Latest / In The Loop / Why OpenAI's Atlas Browser Won't Take Down Google

Transcript

Why OpenAI's Atlas Browser Won't Take Down Google · In The Loop

• 0:00
  So Open AI have just brought out their own AI powered browser
• 0:03
  just months after Perplexity brought theirs out.
• 0:06
  Many people online have tried to claim that this is the start of
• 0:09
  a new browser war or that Google is suddenly going to get
• 0:13
  dethroned. And we actually saw a huge
• 0:16
  amount, billions in fact, of their value as a company
• 0:19
  collapse in the space of just 48 hours.
• 0:22
  As always, there's been lots of hype with this new release and
• 0:25
  in today's episode I'm going to break down what I really think
• 0:27
  about it and why I do not think this browser is going to take on
• 0:32
  Google. This is in the loop with Jack
• 0:34
  Horton. I hope you enjoy the show.
• 0:47
  So as the title and intro I think of this episode explains,
• 0:52
  I don't actually think Open a Eyes browser called Atlas or the
• 0:55
  Perplexity browser is going to dethrone Chrome.
• 0:59
  You know, I don't think it's going to capture even 20% of the
• 1:03
  market. And I may look back in six, 1218
• 1:06
  months, three years and eat my words, but I don't even think
• 1:10
  it's going to become part of most people's process or daily
• 1:15
  workflows. But in many ways, I also think
• 1:17
  that's the wrong way of looking at it.
• 1:18
  And it's definitely the wrong question because I don't think
• 1:21
  that question should be, is this going to be Chrome?
• 1:25
  I think the more useful question is, does Atlas solve specific
• 1:29
  problems we have today? And I guess what can you learn
• 1:32
  about testing it? So when that was launched just
• 1:34
  over a week ago, Google's stock price plummeted.
• 1:38
  It closed about 2% down overall. At one point, it dropped by
• 1:42
  4.8%. You know, some people just made
• 1:44
  a lot of money very quickly if they bought at the dip.
• 1:48
  But that's roughly $18 billion in market value that just
• 1:51
  vanished overnight. And, you know, headlines have
• 1:53
  been talking about the return of the browser wars, but within 48
• 1:58
  hours, it's been confirmed by TechCrunch and many others that
• 2:01
  Atlas is actually built on Chronium, which is Google's open
• 2:04
  source infrastructure for building web browsers.
• 2:06
  I'll get into why I think this is really important in more
• 2:09
  depth throughout the episode, but it just shows, I don't think
• 2:12
  Open AI is even competing with Chrome at an infrastructure
• 2:15
  level. They're trying to build this new
• 2:16
  interface into the Internet. So before we really get into the
• 2:33
  the meat and potatoes of the discussion today, I think it's
• 2:36
  worth summarising what Atlas is, what they're doing differently,
• 2:40
  whether or not you should even pay any attention.
• 2:42
  So when this was actually launched, Sam Altman described
• 2:45
  Atlas as basically representing this like Rare wants in a
• 2:48
  decade, opportunity to rethink what browsers could ultimately
• 2:51
  be. And the actual product itself
• 2:54
  centres on really 2 main capabilities and features, which
• 2:57
  is context aware Chat Chi BT integration and it's agent mode.
• 3:02
  So that context aware piece basically means that Chat Chi BT
• 3:05
  lives in the sidebar of Atlas, which is very similar to what
• 3:09
  Perplexes did with their browser called Comet.
• 3:11
  And it essentially means that as you speak to Chat Chi BT in this
• 3:14
  sidebar, it understands what's on the screen and I guess it
• 3:16
  understands the context of what you're talking about.
• 3:18
  So essentially it's just removing the need to copy and
• 3:20
  paste. So if you're looking at a web
• 3:22
  page and you say and ask a question, it already knows what
• 3:24
  you're referring to on that web page.
• 3:27
  So it's not entirely new technology.
• 3:28
  As I said, Perplexity launched it in July, which has many
• 3:31
  similar features. But the integration with ChatGPT
• 3:34
  obviously means that that browser Atlas is now in the
• 3:38
  hands of potentially 800 million users because that's how many
• 3:42
  people use Open AI's ChatGPT every single week.
• 3:45
  Now, Agent mode is where things could get both interesting, but
• 3:48
  for me, I found quite challenging open air.
• 3:50
  I tried to pitch it as ChatGPT, being able to take actions on
• 3:54
  your behalf inside the browser. And the examples they included
• 3:57
  are kind of like go do my shopping or adding recipe
• 4:01
  ingredients to a cart or placing specific orders for specific
• 4:04
  things or researching competitors and compelling
• 4:06
  insights and putting them into a document.
• 4:09
  You know, and there's a lot of mixed results from people online
• 4:12
  and early testers and definitely from myself.
