WEBVTT

00:00:00.510 --> 00:00:07.510
[Music]

00:00:22.300 --> 00:00:29.300
SPEAKER 1: We find ourselves
in a new era, one where access

00:00:29.620 --> 00:00:35.617
to information is access
to expertise, from the farm

00:00:35.700 --> 00:00:39.957
to the lab,
from the boardroom to the classroom.

00:00:40.040 --> 00:00:43.977
Now anyone can save time
with a personal assistant.

00:00:44.060 --> 00:00:45.537
SPEAKER 2: By our measurements,
the introduction

00:00:45.620 --> 00:00:50.487
of the AI has meant that it now takes
about 40% less time to access data.

00:00:50.570 --> 00:00:54.427
SPEAKER 3: We've had operational staff
that write policies and procedures,

00:00:54.510 --> 00:00:56.977
and that normally takes
between two and three weeks,

00:00:57.060 --> 00:01:00.367
and they've found a way to do
that within two to three days.

00:01:00.450 --> 00:01:04.627
SPEAKER 1: This is a new way
to analyze with a personal coach.

00:01:04.710 --> 00:01:06.127
SPEAKER 4: AI Insights
gives us the ability

00:01:06.210 --> 00:01:08.317
to supercharge someone's
knowledge of the game,

00:01:08.400 --> 00:01:11.737
make them feel part of a community
and part of the sport.

00:01:11.820 --> 00:01:14.047
SPEAKER 5: Cognates is a Socratic
tutoring support system

00:01:14.130 --> 00:01:18.997
that allows us to have
a 24-hour tutor in every subject.

00:01:19.080 --> 00:01:22.707
SPEAKER 1: AI is unlocking
creativity for us all.

00:01:22.790 --> 00:01:28.717
SPEAKER 6: Descriptions are so detailed in
my imagination, I can paint the artwork.

00:01:28.800 --> 00:01:30.097
SPEAKER 7: One of the biggest pain points

00:01:30.180 --> 00:01:32.467
about the creative process
is just getting those ideas

00:01:32.550 --> 00:01:35.047
out of your head, and it gave
me four different examples.

00:01:35.130 --> 00:01:39.937
I was able to draw inspiration from that
and create an embroidered product based

00:01:40.020 --> 00:01:45.437
on these initial prompts. SPEAKER 1:
And we are just getting started.

00:01:45.520 --> 00:01:50.427
[Music]

00:01:50.510 --> 00:01:54.070
SATYA NADELLA: Right. Good morning.

00:01:56.240 --> 00:01:57.637
Good morning.

00:01:57.720 --> 00:02:03.577
It's fantastic to be back
in Sydney and in Australia.

00:02:03.660 --> 00:02:06.367
This morning, I got up,
and then I tried to sort of go

00:02:06.450 --> 00:02:12.457
down memory lane, thanks
to Copilot, and try and get back

00:02:12.540 --> 00:02:17.737
to the first time I showed up in Sydney
for one of the developer conferences.

00:02:17.820 --> 00:02:21.687
And it was, I think, in 1993,

00:02:21.770 --> 00:02:25.827
and I remember distinctly
getting through the airport

00:02:25.910 --> 00:02:29.997
with my luggable Compaq computers,

00:02:30.080 --> 00:02:35.677
and the conference was all about
-- we had introduced, I think,

00:02:35.760 --> 00:02:40.627
for the first time back then,
Visual Basic for Applications.

00:02:40.710 --> 00:02:44.677
That was like we replaced
the macros in Excel 4.

00:02:44.760 --> 00:02:48.787
Excel 5 had come out,
and we had Visual Basic.

00:02:48.870 --> 00:02:54.007
And I distinctly remember,
in fact, having a session.

00:02:54.090 --> 00:02:58.237
This time around, there a few
more people for my session,

00:02:58.320 --> 00:03:00.307
and then I remember
going to one of the banks

00:03:00.390 --> 00:03:03.727
and having a big discussion
on going from macros

00:03:03.810 --> 00:03:06.277
to Visual Basic -- is it
a good thing, or what have you?

00:03:06.360 --> 00:03:10.837
But little did I know that
whatever 30 odd, 35 years nearly

00:03:10.920 --> 00:03:15.847
after that, the hardest thing
in tech is, again, guess what?

00:03:15.930 --> 00:03:22.057
Agents in Excel and or
command line interfaces

00:03:22.140 --> 00:03:28.267
for coding agents. I mean, it's
fascinating that as the frontier moves,

00:03:28.350 --> 00:03:33.697
so does the arrival back
to where you started,

00:03:33.780 --> 00:03:39.037
to sort of use that T.S. Eliot metaphor.

00:03:39.120 --> 00:03:44.317
To me, I want to spend most
of my conversation today

00:03:44.400 --> 00:03:49.027
and time today to talk a little
bit about that new frontier,

00:03:49.110 --> 00:03:53.917
and how are we all collectively
going to push that forward

00:03:54.000 --> 00:03:57.307
with essentially what is a new input?

00:03:57.390 --> 00:04:02.047
The outcomes that matter still
remain very much the same, right?

00:04:02.130 --> 00:04:07.327
When we think about any
organization, any institution,

00:04:07.410 --> 00:04:11.557
even individually, what we
care about is the excellence

00:04:11.640 --> 00:04:15.007
in outcomes, right, which
is how do we show up and change

00:04:15.090 --> 00:04:18.607
that engagement with customers,
or the employee experience,

00:04:18.690 --> 00:04:22.687
or the efficiency with which
any given business process gets

00:04:22.770 --> 00:04:27.487
completed, or even bending
the curve in innovation, right?

00:04:27.570 --> 00:04:29.857
Whether it's software
development or drug discovery,

00:04:29.940 --> 00:04:32.647
how are we going to truly change that?

00:04:32.730 --> 00:04:36.957
And that's sort of the frontier
that we are trying to move.

00:04:37.040 --> 00:04:41.887
And if you look back at it, right,

00:04:41.970 --> 00:04:46.717
throughout these transitions
that we've had in technology,

00:04:46.800 --> 00:04:51.247
a lot of this has come down
to us changing the way we work

00:04:51.330 --> 00:04:55.507
with technology, but more than
the way we work with technology,

00:04:55.590 --> 00:04:58.627
how technology reshapes the work we do.

00:04:58.710 --> 00:05:03.997
That's sort of been the story
of sort of the movement

00:05:04.080 --> 00:05:06.517
on the frontier,
and it starts, quite frankly,

00:05:06.600 --> 00:05:09.157
with that mindset, right, that curiosity,

00:05:09.240 --> 00:05:12.847
the posture that we need to have
that when there's something new,

00:05:12.930 --> 00:05:15.427
you go all in; you learn it.

00:05:15.510 --> 00:05:18.037
That means you really need
to think about your skills.

00:05:18.120 --> 00:05:23.677
You need the new tools. And perhaps
one of the most interesting things

00:05:23.760 --> 00:05:30.157
and important things that we'll talk about
in the AI era is codifying the outcome,

00:05:30.240 --> 00:05:35.797
perhaps even more robustly going
forward into these evals, right?

00:05:35.880 --> 00:05:40.387
If you think about AI, anyone who's
sort of doing applied AI starts by sort

00:05:40.470 --> 00:05:45.397
of with this mantra called
the product is the eval

00:05:45.480 --> 00:05:48.637
and that hill-climbing on evals, right?

00:05:48.720 --> 00:05:52.147
And that's that subtle, nuanced way all

00:05:52.230 --> 00:05:57.277
of us judge what great looks
like, good looks like inside

00:05:57.360 --> 00:06:00.997
of your organization for any
outcome becomes the real

00:06:01.080 --> 00:06:04.897
currency through which
we can then change the use

00:06:04.980 --> 00:06:09.217
of technology and have it reshape
our organizational frontier.

00:06:09.300 --> 00:06:14.917
And so to that end, we are building
this end-to-end tech stack, right?

00:06:15.000 --> 00:06:19.537
It's got three parts to it,
and for the rest of my talk,

00:06:19.620 --> 00:06:23.917
I want to sort of unpack
each of these in some detail,

00:06:24.000 --> 00:06:29.347
and the true line across all
of this is going to be two things:

00:06:29.430 --> 00:06:33.067
intelligence and trust.

00:06:33.150 --> 00:06:38.107
Because ultimately, you really
want the intelligence to show

00:06:38.190 --> 00:06:40.567
through across each layer of the stack

00:06:40.650 --> 00:06:43.537
and amplify what you're
doing, move that frontier,

00:06:43.620 --> 00:06:46.837
move that outcome,
but you really need trust

00:06:46.920 --> 00:06:51.307
because it quickly can become
vicious cycle versus a virtuous cycle

00:06:51.390 --> 00:06:53.197
if you really don't have
intelligence and trust.

00:06:53.280 --> 00:06:57.307
And so we'll sort of talk through
how those all come together.

00:06:57.390 --> 00:07:00.127
So let's start with sort
of the top of the stack,

00:07:00.210 --> 00:07:03.007
which is around these high-value scenarios

00:07:03.090 --> 00:07:07.267
and high-value applications
or agentic systems.

00:07:07.350 --> 00:07:12.397
And for us, it starts with the way
we are architecting Copilot.

00:07:12.480 --> 00:07:15.847
If you think about Copilot,
whatever, three years ago or so,

00:07:15.930 --> 00:07:18.847
we started with mostly
a chat-based response,

00:07:18.930 --> 00:07:22.777
request-response type
of pattern, and especially

00:07:22.860 --> 00:07:26.317
with GPT-4, it became a new way for us

00:07:26.400 --> 00:07:30.277
to discover knowledge inside
the organization or on the web.

00:07:30.360 --> 00:07:33.577
And then subsequently,
we had reasoning models come

00:07:33.660 --> 00:07:36.907
out that sort of helped us
not just have essentially new

00:07:36.990 --> 00:07:39.847
search, but we had
reasoning plus search, right?

00:07:39.930 --> 00:07:45.697
Even the chat interface started to pretty
drastically change with reasoning models.

