Keynote by Vishwas Lele || Azure Virtual Conference

Well, you know, I was looking at the agenda for today's conference and they're great speakers

They'll be going into a lot of detailed topics. And my objective here is to motivate a discussion

about the sessions that you're about to see later today. You'll drill into a lot of information

Mahesh already talked about and Magnus together talked about the sessions that are there on the

docket, I just wanted to take a few minutes to motivate a couple of things. So here's my first

slide, and this is my agenda slide. Quite unusual for an agenda slide to see a diagram, but these

are the three things that I really want to focus on. And I have limited time, so I won't get into

as much code as I would like to, but hopefully I can give you a sense for why this is important

So the three things that I want to focus on today, first is cloud native constructs

So what do I mean by that? And what does cloud native mean? And I'm sure you've heard of many terms

There is a whole foundation called the Cloud Native Computing Foundation, CNCF

There are a whole host of definitions about cloud native. I'll just give you my simple definition, which is, are you trying to build applications in the cloud that take advantage of the latest innovations and capabilities in the cloud, whether it is serverless, whether it is a managed Kubernetes environment, whether it is a consumption-based SQL server or a global database like Cosmos DB or something like that

The idea is that you're taking advantage of the latest and greatest innovation

Many of these services that we talked about, they've been designed with cloud in mind

whether it is multiple regions, failure, failover capability, elasticity, and whatnot. So if you're taking advantage of those things, I like to call them cloud native constructs

you know, whatever your definition may be. So keep that in mind and we'll drill into that

in a moment as well. So that's one. The second thing which is becoming increasingly important

is consumption-based pricing. So what do I mean by that? As large companies are moving to the cloud

as they are moving their existing applications to the cloud, it is becoming very clear

that now the next step of the journey is how do we optimize our footprint in the cloud? How can

we optimize the consumption in the cloud, right? This is going to be really important. And there's

a fundamental shift that this is happening, right? It is fine for you to move to the cloud

in a pre-provisioned capacity model. What do I mean by that? Well, I'm moving to the cloud

and I need these many VMs. I need this much storage. I need this much compute capability

or network or whatnot, I'm pre-provisioning that capacity. And of course, there's auto-scale

don't get me wrong, I'll provision it to a certain number, and then I can auto-scale up and down

Many applications can't take advantage of auto-scaling because, you know, their business is

consistent across the 24 by 7 time frame, may not be a peaks, they may have some peaks

but it's consistently used. There are many applications that see episodic use

a few users in the morning and then nobody in the day, maybe in the evening. So you have to keep

those systems up. And if you think of those in terms of consumption-based pricing, you are having

to pay for all the time that you have reserved the capacity. And what we are finding now in 2020

that the option for only paying for what you use is becoming really important

And that's what I mean by consumption-based pricing. Only use resources and only pay for resources

that you are using. Let me take a couple of examples to point this out. This has been around for a long time

but this is becoming more and more important. Let me give you a couple of examples

Everybody knows Azure Functions has a consumption-based pricing model. So you don't pay for a pre-provision capacity

You're paying for how many times you're invoking that piece of code. That's essentially what a consumption-based pricing is

But if you look at it today, Cosmos DB team has announced a serverless model

SQL Server in Azure. Azure SQL database has a tier, consumption tier

where you run a query and based on the kind of query that you're running

they can scale up the number of cycles they make available for your query to run efficiently

Take API management for example, they have a consumption tier. So previously, we used to have

hey, provision an API management instance, and it gives you 2 million calls

or whatever million calls it gives you, and customers are finding that, hey, I'm using 200,000 calls

I'm not quite using the quota, should I be paying for that amount

Wouldn't it be nice if I could pay for the API gateway that works in a consumption-based model

where I only pay for the number of API calls I'm servicing

So that's consumption-based pricing. Very important concept. And I know you cannot move to consumption-based pricing in all cases

And in some cases, provision capacity would make it more cost-effective. If you have a SQL server that is running at a consistent capacity for most majority of your month

then it might be advantageous for you to go with a pre-provision capacity. In fact

it might be advantageous for you to go with something called the reserved instances where

you say Microsoft or Amazon or whoever, I want this capacity for a period of time

