Seven Key Product Announcements from Google I/O 2024

AI Agents

Of all the things mentioned during the Keynote, this was most interesting for both good and bad reasons. An AI agent is geared to work on your behalf while you’re doing other things. In the example, they posited the issue of returning a pair of shoes. In such a case, the agent will contact the seller, fill out the refund, and arrange all the shipping details without you having to do anything.

In another example, you may be leaving for a business trip. The AI could book your flight, rental car, and hotel for you and even set up things to do on your days off. The agent would handle all the details if an issue required you to reschedule the trip.

While this sounds great in practice, one thing not mentioned during the conference is AI “hallucinations” and AI accuracy in general.

Gemma

While Gemini is great for generalized information, you may want to create a model based on your data. Gemma is designed to be a lightweight model that developers can customize based on their needs. It’s built on the same models used to build Gemini.

This model is designed to run on upcoming Nvidia chips, which are expected to be released at the end of the year. It also has the flexibility to run on devices with limited power, which lets you embed your own model into an app.

Gemma also provides a way to share and collaborate with other models on ai.google.dev/gemma. For example, there’s CodeGemma to incorporate code completion in your app, PaliGemma for a vision-language model, and RecurrentGemma to leverage recurrent neural networks.

Synth ID

The Google I/O keynote contained many token mentions of privacy and responsibility. Thankfully, one announcement really addressed the concept of responsibility: Synth ID.

Synth ID is a collection of technologies that aim to watermark AI-generated content regardless of its type. This watermarking would not be obvious to the end user but detectable using tools.

This initiative aims to curb disinformation. For instance, a bad actor may produce an AI-generated video to embarrass a political rival. With Synth ID, it would be clear that it was an entirely fake video.

Synth ID works with video, audio, and even text while undetectable to the end user. Here’s an example of text watermarking from Google:

Google is working with large AI companies to develop and implement this standard. How this works in practice is yet to be seen, but it’s something to watch in the future.

General Developer Improvements

While Google has gone “all-in” with AI, the developer keynote also featured improvements to their various frameworks and languages.

This part of the conference started with a large endorsement of Kotlin Multiplatform. Kotlin Multiplatform is a framework used to create cross-platform apps using both Kotlin and Jetpack Compose. With Google putting its full weight behind it, there will be many improvements in the future. You can read our Kotlin Multiplatform book here.

Google Chrome is receiving a view transitions API. This will allow your web app’s transitions to flow from page to page as if using a native app. Of course, Chrome is also an AI model that helps diagnose and fix errors.

Firebase announced web hosting, and they even feature a new logo. Firebase now also features a way for Firebase to work with SQL databases with Firebase Data Connect.

Finally, Flutter made a small appearance with WASM web support and the integration of the Gemini API. While Flutter didn’t have a big showing during the keynote, at least there were lots of session videos produced by the team.

Where to Go From Here?

Many announcements were made during this year’s Google I/O, with many marked as “coming this summer” or “coming soon”. That said, you can find a rich assortment of session videos on the official Google I/O developer site. You can also keep an eye on Google’s official developer blog. Finally, keep checking back with Kodeco for more information on Google development tools and how to use them!