00:07:45.780 --> 00:07:50.287
Then we started getting better
at sort of long-run agents

00:07:50.370 --> 00:07:52.657
that you could send tool-calling.

00:07:52.740 --> 00:07:57.577
And so that started sort
of really the beginning of agents

00:07:57.660 --> 00:08:00.457
like Researcher and Deep
Research and what have you.

00:08:00.540 --> 00:08:04.297
So the first thing we started
doing was to say, okay,

00:08:04.380 --> 00:08:08.257
let's take even chat
and that reasoning and make sure

00:08:08.340 --> 00:08:11.497
that you have access
to all the models, right?

00:08:11.580 --> 00:08:16.417
So in fact, when I use
Copilot today, I'm mostly sort

00:08:16.500 --> 00:08:20.557
of doing Auto, and Auto
is picking the best model

00:08:20.640 --> 00:08:25.657
for my intent, and then I have
a dropdown with all the models available.

00:08:25.740 --> 00:08:30.427
And that is sort of how the chat
interface itself has evolved, right?

00:08:30.510 --> 00:08:32.527
You used to be, in the beginning,
you used to have a router,

00:08:32.610 --> 00:08:35.587
but now the router
is becoming smarter and smarter

00:08:35.670 --> 00:08:38.437
in being able to sort of do
the intent classification,

00:08:38.520 --> 00:08:41.827
pick the right models for the intents.

00:08:41.910 --> 00:08:46.447
Next, we've now taken that idea
of having multiple models

00:08:46.530 --> 00:08:49.177
and put them even
in the context of agents.

00:08:49.260 --> 00:08:54.067
So if you go to Researcher,
you now have these new modalities.

00:08:54.150 --> 00:08:57.847
For example, you have
a modality called Critique,

00:08:57.930 --> 00:09:01.087
and I love that a lot because,
after all, how do we work with people?

00:09:01.170 --> 00:09:03.277
We critique each other and our ideas.

00:09:03.360 --> 00:09:06.607
That's the way you work with even models.

00:09:06.690 --> 00:09:09.427
In fact, another mode
that I like a lot is Counsel,

00:09:09.510 --> 00:09:12.547
which is essentially
assigning a particular task

00:09:12.630 --> 00:09:17.437
or a research topic to a group,
and then you get the best

00:09:17.520 --> 00:09:19.557
of all the models come together.

00:09:19.640 --> 00:09:23.097
It's fantastic to see it if you
have not used it to sort of look

00:09:23.180 --> 00:09:25.927
at both the models,
not only doing critique,

00:09:26.010 --> 00:09:31.087
but also finding biases in each
model, and then, in fact,

00:09:31.170 --> 00:09:32.737
insights from each model, right?

00:09:32.820 --> 00:09:35.617
So when even the research
report comes back,

00:09:35.700 --> 00:09:41.707
you gain deeper insights because that you
have multiple models at your disposal.

00:09:41.790 --> 00:09:45.277
Now, the thing that I'm really excited

00:09:45.360 --> 00:09:50.467
about is this next big thing
that we've done, and in fact,

00:09:50.550 --> 00:09:56.347
today's the general availability
of agent mode across Word,

00:09:56.430 --> 00:10:00.997
Excel, and PowerPoint. Yeah. [Cheers]

00:10:01.080 --> 00:10:04.840
Yeah, it's kind of like 1993 again.

00:10:07.230 --> 00:10:13.147
And it's actually, in some sense,
all of my life's work is right here.

00:10:13.230 --> 00:10:15.337
And the reason I say
that is because if you think

00:10:15.420 --> 00:10:22.420
about what's happened is Excel
became a super popular tool,

00:10:22.610 --> 00:10:27.667
because it helped us intuitively
get number sense, right?

00:10:27.750 --> 00:10:33.517
It had representational
power in the canvas, right?

00:10:33.600 --> 00:10:37.897
The rows and columns have
information content, right?

00:10:37.980 --> 00:10:39.127
As soon as you see a spreadsheet,

00:10:39.210 --> 00:10:42.897
you kind of make sense of what
that spreadsheet is, right?

00:10:42.980 --> 00:10:48.787
That's why the canvas has lasted
this many decades, in fact.

00:10:48.870 --> 00:10:51.757
And so now think about
what agent mode does.

00:10:51.840 --> 00:10:58.177
What agent mode does is it takes
that information in the sheet,

00:10:58.260 --> 00:11:05.047
the positional or spatial information
representation, and then mixes it

00:11:05.130 --> 00:11:12.130
with essentially these powerful
reasoning models and the ability

00:11:12.450 --> 00:11:17.287
to take action -- not just
reason, but take action.

00:11:17.370 --> 00:11:20.917
And so when you put that together,
it's magic, right,

00:11:21.000 --> 00:11:27.097
which is you now are creating a very
-- you're instructing the model

00:11:27.180 --> 00:11:31.897
to create a very sophisticated
model, an Excel model.

00:11:31.980 --> 00:11:34.837
And then when it creates
the model, it's represented

00:11:34.920 --> 00:11:38.377
in this canvas, which you know
how to even spatially inspect,

00:11:38.460 --> 00:11:41.347
and you know what's driving what.

00:11:41.430 --> 00:11:46.537
What's next to what? And then you
can then further reason about it.

00:11:46.620 --> 00:11:49.147
And in fact, you can take
that entire reasoning trace,

00:11:49.230 --> 00:11:51.877
plus the sheet, plus the model,
and share it, right?

00:11:51.960 --> 00:11:54.547
I mean, think about
the cognitive work inside the

00:11:54.630 --> 00:11:57.757
company and how it's getting reshaped.

00:11:57.840 --> 00:12:01.807
It has familiarity of a spreadsheet,

00:12:01.890 --> 00:12:05.917
but everything around it is changed.

00:12:06.000 --> 00:12:10.957
And that, I think, is when you know you
have a powerful sort of set of technology.

00:12:11.040 --> 00:12:13.447
And so we're really, really
excited about this agent mode.

00:12:13.530 --> 00:12:15.247
And of course, we're not
stopping there, right?

00:12:15.330 --> 00:12:20.017
The next thing that is also
built into Copilot now is Cowork,

00:12:20.100 --> 00:12:25.057
the ability to now assign
a synchronous task in a queue

00:12:25.140 --> 00:12:27.157
to Cowork, which, by the way,

00:12:27.240 --> 00:12:30.787
could be creating a Word
document or an Excel spreadsheet

00:12:30.870 --> 00:12:34.687
or what have you, and that then
you can open up in agent mode

00:12:34.770 --> 00:12:37.537
and round trip this entire way of working.

00:12:37.620 --> 00:12:42.487
It does feel like when we move
from faxes and interoffice memos

00:12:42.570 --> 00:12:47.287
to emails and spreadsheets and documents
floating around the enterprise.

00:12:47.370 --> 00:12:52.327
It feels like that time
again where the way we work,

00:12:52.410 --> 00:12:56.827
the artifacts we create,
the workflows that we are involved in,

00:12:56.910 --> 00:12:59.827
are fundamentally going
through a sea change.

00:12:59.910 --> 00:13:04.147
And to kind of give you a little bit of a
flavor for this Copilot ecosystem at work,

00:13:04.230 --> 00:13:05.737
I wanted to introduce my colleague, Nirja.

00:13:05.820 --> 00:13:09.437
Nirja, take it over.
NIRJA TRIVEDI: Thanks, Satya.

00:13:09.520 --> 00:13:13.227
[Applause] Hello, everyone.

00:13:13.310 --> 00:13:18.247
I am so excited to be here with you
all in Sydney, and I'm even more excited

00:13:18.330 --> 00:13:21.697
to show you the latest
Microsoft 365 capabilities

00:13:21.780 --> 00:13:24.877
that Satya was just talking
about, including Cowork,

00:13:24.960 --> 00:13:28.087
multimodality,
and using agentic experiences

00:13:28.170 --> 00:13:30.577
in the apps you use every single day.

00:13:30.660 --> 00:13:35.497
We have a lot to show you. So I'm going
to get started, starting with Cowork.

00:13:35.580 --> 00:13:39.577
So Cowork lets you build out assets

00:13:39.660 --> 00:13:43.867
in parallel using Work IQ data.

00:13:43.950 --> 00:13:46.897
Work IQ is more than just work content.

00:13:46.980 --> 00:13:50.407
It is the intelligence layer
that powers Copilot and the apps

00:13:50.490 --> 00:13:52.717
that you use every single day.

00:13:52.800 --> 00:13:55.447
It makes sure that it
understands your reasoning

00:13:55.530 --> 00:13:57.637
and understands your decisions,
the workflows,

00:13:57.720 --> 00:14:02.317
the people that you talk to, so your work
is tailored and personalized to you.

00:14:02.400 --> 00:14:05.107
With Cowork, you also
have access to a breadth

00:14:05.190 --> 00:14:08.347
of different prompts,
like organize my inbox,

00:14:08.430 --> 00:14:12.457
organize my week, prep for a meeting,
and research a company.

00:14:12.540 --> 00:14:15.007
So really a range of different prompts.

00:14:15.090 --> 00:14:18.037
If I scroll down even more,
you see that I have access

00:14:18.120 --> 00:14:22.237
to all the lists of tasks that are running
as well, the ones that have been completed

00:14:22.320 --> 00:14:27.187
and that I can reference the assets being
made, like this Excel doc, as well as ones

00:14:27.270 --> 00:14:29.767
that need user input for me to continue.

00:14:29.850 --> 00:14:32.167
And then finally,
the ones that are in progress

00:14:32.250 --> 00:14:34.897
that I can slam my laptop
shut and then come back,

00:14:34.980 --> 00:14:37.837
and Cowork will still keep working.

00:14:37.920 --> 00:14:40.867
So for this demo, I'm going
to use Cowork to help me build

00:14:40.950 --> 00:14:44.467
out the version 2 of my product launch.

