does DEO look like?
- [27:09]
Naman Pandey
Awesome. So I will say before I actually jump into Diya and pull it up here, of the three, if you had, if you had me rate them just purely on the basis of aesthetics, you know, I don't know, like product polishing. I don't know the right word. Maybe you can tell me Aakash, because you're from that space. Actually DIA wins like it is experiencing a product unlike any that I've seen not only in the agentic AI browser space, but any product at all. And it's funny because I was just trying out Google's Anti Gravity yesterday, which honestly as I think about it would be a contender for this video as well given its agentic capabilities even though it's not really a browser. But nothing has come close for me in terms of what a joy it has been to get onboarded to DIA as a user. So for any PMs out there taking notes in terms of what is a good customer onboarding experience, check out dia. It is an absolute case study that you can use to for building out bullets that make that experience such a joyful one. So I just wanted to call that out real quick. So yeah, now in terms of the use case itself, I have here one of your videos Akash. And what I wanted to do here was there's a few things that it unlocks when it comes to context. Right. So you know, we'll have get started with Diya. We have that and we have your channel here. So if I just go on my new tab and all I'll say is so you can specifically select what tabs you want. But we won't be doing that actually. We'll just say based on the two YouTube videos, I'm trying to make a video on how to become an AI pm, help me write the script for this video with a strong hook, etc. That is AI. So as you can see, obviously I've been not the greatest prompter here again, but from what has been my experience it'll. And keep in mind that these videos aren't even playing right now. That one actually hasn't even gotten to the video. It's still just the ad before the video. But what I had found before was because it has all of the context stored in across the two videos that were open here in which. In which in our case were the AIPM video and the other one which is here. Like it obviously goes through the entire thing. And because it's your video Akash, maybe you can tell me if this stuff even tracks. So it's talking about. Forget the hype. AI PMs aren't prompt magicians. They decide when not to use AI. Is this sounding like stuff that you had covered in that video, would you say?
- [29:37]
Akash
Uh huh. Yeah.
- [29:38]
Naman Pandey
I mean as you can see I think this is pretty comprehensive. I think it had. It has done a really good job of trying to reconcile that information. This is really well detailed. Actually I didn't even want these many details. So yeah, so I mean that is one thing that it really excels at. I was also trying to look through other use cases in terms of Gmail itself. So I gave it the same use case as I did with Atlas and it like it was not able to do that pretty much. It was like way more restrictive. I think it wanted me to upload all of the bank statements and such like manually, which that defeats the entire point. Like I don't know why I would do that in the first place right
- [30:17]
Akash
Today's podcast is brought to you by Pendo, the leading software experience management platform. McKinsey found that 78% of companies are using Gen AI, but just as many have reported no bottom line improvements. So how do you know if your AI agents are actually working? Are they giving users the wrong answers? Creating more work instead of less, improving retention or hurting it? When your software data and AI data are disconnected, you can't answer these questions. But when you bring all your usage data together in one place, you can see what users do before, during and after they use AI, showing you when agents work, how they help you grow, and when to prioritize on your roadmap. Pendo Agent analytics is the only solution built to do this for product teams. Start measuring your AI's performance with agent analytics at Pendo IO Akash that's P E N D O IO A A K A S H I hope you're enjoying today's episode. Are you interested in becoming an AI Product Manager? Making hundreds of thousands of dollars more joining OpenAI and Anthropic then you might want to do a course that I've taken myself. The AIPM Certificate ran by OpenAI product leader Mikdad Jaffer. If you use my code and my link, you get a special discount on this course. It is a course that I highly recommend. We have done a lot of collaborations together on things like AI product strategy, so check out our newsletter articles if you want to see the quality of the type of thinking you'll get. One of my frequent collaborators, Pavel Hearn is the Build Labs leader. So you're going to live build an AI product with Pavel's first feedback if you take this AIPM certificate so be sure to check that out. Be sure to use my code and my link in order to get a special discount. And now back into today's episode. Are you looking to land your next product management job? I am accepting a group of just 30 product managers into a 12 week cohort led by me where every Monday for 90 minutes I help you through your job search. Creating your candidate market fit, updating your LinkedIn, updating your base resume. You're going to get personalized feedback and one on one mentorship sessions with my co teachers Ankit Vermani who is an AIPM at Atlassian and was a group Product manager at Meta. Prasad Reddy who is a CPO and has been in product for over 26 years as well as my other live instructor Bart Jaworski who's going to run another 90 minute session per week. So if you want coaching from me to land a PM job, this cohort is a no brainer. It is a premium priced product. It is more expensive than the average product out there, but the return is huge. Most people who join the cohort see a salary raise anywhere from 10 to $100,000 in the first year and so the ROI will be there within a year and we guarantee two plus interviews. So if you don't get two interviews after completing the 12 week program and following all the steps, we will refund the money to you. So it's a no brainer. Check it out@landpm job.com and now back in today's episode.
