Why Google Has So Many Half-Baked AI Products, And The Key Difference Between Google And Apple

"Working backwards to technology" vs. "hoping that the dots will connect somehow in the future"

Google’s AI Disarray

How is Google going after the AI consumer space, exactly?

There’s the Gemini app, of course. It offers chat, image generation, and can access your Google Drive. Gemini can answer questions about PDF documents or information it pulls from the web, and perform Deep Research. Through Canvas, it can also create code and documents.

Or maybe it’s NotebookLM, which is only available on the browser (though an Android app is presumably coming soon). This one, too, supports uploading documents (or pulling them from Google Drive), but for the purpose of generating an Audio Overview – a podcast, basically – discussing those documents. Oh, right, there is also chat – just like with Gemini – for asking questions about the input documents; and – as of last week – NotebookLM can “discover sources from around the web”. Just like, again, Gemini. In addition, though, NotebookLM offers its document creation and audio overviews features – which by the way were also just launched in Gemini.

You can see where I’m going with this.

Then there is AI Studio; it’s geared more toward developers, so it’s much more configurable, and allows you to eventually build an app through the Gemini API. But it also has chat, and you could upload documents, or photos, or – new feature! – link a YouTube Video, which Gemini API apparently supports, but the Gemini *app* does not. (NotebookLM does, however). AI Studio can search the web, but it doesn’t generate audio overviews nor briefing documents.

Using good ol’ Google Search often triggers an AI Overview; that’s a Generative AI answer that may be exported to a doc or an email, but there’s no chat. There isn’t even a “Continue the conversation” button that may take you to Gemini (or NotebookLM). There is also a family of experimental tools built by Google Labs, for generating images, videos or music; and, of course, many AI touch points across all existing Google products.

It just feels like a mess?

Meanwhile, OpenAI – Google’s most prominent challenger – has been focused on one product¹. ChatGPT is advanced, well-polished, and offers both mobile as well as desktop apps. It’s easy to discover new features, and OpenAI gives cool and concise demos; why is Google appears to be doing the exact opposite?

What’s the thinking behind maintaining a fragmented array of products, under different brand names, at different levels of maturity and partially overlapping features?

The answer, I believe, is that this is simply the way Google does things. And it might be easier to understand it, by contrasting it with Apple.

Steve Jobs On Working Backwards

John Gruber had complained about Google launching too many half-baked beta products, after yet-another cool prototype was announced back in 2014:

My quip today that Google is beginning to remind me of pre-NeXT Apple in the ’90s — announcing more cool R&D prototypes than they release actual cool products — brought to mind one of the segments from this session.

The session he is referring to is the famous Q&A session held by Steve Jobs at Apple’s 1997 WWDC. This was after he had just returned to the company, which was on the brink of failure at the time.

Gruber explains:

It starts with a testy remark from an attendee upset about Apple having killed OpenDoc. Attendee: “It’s sad and clear that on several counts what you’ve discussed, you don’t know what you’re talking about.” Jobs deftly laughs off the insult, and goes on to explain that he has no doubt that OpenDoc contains some great technology — that it allowed for things no other technology could accomplish. But that that alone was not enough.

Then he says this:

One of the things I’ve always found is that you’ve got to start with the customer experience and work backwards to the technology. You can’t start with the technology and try to figure out where you’re going to try to sell it. And I’ve made this mistake probably more than anybody else in this room. And I got the scar tissue to prove it.

If it wasn’t obvious then, it’s certainly obvious in hindsight that NeXT itself was the biggest of those scars. Amazing technology — an operating system and developer frameworks that debuted in 1988 and today serve as the foundation for almost the entirety of Apple’s product line. But NeXT never turned that technology into a successful product. Without a focus on products, new technologies are a crapshoot.

Start with the customer experience, and work backwards to the technology.