00:14:44.550 --> 00:14:47.947
To do that, Cowork is going
to look through some of my assets.

00:14:48.030 --> 00:14:50.587
It's going to analyze a recap
email that was sent earlier,

00:14:50.670 --> 00:14:54.367
and then build out four
different assets in parallel.

00:14:54.450 --> 00:14:58.657
The first one that it's going to build out
is a launch budget using a previous launch

00:14:58.740 --> 00:15:03.187
that I already had. It's going to take
some assumptions from that launch as well.

00:15:03.270 --> 00:15:05.017
The second thing it's going
to do is that it's going

00:15:05.100 --> 00:15:07.447
to create a new product pitch deck for me,

00:15:07.530 --> 00:15:10.387
using a technical
specification doc that I have.

00:15:10.470 --> 00:15:15.187
And then I also want it to include
that second version launch budget as well.

00:15:15.270 --> 00:15:18.607
Lastly, I wanted to mobilize
my team by sending a chat message

00:15:18.690 --> 00:15:22.657
to my v-Team with highlights
of the news and schedule a meeting

00:15:22.740 --> 00:15:25.327
with two of my stakeholders
this afternoon.

00:15:25.410 --> 00:15:29.617
Okay, that was quite a bit
of different things I want Cowork to do.

00:15:29.700 --> 00:15:32.977
Awesome. It hasn't even started yet.

00:15:33.060 --> 00:15:37.267
But I wanted to do four different things,
requiring many different skillsets.

00:15:37.350 --> 00:15:40.597
To get us all on the same page,
it's going to build me a launch budget.

00:15:40.680 --> 00:15:43.357
It's going to create a product
pitch deck, send a message

00:15:43.440 --> 00:15:45.547
to my team, and schedule a meeting.

00:15:45.630 --> 00:15:52.147
All right, let's get started. As Cowork
gets to work, you'll see on the right side

00:15:52.230 --> 00:15:55.057
that I can track its
progress as it continues,

00:15:55.140 --> 00:15:57.787
as well as all the tasks
that it has as well.

00:15:57.870 --> 00:16:01.777
So the ones that I went over,
like the email recap as well

00:16:01.860 --> 00:16:05.497
as the budget,
the pitch deck, all of those.

00:16:05.580 --> 00:16:08.407
And then on the chat pane,
you can also see its chain

00:16:08.490 --> 00:16:11.077
of thought as it goes
through each of the tasks.

00:16:11.160 --> 00:16:13.867
And then finally,
it tells me exactly the summary

00:16:13.950 --> 00:16:17.947
of all four tasks, again,
that are running in parallel.

00:16:18.030 --> 00:16:19.477
Now, this is sped up for the sake of time,

00:16:19.560 --> 00:16:23.197
but you could see here that it
also includes the first action

00:16:23.280 --> 00:16:28.267
item, and this is the message, the Teams
message that will be sent to my team.

00:16:28.350 --> 00:16:32.617
Again, I always have access
to review before anything is sent.

00:16:32.700 --> 00:16:38.137
Be it a Teams message, an invite,
an email -- I can always review.

00:16:38.220 --> 00:16:40.357
All right, this looks good to me.

00:16:40.440 --> 00:16:44.887
I can also interrupt Cowork.
So let's say, like we always do,

00:16:44.970 --> 00:16:47.587
we think of a last-minute
task as we're working.

00:16:47.670 --> 00:16:51.967
I can send that to Cowork. For example, I
want to send an email with an exec summary

00:16:52.050 --> 00:16:53.827
with the budget recommendations.

00:16:53.910 --> 00:16:55.687
I realize that I've sent many messages,

00:16:55.770 --> 00:16:58.927
but I haven't really
included that net new asset.

00:16:59.010 --> 00:17:01.207
Cowork will work on sending that email,

00:17:01.290 --> 00:17:03.727
but it won't stop
the previous list of tasks.

00:17:03.810 --> 00:17:06.457
Everything will continue moving forward.

00:17:06.540 --> 00:17:10.297
And then you can see here that
that next step is already ready.

00:17:10.380 --> 00:17:13.087
It's found some time for me
to meet with my stakeholders,

00:17:13.170 --> 00:17:16.297
and this looks great to me,
so I'm going to hit "Create" there,

00:17:16.380 --> 00:17:19.687
and I just have a few
more assets to continue.

00:17:19.770 --> 00:17:23.167
On the right side, you can see
that my progress bar is continuing.

00:17:23.250 --> 00:17:29.947
I have four out of six items done, as well
as different output folders as well.

00:17:30.030 --> 00:17:32.797
Finally, the last task is ready.

00:17:32.880 --> 00:17:37.987
It has that email, again with
that launch budget attached to,

00:17:38.070 --> 00:17:39.667
and that looks good for me to send.

00:17:39.750 --> 00:17:45.247
That's the last thing. And then just
a few moments, the task is complete.

00:17:45.330 --> 00:17:49.147
So you'll see here on the right
side, I can verify all the tasks

00:17:49.230 --> 00:17:52.507
that are complete,
all the folders that have the output

00:17:52.590 --> 00:17:57.847
of all the tasks that I have,
finally the input folders as well.

00:17:57.930 --> 00:18:01.957
Let's dive in to the first
asset and see how it looks.

00:18:02.040 --> 00:18:08.227
All right, not bad. This is the pitch
deck that was ingesting that tech-spec doc

00:18:08.310 --> 00:18:12.247
as well as the launch budget. So you
could see here, there's an exec summary.

00:18:12.330 --> 00:18:14.577
It's really beautifully designed.

00:18:14.660 --> 00:18:21.237
It also includes my brand colors as well
as the version 2 opportunity as well.

00:18:21.320 --> 00:18:24.267
And then finally, it includes
a bit of that launch budget as well.

00:18:24.350 --> 00:18:27.817
So this is a net new creation by Cowork.

00:18:27.900 --> 00:18:31.177
All right, that looks good.
Now that I have the first asset,

00:18:31.260 --> 00:18:35.347
let me run over to my Teams
meeting to see if any

00:18:35.430 --> 00:18:39.577
of my stakeholders have
responded to my chat.

00:18:39.660 --> 00:18:41.407
And I see a couple of them
have started to respond.

00:18:41.490 --> 00:18:45.367
So here's that chat message.
I see a few of them are responding,

00:18:45.450 --> 00:18:48.637
and let's check this one out. So one
of my team members said this is great,

00:18:48.720 --> 00:18:53.017
but I want to know more about
the improvements of v1 versus v2.

00:18:53.100 --> 00:18:57.277
Makes sense. Good question.
So I could either go to Copilot,

00:18:57.360 --> 00:18:59.827
compare the two technical
specification docs,

00:18:59.910 --> 00:19:03.427
copy and paste it back into
Teams, or I can use Teams

00:19:03.510 --> 00:19:06.307
for what it's used for, a canvas to work

00:19:06.390 --> 00:19:08.947
with both my collaborators
as well as my agents.

00:19:09.030 --> 00:19:14.287
So I'm going to do just that. I'm going
to go up, and I'm going to add agents.

00:19:14.370 --> 00:19:19.657
And so I want to add Copilot
to this response, and I'm going

00:19:19.740 --> 00:19:22.327
to add Copilot to this chat.

00:19:22.410 --> 00:19:24.817
I can at mention Copilot
as well as any of the agents

00:19:24.900 --> 00:19:31.687
that you saw up there, and I can ground it
on the specification doc and click "Send."

00:19:31.770 --> 00:19:36.607
Within a few moments, Copilot will
pull in the right responses a lot richer

00:19:36.690 --> 00:19:38.767
than if I quickly gave the answer myself,

00:19:38.850 --> 00:19:42.187
or if I had to go back
and copy and paste it from chat.

00:19:42.270 --> 00:19:46.687
So you can see here, it's a great
response, and this looks good to me.

00:19:46.770 --> 00:19:51.727
I did notice there was one last
stakeholder that had some news,

00:19:51.810 --> 00:19:53.737
and this is actually
really good news for me.

00:19:53.820 --> 00:19:56.677
She says there's about 100k
underspend this quarter,

00:19:56.760 --> 00:19:59.197
and she wants to put that
towards the marketing budget.

00:19:59.280 --> 00:20:02.737
You don't see that a lot,
100 extra k, so this is awesome.

00:20:02.820 --> 00:20:05.517
So I'm going to go back to Cowork,
and I'm going to figure

00:20:05.600 --> 00:20:08.187
out what am I going to do
with this extra budget?

00:20:08.270 --> 00:20:12.777
So here I am back at Cowork with that
first task that Cowork completed for me.

00:20:12.860 --> 00:20:17.937
That was the Excel doc. So I'm going
to open that up, that Excel sheet,

00:20:18.020 --> 00:20:21.687
and right when I open it up,
I want to call it a couple things.

00:20:21.770 --> 00:20:24.387
The first thing is right
when I open up the sheet,

00:20:24.470 --> 00:20:26.937
I see that it's automatically
saved into my OneDrive,

00:20:27.020 --> 00:20:29.907
so I don't need to go and find
it in the task if I need to.

00:20:29.990 --> 00:20:31.857
The second thing that I
want to call out is

00:20:31.940 --> 00:20:35.947
that it's also applying
the security label as well.

00:20:36.030 --> 00:20:39.517
So here I can see that it
has the right security label

00:20:39.600 --> 00:20:41.587
that is using the underlying assets.

00:20:41.670 --> 00:20:44.887
So Satya mentioned trust
and security are part of everything

00:20:44.970 --> 00:20:46.747
that we do, and you see that here as well,

00:20:46.830 --> 00:20:51.127
using that underlying
confidentiality label.

00:20:51.210 --> 00:20:53.437
And now I'm going to go
and activate Copilot

00:20:53.520 --> 00:20:56.047
so we could use it in the
agentic apps that we use.

00:20:56.130 --> 00:20:59.347
So here I have access
to multimodal functionality.

00:20:59.430 --> 00:21:04.447
I could choose whichever model I want for
the job, including our latest GPT models,

00:21:04.530 --> 00:21:07.357
as well as our latest Claude models.