• 4:14
  For example, one security researcher called Simon
• 4:17
  Williamson wrote that actually watching agent mode A felt like
• 4:22
  watching a toddler try and use a computer for the same first
• 4:24
  time, but actually realized that you could very easily inject
• 4:30
  prompts and sabotage the browser and the agent whilst it's
• 4:34
  performing tasks. So it's actually a big security
• 4:36
  concern. Also there's many reports on how
• 4:39
  it's just incredibly slow, it gets stuck in loops and
• 4:42
  sometimes just fails at very simple tasks that would have
• 4:44
  taken a human less than 30 seconds to complete themselves.
• 4:47
  And that's definitely been my experience.
• 4:49
  I've not really been a fan of it to be honest so far.
• 4:53
  I hope it gets better because it definitely represents some
• 4:55
  really cool ideas. But as I've said in previous
• 4:58
  episodes, it taking control of the screen just feels a little
• 5:02
  bit pointless when you just use APIs.
• 5:05
  So transferring data behind the scenes for you between agents,
• 5:08
  it's just quicker. It's the type of experience that
• 5:11
  I look for. Not this, you know, clunky
• 5:14
  browser that pops up with an agent confused trying to
• 5:17
  navigate websites. But that that's quite a
• 5:19
  pessimistic view on this new capability.
• 5:21
  Of course, they've also added a memory feature Atlas.
• 5:25
  So the Browser Atlas now starts to track your browsing behaviour
• 5:30
  and of course, build up more context over time.
• 5:32
  And according to Open AI, this will enable you to search and
• 5:36
  say, find all the job postings I looked at last week or summarize
• 5:40
  all the industry trends and reports that I've been reading
• 5:42
  over the past month. And obviously that means it's
• 5:44
  just very personalized and the trade off obviously being
• 5:49
  massive privacy implications. I know I've said this again and
• 5:52
  I keep on saying it. I have a tinfoil hat at times
• 5:55
  and I'm going to put it on when ChatGPT is using my conversation
• 5:58
  history, some of which might be private.
• 6:01
  It's using it's memory about all the things that I've talked
• 6:03
  about. It's maybe using context I
• 6:05
  provided to it in the form of files and documents.
• 6:09
  I don't want it to use that to sell more things to me and
• 6:12
  advertise more things to me, to be honest.
• 6:14
  Now, if we actually then look at what's working with today's
• 6:17
  early users. So the context aware assistance
• 6:21
  definitely helps with some things.
• 6:22
  I've definitely enjoyed using it in some tasks.
• 6:24
  Now it's only available for Mac users.
• 6:26
  So sorry everyone that's on Windows.
• 6:29
  Windows is rubbish anyway. So, you know, it's a, it's a
• 6:32
  lesson there, but there's been lots of people on Twitter
• 6:34
  basically getting Atlas to help with improving landing pages,
• 6:38
  suggesting improved thumbnails or without kind of switching
• 6:42
  context. So just setting it a task in the
• 6:43
  browser and it going doing it. Another Twitter user called Raza
• 6:46
  Martin actually compared both Atlas from Open AI and Comet
• 6:50
  from Perplexity on specific tasks such as tracking her
• 6:55
  daughter's school assignments across many different classes
• 6:57
  and apparently reported that Atlas completed it much more
• 7:00
  accurately whilst Comet really struggled.
• 7:03
  Now, I do recommend you guys go test it.
• 7:05
  And if you're going to go test it, I'd approach it somewhat
• 7:08
  systematically. It's quite a big ask to change
• 7:11
  browsers actually. It's a weird one.
• 7:12
  It's like trying to change mobile phone providers.
• 7:15
  There's definitely a, for me, a resistance to doing it.
• 7:18
  I'm so used to using certain things over others, even as
• 7:21
  somebody that's quite an early adopter of new technologies.
• 7:24
  But if you do want to test it, I'd recommend trying it for a
• 7:27
  few days to maybe a week and trying to focus on specific
• 7:30
  things that you would use that context aware sidebar for.
• 7:35
  So without touching agent mode, you know, draft social media
• 7:38
  posts using this information, IE the web page or, or use this
• 7:43
  page to write a summary e-mail and send it to John or improve
• 7:46
  my presentation that I'm working on right now.
• 7:48
  You could also maybe use it for learning, you know, on YouTube,
• 7:51
  you can ask questions about what's on the YouTube video.