- [33:04]
Naman Pandey
So that was the first use case. One of the major things about DIA that makes it a really good use case for enterprise that unfortunately I had no way of testing was that it directly integrates with Jira, which is obviously another Atlassian product. So where this is super relevant for PMs is if your company allows you to install external software, what you what that enables you to do is and I wish I could demo this but there was really no way I could do that. But if you have like a GitHub repository and if you have your Jira logged in, you can go into the repository and ask it to scan any open bugs that maybe others have submitted or maybe that you have found and it will automatically generate rate that Jira ticket for you along with all of the details that are provided on your GitHub repository. So there were a lot of use cases doing that. But my organization doesn't use Jira or you know, any Atlassian product. So there was no way for me to demo this, but I did want to flash that because I know that most PMs are really in the weeds with Jira tickets a lot and this is something that's kind of custom made to integrate really seamlessly with Jira. So yeah, that was the use case that I wanted.
- [34:11]
Akash
Atlassian Suite DIA is really, really interesting and Does DIA expose what models it's using under the hood?
- [34:18]
Naman Pandey
I tried asking it actually and it just, it didn't tell me earlier. So we'll, we'll ask it.
- [34:23]
Akash
All right. It's a pretty good job of those YouTube videos, I feel like there might be some Gemini. Maybe it has some like smart model usage too where, like when it's YouTube videos that uses Gemini, then maybe uses different models at different times.
- [34:37]
Naman Pandey
That's a good call. It says it's using GPT4 class model, which is pretty concerning because why are we using GPT4 in November 2025? Like that is, you know, problematic to say the least. But I will say I do want to caveat that it probably doesn't mean much because if it's built especially for Atlassian, as we know now, because it was acquired for its specific internal use cases, I can see how GPT4 would actually be good enough, given that specific training and context engineering that they would have done on this, for it to be. For it to do a reasonably good job. Yeah. So the other thing that had come up in my analysis of DIA was a lot of PMs also obviously work with LOOM videos, right. So this can be a new product launch. This can be you're trying to show a bug. Say somebody shares a loom video of them walking through what a bug is in your tool. If you have this ready again, it has the context to sift through your entire loom video, generate the transcript for it, and then fill out a JIRA ticket automatically based on what the loom video did. This I thought was really cool. And I mean this is really niche when it comes to a use case. Again, do most people need or use this in their day to day work? Probably not, but because this is a podcast for PMs, this is something that is right up your all Sally. So I did want to flash that particular niche use case out as well.
- [35:59]
Akash
Nice. So anything Atlassian Suite DIA is great.
- [36:01]
Naman Pandey
Exactly. And then last but not least, I think add one other thing. So in terms of as far as putting all three of these head on, I did find through my various analysis, which it's kind of hard to demo this particular piece. This is only something you learn when you do enough of this. But when it comes to having say you have 12 tabs open, all for different things. So I have a YouTube channel, I like to write scripts for my videos and a lot of tabs, say seven out of those 12 tabs was. Were relevant for my video. What I found again and again was. Of the three browsers, DIA was consistently the best at making sure that it understands which seven tabs I'm talking about when it comes to my specific video. So what I mean by that is if I just open Google Docs here, just say docs new, and like, as you saw there, it can be a slightly, somewhat steep learning curve to figure out saying what opens a website versus just making it search. It can be really annoying. But once you use enough, you kind of get the hang of it. So what I was saying here is, while you say, continue to type out like your script. Right. If you open chat with this document again, a pretty common use case for most people is to consolidate information across their various open tabs into whatever document that they're writing now. Because in my use case, I was trying to write a YouTube video and I had other tabs open that were not relevant, I find it annoying to tell it and point it to which tabs it should look at. I think it should be the job of a really smart agent AI browser to do that thinking for me. Right. And I found again and again across the three, that DIA did the best job at figuring out that These are the 7 out of 12 tabs that are pertinent for this video or this document. So let's really do a good job at segregating those seven from the other five. What I found with Atlas and Perplexity were that for tab context, that's kind of the technical term for this, they struggle a little bit in terms of, like, it'll drop, like maybe the last tab, or it'll include like a random Amazon link that has nothing to do with my video and start suggesting that I write Amazon stuff on my video, which, as you can understand, that can be pretty annoying. So that was really the last thing I wanted to call out in terms of d specific behaviors when it comes to the AI browser.