00:21:07.440 --> 00:21:09.157
And then I'm going to have
Copilot help me figure

00:21:09.240 --> 00:21:12.337
out what am I going to do
with that extra 100k?

00:21:12.420 --> 00:21:15.697
So as I click "Send,"
Copilot is going to help me build

00:21:15.780 --> 00:21:18.517
out a dashboard with clear
charts and visuals,

00:21:18.600 --> 00:21:21.847
and it's also going to show me
clear recommendations about how

00:21:21.930 --> 00:21:26.167
to distribute across
the marketing tactics as well.

00:21:26.250 --> 00:21:29.287
And as it starts building out,
I can see on the right side

00:21:29.370 --> 00:21:31.957
that I can build the dashboard,
creating new sheets

00:21:32.040 --> 00:21:33.817
with recommendations and charts.

00:21:33.900 --> 00:21:36.577
And I can also see its
chain-of-thought reasoning

00:21:36.660 --> 00:21:39.817
as it goes through this analysis.

00:21:39.900 --> 00:21:42.667
And then finally,
I get this beautiful image

00:21:42.750 --> 00:21:49.207
with pivot tables, Excel native charts, as
well as a full in-depth analysis here too.

00:21:49.290 --> 00:21:56.137
So if I scroll down, I can see here
a rich analysis, including key numbers.

00:21:56.220 --> 00:21:59.497
It tells me exactly what
each dashboard contains.

00:21:59.580 --> 00:22:03.967
And then finally, it tells me
what the core strategy is as well.

00:22:04.050 --> 00:22:08.167
And if I go back to the grid,
I can click on any single cell,

00:22:08.250 --> 00:22:10.597
and I have the ability to edit a formula

00:22:10.680 --> 00:22:15.007
or change any single formula
as well without breaking any

00:22:15.090 --> 00:22:18.067
of the grid because it's Excel native.

00:22:18.150 --> 00:22:22.927
So we went through quite a few different
things today, four different things.

00:22:23.010 --> 00:22:27.727
We were able to build a completely new
pitch deck with using Excel imagery.

00:22:27.810 --> 00:22:30.697
We were also able to build
a completely new Excel sheet

00:22:30.780 --> 00:22:33.937
and then finally mobilize
my team as well through a series

00:22:34.020 --> 00:22:36.457
of Teams messages as well as Outlook.

00:22:36.540 --> 00:22:38.437
Now, I hope this gets you excited

00:22:38.520 --> 00:22:41.347
because you could see this is
just one way that Copilot fits

00:22:41.430 --> 00:22:46.087
into your workflow, but I hope you can see
how it fits into your workflow as well.

00:22:46.170 --> 00:22:48.967
Now, I hope this gets you
excited to use Cowork right

00:22:49.050 --> 00:22:53.377
in the Copilot app and Copilot
in the apps that you use every single day.

00:22:53.460 --> 00:22:54.437
Thank you.

00:22:54.520 --> 00:23:01.337
[Applause]

00:23:01.420 --> 00:23:02.447
SATYA NADELLA: Thank you so much, Nirja.

00:23:02.530 --> 00:23:06.817
It's so wonderful to see
even the evolution of Copilot

00:23:06.900 --> 00:23:08.977
in just the last few years, right?

00:23:09.060 --> 00:23:13.507
I mean, in the first instance,
it was a bit of a UI on a model,

00:23:13.590 --> 00:23:17.347
but now you can think
of even the Copilot system

00:23:17.430 --> 00:23:24.430
as essentially a cognitive loop
that or a harness that you have

00:23:24.810 --> 00:23:29.677
at your disposal for you
to be able to have more agency,

00:23:29.760 --> 00:23:33.127
more ambition on what you want
to create, how you want to,

00:23:33.210 --> 00:23:37.897
in some sense, initiate
the trajectory of a task.

00:23:37.980 --> 00:23:42.637
And that, I think, is what you
see in all of those features,

00:23:42.720 --> 00:23:45.607
and then it's also becoming
quickly multiplayer, right?

00:23:45.690 --> 00:23:50.077
So you're not just single-player
mode, but when you're in Teams,

00:23:50.160 --> 00:23:53.587
you're interacting with not just Copilot,
but all the other agents as well.

00:23:53.670 --> 00:23:56.017
And that, I think, is what is going

00:23:56.100 --> 00:24:00.907
to probably unfold very rapidly
throughout this calendar year.

00:24:00.990 --> 00:24:03.967
Now, we are seeing fantastic momentum.

00:24:04.050 --> 00:24:07.087
The great news is this is not
about sort of this rate

00:24:07.170 --> 00:24:09.787
of diffusion here happening slowly.

00:24:09.870 --> 00:24:15.127
It's in some sense happening all very
fast, everywhere, at the same time.

00:24:15.210 --> 00:24:18.757
In Australia, we've now
got some fantastic momentum

00:24:18.840 --> 00:24:23.107
around Copilot usage across
all of the top companies

00:24:23.190 --> 00:24:24.637
that are represented on this slide.

00:24:24.720 --> 00:24:30.877
I think 17 of the 20 top companies in the
exchange here are using Copilot deeply.

00:24:30.960 --> 00:24:32.647
All the banks are using it.

00:24:32.730 --> 00:24:36.937
Multiple. I think there is 57
government agencies using it.

00:24:37.020 --> 00:24:38.887
So it's fantastic to see the momentum,

00:24:38.970 --> 00:24:43.237
and I had a chance yesterday
to meet with this team

00:24:43.320 --> 00:24:48.337
at Australia Post that is
sort of looking at Copilot

00:24:48.420 --> 00:24:51.967
and the adoption of Copilot,
and it's a complete bottoms-up way.

00:24:52.050 --> 00:24:54.937
They have all these Copilot
champions and team captains

00:24:55.020 --> 00:25:00.007
who are really making sure
that everyone is trained on it

00:25:00.090 --> 00:25:01.897
and is getting the best out of it.

00:25:01.980 --> 00:25:07.357
But the thing that stood out
for me is how much care they're sort

00:25:07.440 --> 00:25:11.317
of taking in making accessibility one

00:25:11.400 --> 00:25:15.217
of the big beneficiaries of Copilot.

00:25:15.300 --> 00:25:17.617
In other words, they've used Copilot

00:25:17.700 --> 00:25:22.327
and its native technology,
but it's really driving everyone

00:25:22.410 --> 00:25:25.987
with all the abilities to be
able to fully participate

00:25:26.070 --> 00:25:28.567
and give them more agency
inside of the enterprise,

00:25:28.650 --> 00:25:31.807
inside of a Teams
conversation, in a meeting,

00:25:31.890 --> 00:25:35.317
inside of being able
to sort of use Copilot

00:25:35.400 --> 00:25:39.127
to really get things done
in a much more seamless way,

00:25:39.210 --> 00:25:41.197
and it was fantastic to see that.

00:25:41.280 --> 00:25:43.957
So that was the first part of the stack.

00:25:44.040 --> 00:25:49.570
So we go to the next stage,
which is the AI platform

00:25:50.760 --> 00:25:57.577
or the agent platform. Now,
the best way to think about this is

00:25:57.660 --> 00:25:59.557
in a couple of different layers.

00:25:59.640 --> 00:26:02.767
There is models. There's context.

00:26:02.850 --> 00:26:07.567
There is agents. There's harness, and
then there's security around it, right?

00:26:07.650 --> 00:26:10.507
So you want to have
access to all the models.

00:26:10.590 --> 00:26:14.677
You want to be able to have
access to all of the rich context.

00:26:14.760 --> 00:26:18.277
You then want the agents to be
built on top of that, right,

00:26:18.360 --> 00:26:22.657
using any model with all the context, and
then you want the harness that knows how

00:26:22.740 --> 00:26:24.337
to orchestrate all of it, right?

00:26:24.420 --> 00:26:26.527
That's sort of the core of it,
and then, of course,

00:26:26.610 --> 00:26:29.407
you need the security
and compliance around it.

00:26:29.490 --> 00:26:33.697
So I want to unpack a little
bit of this platform.

00:26:33.780 --> 00:26:39.757
And for us, the organizing
sort of piece here is Foundry,

00:26:39.840 --> 00:26:44.497
and Foundry brings all
of these things together,

00:26:44.580 --> 00:26:47.707
whether it's the agent
and the multiagent run times,

00:26:47.790 --> 00:26:50.527
whether it's models,
whether it's the context layer.

00:26:50.610 --> 00:26:54.937
And so I kind of want to walk through
each one of these pieces in some detail.

00:26:55.020 --> 00:26:57.877
The first is the models.

00:26:57.960 --> 00:27:01.987
Of course, you have the frontier
models, but the reality is,

00:27:02.070 --> 00:27:09.070
we have 11,000 models in Foundry
today, and this is sort

00:27:09.210 --> 00:27:14.707
of the base material for you to be able to
build your own agent in agentic systems.

00:27:14.790 --> 00:27:17.587
In fact, all the frontier
models, as I said, are there,

00:27:17.670 --> 00:27:19.717
and they're rapidly evolving.

00:27:19.800 --> 00:27:22.207
We ourselves, with MAI have
launched, even in the last,

00:27:22.290 --> 00:27:26.107
I think, month and a half,
a fantastic transcription model,

00:27:26.190 --> 00:27:28.177
an image model in two varieties.

00:27:28.260 --> 00:27:33.067
One is a very high-performance
and low-cost image model,

00:27:33.150 --> 00:27:39.487
a new voice model, again.
And so it's fantastic to see the, really,

00:27:39.570 --> 00:27:44.707
not just the big frontier models, but even
the helper models we use all the time

00:27:44.790 --> 00:27:49.807
for different modalities inside of our
applications are all rapidly evolving.

00:27:49.890 --> 00:27:54.157
But I wanted to talk about
this one model that struck a chord

00:27:54.240 --> 00:28:00.637
for me because what is the
innovation agenda around models,

00:28:00.720 --> 00:28:04.867
and how does one go about it
and build this as a capability

00:28:04.950 --> 00:28:08.497
that truly empowers
the next level of innovation?

