Bolting AI vs Reimagining: Is OpenAI Netscape Or Google

Sam Altman on bolting AI onto existing services vs. reimagining them with AI, Google's reimagining and Chrome displacing Internet Explorer, ChatGPT Atlas, Google Disco, and Netscape's cautionary tale

Bolting AI vs. Reimagining

The “OpenAI is Netscape” narrative has been gaining traction lately (popularized by my new fellow substacker Michael Burry); playing the role of Microsoft, who famously crushed Netscape, is, of course, Google. Sam Altman recently made an interesting point on The Big Technology Podcast, when asked about this threat:

Google has probably the greatest business model in the whole tech industry. I think they will be slow to give that up [...] Bolting AI into web search [...] I don’t think that’ll work as well as reimagining the whole… This is actually a broader trend I think is interesting.

Bolting AI onto the existing way of doing things, I don’t think is going to work as well as redesigning stuff in the AI-first world. This was part of why we wanted to do the consumer devices in the first place, but it applies at many other levels.

If you stick AI into a messaging app that’s doing a nice job summarizing your messages and drafting responses for you, that is definitely a little better. But I don’t think that’s the end state. That is not the idea of you having really smart AI that is acting as your agent, talking to everybody else’s agents and figuring out when to bother you, when not to bother you, and how to, you know, what decisions it can handle and when it needs to ask you.

So similar things for search, similar things for like productivity suites. I suspect it always takes longer than you think, but I suspect we will see new products in the major categories that are just totally built around AI rather than bolting AI in.

Exactly. I am, as avid readers may expect, very much in agreement with Altman here. Benedict Evans described the `bolting AI` approach as ‘legacy thinking’ as early as 2023, a time when companies were mostly bolting ChatGPT API calls into their products:

Whenever we get a new tool, we start by forcing it to fit our existing ways of working, and then over time we change the work to fit the new tool. We try to treat ChatGPT as though it was Google or a database instead of asking what it is useful for. How can we change the work to take advantage of this?

My ongoing fascination with these types of stories is a recurrent theme around here; most recently we discussed how Oracle failed in the cloud because it bolted an internet connection onto its IT product suite while Amazon reimagined the entire stack with AWS, and how, around the same time, the iPhone eclipsed Windows while Microsoft was forcing its PC operating system onto a phone¹. BlackBerry, as a poor response to the iPhone, literally bolted a glass touch screen on top of its physical keyboard, resulting in the worst phone ever made: the (Shit)Storm.

So, naturally, I was persuaded by Altman’s argument; he had me at ‘reimagining’. The question I was asking myself, however, was what is it, exactly, that OpenAI is reimagining?

It was the next thing Altman said that made me raise an eyebrow:

… We will see new products in the major categories that are just totally built around AI rather than bolting AI in.

And I think this is a weakness of Google’s even though they have this huge distribution advantage.

My (perhaps unpopular) opinion is that, for now at least, that doesn’t appear to be a “weakness of Google’s”, but rather of OpenAI’s. It doesn’t strike me as though much reimagining went into ChatGPT Atlas, for example. Benedict Evans wrote after OpenAI launched the browser in October:

Chrome was a huge upgrade from Internet Explorer, but Chrome today is already good, so what is Atlas offering? It doesn’t actually change web browsing itself at all — it adds some ChatGPT stuff around the sides, and a few obvious features that Chrome should (LLM-search of your history, say), but the web is still the web.

[...] Some of this looks like people doing what they know how to do. Sora is the Meta growth playbook, and there’s a device and a browser because devices and browsers are what you do. But how can you pull the whole thing inside out, and break through the elemental LLM problem that you have to think of what to ask? There’s very little sign of that kind of thinking in the Atlas launch — for heaven’s sake, the first two demos they offer are searching Google Docs and (sigh) writing code. This feels like people who said they needed a browser, so did a browser, not people challenging their comfort zones and pushing to new thinking.

Or, in other words, OpenAI has bolted AI onto a web browser; it wasn’t even the first (nor second) company to do so, and none of them seem to do anything remotely close to reimagining.

Chrome And Reimagining Applications

“When I joined Google in 2001, Larry and Sergey immediately said, ‘We should build our own browser,’” [then-CEO Eric] Schmidt says. “And I said no.”

- “Inside Chrome: The Secret Project to Crush IE and Remake the Web”, Steven Levy (2008)

Chrome being a huge upgrade from Internet Explorer is another interesting case of bolting the new thing onto the older paradigm, as opposed to completely reimagining.