- [38:18]
Akash
So. So what I've learned so far is that the tab context is really the superpower of these Agentica browsers. And for Atlas and Comment, you can expect to be very controlled with your tabs and not have open tabs that are going to be polluting that context with unnecessary context. But for dia, you can have those and it'll still figure out the relevant ones.
- [38:37]
Naman Pandey
That's exactly right. Yep. And I'm sure there are people much more disciplined than me that do a good job of closing tabs. I just really struggle with that and I've just given up trying now. So that made D really useful for my use case. It was Just really cool to tell it and have it almost read my mind in a sense. And it was just incredibly satisfying to get that result every single time where, you know, no balls were dropped and all of the context from my relevant tabs was retained.
- [39:07]
Akash
Awesome. So we just went through the strengths of all the browsers. What are the weaknesses? What should you not be using these browsers for?
- [39:13]
Naman Pandey
Yep. So for each, I can go down the list one by one. If you're trying to do anything really quickly, like if you're trying to finish an operation and time is of essence, Atlas is not usually the best choice for that, just because as we saw, it needs to in its own way, see through everything on your screen and go through each field one by one. So while it does a pretty robust job, it's not the fastest. So if time is of essence, probably Atlas is not the way to go. The other two, especially dia, are much, much faster. For Perplexity. If you're trying to get through any of the more navigational type uses where you're trying to, maybe you're going to be hit with a captcha of some sort, or you might have to go through dark patterns that aren't the most intuitive to recognize. I found it struggle a little bit. This was not true 100% of the time, but sometimes it can struggle with a long chain of navigation before it gets to where it's trying to get to. This usually doesn't happen as much for most research because you're just a Google search away or a PDF report or what have you. But keep that in mind for more navigational tasks, it might be better to use either Atlas or dia. Now for DI itself, the way it grabs all of your context from the tabs and such, I found that there is no really good way to keep out any private information that you don't want exposed. Compared to Perplexity and Atlas, that do a good job of keeping all of your records on your device right, which is why they make you install these things. Now, I have no way of checking this, but in my analysis I found a lot of reports online where people did say that their information had found a way to get outside. I don't know again if this happened for sure. I just wanted to call out this finding just in case any user might be dealing with especially sensitive information. Probably best to avoid DEA at this stage when it comes to that.
- [41:06]
Akash
I wanted to ask about that. I mean, should somebody be using an AI browser, there seem like there's a lot of privacy and security. Security concerns.
- [41:12]
Naman Pandey
It Kind of depends on what your overall privacy appetite is. As I like to think about it, like that LinkedIn scraping thing. You can do the same exact thing on chatgpt.com browser as well. It's not like exactly the same, but the agent mode still makes you log in. It still has access to your entire screen. It kind of for me just depends on the individual person's appetite for what they feel is at risk. A lot of people believe that the moment you make a Gmail account, like just kiss your web privacy goodbye forever. I don't hold that opinion. But yeah, like I'm not trying to say that it is the most secure and you can feel completely at ease with all of your information floating around or not. It's for each individual person to kind of determine that. But there is no indication that these are especially privacy first or privacy focused across the board, any of the three browsers. I would not be counting on that. If that was a deal breaker for me, I would not be using any of these three actually.
- [42:09]
Akash
And are they overhyped or do you think they're correctly hyped?