00:28:08.580 --> 00:28:13.537
So to showcase that, let me
just roll the video on GigaTIME.

00:28:13.620 --> 00:28:16.597
HOIFUNG POON: Immunotherapy
is really the most promising

00:28:16.680 --> 00:28:19.657
direction for taming cancer,
once and for all.

00:28:19.740 --> 00:28:23.647
The whole idea is using
the immune system to fight cancer.

00:28:23.730 --> 00:28:26.317
CARLO BIFULCO: The big challenge we have
in the field right now is to understand

00:28:26.400 --> 00:28:30.697
which patients will respond. HOIFUNG POON:
Spatial proteomics can simultaneously

00:28:30.780 --> 00:28:36.097
measure multiple protein and create
multiplex immunofluorescent images.

00:28:36.180 --> 00:28:38.827
They can tell you whether
the patient might respond

00:28:38.910 --> 00:28:40.987
to immunotherapy or not,
but unfortunately,

00:28:41.070 --> 00:28:46.990
to generate this kind of images,
takes days and costs thousands of dollars.

00:28:49.500 --> 00:28:52.567
In collaboration with Providence
and University of Washington,

00:28:52.650 --> 00:28:55.717
we developed this multimodal
AI called GigaTIME

00:28:55.800 --> 00:28:59.047
to generate spatial proteomics at scale.

00:28:59.130 --> 00:29:00.517
SHENG WANG: So we have
trained the model using

00:29:00.600 --> 00:29:04.147
large-scale data,
analyzing millions of cells.

00:29:04.230 --> 00:29:06.607
The input image has limited data.

00:29:06.690 --> 00:29:09.547
We don't know where the protein is active.

00:29:09.630 --> 00:29:11.887
GigaTIME can convert this image

00:29:11.970 --> 00:29:15.757
to a colorful multiplex
immunofluorescent image.

00:29:15.840 --> 00:29:18.337
The color is the protein activation.

00:29:18.420 --> 00:29:22.237
They are the key factor drives
diseases, especially in cancer.

00:29:22.320 --> 00:29:25.147
This can help us to predict
patient drug response.

00:29:25.230 --> 00:29:27.427
HOIFUNG POON: We are now
taking the first meaningful step

00:29:27.510 --> 00:29:31.540
to be able to generate this kind
of data for everyone in the globe.

00:29:32.850 --> 00:29:37.297
The GigaTIME model is now available
to everyone through Microsoft Foundry.

00:29:37.380 --> 00:29:43.387
Our hope is to empower every secondary
hospital in every tertiary city to be able

00:29:43.470 --> 00:29:46.537
to now conduct their own research.

00:29:46.620 --> 00:29:48.427
CARLO BIFULCO: We look forward to a future

00:29:48.510 --> 00:29:50.857
where we can actually
bring this to the clinic,

00:29:50.940 --> 00:29:54.397
impact patients' lives, and bring the next
generation of immunotherapies to places

00:29:54.480 --> 00:29:57.070
where this would be otherwise impossible.

00:29:58.520 --> 00:30:03.907
[Applause]

00:30:03.990 --> 00:30:08.317
SATYA NADELLA: What was said
in the video sort of makes a ton

00:30:08.400 --> 00:30:14.887
of sense, that you can
take a technology like this,

00:30:14.970 --> 00:30:18.037
take what is a complex test
that costs a lot of money,

00:30:18.120 --> 00:30:20.737
takes a lot of time,
and truly democratize it, right,

00:30:20.820 --> 00:30:24.397
that point about every tertiary hospital

00:30:24.480 --> 00:30:28.117
in every secondary city
now with just a pathology,

00:30:28.200 --> 00:30:32.677
a regular slide pathology can
get this test done, essentially,

00:30:32.760 --> 00:30:38.917
is a pretty big breakthrough.
But the question is how that came about.

00:30:39.000 --> 00:30:41.587
The way it came about
is two research institutions

00:30:41.670 --> 00:30:45.037
that have had this knowledge,
right, this ability.

00:30:45.120 --> 00:30:48.577
They've obviously had all this data,
and they were able to take

00:30:48.660 --> 00:30:53.497
that data, train a model,
a generative model,

00:30:53.580 --> 00:30:58.657
to essentially do simulation
of a test using compute

00:30:58.740 --> 00:31:00.967
and share it with the world, right?

00:31:01.050 --> 00:31:03.397
That, to me, is what's happening, right?

00:31:03.480 --> 00:31:06.127
This is not about one frontier
model company doing things.

00:31:06.210 --> 00:31:09.007
It's about all of us as an ecosystem

00:31:09.090 --> 00:31:14.917
at the frontier have the ability
to use these new techniques,

00:31:15.000 --> 00:31:17.287
like generative AI, these new algorithms,

00:31:17.370 --> 00:31:20.917
this new compute
infrastructure that's there,

00:31:21.000 --> 00:31:25.387
and turn what is essentially
the tacit knowledge we have,

00:31:25.470 --> 00:31:29.167
the information we have,
the data edge we have,

00:31:29.250 --> 00:31:31.777
and turn that not only
to move the frontier inside the

00:31:31.860 --> 00:31:37.087
organization, but you can move
the frontier for the society at large.

00:31:37.170 --> 00:31:39.727
And that, I think,
is what's exciting about sort

00:31:39.810 --> 00:31:44.287
of the true democratic force
that's underlying all this.

00:31:44.370 --> 00:31:49.267
The next thing that I want to talk
about beyond the models is context.

00:31:49.350 --> 00:31:53.737
In some sense context around
the models is perhaps one

00:31:53.820 --> 00:31:58.207
of the core assets that's,
quite frankly, distributed, right?

00:31:58.290 --> 00:32:03.127
Every organization has unique
context, and you want to be able

00:32:03.210 --> 00:32:07.207
to organize that context such
that you can put it together

00:32:07.290 --> 00:32:11.377
with these models to both
reason and take action.

00:32:11.460 --> 00:32:15.277
And that's sort of one of the
top engineering considerations

00:32:15.360 --> 00:32:18.697
and architectural considerations,
Work IQ, right?

00:32:18.780 --> 00:32:25.780
So Work IQ is that rich database
underneath Microsoft 365, right?

00:32:27.000 --> 00:32:29.647
Everything Nirja just showed
around Copilot was tapping

00:32:29.730 --> 00:32:34.627
into Work IQ, and that Work
IQ is available headless

00:32:34.710 --> 00:32:39.157
through an MCP interface
for any app developer to be able

00:32:39.240 --> 00:32:43.177
to build an agentic system around
it, and this -- think about this.

00:32:43.260 --> 00:32:47.167
Up to now, I've always said that one
of the most important databases inside

00:32:47.250 --> 00:32:49.477
of any organization is the
database we don't talk about,

00:32:49.560 --> 00:32:53.977
right, which is the database
that contains all of the people

00:32:54.060 --> 00:32:57.487
and their relationship to other
people inside the organizations,

00:32:57.570 --> 00:33:02.797
the communications graph and data,
the document libraries,

00:33:02.880 --> 00:33:07.747
project files -- all that is in Work IQ.

00:33:07.830 --> 00:33:11.467
Similarly, another very
important database when it comes

00:33:11.550 --> 00:33:16.027
to thinking about outcomes
is what's in Fabric, right?

00:33:16.110 --> 00:33:19.327
If you think about all
the various dashboards

00:33:19.410 --> 00:33:23.587
that have gotten created over
the decade, and if you think

00:33:23.670 --> 00:33:28.447
of any dashboard is some outcome,
some metric, someone's measuring.

00:33:28.530 --> 00:33:32.677
So there's a rich semantic
model that was created.

00:33:32.760 --> 00:33:37.267
That's, again, super
important context to be able

00:33:37.350 --> 00:33:44.350
to give these LLMs and models
as you try to drive your workflows.

00:33:45.360 --> 00:33:49.087
And by the way, Fabric now
has the ability to reach

00:33:49.170 --> 00:33:53.377
out to any data, not just in Azure,
but everywhere, right?

00:33:53.460 --> 00:33:54.937
You can connect to AWS.

00:33:55.020 --> 00:33:58.897
You can connect to GCP, bring
all of the enterprise data asset

00:33:58.980 --> 00:34:02.917
as context inside of this Fabric IQ.

00:34:03.000 --> 00:34:06.307
And then, in fact, the outer loop
of this can all be Foundry IQ.

00:34:06.390 --> 00:34:11.857
What Foundry IQ does is essentially even
not only gets you that document database.

00:34:11.940 --> 00:34:13.957
So if you have a lot
of unstructured information,

00:34:14.040 --> 00:34:19.237
you can then vectorize it,
make it a part of your context.

00:34:19.320 --> 00:34:24.387
But you can also use it as an
outer planner to be able to plan

00:34:24.470 --> 00:34:28.077
across Work IQ, Fabric,
and Foundry, right?

00:34:28.160 --> 00:34:31.477
That's sort of how these
things come together.

00:34:31.560 --> 00:34:37.087
So context and this IQ layer
beyond the models becomes the

00:34:37.170 --> 00:34:39.757
other second important piece.

00:34:39.840 --> 00:34:43.027
Now the third thing I want
to talk about when it comes

00:34:43.110 --> 00:34:48.337
to agent platform is the agent
service itself, right?

00:34:48.420 --> 00:34:54.957
With every shift in platforms,
you kind of need a runtime,

00:34:55.040 --> 00:34:56.287
and that's what this is.

00:34:56.370 --> 00:35:01.987
The new runtime for AI agents
is this Foundry agent service,

00:35:02.070 --> 00:35:05.827
and really exciting announcements
today up and down the stack, right?

00:35:05.910 --> 00:35:08.647
We have now persistent memory.

00:35:08.730 --> 00:35:15.577
We have tools use. So basically models
that then can be attached to knowledge

00:35:15.660 --> 00:35:20.137
that can then also have attached to tools.