Please give me a discount for a long-term usage. It might make sense, but I want you to focus on

consumption-based pricing. The third and final thing I want to talk about here really is developer

productivity. So what do you mean by that? Well, so we can have cloud native constructs

we can have consumption-based pricing, but it doesn't really help us if we can't build these

applications quickly. What are the sort of the developer productivity benefits that are coming

out from the various teams and how can we leverage that to build these applications quickly

So think about these three things as you are embarking on your journey in the cloud or you're already in the cloud as you're optimizing your presence in the cloud

Think of these three things as we progress further. Okay, so that said, I want to go quickly to Vishwas

Now, you've talked about these three motivations, consumption, cloud native, and productivity

But before you go there, let me ask a question. Who's using the cloud

Why is it important for me to learn this? Can you tell me about it? And here's a slide

I took this from Scott Guthrie's keynote back in June and built

More than 95 of Fortune 500 companies are using Azure public cloud some sort So I often still get questions from people that I interested in the cloud but I really

didn't have an opportunity to work on it because my employer will not let me go to the cloud

or my customer is very concerned about the cloud. We want to demystify that if 95% or greater Fortune 500 companies are using this, there's

a very good chance that the use case that you're working on is probably be acceptable in the cloud

And of course, during my introduction, I talked about this earlier. I live in the Washington

D.C. area. We do a fair amount of work on the Azure government space. And if some of the agencies

which do classified work are going to Azure government and public cloud in general

There are very few use cases out there where you can say, I can't run in the cloud anymore

So I want to take this back. Here's another slide very quickly

This slide is, again, gives you answering the question, who's using the cloud

Well, if Microsoft was not seeing this kind of adoption, they wouldn't be creating these

new regions, including the one in New Zealand. You can see at the bottom right, brand new region, 61 regions and growing

And the interesting number is that it's a worldwide presence. This is more than AWS and Google combined

Pretty interesting statistic. Okay, so we've talked about who's using the cloud

and why should you be and what are the kind of customers

The next thing we want to talk about is, so Vishwas, you said cloud-native applications

and you talked about consumption-based pricing and you talked about Azure Functions

that sounds like an interesting pattern but who is building real world enterprise grade

applications in that realm who's using these kinds of constructs like serverless and

consumption-based pricing to build enterprise grade applications well i want to show you a

quick example this is a chart that you may have seen and just look at the hockey stick growth

Back in March, Teams usage was around 32 million active users. In April, it reached to about 75 million active users

So you're looking at this chart and saying, yeah, that sounds great. But, you know, every collaboration remote meeting tool is growing at this pace

So why are you talking about this? Well, I'm talking about this because of this slide

Teams, whether some of you may be using it, some of you may be using some other tool

So Teams uses the kinds of services that I was talking about earlier, which is all the serverless kinds of things

It uses Kubernetes. It uses Cosmos DB. It is using Redis cache or service fabric and things like that

And look at the chart on the left hand side of the screen where because it was built on these cloud native constructs

they were able to quickly scale this application up. Going from 35 million to 75 million within a span of 30 days

is quite an incredible story. So the point I'm trying to make through this slide is

that use of serverless technologies together is leading up to really significant enterprise-grade applications

And having 75 concurrent or active million users, 75 million active users is a pretty significant number by any measure

Okay, so moving on, the next thing I want to talk about the developer productivity

So it's great to have serverless, it's great to have consumption pricing. What about developer

productivity? What are the kinds of things that we should be thinking about? So first and foremost

when we talk about Azure or when we talk about the Microsoft cloud, we are not just talking about

Azure. And as we go through the presentation, you'll realize this, that it is really the confluence

of three clouds that are giving us this hyper-productivity kind of experience. It's the

Azure platform, which powers it all. And then on top of that, you have the power platform

which is the low-code, no-code platform. And I'll give you a little hint. So there's a lot

of talk about low code, no code, and there's a lot of talk about citizen developers building

applications and not having a need for professional developers necessarily to write all these

applications. And that's certainly true. Power Platform has enabled citizen developers to dip

their toes into the world of building applications. But I'll give you a little tip. I see many

professional developers dipping into the power platform because they're able to crank out

applications such a much rapid clip than they would if they were using the traditional tools

So I want you to take this important lesson away. When we talk about productivity, we're talking about low code in the context of professional developers as much as we are

talking about the citizen developers. So our platform, Dynamics 365, you're building an

application for a customer. Maybe you want a single view of the customer. Dynamics could be great. You

don't want to create this up from scratch. And then, of course, we have the Microsoft 365 or M365

where you have the whole collaboration platform. You have the Teams platform, where you can embed

