Data Driven

[bright upbeat music]

Alexa, could you book me a reservation

for four somewhere fun in Queen Anne?

[Alexa] Grappa has openings for four tonight

from 5:15 p.m. to 8:45 p.m.

What time works best?

How about 5:30?

[Alexa] Perfect.

I've booked a table for four at Grappa tonight at 5:30 p.m.

It's on your calendar too.

Alexa+ changes the game.

Alexa+ will be reminiscent of the Alexa experience

that you once knew, and otherwise it is entirely different.

[upbeat music]

Alexa makes it easy to just have a conversation

and then our AI systems are able

to make sense of what is the customer after

and what are the best tools or ways to accomplish that.

It involves completely rearchitecting the system

around generative AI and large language models,

but also bringing along the core capabilities,

the tens of thousands of devices

and services that we support within Alexa

need to come along to Alexa+.

So building these unbelievably powerful generative AI

utilities at scale,

you need the right core technologies.

AWS having access to those tools

and the expertise,

we both collaborated to make our vision reality

and also learned along the way.

In the making of Alexa+,

we were faced with so many challenges.

Alexa has a lot of features from smart home

to music, timers and alarms, information,

and it turns out LLMs are easily confused.

So we had to decompose this range of features

into units we called experts.

The LLM does the conversation

and when it detects that we're in the vicinity of music,

for example, we just gave the LLM the music expert.

The next challenge was there's a lot of models out there,

different types of language models,

and no one model is right for all jobs.

So we built an orchestrator of models

so we can use the best model for each job.

Amazon Bedrock is an unbelievable source

for basically the best state-of-the-art models

and we use basically the full range

of models available in Amazon Bedrock,

but really anything you might need for any particular job.

The third problem is latency.

When a user asks us to do something, we want to do it fast.

And when you take those skills

and you deliver them fast,

there's a magic effect that really delights the user.

I mean, we're sitting here chatting, right?

And if I hang up for a little while,

you probably get pretty frustrated.

And the same's true of Alexa,

if it's gonna be a natural voice interaction,

it's super important for us to be fast.

I'd say the best way to summarize it

is we had to find that operating point

that balances accuracy, latency, and efficiency.

When we brought all of this together

that made a personal assistant based on LLMs possible.

[Alexa] What kind of food are you in the mood for tonight?

Let's do that sushi place.

[Alexa] Great choice.

Just tell me what you'd like

and I'll find it on the menu as you speak.

Alexa is smarter.

Alexa will bring more personalization to your experience,

and Alexa can get a whole new world of things done.

[Alexa] I see that Japonessa Sushi Cocina delivers

through both Grubhub and Uber Eats.

Which one would you prefer to use?

Let's do Grubhub.

There are tons of partners that are ready to go with APIs

and most of what Alexa gets done today is through those,

but it's not always through APIs

and so we wanted to build utilities

so that we could plug into anything.

There were some partners that did not have a scalable API

or their main interface was through websites.

And for those,

we came up with some agentic capabilities called Nova Act.

We taught Alexa how to navigate to websites

and take actions on behalf of the user.

With Amazon Nova Act,

we're actually able to unlock a whole new set

of integrations through partner websites

and services that they have up online.

Alexa, I need someone

to fix my leaky faucet tomorrow morning.

Can you help?

[Alexa] Perfect.

I can set you up with Raptor Plumbing & Sewer

for your leaky faucet repair tomorrow morning.

Nova Act, it's able to navigate a website for you

as if you are doing it yourself.

It find someone who meets my needs,

find when they're available, compare that to my calendar,

book a repair, and tell me when it's done.

[Alexa] They respond super fast, within three minutes,

and have a 4.9 rating.

Ready to send the request?

Now we're able to unlock more services

that don't have an API

that our customers can now access with Alexa+.

What I love about the team here

and what we have under one roof at Amazon

is you have AWS Amazon Bedrock,

Amazon Nova working on systems

that can actually get things done for customers

to pay off a vision as big as Alexa+.

This promise of technology

that supports you without getting in your way,

without taking you out of the flow or the conversation,

we're still on the journey to realize that vision

and really let the technology fade into the background

and just support you in your everyday life.

[Alexa] Absolutely. You got it.