00:35:20.220 --> 00:35:23.707
You can then, on top of it,
build these multiagent

00:35:23.790 --> 00:35:27.357
and agentic systems that you
can orchestrate using any

00:35:27.440 --> 00:35:32.247
orchestrator you choose.
You can publish them with one click.

00:35:32.330 --> 00:35:35.337
And then, as I said,
you have memory that's persistent

00:35:35.420 --> 00:35:38.157
across all of this,
and they're long-living agents,

00:35:38.240 --> 00:35:43.107
and they're stateful agents.
And so the one thing that I'm most excited

00:35:43.190 --> 00:35:48.777
about is the hosted agent service
that now goes into preview mode as well.

00:35:48.860 --> 00:35:51.327
So it's really exciting to see
this because now what do you have?

00:35:51.410 --> 00:35:55.887
Okay, all right. [Cheers]

00:35:55.970 --> 00:36:00.927
In fact, this is like that day
we started the cloud journey the

00:36:01.010 --> 00:36:03.747
first time we talked about
containers and said, oh,

00:36:03.830 --> 00:36:06.327
there's a new type of virtual
machine, effectively,

00:36:06.410 --> 00:36:07.977
for cloud-native applications.

00:36:08.060 --> 00:36:13.717
So think of this as the birth
of this new generation

00:36:13.800 --> 00:36:18.487
of AI-native applications
with a hosted service.

00:36:18.570 --> 00:36:22.117
So literally, you have
a sandbox per session

00:36:22.200 --> 00:36:24.997
that can bootstrap itself fast, right?

00:36:25.080 --> 00:36:28.087
So it's instant on. So in fact,
when you look at the service,

00:36:28.170 --> 00:36:30.637
you will instantly be
able to bring up a sandbox.

00:36:30.720 --> 00:36:33.217
The sandboxes are isolated.

00:36:33.300 --> 00:36:38.377
You have persistent memory that
is session state is persistent.

00:36:38.460 --> 00:36:41.887
And then, of course, in like
any other service in the cloud,

00:36:41.970 --> 00:36:44.047
you can scale up and down
elastically, right?

00:36:44.130 --> 00:36:47.677
So this becomes, I think,
a massive boost because one

00:36:47.760 --> 00:36:50.317
of the things I've been
talking to a lot of developers,

00:36:50.400 --> 00:36:55.657
and they come and say, look, when can
we have a hosted service for agents?

00:36:55.740 --> 00:36:57.937
Because agents are new software, right?

00:36:58.020 --> 00:36:59.557
People talk about what's
happening to software.

00:36:59.640 --> 00:37:05.137
Guess what? All software is going agentic,
and you need to host these agents.

00:37:05.220 --> 00:37:08.707
You need to have them have
their own computer, essentially,

00:37:08.790 --> 00:37:13.887
and that's what this hosted
container sandbox is.

00:37:13.970 --> 00:37:17.337
And then it needs to have state, so that's
kind of where the persistence comes.

00:37:17.420 --> 00:37:19.827
You need to be able
to scale them up and down.

00:37:19.910 --> 00:37:23.667
And so we are very, very excited
about this new Foundry service

00:37:23.750 --> 00:37:25.857
with hosted agents and what you all can do

00:37:25.940 --> 00:37:29.047
to scale the agentic
systems going forward.

00:37:29.130 --> 00:37:31.717
Now then you have the runtimes,
you have the models,

00:37:31.800 --> 00:37:36.937
you have the context -- you want
to start building these applications,

00:37:37.020 --> 00:37:41.257
and that's where we have the richness
of multiple different tools.

00:37:41.340 --> 00:37:44.017
You can start right inside
of Copilot with something

00:37:44.100 --> 00:37:49.207
like App Builder, hit a prompt, and here's
a career path app that you just built

00:37:49.290 --> 00:37:52.027
with a single prompt
inside of a chat session.

00:37:52.110 --> 00:37:54.937
It's a full stack app that gets generated.

00:37:55.020 --> 00:37:59.347
You can share it, or you can
go to Copilot Studio, again,

00:37:59.430 --> 00:38:02.707
build an agent using
a simple set of prompts.

00:38:02.790 --> 00:38:05.497
In this case,
it's an onboarding app that got built

00:38:05.580 --> 00:38:10.237
as our onboarding agent that got
built using Copilot studio.

00:38:10.320 --> 00:38:13.957
And then, of course,
you have the full richness

00:38:14.040 --> 00:38:17.947
of the professional developer
tools, right, with VS Code

00:38:18.030 --> 00:38:22.177
and GitHub and now all the agentic tools.

00:38:22.260 --> 00:38:25.027
In fact, one of the most
exciting things is right here

00:38:25.110 --> 00:38:28.807
in Australia, we now have, in GitHub,

00:38:28.890 --> 00:38:34.747
2.3 million strong community. [Applause]

00:38:34.830 --> 00:38:36.217
Yeah, it's and it's growing, right?

00:38:36.300 --> 00:38:38.377
It's 26% growth year over year.

00:38:38.460 --> 00:38:43.857
It's great to see the community
continuing to scale.

00:38:43.940 --> 00:38:48.457
And GitHub is not just
about the code repo,

00:38:48.540 --> 00:38:54.787
but it's also now becoming the place,
the control plane for all the agents.

00:38:54.870 --> 00:38:59.977
So we described it as Agent HQ,
so you have every coding agent

00:39:00.060 --> 00:39:05.227
that that you use will be available
to you right next to your repo, right?

00:39:05.310 --> 00:39:08.677
So you have your code.
You have your coding agents.

00:39:08.760 --> 00:39:11.767
And in fact, one of my
favorite new features is,

00:39:11.850 --> 00:39:16.527
not only do I have my source code,
but I have my session traces

00:39:16.610 --> 00:39:19.767
as a first-class thing
inside of GitHub, right?

00:39:19.850 --> 00:39:21.627
So because, after all, going forward,

00:39:21.710 --> 00:39:26.157
it becomes very important
for us to not only have the code,

00:39:26.240 --> 00:39:29.347
but also the traces
that generated the code

00:39:29.430 --> 00:39:31.747
because of the coding
agents right with the code.

00:39:31.830 --> 00:39:36.607
And I think this become, in fact,
for true provenance of code going forward,

00:39:36.690 --> 00:39:39.667
that becomes a pretty important attribute.

00:39:39.750 --> 00:39:45.937
And then we're also exposing the tooling
through all the form factors, right?

00:39:46.020 --> 00:39:49.777
As I said, CLI is back.
Everybody loves it.

00:39:49.860 --> 00:39:51.487
And the reason why is not just

00:39:51.570 --> 00:39:54.367
because a command line is powerful,
but it's command line

00:39:54.450 --> 00:40:00.357
when coupled with rich agentic
models and natural language

00:40:00.440 --> 00:40:01.827
that have access to everything.

00:40:01.910 --> 00:40:04.237
In fact, one of the coolest things

00:40:04.320 --> 00:40:08.557
that I found myself doing is you
can literally go to your repo

00:40:08.640 --> 00:40:12.727
in GitHub CLI and then
you can connect to Work IQ

00:40:12.810 --> 00:40:17.767
through something like an MCP and say,
hey, I was in a design meeting yesterday.

00:40:17.850 --> 00:40:20.257
We had a discussion on this topic.

00:40:20.340 --> 00:40:23.617
Can you make sure that my repo
is in sync with the discussion?

00:40:23.700 --> 00:40:25.507
I mean, think about that.
How crazy is it, right?

00:40:25.590 --> 00:40:28.477
With a natural language
utterance, essentially,

00:40:28.560 --> 00:40:31.657
you're able to relate a design meeting

00:40:31.740 --> 00:40:36.937
that was mostly a transcription
-- that to source code,

00:40:37.020 --> 00:40:39.127
all being done in a command line, right?

00:40:39.210 --> 00:40:44.227
That's the type of richness
in that form factor that you can bring.

00:40:44.310 --> 00:40:47.047
And then, of course, IDEs still matter.

00:40:47.130 --> 00:40:49.387
In fact, IDEs are becoming ADEs, right?

00:40:49.470 --> 00:40:52.027
So they're now evolving to, basically,

00:40:52.110 --> 00:40:57.097
now the sessions app inside
of VS Code is basically an ADE.

00:40:57.180 --> 00:41:01.297
And sometimes when you
go into an agent mode and

00:41:01.380 --> 00:41:04.267
or the agentic output,
it says please open up in VS Code.

00:41:04.350 --> 00:41:07.177
And if you want to inspect any code,
you want that profiler there.

00:41:07.260 --> 00:41:09.397
You want the debugger there.

00:41:09.480 --> 00:41:14.377
And so we have all the form
factors that developers need

00:41:14.460 --> 00:41:19.777
and want and want to use as part
of a workflow all available to them.

00:41:19.860 --> 00:41:22.957
And so to show you all
of this developer richness,

00:41:23.040 --> 00:41:26.677
I wanted to introduce Australia's
own Damien to the stage.

00:41:26.760 --> 00:41:33.337
Damian, take it away.
[Applause] DAMIAN BRADY: Thanks, Satya.

00:41:33.420 --> 00:41:37.417
Thanks, Satya. My name is Damian,
as Satya mentioned, and I am one

00:41:37.500 --> 00:41:41.677
of the team at GitHub, and also one
of the 2.3 million developers right here

00:41:41.760 --> 00:41:45.727
in Australia, from Brisbane.
Yeah, thank you.

00:41:45.810 --> 00:41:48.547
All right.
One of the projects I've been working

00:41:48.630 --> 00:41:52.687
on lately is Tailspin Transport,
and it is a little tool.

00:41:52.770 --> 00:41:56.257
It's our EV company, and so I'm going
to be working on the website today.

00:41:56.340 --> 00:41:58.137
I'm going to show you
what that looks like.

00:41:58.220 --> 00:42:00.777
So let's have a look at the website.

00:42:00.860 --> 00:42:03.567
This is the Tailspin Transport website.

