A Google expert explains full-stack AI and full-stack development

So it started with apps, and now it’s on to AI?

Right. We’ve taken that exact same end-to-end principle and applied it to AI. If you’re trying to deliver value with AI, you can either buy a bunch of disparate parts from different vendors and try to stitch them together yourself, or you can look for an integrated system where everything you need is already connected.

What disparate parts can someone stitch together to make a full AI stack?

An intentional AI stack needs a cohesive combination of layers to get a job done: compute infrastructure, an AI model, an orchestration platform and the user interfaces. At Google, we’ve deliberately invested in every single layer. We provide the hardware like Tensor Processing Units (TPUs), frontier models developed by Google DeepMind like the Gemini family of models, the Gemini Enterprise Agent Platform and the interfaces people use daily, like Maps and Gmail. We’ve essentially done the hunting for you and put all the necessary components right inside the box.

Did we know we wanted to have a full-stack approach way back when Google first started working on AI?

It was absolutely a deliberate, decades-long strategy. For instance, our bet on custom TPUs is already over 10 years old. We recognized early on that there’s massive value in owning our own supply chain and raw infrastructure when serving up the world’s most important internet services. Owning that thread throughout the entire stack lets us deliver a level of service, performance and reliability that’s very hard to achieve if you’re at the mercy of multiple parties.

On the flip side, does adopting a full-stack platform limit builders in some way?

That’s a very fair concern, but locking people in doesn’t align with our ethos. No company does open source quite like Google; we regularly give away foundational technology and source code that the entire industry depends on.

We like to describe our AI platform as “opinionated but extensible” and “batteries included” — meaning everything you need to build and run an application is ready to go out of the box. However, if you want to use another company’s AI model instead of Gemini, or hook up different software instead of Google Workspace, you can plug those right in. We want you to use our products every day based on the completeness of our platform, not because we forced you into a closed choice.