That principle has been guiding Apple since Jobs returned in 1997, shaped by the NeXT and the Macintosh scars. It was unique – in a valley crowded with technologists rushing to fund and ship almost every possible demo deemed cool – and instrumental in the phenomenal success Apple has experienced.

Google, one may argue, follows the exact inverse approach: while Jobs made Apple work backwards to the technology, Google starts with technology, and explores every possible path forward. That’s likely because the Google founders were shaped by different experiences.

Nikola Tesla and Internet Distribution

I'm going to try to be a little bit controversial … I think there's a lot of opportunity out in the world by doing things a bit differently than other companies have done them. I'm sort of amazed more people don't do this, and I'm going to use Google as an example of a way of doing things differently.

This is how Larry Page, co-founder of Google, opened his talk at the Stanford Graduate School of Business back in 2002. Just four years after he and Sergey Brin had moved Google off campus (by renting Susan Wojcicki’s garage), Page was still a Stanford PhD student on leave.

And how was he doing things differently than other companies? Page started by mentioning Nikola Tesla as a source of inspiration for him growing up:

At a pretty young age I wanted to be an inventor. Actually I read an autobiography of Tesla … the guy who invented AC power and generators and all sorts of things that we use today.

He barely was able to support his work - he's probably one of the greatest inventors of all time, and he's sort of pathetic in that, you know, he would have been able to do more if he was better at commercializing his inventions.

That little story pretty much distills the Google blueprint: invent amazing things – just like Nikola Tesla did – but, unlike him, figure out how to commercialize them (don’t be pathetic). But how would Google avoid the financial misfortunes of Nikola Tesla?

I think Google's a pretty good example of why product technology is important […] The common wisdom is that marketing is what really matters, based on product technology not mattering. I think that sort of misses one key thing which is the internet.

The internet is changing the world. If you look at the trend which the internet represents, with the internet you'll probably have better information than you do now on products. If you assume people have pretty good information on products, are they going to buy the right product? Probably. They'll probably use that information rather than your advertising. … with the internet, you can have a lot of distribution without spending very much money.

That’s the second piece; the Google founders had strong distaste for traditional promotion and brand building, especially in the early days. They classified everything a company could do to improve business as either product technology (good) or marketing (bad). The book I’m feeling lucky gives more context:

Scott Epstein, the consultant Google hired in early 1999 to get marketing on track, explained that they asked for feedback in traditional promotion and brand building. “They rejected that for one important reason—they believed in a ‘viral’ approach: ‘If the product was so good, they’d get an audience anyway through word of mouth.’” As a PR person, Cindy told me they were equally dismissive of typical brand spending. The result was to rely on an audience through word of mouth

So that’s how things worked: intense focus on inventions, on improving the product and its underlying technology, while relying on the distribution potential of the internet. No need to waste money on brand advertising or paid promotions; if it’s good, internet users would discover it – an option that wasn’t available for Tesla back in the 19th century – and it would go viral. By avoiding advertising spend, Google became immensely profitable from very early on².

Google from the very beginning was, and in many ways still is today, all about starting with technology, and pushing forward from there.

Connecting The Dots

It was Steve Jobs, though, who so perfectly articulated – without meaning to – what is essentially the essence of Google. Also at Stanford – the birthplace of Google – Jobs gave an inspiring commencement address in 2005, where he told the story of how, after dropping out of college, he stumbled into taking a calligraphy class:

I learned about serif and sans-serif typefaces, about varying the amount of space between different letter combinations, about what makes great typography great. It was beautiful, historical, artistically subtle in a way that science can't capture, and I found it fascinating. None of this had even a hope of any practical application in my life.

But ten years later, when we were designing the first Macintosh computer, it all came back to me. And we designed it all into the Mac. It was the first computer with beautiful typography.

[...] If I had never dropped out, I would have never dropped in on this calligraphy class, and personal computers might not have the wonderful typography that they do.

Of course it was impossible to connect the dots looking forward when I was in college. But it was very, very clear looking backwards ten years later.