00:42:03.650 --> 00:42:05.397
We have our EV vehicles.

00:42:05.480 --> 00:42:08.637
We've got a Vehicles page as well,
very comprehensive FAQ,

00:42:08.720 --> 00:42:10.237
which I'll get to in a second.

00:42:10.320 --> 00:42:14.437
And I'm going to do some work today to add
a search feature to the Vehicles page.

00:42:14.520 --> 00:42:17.047
And the reason I'm doing that now
is because I've got an email

00:42:17.130 --> 00:42:19.927
from one of our product managers
giving me some information

00:42:20.010 --> 00:42:24.187
that I've been waiting for about
exactly how they want this to behave.

00:42:24.270 --> 00:42:27.097
So I use Visual Studio Code,
and I use Visual Studio,

00:42:27.180 --> 00:42:29.977
but lately I've been using
the GitHub Copilot CLI,

00:42:30.060 --> 00:42:32.107
so that's what I'm going to use today.

00:42:32.190 --> 00:42:35.707
So let's kick off the CLI,
and the first thing I want to do

00:42:35.790 --> 00:42:38.677
when I get in is have a look
at the model that I'm using.

00:42:38.760 --> 00:42:43.567
So I have been playing with GPT-5.4,
you can see right down the bottom.

00:42:43.650 --> 00:42:45.727
But this is a relatively
straightforward task.

00:42:45.810 --> 00:42:49.317
So I'm actually going to use
a different model, and I'm going

00:42:49.400 --> 00:42:53.937
to use Claude Sonnet 4.6. Now, you can
see just how many models I have available

00:42:54.020 --> 00:42:58.647
to me just with the one Copilot
subscription, even Opus 4.7,

00:42:58.730 --> 00:43:00.727
which came out incredibly recently.

00:43:00.810 --> 00:43:03.847
So we use 4.6,
and then I'm going to ask it

00:43:03.930 --> 00:43:05.887
to gather some information together for me

00:43:05.970 --> 00:43:08.617
so I know I'm doing
exactly the right work.

00:43:08.700 --> 00:43:10.507
Now, here's a prompt I prepared earlier.

00:43:10.590 --> 00:43:12.337
You don't want to see me type that.

00:43:12.420 --> 00:43:15.937
And what I'm doing is I'm
asking Work IQ to find that email

00:43:16.020 --> 00:43:18.907
from Evie Smith, the PM,
and pull that information

00:43:18.990 --> 00:43:22.537
in together with information
from the GitHub issue

00:43:22.620 --> 00:43:25.377
and build a single set of requirements.

00:43:25.460 --> 00:43:27.417
And so let's have a quick
look at that email.

00:43:27.500 --> 00:43:30.027
We can see that there's
a requirement for a search bar

00:43:30.110 --> 00:43:33.727
with simple search,
and then Azure AI Search,

00:43:33.810 --> 00:43:36.397
the category sidebar,
the price range filter.

00:43:36.480 --> 00:43:39.667
And as much as a developer I
would love Evie to have put this

00:43:39.750 --> 00:43:43.717
in our GitHub issue, which is right
here, I realized that that's not

00:43:43.800 --> 00:43:46.567
where communication always
happens in our organization.

00:43:46.650 --> 00:43:50.737
So I could copy and paste this,
but instead, I'm using MCP to go

00:43:50.820 --> 00:43:55.057
and reach those two surfaces
to pull the information together.

00:43:55.140 --> 00:43:57.937
So Work IQ allows me to reach
into the M suite and pull

00:43:58.020 --> 00:44:01.507
that information out,
and the GitHub MCP server also lets me

00:44:01.590 --> 00:44:04.237
reach into GitHub to pull
that information out.

00:44:04.320 --> 00:44:07.267
And what's really cool is it's
worked out it can do both

00:44:07.350 --> 00:44:12.427
of those things in parallel, which
is something that I am incapable of doing.

00:44:12.510 --> 00:44:15.847
I am decidedly single-threaded,
unfortunately.

00:44:15.930 --> 00:44:19.477
So as that's coming back,
it's going to build this one set

00:44:19.560 --> 00:44:24.217
of requirements for me, after asking
Work IQ and getting that information back.

00:44:24.300 --> 00:44:26.677
And I could continue
with that set of requirements

00:44:26.760 --> 00:44:32.677
and work synchronously here in the CLI
or switch to the Visual Studio Code

00:44:32.760 --> 00:44:35.107
and keep working synchronously there.

00:44:35.190 --> 00:44:38.017
Here we go. Starting
to get the information back.

00:44:38.100 --> 00:44:39.217
And if I wanted to do that,

00:44:39.300 --> 00:44:44.437
I could also control it remotely using
this new remote command in the CLI.

00:44:44.520 --> 00:44:46.927
So even though the work
is happening on my machine,

00:44:47.010 --> 00:44:51.457
I could adjust it and make
changes using my mobile or the web.

00:44:51.540 --> 00:44:54.817
But instead, I want
to show you a new feature,

00:44:54.900 --> 00:44:58.417
a new experimental feature
called the rubber duck agent.

00:44:58.500 --> 00:45:02.527
So this will take a sec, so I'm going
to go to one I've already done earlier.

00:45:02.610 --> 00:45:05.617
And all I'm doing here is asking
can you give me a rubber duck

00:45:05.700 --> 00:45:07.987
review of these requirements?

00:45:08.070 --> 00:45:12.037
And what the CLA is doing is it's
kicking off the rubber duck

00:45:12.120 --> 00:45:16.957
agent here, and it's reviewing
what work has just been done.

00:45:17.040 --> 00:45:19.627
Now, the cool thing about
this is I was using Claude Sonnet

00:45:19.710 --> 00:45:23.107
4.6, but to do the review,
it's going to choose a model

00:45:23.190 --> 00:45:27.427
from a different model family,
so we get a real second opinion on this.

00:45:27.510 --> 00:45:29.557
And it's lucky we did.
We found a few issues here.

00:45:29.640 --> 00:45:32.587
Range is ambiguous, type versus category.

00:45:32.670 --> 00:45:34.927
The fallback spec is too vague,
and a few other things.

00:45:35.010 --> 00:45:38.677
So I would obviously want
to clarify some of these

00:45:38.760 --> 00:45:40.627
for it before I started
work, but this is a demo,

00:45:40.710 --> 00:45:42.697
so let's just pretend it's perfect.

00:45:42.780 --> 00:45:44.047
I'm going to continue with this.

00:45:44.130 --> 00:45:48.217
I could do it locally, but instead,
I'm going to delegate this to GitHub,

00:45:48.300 --> 00:45:51.487
and that will send this
information to the cloud,

00:45:51.570 --> 00:45:54.457
and the Copilot cloud
agent will pick up that work

00:45:54.540 --> 00:45:58.027
and start working on that, while
I can move on to something else.

00:45:58.110 --> 00:46:00.637
So the thing I'm going to move
on to is the other thing

00:46:00.720 --> 00:46:03.247
that we do as developers,
which is a code review.

00:46:03.330 --> 00:46:06.037
Now, I mentioned that FAQ
page was really comprehensive.

00:46:06.120 --> 00:46:08.647
The downside of that is
sometimes it's hard to scroll

00:46:08.730 --> 00:46:11.677
through all that information,
and so it'd be great

00:46:11.760 --> 00:46:14.707
if we had an agent that we
could just ask questions of,

00:46:14.790 --> 00:46:17.437
and that's what my colleague
has put together.

00:46:17.520 --> 00:46:21.397
So he's used Microsoft Foundry
and created a custom agent.

00:46:21.480 --> 00:46:24.637
It's based on GPT 5.4 mini,
which means that it's going

00:46:24.720 --> 00:46:27.637
to be responsive and relatively cheap.

00:46:27.720 --> 00:46:33.637
We've got a set of instructions here, and
importantly, we also have a file search.

00:46:33.720 --> 00:46:36.817
Now, this is one of many ways
you can insert real information

00:46:36.900 --> 00:46:40.717
in, but he just uploaded a markdown
document with our FAQ page.

00:46:40.800 --> 00:46:45.367
So that means when we ask
a question of the agent here,

00:46:45.450 --> 00:46:48.427
we get real answers, grounded in truth.

00:46:48.510 --> 00:46:50.497
So they're grounded in that FAQ document,

00:46:50.580 --> 00:46:53.797
rather than potentially
hallucinating them.

00:46:53.880 --> 00:46:55.987
But we can't just trust
that that's going to work.

00:46:56.070 --> 00:46:57.757
So he's also done some evaluation

00:46:57.840 --> 00:47:01.837
and used automatic evaluation
to have a look at how successful

00:47:01.920 --> 00:47:04.597
that agent is across
a number of different areas,

00:47:04.680 --> 00:47:08.287
so things like tool selection,
output utilization,

00:47:08.370 --> 00:47:11.647
even things like relevance
and at task adherence.

00:47:11.730 --> 00:47:14.077
So we've really validated
whether that agent is going

00:47:14.160 --> 00:47:15.637
to do the job we want it to do.

00:47:15.720 --> 00:47:17.437
Okay, that's the agent. That looks great.

00:47:17.520 --> 00:47:19.537
Let's look at the actual pull request.

00:47:19.620 --> 00:47:20.977
Now, because I've been doing other work,

00:47:21.060 --> 00:47:24.307
the first thing I did was assign
this to GitHub Copilot.

00:47:24.390 --> 00:47:28.597
And Copilot Code Review has looked through
this pull request, knowing the context

00:47:28.680 --> 00:47:32.047
of our entire repo,
and it's made some suggestions.

00:47:32.130 --> 00:47:34.087
It's made quite a few
suggestions, actually.

00:47:34.170 --> 00:47:36.517
We've got nine comments
generated, and I could scroll

00:47:36.600 --> 00:47:39.217
through and fix them
individually if I wanted,

00:47:39.300 --> 00:47:42.877
or I could just ask Copilot
to fix them all in one batch.