- [42:12]
Naman Pandey
So they are under hyped actually for the very specific and small list of use cases that they're really good at. Then it becomes kind of a function of. Is that something that you're trying to do? Right. And as I said, that list is really small. For that small list, and I should add ever growing small list, I would actually say they're underrated. Like there are just personally if I had to go through each person's LinkedIn and find out guests for my particular niche, that's hours of work that this just did in like 15 minutes right now. Are most people trying to do aggressive outreach from their day jobs or what have you? I don't know. Right. That depends on the person. But if that is something that you're trying to do. Yeah, like why would you not save two and a half hours? And this is just one time right? Now imagine if you're doing that multiple times for multiple users use cases. So these are not to be portrayed as a, you know, like the forthcoming of or the best thing since sliced bread. Right. Fail at most general things that other things might be better at. But if you're trying to do that specific niche use case type thing, which actually I'm happy to share a list of that for these use cases, you have to be at least considering using one of these tools that we went over today.
- [43:26]
Akash
Let's do something together. Let's build a mind map. So I want to build the AI agentic browser use cases, let's build these together. So what are. Maybe we'll do this in a couple of different categories. So let's do the PM use cases and then general and then on the other side of this let's also do kind of the non use cases like when you should use something else. So if you were to start to think about this at the very top, what are the PM use cases that PMs must be using it for?
- [44:02]
Naman Pandey
Just one thing that comes top of mind is it's like a umbrella of things. Right. That you can do just with that information. So navigational note taking in a structured way towards an end that you define, I would say would be a good way to go about this. Exactly. Competitor analysis. You can even have it look at like onboarding stats and such. You can even do a sentiment analysis. Like what are people on Twitter saying over the last few months? Like we can do one with Gemini 3 Pro's overall sentiment across the board on Twitter just based off, you know, last week. And I think it would be cool to compare that versus three months ago where everybody kind of was like, oh, Google is over, you know, they've fallen behind, blah blah. So it will be able to generate that gap or that gulf in the sentiment that exists. So all of those type of navigational use cases it would really do a wonderful job of.
- [44:57]
Akash
And you said it is even good for data analysis.
- [45:00]
Naman Pandey
Yep. So for perplexity specifically it does a really good job of sheets and I think it's because they have like some sort of tie up. This is, I don't know this for a fact, but especially for Google sheets and say you have like three, four more PDFs that are talking about stuff or data in your sheet. They can all talk to each other and you can update your sheet based on the findings of the PDF or like three other tabs that you have open. Yep, those are the top use cases that PM should definitely be checking out, I would say.
- [45:29]
Akash
Okay. And then the top general use cases. So you started to go over some of these. Right. I think it was like email related email sorting. I think you went through like Shopping Companion essentially. Could I finding you things. What other big use cases should people be using this for?
- [45:46]
Naman Pandey
Scraping. Anytime you're trying to scrape anything, it is like again, I come from computer science, I know how to write scrapers. They're not easy. Most people do not want to be writing scrapers, I promise you that. Yeah, but now you can scrape anything. Again, it takes time. I Do as I caveat it earlier, it's not the fastest compared to a scraper bot that you might write. It will do the job for you though and it will do a really good job for you.
- [46:11]
Akash
Nice. Anything else people should be remembering if I want to do X use a
- [46:15]
Naman Pandey
AI agentic browser actually for Akash, for the PM use cases. Documentation is a really big one. So at any point if you're trying to document a flow, I actually think maybe we covered that a little bit. But say you have a new product, you're trying to write documentation for users that are yet to use it, you can instruct these browsers to go through page by page, end up on every page or any flow that exists and write documentation along with screenshots for your entire app. So they can have all of your journeys mapped out individually in different swim lanes just because you instructed them to do so.
- [46:50]
Akash
That's awesome. All right, and then what are the non use cases? What should you not be using these for? It feels like they were pretty slow, right? So anything you can do in less than five minutes, probably not do.
- [46:59]
Naman Pandey
I would say slow plus again I think privacy. Right. That's a big one because if you're trying to do anything useful, more often than not it will make you sign into it'll, you know, like Amazon, LinkedIn. I tried doing Instagram. That was a really cool case where meta actually has inbuilt code on their on the web version of Instagram to prevent any bot like usage. So I spent almost an hour trying to break it which I'm sure people smarter than me probably have broken it. But for Atlas, which is, you know, I didn't find anything that it was not able to do. It was not able to scrape Instagram for me which I thought was very interesting. But for yeah, pretty much anywhere where you're not comfortable logging in or where you know, you might run into captchas or you might actually I do want to say a lot of them weirdly sometimes do well with your generic captchas. It's the stuff where you have to drag and rotate something. I don't know if you've seen those ever. Akash, of course. Yeah, some of the more like niche or cooler captchas they all struggle with. So if you're going to run into that they'll be pretty useless. But yeah, they crush normal captures like traffic lights. Oh you know they eat them for breakfast. So those are not concerns. Yeah but so yeah, I would say deeper navigations are like. It was funny. I tried to get it to cancel Amazon subscription for me. And all of them struggled. Okay? Which I hope you see the irony in that because, you know, the dark patterns are so dark that it throws away even AI tools to in order to do that. So, yeah, that's the biggest thing that comes to mind where not the greatest at following long chains of navigation.