Power Platform applications. So, you can do all kinds of things to very quickly build these kinds

of applications. So moving on, here is a typical application. We have PowerApps running inside

Teams. So you can just build a low-code, no-code application and snap it into a Teams tab

which in turn calls API management. Teams can call an Azure service. It'll call an API management

And then API management can in turn call a microservice hosted in Azure Kubernetes service

and that Kubernetes service can then in turn talk to Cosmos DB

And, you know, there are all kinds of interesting things there related to the Kubernetes community doing things like the service broker

which automatically brokers a connection to Cosmos DB. So, you know, you are within the cloud native constructs, essentially

The next thing I want to talk about here really quickly is the data platform, right

You just cannot have this application generate a lot of data and not have good tools to yze this data

What do I mean by that? So on the left-hand side of the screen, you have data coming in from your on-premises application

Maybe you have some SaaS services. Maybe you have some IoT story

What have you? You have all of this data coming in, and then you now want to yze this data

and once again the concept of serverless and consumption-based pricing applies here as well

and this is sort of the important thing to take away from this slide so you bring all of this data

you store it some sort of a data lake and then once you have the data in the data lake you can now process this data using a relational data warehouse model which is the SQL pool capability previously known as the Azure data warehouse

Azure SQL DW is now called the SQL pool capability. So you have this data coming in into some sort of a storage data lake

And now you want to run some sort of an ytic query

in a relational model where, well, you're not paying for any SQL compute capability

if you don't want to. You can use something called the on-demand pool

and just run your query and your ytics query is running and you're only paying for the time

that ytics query is running. That's one part of Azure Synapse, right

You may have data which does not lend itself to a relational warehouse technology

and you really need to process This is using some sort of a Spark script

Well, Azure Synapse combines the two models of ysis. So on one hand, you have the relational model

and on the other hand, you have a Spark pool available. So all the goodness of Spark is available to you

So both these models can be brought up, can be elastic, can be on demand, and can be consumption-based pricing

So I want to make sure that our theme of all the way of building application

in a serverless consumption model continues all the way down to data ysis

The last two things I want to talk about in this slide is, of course, Azure Machine Learning

Azure Machine Learning is, again, it can be connected to the same data lake

You can run a training algorithm. And once again, in the same model, in this case

you might be running a deep learning algorithm. and that deep learning algorithm may require some GPU compute versus a CPU compute

You can now train your model using a GPU compute, pay for the time you're training it

And once you're done training, you can get rid of the GPU compute. You can take your trained algorithm, containerize it, and run it inside another consumption-based pricing called ACI or the Azure Container Instances

So you see where I'm going with this, that if we think about this whole model, we can have the richness of the platform, the productivity of the platform, at the same time, have a lower total cost of ownership

Okay, so that's the Azure ytics part. So with that said, I wanted to quickly show you a very simple application and I'm not going to get into as much detail because there are lots of sessions to go into the details

My goal here is really to motivate the things that I talked about

As a developer, I always like to have some code working to sort of bring home the point

But we will very quickly just go through this model. And then let's see, maybe leave some time for questions if there are any

So I have a simple reference application. And two of my colleagues, Gaurav Mantri, some of you may know him, and Vansli

This application was really put together in the last 48 hours because I thought it would be good to have an application

So I wanted to thank them, call them out. Here is what the application does

so we have a trip tracker application very nothing fancy we have simple application so

i'll walk you through this application right here starting here so first thing we wanted to do is

we wanted to have a mobile application that works across all mobile platforms and this is where

power app shines so we built a canvas app which can then be deployed to ios android what have you

So built a power app very quickly within hours, literally, to build a power app

That power app calls an Azure function and we'll take a look at the code

And what in this case is happening is when, let's say, I'm an Uber driver or I'm a taxi driver, I start my trip

Application comes, my mobile app starts sending GPS coordinates. Right here, start sending GPS coordinates to an Azure function

this function is HTTP triggered, so I'm only paying for the times that is coming in

The Power App calls the Azure function. The Azure function then in turn writes it to Cosmos DB

and then right here goes into the Cosmos DB. There's another Azure function here

which is triggered by the Cosmos DB change feed. So you may be familiar with the change feed

a very powerful model. All the things that are happening in a Cosmos DB instance

are available to you in the form of a change feed where you can do some processing on that data