Microsoft has, famously, leveraged the huge distribution advantage it had with Windows, and blocked Netscape (the original OpenAI?) by, essentially, bolting the World Wide Web as yet another desktop app. You double-clicked on the Microsoft Word icon for editing your documents, Microsoft Excel for spreadsheets, and Internet Explorer for, you know, the internet.

Google, on the other hand, was reimagining the entire concept of a software application. From A Brief History of Ajax:

[...] the bubble burst and web development crashed. Not, however, before Microsoft added a little-known function call named XMLHttpRequest to IE5. Mozilla quickly followed suit and, while nobody I know used it, the function stayed there, just waiting to be taken advantage of.

XMLHttpRequest allowed the JavaScript inside web pages to do something they could never really do before: get more data.1 Before, all the data either had to be sent with the web page. If you wanted more data or new data, you had to grab another web page. The JavaScript inside web pages couldn’t talk to the outside world. XMLHttpRequest changed that, allowing web pages to get more data from the server whenever they pleased.

Google was apparently the first to realize what a sea change this was. With Gmail and Google Maps, they built applications that took advantage of this to provide a user interface that was much more like a web application.

[...] With Gmail, for example, the application is continually asking the server if there’s new email. If there is, then it live updates the page, it doesn’t make you download a new one. And Google Maps lets you drag a map around and, as you do so, automatically downloads the parts of it you want to look at inline, without making you wait for a whole new page to download.

While Microsoft had invented XMLHttpRequest in the first place, it was Google who pushed the limits on the new possibilities it unlocked. To the extent that Internet Explorer was struggling to keep up with the rich web applications launched by Google in the mid 2000s. It was increasingly frustrating to visit Gmail, or YouTube, or Google Maps or Google Docs using Internet Explorer; Microsoft’s browser was built for the static HTML web pages of the 1990s. Bolting, essentially, the web onto the desktop app paradigm, which Microsoft dominated since the 1980s.

While bolting the internet onto the Windows desktop may have been enough to defeat Netscape, Microsoft’s browser was stretched beyond its limits when Google started building complex apps meant not for Windows, but rather for the browser itself.

Internet Explorer would often freeze, lag, or show a “script taking too long” error, as it tried to render Gmail or Google Maps. This is what enabled Chrome – launched in 2008 – to displace IE as the most used browser within four short years. The “Want a Faster Browser?” banner, displayed on top of Search and other Google apps, was so incredibly effective as Google Maps or YouTube was giving your IE browser a run for its money. Steven Levy summed it up well: “Speed may be Chrome’s most significant advance. When you improve things by an order of magnitude, you haven’t made something better — you’ve made something new.”

---

A key point in this story is that there was never a plan. Then-CEO Eric Schmidt was actually very clear on his objection to Google building a browser and taking on Microsoft. It wasn’t part of some 5-year roadmap. The Google Maps team was never told “make IE break so we could ship our own browser.”

What actually happened was in line with the “follow your curiosity, and trust that the dots will somehow connect in your future” approach (which, ironically, was so beautifully articulated by Steve Jobs, but it’s Google that these words describe so well).

Google’s deep obsession and fascination with the web, coupled with the founders’ limitless ambition, led it to imagine complex apps running inside a web browser. Once the dots started to connect, a new browser, rebuilt from the ground up around a blazing fast JavaScript engine, became a critical component of the new order (and of Google’s product suite).

This was far from obvious at the time, as even Schmidt was opposed² to Google launching a browser. In starting Chrome as a skunkworks project hidden from the CEO, Larry and Sergey probably followed their curiosity and trusted that the dots would somehow connect. Looking backward, it is now clear how they did.

AI And Reimagining The Browser

Fast forward back to today: Google has indeed been bolting AI onto each and every app lately, including web search (with AI Mode), and, of course, Chrome; “Reimagine with AI” was the exact wording with which Google announced the integration of Gemini into Chrome, a few weeks ahead of ChatGPT Atlas³. Not to diminish the engineering effort that went into this integration, it’s just that an “ask your AI assistant about your various open tabs” feature falls on the bolting AI end of the spectrum.

At the same time, however, Google is also making a real attempt at reimagining things with AI. From Google Labs’ blog earlier this month:

The web is a vast collection of applications and information, making it an incredible engine for discovery and learning. Yet, as our online tasks have grown more complex we’ve all felt the frustration of juggling dozens of open tabs to research a topic or plan a trip. We believe the web itself has the opportunity to adapt to the complexity, which is why we’re introducing Disco, featuring GenTabs, the newest experiment from Google Labs.

Disco is our new “Disco”very vehicle designed to reimagine browsing and building for the modern web. Disco will help us learn faster and work together with AI enthusiasts to shape the future of web browsing. And the first feature we’re testing is GenTabs, which was built with Gemini 3, our most intelligent model.