• 7:54
  You could ask clarifying questions about what you're
• 7:56
  reading without obviously having to explain or put web pages and
• 7:59
  provide any context. And then obviously later I'd
• 8:01
  recommend trying agents probably for low friction or quite simple
• 8:05
  tasks. So form filling basic research
• 8:09
  asks like find the top five articles about X, maybe maybe
• 8:12
  some form of data collection across different websites or
• 8:15
  simple comparisons. And you want to be watching for
• 8:18
  how quickly the agent can complete that task versus you
• 8:21
  just doing it yourself. So yeah, have a test of it and
• 8:25
  trying to take away learnings about a, the future direction of
• 8:28
  this space, but also whether you can imagine users not using this
• 8:31
  on a a day-to-day basis Anyway, now we're going to I
• 8:45
  guess what ChatGPT is the new features and capabilities.
• 8:49
  Let's spend a minute discussing, I guess that strategic context
• 8:53
  about because I think understanding why Atlas exists
• 8:56
  and I guess the constraint it's going to have to operate under
• 8:58
  will give us quite a good indication about how successful
• 9:02
  it is. And you'll probably start to see
• 9:04
  why I think what I think around open AI not becoming the go to
• 9:09
  browser and dethroning Chrome. So from a technical point of
• 9:12
  view, Cronium, so the infrastructure provided and
• 9:15
  built by Google to allow other companies to build web browsers,
• 9:18
  that's Cronium powers about 83% of global browsers as of 2025 S
• 9:24
  Atlas runs on Cronium, so does Edge, Brave, Opera, Arc.
• 9:28
  The exceptions are literally only Safari, which has 16%
• 9:31
  market share and Firefox at 2.37% market share.
• 9:35
  Well, I think it's quite interesting is that Open AI have
• 9:38
  hired Ben Guja as Atlas's head of engineering.
• 9:41
  And for context of why this is important, Guja built Firefox at
• 9:45
  Mozilla and then Chrome at Google.
• 9:47
  So his third big browser project, he chose to build
• 9:50
  Chrome on top of Cronium. So Google's open source
• 9:54
  infrastructure for building new web browsers rather than
• 9:57
  creating a new engine. You know, even Microsoft
• 9:59
  couldn't justify fighting this fight.
• 10:01
  They spent many years developing what's called Edge HTML, which
• 10:05
  is, I guess, their own proprietary browser
• 10:07
  infrastructure. And you can guess what happened
• 10:10
  in in December 2018, Microsoft announced they were giving up on
• 10:15
  the Edge HTML project and switching to Chronium.
• 10:18
  And by January 2020, that transition is done.
• 10:20
  So that's one of the biggest, wealthiest technology providers
• 10:24
  in the world with almost unlimited resources, really
• 10:28
  looked at that competitive landscape and say we're not
• 10:30
  really going to invest here. We don't think we can win.
• 10:32
  You know, building a browser engine is difficult and very
• 10:35
  expensive and arguably difficult to have any real IP or
• 10:40
  competitive advantage by owning that infrastructure layer.
• 10:42
  You have to build rendering pipelines, JavaScript
• 10:45
  interpreters in preventing thousands and many thousands of
• 10:48
  web standards, handling security updates, maintaining,
• 10:53
  maintaining compatibility across different devices and our
• 10:55
  operating systems. You know, the cost is just
• 10:58
  astronomical and the use cases that you have to get right is
• 11:02
  also astronomical. And Microsoft, obviously the
• 11:05
  Marfan for there's no point doing this.
• 11:07
  Now coming back to Ben Gudger and why I think this is
• 11:09
  especially interesting on top of the fact that he chose to use
• 11:12
  Cronium. Now there are many different
• 11:14
  ways to interpret this decision. One perspective might be that
• 11:17
  building a browser from or browser infrastructure from
• 11:21
  scratch would require hundreds of millions of dollars in years
• 11:23
  development and deliver very limited additional value to any
• 11:26
  users. But by building on Chronium,
• 11:29
  Open AI can just focus on the AI integration, which is the actual
• 11:33
  innovation here and that's arguably smart resource
• 11:36
  allocation. Another perspective is that it
• 11:38
  creates a big structural dependency on Google's
• 11:42
  infrastructure. Well, really just trying to
• 11:44
  compete at an application layer. Every Cronium update would
• 11:48
  favour Chrome, which obviously is a massive strategic
• 11:51
  vulnerability. And I think both perspectives
• 11:53
  capture something real and important.
• 11:56
  And the real question is which dynamic will prove more
• 11:59
  important to success. You know, And I think it was
• 12:03
  definitely an argument to say that Open AI financial situation
• 12:06
  is a driver here because Open AI are going to lose $44 billion
• 12:10
  between now and 2029 before maybe reaching profitability.
• 12:14
  And only about 5% of their 800 million users actually pay for
• 12:17
  subscriptions. And some Altman in a Fortune
• 12:20
  interview actually acknowledge that they're losing money on a
• 12:23
  $200 a month Pro subscription because the amount of usage
• 12:26
  exceeds their initial projections.