So we have another Azure function. So this first function is triggered by HTTP

The second function is triggered essentially by the change feed. So once we write to a Cosmos DB instance

this Azure function is going to get triggered. this Azure function is then going to use SignalR

to essentially send an update back to an administrative app. So here's an administrative app

which can see where my taxi drivers are, and then they can essentially look at that data

So pretty simple model, but at the same time, not only is it simple, it has two other characteristics

Let's just talk about this for a moment. So it's a pretty simple model

but I'm of course paying consumption pricing but at the same time I have so

what happened here so I'm not sure what just happened okay I'll keep please tell me if you can see

this the screen suddenly change on me a little bit hopefully you can see it

Okay, I'll assume. Yeah, Vishwas, we can see it clearly. Okay, very good

Thank you, Stephen. Just something happened, and I just wanted to make sure you're able to hear me okay

So very quickly, the point I was trying to make here was, you know, all these pieces of code are, as I mentioned, they are serverless model

But at the same time, we have sort of outsourced the underlying complexity and the resilience

So what do I mean by that? Let me just explain this point a little bit. So because we are writing to Cosmos DB

Cosmos DB is a geo-replicated database. If we have lots and lots of new drivers show up

all we have to do is, you know, it's really up to Cosmos DB to scale up and down

It's not our responsibility. And let's say we have to spin up this application

in more than one region. We can have a geo-replication turned on

So we really have outsourced the underlying complexity of scalability and resilience

Same thing with the Azure functions. If more load comes in, we are not in the business of writing auto-scale logic

Somebody else's job it is. We just give them a range. They scale us up and down

And similarly with the websites here, since this is a single-page application

this can be hosted inside the static websites capability that Microsoft just announced And the idea is you really not paying for a web server because the entire code can be

on inside a CDN, if you will, and all of the dynamic aspects of the code are really some

JavaScript and the signals that are coming in here as part of that

So this may seem like a toy application, which it is. It was built in a few hours, but it at the same time exhibits the characteristics of an enterprise application in terms of scalability, resilience, and things like that

So with that said, let's go over and just take a quick walkthrough of this

Let me start out with, and we will make, what I'll do is after the session is over, I'll tweet the link to this code so you can spend more time looking at it if you like to

So the code is not very interesting, I have to say. Let's just see a couple of things here

So this is the trip tracker function. This is where our PowerApp is going to be

When HTTP trigger comes in, our PowerApp will come in here. We capture some data and we simply write it to the Cosmos DB service

Nothing fancy going on here. Let's look at the other Azure function right here

which is going to be triggered when the change feed triggers, okay, right here on line 15

And when that triggers, all we are going to do is do an add async for SignalR right here

which will then communicate with our administration website and update the location of that taxi driver

or an Uber driver, whatever. Okay. So those are the two important pieces of code right there

And then here's some code which basically essentially sets up the signal R right here

If you go up here, this is where we are setting up signal R. This is the startup.cs code

And this is, of course, in the function app startup code. This will enable the signal R

So pretty simple code base. I'm just trying to look at anything else I want to show here

There's some simulator. Of course, let's just look at the admin website here very quickly

So you can see the admin website. These are all the pieces of code

Not very, very interesting right here. You can see it's essentially a single page application

that can be hosted inside the static websites. Okay, so since I'm going to go run out of time very quickly

so let me just also show you this application in action. and so what I want to do here is I'll come to my resource group in a minute

All of the resources for that entire application are inside this one resource group

So we have the app service. We have the consumption plan where our Azure functions are hosted

We have a bunch of app insights. We have Cosmos DB, and we will go into this a little bit

We have a Cosmos DB instance, and there's a capability within Cosmos DB called the Synapse Link

And remember when I was talking about Synapse as your way to yze this data

Where inside Cosmos DB, we have something called Synapse Link. We can go into Cosmos DB and establish a connection so that that data is automatically transferred into an ytic store

So all of the hard work and error-prone logic that we had to write in ETL code is sort of greatly simplified by having that Synapse link

And I'll show that in a moment. And then I also have the Key Vault, of course

I have the machine learning service. We will go into that. Once you have collected the data, once we have done some ytics, we can also do some machine learning on that

Then I have the SQL pool. This is what I was talking about, where I can run my ytics

I have a bunch of storage accounts, of course, that you would expect. And here are the two Synapse workspaces that I have, essentially