Again, you can't connect the dots looking forward; you can only connect them looking backwards.

So you have to trust that the dots will somehow connect in your future.

Jobs never referred to Google in his speech, of course; but it’s remarkable how well this beautiful story reflects, in fact, the way Google works.

Follow your curiosity, and trust that the dots will somehow connect in your future.

To the angry member of the audience, Jobs explained in 1997 that “Focusing is about saying no. And you’ve got to say no, no, no. When you say no, you piss off people”. Google, however, didn’t have Jobs, and wasn’t burdened by the type of scars Jobs had endured during his first tenure at Apple and the NeXT exile³.

Google mostly says “yes, yes, yes”, and has launched an endless array of beta products over the years, leaving the job of saying “no, no, no” to, well, the people of the internet. That was basically the purpose of programs like Labs or X: following curiosity and trusting that dots may somehow connect; in most cases they never do – the internet says “no” – and the product ends up in the infamous Google Graveyard. But in some cases the dots are connecting, which was the case for NotebookLM’s Audio Overviews.

NotebookLM and Gemini

NotebookLM was first announced in summer of 2023 – alongside a line of AI endeavors incubated within Google Labs – to “reimagine what notetaking software might look like if you designed it from scratch knowing that you would have a powerful language model at its core”. But you probably haven’t heard of it until 15 months later, when this invention experienced its viral moment: Audio Overviews rolled out in September of 2024, which a Wall Street Journal article dubbed “Google’s ChatGPT moment”.

The NotebookLM team never started with a specific goal of auto-generating podcasts for documents, and worked its way back to the technology; without the charter to openly explore how generative AI may power a research app, they would have never stumbled onto the podcast idea in the first place. This is one of the rare and serendipitous cases that make the Google Graveyard worthwhile.

Since this wasn’t fully planned, though, what is to be done once it starts to gain momentum? NotebookLM was built under Google Labs, which appears as a separate organization than Gemini and DeepMind, reporting to a different VP with different OKRs and different resources allocated to meet them (and, potentially, different considerations about promoting their career at Google). I suppose that leads to a situation where the NotebookLM team is scrambling to capitalize on its early success, developing it into a mature product on one hand – which also means incorporating capabilities that overlap with Gemini – and, on the other hand, embedding its Audio Overviews feature across Google’s product portfolio. Including Gemini.

While the romantic “dots will somehow connect” argument can make Google’s AI product chaos understandable, the competition between Gemini and NotebookLM is getting silly at this point. They should probably be merged together, which is perhaps the backstory behind this news report from Reuters last week:

Google is replacing Sissie Hsiao, who led the development of the artificial intelligence chatbot Bard, now known as Gemini, the company told staff in its AI division on Wednesday.

Hsiao will step down immediately, a company spokesperson confirmed. Josh Woodward, who leads Google Labs and oversaw the launch of NotebookLM — the company's popular tool that can turn text into a podcast-like show — will replace her.

[...] Woodward will retain his role as head of Google Labs while shaping the next chapter of Gemini, the spokesperson said.

This seems like a healthy organizational change; Google Labs is actively exploring many more directions, hoping the dots somehow connect in the future. When that happens again, instead of sparking a competition with Gemini, the successful experiment could simply mature into a feature on the Gemini app.

While Larry Page was spot-on with his 2002 insight that “with the internet, you can have a lot of distribution without spending very much money”, it’s also a good idea to consolidate that distribution into one place (i.e. Gemini); as John Gruber so bluntly articulated it: “Without a focus on products, new technologies are a crapshoot.”

1. I acknowledge that Sora was launched as a standalone product; for the most part, though, OpenAI seems almost entirely focused on ChatGPT.

2. Yes, I see the irony in that Google was (and still is) making money mostly through offering other companies a way to promote and market their products and services, something that Google itself was reluctant to do.

3. It also helped, of course, that launching an internet product requires very low marginal costs, unlike the physical devices Jobs was building.