• 12:28
  So another interpretation is that Open AI is investing very
• 12:32
  heavily in its growth phase, similar to how Amazon operated
• 12:36
  at a loss for years and Uber and many big companies.
• 12:39
  And obviously therefore Atlas could be a very clever
• 12:42
  distribution strategy to 800 million users.
• 12:45
  But you could also argue that the amount of financial pressure
• 12:47
  on open AI is leading to strategies which are a bit
• 12:51
  pointless. A good example of this is Sam
• 12:53
  Altman described advertising as a complete last resort at a
• 12:57
  Harvard Business School interview in May 2024.
• 12:59
  Yet by 2025 of this year, he's been telling structures.
• 13:04
  So an advertising media platform led by someone called Ben
• 13:07
  Thompson saying that he loves Instagram ads.
• 13:10
  An ad week reported that Open AI hired one of Google's ex head of
• 13:15
  search advertising as the VP in May 2025.
• 13:18
  And apparently Open AI targeting about $1 billion in advertising
• 13:22
  revenue by 2026. Maybe this just shows that
• 13:25
  they're burning so much cash, they're maybe pivoting to other
• 13:29
  revenue models to cover costs. It wouldn't be that surprising.
• 13:33
  As a consumer platform scales, it often adds advertising
• 13:36
  revenues to its business model. And time will tell which
• 13:40
  interpretation matters. But also there are some points
• 13:42
  on data collection and context capturing that maybe highlight
• 13:46
  why they've done this as well. You know, the Perplexity CEO has
• 13:49
  said very explicitly that the reason that AI companies are
• 13:52
  producing browsers is because they want more context and data
• 13:55
  on you because they want to know what hotels you're buying, what
• 13:58
  restaurants are going to, what you look at on the Internet
• 14:02
  because it tells them more about you and they say to personalized
• 14:06
  interactions. And it coming down to the killer
• 14:08
  point of why I don't think they can be that successful is just
• 14:11
  looking at that market structure.
• 14:12
  The fact that Chrome has almost 70% of market share and the
• 14:16
  companies that Google are competing with are basically in
• 14:18
  Google's pocket. You know, Safari has about 16%
• 14:21
  mainly because they own iPhones and Firefox has 2.45%.
• 14:25
  And we watch Firefox go from about 31% of market share in
• 14:29
  2009 to 2.3% now. Google has invested very
• 14:33
  cleverly in bundling Chrome into all Android devices and
• 14:37
  throughout the Google ecosystem, they've obviously put huge
• 14:40
  amounts of investment into Chrome itself.
• 14:43
  Google also pays 10's and 10s of billions of dollars to Apple to
• 14:46
  have Chrome on every single device.
• 14:48
  And up to 70% of Firefox's revenue is coming from Google to
• 14:52
  use Google services. And even Microsoft, who've got
• 14:55
  lots of Enterprise Products, lots of bundling of Edge into
• 14:59
  Windows products, lots of marketing, they've only reached
• 15:02
  about 5% after years of trying hard.
• 15:04
  Now, this seems very doom and gloom because I do think that
• 15:08
  Atlas could have maybe 1 to 4% market share, which you could
• 15:12
  say seems very small. But that's about 25 to 100
• 15:15
  million users. And by many metrics, that's a
• 15:17
  successful product. So yeah, maybe it becomes a
• 15:19
  specialized tool for 100 million people around the world.
• 15:22
  Maybe it doesn't, but they've got a massive distribution
• 15:25
  advantage because they've got 800 million weekly active users
• 15:28
  there and converting just a few percent would put them into 10's
• 15:31
  and 10s of millions of users, which they could monetize and
• 15:34
  generate 10s of millions or billions of dollars.
• 15:36
  Now I find it quite depressing. I wish we were moving away from
• 15:39
  an advertising model. It seems that it's creeping back
• 15:41
  as per usual. But anyway, let's pull this
• 15:55
  together. I think open our eyes Atlas
• 15:57
  browser is unlikely, as I said, to achieve this revolutionary
• 16:00
  market transformation. I think mainly because I think
• 16:03
  just market dynamics, historical precedent, many structural
• 16:06
  constraints all point towards that outcome.
• 16:09
  But I don't think that means it's not worth engaging with
• 16:12
  market share and individual utility.
• 16:14
  I think two different things. Maybe Atlas serves up to 100
• 16:17
  million users and has and becomes a niche product.
• 16:21
  Maybe it doesn't. But for you as an individual,
• 16:23
  market outcome shouldn't really matter too much.
• 16:25
  What matters is whether Atlas solves a pain point that you
• 16:29
  have today more than your current setup.
• 16:32
  And the only way to learn and find that out is to go test.
• 16:36
  So give it one to two to three weeks of testing and see what
• 16:38
  you think. Anyway, that's it for today.
• 16:42
  I hope you enjoyed the episode and I'll see you next week.