GenTabs helps you navigate the web by proactively understanding your complex tasks (through your open tabs and chat history) and creating interactive web applications to help you complete the tasks.

A browser that isn’t only rendering web pages, but also generating customized user experiences on the fly, is very thought-provoking. I expanded last summer about why I thought a similar concept – Voidware, interacting with an app that doesn’t exist – was a big deal. One manifestation of that concept was launched alongside Gemini 3: Generative UI⁴ can create “a rich, custom, visual interactive user experience for any prompt”. The Disco browser is probably powered by Google’s Generative UI.

It’s way too early to tell if Generative UI ends up as big a deal as XMLHttpRequest did 25 years ago, or if Disco is even going to succeed (I’m still on the waitlist, by the way, in case any Googlers reading this care to help me jump the line). The point is that Google is following its curiosity, trusting that the dots will connect somehow, with this being only one in a series of AI experiments coming out of Google Labs.

If anything, Google’s weakness is the opposite of what Altman claimed it to be: it might be reimagining too much, leading to a disarray of incomplete AI products. Consider Generative UI, which can be experienced through Gemini’s “Visual Layout” mode, but also through a new browser powered by it. This mess and confusion – which is ridiculed by many Google critics – is a necessary side effect of Google’s curiosity following. I made this argument back in April, in Why Google Has So Many Half-Baked AI Products.

OpenAI, in comparison, has been excellent in fully baking its products. ChatGPT is far more polished than Gemini. But OpenAI’s strength might turn out as a weakness; what if ChatGPT isn’t the final thing, but rather just an intermediate stepping stone (akin to Netscape Navigator and static HTML pages); Sam Altman himself suggested something along these lines during his Big Technology Podcast interview:

To be perfectly honest, I expected that by this point ChatGPT would have looked more different than it did at launch [...] I just thought that the chat interface was not going to go as far as it turned out to go [...] It looks better now, but it is broadly similar to when it was put up as a research preview.

[...] AI should be able to generate different kinds of interfaces for different kinds of tasks. So if you are talking about your numbers it should be able to show you that in different ways and you should be able to interact with it in different ways. We have a little bit of this with features like canvas. It should be way more interactive.

It sounds like he is describing Google’s Disco; but Google is the one launching experiments, while OpenAI is recently doubling down on improving ChatGPT and delaying other initiatives.

---

Netscape would soon “reduce Windows to a poorly debugged set of device drivers,” co-founder Marc Andreessen boasted in 1995, having blindsided Bill Gates with the explosive early growth of Netscape Navigator. “The browser will be the operating system,” Andreessen went on to predict. While he got that latter part right, Netscape itself was reduced to yet another Microsoft casualty within two years of Andreessen’s premature victory lap. It was Google who eventually reimagined its way into reducing Internet Explorer to a downloader for Chrome.

More recently, ChatGPT has indeed blindsided Google with its explosive early growth. But the cautionary tale of Andreessen’s hubris is that firing derogatory comments at the incumbent might not be sufficient. OpenAI would have to actually “redesign stuff in the AI-first world” – rather than just talk about it – if it wants to recreate the kind of success Google experienced in the early 2000s, and evade the Netscape fate.

---

For new posts to land directly in your inbox:

Disclosure: Long GOOG. Not financial advice. This post is for educational and general purposes only and should not be relied upon for investment decisions.

1. Before venturing into IT, Amazon had reimagined commerce, while Barnes & Noble bolted a website onto its brick-and-mortar bookstore chain. Despite everyone’s expectations, the B&N website did not crush Amazon. The same thing happened with Walmart who also bolted a website, only to find out that an e-commerce logistics network requires a completely different architecture than the one optimized to serve a national chain of stores. Blockbuster discovered the same harsh lesson, after bolting a DVD rental website onto its store network, but failing to beat Netflix. A decade later, HBO bolted a streaming service (HBO Go) onto its cable channels in 2010, while Netflix (dismissed by Time Warner CEO as ‘the Albanian army’), was reimagining television.

2. Schmidt was probably scared of taking on Microsoft because of his previous experience running Novell, whose fate was somewhat similar to Netscape’s. What Schmidt was slow to realize was that Microsoft’s source of power – Windows dominance – was being undermined by the emergence of rich web applications, running in the browser.

3. Which again demonstrates that an AI browser was a pretty obvious move on OpenAI’s part.

4. It was also fun for me to discover that Generative UI was built by my former and very talented colleagues from a million years ago when I worked on Google Search.