- [48:40]
Akash
Okay, so this, you guys, is the mind map. This is how you should be thinking about using your agentic browser versus an LLM. Now let's go to the moment of truth. How would you rank these browsers?
- [48:54]
Naman Pandey
So obviously most people I think would be more interested in the usability piece. I found that for most tasks that everyday people are doing and if your goal is to save time in order, I would rank ChatGPT Atlas to be the winner here. Just its scraping feature itself unlocks so many absolutely amazing possibilities for me that it's just hard to question the utility. At least time wise Perplexity is a close second. Just because you could argue that a lot of time is spent doing research, which it completely automates really well. And then finally you'll have dia, which is not to say that D is not, you know, very good at saving time because as I showed you guys, you don't have to watch three one hour YouTube podcasts. You can just have the tabs open and have it summarize all three for you in a neat one pager along with takeaways for you. Right? So all of it like is kind of context dependent a little bit, but for most people, I would imagine, you know, ATLAS to be the winner here. Now, in terms of usability, I had, you know, I might catch some heat for this, but I actually found Perplexity to be a much better overall, you know, experience or user experience compared to both ATLAS and dia. I found it to be much more intuitive the way it responds back to me when it comes with to me with clarifying questions or just trying to understand exactly what I'm trying to do. Especially as you saw with my subpar prompting skills, you know, I need that. But it's like really interesting to grasp the level at which it's able to interface with me almost in like a telepathic level, which I found lacking in both ATLAS as well as dia. So I would place that first on that list. And this is followed by, I think pretty close together ATLAS as well as dia. Probably DIA might edge it a little bit because of its superior product experience, but where the rubber meets the road, you probably care less about that and more about the actual, you know, User experience. But yeah, I would put the two of them pretty close to each other when it, when it comes to that. So overall, overall final ranking, probably ATLAS is what you should be starting with if you've never tried any of these. And then probably you can slowly dip your toes into perplexity and then dia, in that order.
- [51:05]
Akash
So now it comes to the moment of truth. What is the best agentic AI browser out there?
- [51:11]
Naman Pandey
Absolutely. So without any drum rolls, the way I did this was break it up into two categories just because it made the most sense. So when it comes to the overall user experience, I found ATLAS to be the winner here just because for most tasks that most everyday people and product managers are trying to do, it just blows the competition out of the water. Especially if you keep in mind that the other two aren't really that good at scraping or doing actually agentic driven workarounds or work loads for you. Right. So that, that is why I would put Atlas first. This is closely followed by Comet, that does a really good job at when it comes to research, consolidating information from a bunch of open tabs, updating your sheets or what have you, and as we went over using various PDFs even that are open to update information that you have on your sheet and then finally you have dia, which just not to say that it's a terrible experience using it. I just found the use case for which it would be the flagship go to for to be slightly limited and that would be my experience, explanation or reasoning to put it at number three. I will say when it came to usability though, Comet would be what takes the top tier for me. Just because I found when interfacing with it, the questions that it had for me to further understand my instructions or what I was trying to do were much, much more superior than DEA as well as atlas. Now I don't know if I'm the only one that feels this way or if there is some inherent bias, but more often than not I found Comet and myself to be on the exact same wavelength. This was followed by DIA and ATLAS in very close succession, which is not to say that they were completely off or they misunderstood my instructions or didn't hallucinate, although there were some instructions of misunderstanding what I was trying to ask them to do, which seemed to almost never happen with Comet, which is why, you know, those are the rankings that I have overall though for most people, for most tasks that you're trying to do online that can be automated or be converted by an agent ChatGPT, Atlas is probably the browser that you should be trying. If you've never tried any agentic AI browser before, it has what to me appears like the most forgiving learning curve in that if you're familiar with ChatGPT, you will find getting onboarded here super, super straightforward. Comet can do a better job of, you know, being upping their experience. But yeah, again, if you've never tried any of these before, go try out Atlas. And for more research oriented tasks, maybe you can slowly dip your toes into Comet and then dia in that order as well.