So not a trivial resource group in terms of all of the capabilities that are there

Okay, so let's just go over here. Let's start our application right here

And this is, of course, a very, very simple application here. And I can do one thing here which might be interesting for you all to look at

Of course, this app can run inside iOS and Android or whatever. I just turned on the developer tools part just to point out one interesting thing here

Let's just go over here for a second now. So I can come here, right here you can see, let me escape out of this and you can see here

that what this application will render on a pixel, what will it render on iPad X and so on and so

forth, right? So that's the beauty of the Power Platform that you're writing essentially code

which is very similar to an Excel macro, and you are dragging and dropping these widgets

Under the covers, they are taking care of making sure your application can be

hosted across all of these platforms. I just wanted to quickly show you that

That's our application right here. Let's start our trip. If I start my trip here, let's see

we'll give it a moment here. So it'll start sending data to the Azure functions

which in turn will trigger our second Azure function, which in turn will trigger the SignalR code

So if I go into my web app here, web administration application right here

you can see that my chart is moving here. This is the static websites code that I talked about

So it is receiving signals from the SignalR service, updating the coordinates on the map

and as our trip is progressing here. So just imagine the productivity difference

If you had to build an application, this was literally a couple of hours of work

I almost toyed with the idea of building this app during the keynote

but then I said, maybe that'll be too much, but literally two buttons and a timer control

and really a data grid here. And then about 10 or 20 lines of formula-like code

which in turn calls the HTTP trigger for my Azure function. That's literally what this code is all about

And this code right here is pretty simple. A single Google map with signal-r coordinates coming in

and we are updating it here. And then see the taxi stopped moving

because I stopped the trip here. Okay, so that's one part that I wanted to talk about

Let's just now go back to our resource group here. And I just wanted to call out a couple of things

We can come in here. So this is the app service plan

This is the app service right here. And this is where our app service

admin interface is hosted right here so that's that's pretty easy to to understand uh i did

promise you that i'll show you the the cosmos db part and the link part so let's just go over to

the cosmos db this is a brand new capability and really exciting here this was part of the keynote

actually if you watch the build keynote and let's just go over to the to the document explorer

open the document explorer and I have two databases essentially let's just go in here and just let's look at the properties

of this guy here let's look at the properties and you if I go further down here and right here you

can see ytical storage is turned on. That's all you literally need to do to send your OLTP

information. So drivers are sending this information in real time back to Cosmos DB

and this information is getting automatically transferred over the Synapse link, converted into a columnar format, which is conducive for ysis. So now, if you wanted

to do some ysis on this data in real time, not last month's data or last week's data

but if you wanted to do real-time ysis inside Synapse, you can just spin up a Spark pool or a

SQL pool and start yzing this data, how many trips are happening, and what is the average price

Do you need additional drivers? You can take these decisions really, really quickly

This was easily, having worked on many projects related to ETL processing, I can tell you

that the error-prone process, this way of taking your data, converting it into columnar

format, conducive for ytics, is a big step forward. So I just want to show you this right here

So that's one part. So that's the Cosmos DB part. I'm trying to see

Let me also show you here. Let's see if I can bring up the Synapse

one of the Synapse workspace here. So, looks like it's waiting for this refresh to happen, but what I was going to essentially

show you here was all of the essentially a couple of examples of queries that I was performing against the relational database and looks like it not showing this up here because of an RBAC issue that okay

I will I will send you a screenshot of a tweet a screenshot I was not going to run this query but

really the idea here was that I was going to take New York taxi service publishes a lot of

historical data and what i did was essentially took 50 gigs of that data because you know my

app is not generating that data of course so in so took a 50 gigabyte set of data set from that new

york taxi service and brought it inside the synapse studio and then run a sql query against that and

it's pretty powerful to see that those queries can be run within seconds without really having built

a canonical OLAP model or something like that. So that's a really powerful capability

Okay. So the last thing I want to show you here is of course, the machine learning piece

So now that we have the data all available to us, we've performed some ytic queries

We have given our line managers some information that they can do

go into a decision-making process. I can now come in here and within the same data

