Gemini & how ‘all of this has happened before’ with Google Assistant

In writing about how Gemini is not ready to replace Google Assistant on Android, I’m struck by déjà vu. All the features that Gemini needs to be a good phone assistant is something that another Google product already figured out.

Generative AI is pushing Google product development into overdrive. It feels like we’re still at the start of that cycle, but I distinctly remember this happening once before. From 2016 to around 2020, Google was busy building Assistant. Like it does today with Gemini, Assistant felt like the future and number one priority given the frenzied pace of new capabilities.

It’s easy to trivialize, but Google Assistant [U: helped drive] the creation of the smart speaker and display form factor, as well as Google’s entry into headphones. It launched on phones, tablets, watches, TVs, and cars running Android, as well as Chromebooks and a messaging app.

At one point, “Google Assistant” was listed in the operating system section of the Pixel 3 tech specs. Then there was Assistant Driving Mode and how Google Assistant was replacing something that was previously under the domain of Android. 

Assistant truly felt like the thing that connected all Google products. It was a unifying theme and it made the company feel coherent. (If we zoom out, Google is at its best when there’s a company-wide mandate. Some like Google+/social and chasing Facebook ended miserably, but others – like the existential shift from desktop to mobile and Google Now playing a key role in differentiating early Android from the iPhone – were successful.) 

The Assistant push came to an end around 2021 and certainly by 2022. Various initiatives outside Google Assistant’s core voice competency were shut down, including the Driving Mode Dashboard, Snapshot, (nascent) Memory, and third-party support. 

The cynical interpretation is that Google is better at building than maintaining as codified by its internal structure for quantifying performance and promotions.

A more charitable view is that Assistant fundamentally hit the technological limits of what was possible in the early 2020s before large language models (LLMs) came on the scene. 

That said, I previously argued that Google in no way needed generative AI to make a good assistant, especially when you look at what made Google Now so promising. Google focusing and prioritizing what’s new is ultimately incompatible with providing a good experience for products that already have hundreds of millions of users. 

Anyways, that’s the past and we’re now in the Gemini era: 

Six: All of this has happened before…

Baltar: But the question remains: does all of this have to happen again?

Battlestar Galactica, but also peter pan (the movie)

The optimistic take is that LLMs are capable enough to build a truly useful assistant and other smart features. Even if it’s not, the work to get artificial general intelligence (AGI) could keep Google focused and result in other foundational innovations that can still drive meaningful product developments. 

I sure hope that Google now has a good foundation, and that it will keep iterating on Gemini rather than destructively restarting after a few years. That would truly be a new era.