- [53:40]
Akash
So chatgpt is the best AI agentic browser. Is it free? Do you have to pay for it?
- [53:45]
Naman Pandey
Yep, it's totally free. You'll probably get hit with a rate limit at some point, but to test it out and do your first agentic operation, it's totally free of cost. There is no fee to use that.
- [53:54]
Akash
And have you been hitting rate limits on your pro plan?
- [53:57]
Naman Pandey
Never. I've been grinding it pretty much nonstop, just relentlessly and not once has it given me the rate limit error. And I'm really a plus user. I'm not even a pro user, so I have the lowest plan possible and I've never run into a rate limit.
- [54:11]
Akash
Okay, so for 20amonth, guys, you get access to the best AI identic browser out there in the world right now, ChatGPT Atlas. And as no one said these models, this product is only going to get better. There is a lot of alpha out there for you guys to go use ChatGPT Atlas, learn the use cases so that when you encounter a use case in real life, you automatically go do it. Learn the hacks, learn the tips, learn the tricks. And to that point, are there any final tips and tricks or advanced features on Atlas that people should know about?
- [54:43]
Naman Pandey
I would just say continue to monitor the docs and stay updated with your news diet. Because the best way or the best time to utilize any of these features are exactly when they drop. As of recording this actually on the 24th of November, ChatGPT has included a shopping agent experience thingy on the browser itself, which if you'll remember, that was only possible on Atlas like until yesterday. So these tools continue to evolve at breakneck speeds. There's really no telling what's coming next. So my personal, you know, compass with this stuff is as soon as something drops, that's actually the best time to test it out because that is when, as Akash said, that alpha is at its biggest, that's when you're almost incentivized as a user because I promise you in a year from now and you can come back to this video, this stuff will not be worth $20. Like I can almost but guarantee you that part. It is now because they're trying to onboard users so make the most of this opportunity. The window won't last around forever, you know. And thanks for creatives like Aakash that really fix up your content diet so that you have the latest and greatest when it comes to these development. So yeah that that's what I would say on that front.
- [55:55]
Akash
All right guys, if you've watched to this point, you are now a candidate in this giveaway. If you comment below the top three tools overall that we ranked in order and you go find me on LinkedIn, follow me and hit a DM. I have an open profile with the three seven days from the episode publishing so you have to do it in the first seven days of the episode publishing. You have to comment and you have to DM that order. Of the three tools that Nomad just gave you, I'll be giving two of the people who DM. So usually like anywhere from 50 to 100 plus people are going to DM me. Just two. I'm going to be giving you a year free of the paid newsletter even if you don't win the giveaway. You have learned a lot by spending this hour with us. Go out there, Go use these AI agentic browsers. I think that if you're a product leader, you should be thinking about getting your IT department to give people access to ChatGPT Atlas, not just ChatGPT. As we just showed, there are so many different good use cases for your product team out there. So if you're a product leader or a product manager, go out there, get access to ChatGPT Atlas added into your PM AI tool stack. For my money, it is one of the best ROIS you will get on a tool out there. Naman, thank you so much for dropping all the sauce.
- [57:05]
Naman Pandey
Thank you for having me. This has been such a delight.
- [57:08]
Akash
If people want to check you out or find you online, where should they go?
- [57:11]
Naman Pandey
My Instagram is at Ready Set 2 podcast. Please do check out my show as well and give me a follow if you'd like to.
- [57:17]
Akash
He has tons of sick content over there and we'll see you all in the next episode. I hope you enjoyed that episode. If you could take a moment to double check that you have followed on Apple and Spotify podcasts, subscribed on YouTube, left a rating or review on Apple or Spotify and commented on YouTube all these things will help the algorithm distribute the show to more and more people. As we distribute the show to more people, we can grow the show, improve the quality of the content in the production to get you better insights to stay ahead in your career. Finally, do check out my bundle@bundle akashg.com to get access to nine AI products for an entire year for free. This includes Dovetail, Mobin, Linear Reforge, Build, Descript, and many other amazing tools that will help you as an AI product manager or builder succeed. I'll see you in the next episode.