So I'm not moving the data anywhere. It is still in the data lake

or in this case specifically, Azure Data Lake Gen 2. Now I want to do some ysis of

what are the factors that lead up to higher trips on any given day

Is it the number of drivers? Is it the placement of the drivers

Is it the timing of the drivers? If I want to figure it out, if I want to do a predictive ytics around that

this I can do. So once here, I'm inside machine learning workspace right here

And I know I'm going through a lot of services, so I'm trying to be a bit slow about it and not overwhelm you with too many tabs

But I think the key thing to take away from my presentation here really is that these are the capabilities available

and then on your own time, you can then go into each capability and explore it for your own

scenario. That's my hope here. So in this case, I have a Python workbook right here

which loads the data from our data lake Gen 2 architecture. It is processing them some data

As you well know, in order to run machine learning algorithms, you need to cleanse that data

Oftentimes you need to vectorize the data so that machine learning models can pick it up

So all this workbook is doing is doing some cleaning. It is doing some transformation So converting lat long into some real street names and whatnot so we can get a better sense here And then once the data processing has been done then I can show you the other workbook right here which is going to do some regression ysis on this data to figure out which features in this data set were most impactful in determining how we can maximize the number of trips

So we are taking predictive ytics into account. and once again, I can run through this code

and certainly I see an error here because when I started running this

just a minute or two before the keynote started, my compute instance was not running at that time

It is running. So once again, going back to our team, I need the compute instance only while I'm inside this

when I'm running this algorithm and once I've trained this algorithm, I can containerize it, put it on Azure container instances

and I really don't need that compute at any point after that

Unless I have to, of course, retrain my algorithm or change some hyperparameters or something like that

I really don't need that compute available back to me. So let's go back to the slides here really quickly

So we have come essentially a full circle, right? We have shown a power app

which we really built using drag and drop and using controls that work across the platform

And I showed you very quickly that, you know, you can, without having all the devices available to you

I gave you a little trick of how you can test your PowerApp against all of the different form factors

quickly build that application. That app is then connected to Azure Functions

which is nothing but an HTTP triggered in our case, which in turn stores data inside Cosmos DB

and inside Cosmos DB, we had an Azure Synapse link, which was sending the OLTP data into an ytics store

So we can do some ytics on top of that. And the real scale of this architecture comes by decoupling

So one function is writing to Cosmos. Cosmos is sending data to Synapse

Another function completely decoupled, and that's really the hallmark of scalable architecture

that you have things decoupled. The other function is waiting for a Cosmos DB change feed, gets a change feed

And then we are not making our administrative interface load and load again and do refresh or anything like that

We are pushing that data out or a WebSocket or something similar to our open browser connection and making updates

Imagine you have hundreds of cars running. You don't want to be expensive in terms of refresh and things like that

And then once we had the data in the ytics store we could run a Spark query or we could run a SQL query And then we could also run some sort of a regression ysis on that data to do predictive ytics So

in a sense, we have tried to achieve the three objectives that I had laid out. And I usually

like to do this. I like to go back to my agenda slide at the end of my session, always, because

did I hit on the points that I really had promised that I will hit on? So cloud native constructs

serverless programming model, not having to worry about patching and underlying OS upgrades

and having to worry about multi-tenancy and things like that, or writing complex

auto-scaling algorithms. We achieved that through a combination of services that we talked about

Did we achieve consumption-based pricing? Well, we did. We were using Azure Functions

We were using Cosmos DB. Now, Cosmos DB, we're using the free tier of Cosmos DB

and Cosmos DB team has announced a consumption-based pricing later this year, so watch for that

But other than that, every other piece, all the way up to machine learning

was on a need basis only. And then did we achieve the developer productivity

Well, we built a multi-platform platform, multi-form factor application using PowerApps, low-code, no-code ability. We use Synapse Link

to reduce our ETL time, and then we can use some of the prepackaged. I didn't get a chance to go

into that. I was showing you an example of regression ysis that I was conducting

but you may have heard of this. In cognitive services, there is a new capability called the

personalizer service. Some of you may have heard of that. It's essentially, and in fact, they

announced something called the apprentice mode for the personalizer service. So you really didn't

even need to write that custom algorithm. You could essentially feed your data to the personalizer

service, and it is based on an RNN model, a reinforcement learning model. So it can get that

data, and it can start giving you decision choices that you can make based on that data

without having to create your own ML workspace and things like that. So we are also taking advantage of the developer productivity in the process

Okay, so that brings us to 10.58, which is about a minute left in my keynote

I was told to go until 11. So, Stephen, Mahesh, Magnus, back to you guys