00:47:42.960 --> 00:47:45.727
And in fact, you can
send a message to Copilot

00:47:45.810 --> 00:47:49.417
from any pull request,
whether it was created by a human

00:47:49.500 --> 00:47:55.147
or by the cloud agent, and ask
it to do some additional work.

00:47:55.230 --> 00:47:59.437
So in this case, that FAQ
page involves a UI change,

00:47:59.520 --> 00:48:01.897
but I didn't see any
screenshots in the pull request,

00:48:01.980 --> 00:48:03.487
which would have been really nice.

00:48:03.570 --> 00:48:08.857
So I asked Copilot to use Playwright
to take before-and-after screenshots

00:48:08.940 --> 00:48:10.807
so I can see the UI difference.

00:48:10.890 --> 00:48:13.057
And this is another MCP
server we have access to.

00:48:13.140 --> 00:48:15.727
And sure enough, there we go.
We've got a before and an

00:48:15.810 --> 00:48:18.097
after with our fields
to ask the questions.

00:48:18.180 --> 00:48:19.837
Now, that wasn't enough for me.

00:48:19.920 --> 00:48:24.517
I asked for a couple more screenshots, and
I noticed there's actually an issue here.

00:48:24.600 --> 00:48:27.157
We didn't connect
to the Foundry correctly,

00:48:27.240 --> 00:48:29.047
so there's something
wrong in the code here,

00:48:29.130 --> 00:48:31.297
and this is something I might
not have noticed initially,

00:48:31.380 --> 00:48:34.717
if I was just doing a code
review looking at that code.

00:48:34.800 --> 00:48:38.857
Okay, I've shown you a few things today,
so I just want to quickly recap.

00:48:38.940 --> 00:48:42.367
We looked at using the GitHub Copilot CLI

00:48:42.450 --> 00:48:46.687
to consolidate information
using MCP servers like Work IQ

00:48:46.770 --> 00:48:51.877
and the GitHub MCP server to start
work on that new search functionality.

00:48:51.960 --> 00:48:56.767
We looked at using Microsoft Foundry
to build a custom agent grounded in truth

00:48:56.850 --> 00:49:00.307
so that we could really answer
questions that the customers had.

00:49:00.390 --> 00:49:04.567
And then we looked at how Copilot Code
Review can help us review a pull request

00:49:04.650 --> 00:49:08.467
and how Copilot can even improve
that pull request as we go.

00:49:08.550 --> 00:49:11.227
And trust me, I've only
just touched the surface

00:49:11.310 --> 00:49:14.767
of what we can do with Copilot
to help enable our teams

00:49:14.850 --> 00:49:17.887
to complete more work faster.
Thanks, everyone.

00:49:17.970 --> 00:49:19.437
Back to you, Satya.

00:49:19.520 --> 00:49:25.197
[Applause]

00:49:25.280 --> 00:49:28.687
SATYA NADELLA: I must say
we have come a long way

00:49:28.770 --> 00:49:35.017
in our naming as well from Visual
Basic to rubber duck agents.

00:49:35.100 --> 00:49:39.787
I love it. It's pushing the frontier.

00:49:39.870 --> 00:49:45.937
And one of the other super
important considerations, right,

00:49:46.020 --> 00:49:47.677
when you build all these agents, right,

00:49:47.760 --> 00:49:50.407
what Damian showed is just
pretty stunning, right,

00:49:50.490 --> 00:49:54.847
between Copilot Studio plus what
he showed, you can be assured

00:49:54.930 --> 00:49:57.847
of one thing, which is we're
going to have lots and lots

00:49:57.930 --> 00:50:02.107
of agents inside of our organization,
in our tenant, in our network.

00:50:02.190 --> 00:50:04.597
And so one of the key
considerations is going

00:50:04.680 --> 00:50:08.827
to be security and compliance
and observability,

00:50:08.910 --> 00:50:12.607
and that's what's the motivation
for Agent 365, right,

00:50:12.690 --> 00:50:17.107
which is how do we build a new
control plane, just like we did

00:50:17.190 --> 00:50:21.577
for the previous era,
but for the agentic era, right?

00:50:21.660 --> 00:50:24.097
We need an identity. So we extended Entra.

00:50:24.180 --> 00:50:27.607
We needed security. We extended Defender.

00:50:27.690 --> 00:50:33.037
We needed data security.
We extended Purview.

00:50:33.120 --> 00:50:39.397
And so fundamentally, now you have in
Agent 365 a platform and a control plane.

00:50:39.480 --> 00:50:41.287
And by the way, they don't
need to be hosted on Foundry.

00:50:41.370 --> 00:50:45.337
They can be hosted on Azure.
They can be anywhere, again, on AWS, GCP,

00:50:45.420 --> 00:50:47.827
anywhere on your private cloud, even,

00:50:47.910 --> 00:50:51.997
but you have this one
control plane that you can use

00:50:52.080 --> 00:50:54.637
to be able to manage all of these agents

00:50:54.720 --> 00:51:00.427
and have real visibility into all of these
agents, and set governance policies

00:51:00.510 --> 00:51:03.457
and enforce those governance policies.

00:51:03.540 --> 00:51:10.027
Now, one other consideration
that perhaps in the months

00:51:10.110 --> 00:51:12.067
to come and definitely in the years

00:51:12.150 --> 00:51:17.977
to come will probably become
the top consideration is in this age

00:51:18.060 --> 00:51:25.060
of AI where you no longer
just have a platform,

00:51:25.470 --> 00:51:32.077
but you have a platform that
extends into a learning system.

00:51:32.160 --> 00:51:37.477
So then the question
for any software company,

00:51:37.560 --> 00:51:43.567
any new AI startup, or,
for that matter, any bank, insurance,

00:51:43.650 --> 00:51:47.917
energy, telco, whatever -- pick
your favorite sort of sector

00:51:48.000 --> 00:51:51.187
of the economy -- is, guess what?

00:51:51.270 --> 00:51:54.907
There's no limits
to these learning systems.

00:51:54.990 --> 00:51:59.437
They can pick up what's novel
and new and interesting to know

00:51:59.520 --> 00:52:04.207
about any business and just make
it part of their base weights.

00:52:04.290 --> 00:52:07.867
And so this is a pretty new
thing that's never happened

00:52:07.950 --> 00:52:13.837
before because it's not like the last
time you thought about a digital system.

00:52:13.920 --> 00:52:18.127
That digital system was a
self-improving learning system.

00:52:18.210 --> 00:52:22.177
It was a system that you could
sort of use with your context

00:52:22.260 --> 00:52:26.947
and your workflows and make
sure that that IP that you create

00:52:27.030 --> 00:52:33.067
on top of it is yours. And that
is what you now and all of us have

00:52:33.150 --> 00:52:34.927
to do, even in the AI era.

00:52:35.010 --> 00:52:37.897
And to do that, that means
the first thing is to really think

00:52:37.980 --> 00:52:42.637
about each of these layers
of the agent platform and make sure

00:52:42.720 --> 00:52:47.647
that you start with your evals,
right, after all,

00:52:47.730 --> 00:52:51.907
your hill-climbing with any
agent system you build

00:52:51.990 --> 00:52:55.747
on some eval and making sure
that those evals are private.

00:52:55.830 --> 00:52:59.527
In fact, perhaps one of the most
important IP of an organization

00:52:59.610 --> 00:53:03.547
or a software company that is
doing something unique is going

00:53:03.630 --> 00:53:05.857
to be captured in some
private benchmark, right?

00:53:05.940 --> 00:53:07.627
In fact, all the public benchmarks are

00:53:07.710 --> 00:53:12.337
so saturated they're more
or less becoming sort of noise.

00:53:12.420 --> 00:53:15.697
And so therefore the real
game for anything is

00:53:15.780 --> 00:53:20.557
about hill-climbing on benchmarks,
and private benchmarks you create,

00:53:20.640 --> 00:53:25.507
private evaluation sets you create,
are going to become the new IP.

00:53:25.590 --> 00:53:28.387
The next thing you want to make
sure is you decouple, right?

00:53:28.470 --> 00:53:30.607
You want to have access
to multiple models.

00:53:30.690 --> 00:53:33.637
Like rubber duck is cool,
or critique is cool,

00:53:33.720 --> 00:53:40.177
but you want the harness, the context,
and the models to all be decoupled.

00:53:40.260 --> 00:53:43.837
It's tempting to couple them
to get some gain on an eval,

00:53:43.920 --> 00:53:48.937
but then you're stuck with that
one particular local maxima.

00:53:49.020 --> 00:53:51.547
But you're not going for the
global maxima, where you want

00:53:51.630 --> 00:53:56.917
to be able to ride every frontier
gain or your own model development.

00:53:57.000 --> 00:54:03.427
You want that context
that's there with your --

00:54:03.510 --> 00:54:08.977
you want to externalize
as much of the expertise inside

00:54:09.060 --> 00:54:11.887
of your organization into skills, right?

00:54:11.970 --> 00:54:14.197
Simple markdown files like in Cowork,

00:54:14.280 --> 00:54:17.917
one of the cool things is you
start writing these skills,

00:54:18.000 --> 00:54:20.407
and you even compound
the skills individually

00:54:20.490 --> 00:54:21.817
and at an organizational level.

00:54:21.900 --> 00:54:25.927
So it's kind of externalizing
the expertise of an organization.

00:54:26.010 --> 00:54:28.807
You want to keep that,
again, very private.

00:54:28.890 --> 00:54:33.727
You want to take all the tacit
knowledge of the organization.

00:54:33.810 --> 00:54:35.827
In fact, the tacit knowledge
of the organization is

00:54:35.910 --> 00:54:39.487
with the people and the way they
work and the way they work with AI.

00:54:39.570 --> 00:54:43.057
Those traces when I talked
about in GitHub, now in sessions,

00:54:43.140 --> 00:54:47.947
you have the traces and the
trajectories of all of your agents.

00:54:48.030 --> 00:54:54.097
That's super important data. Not just your
source code, but that data is your data.

00:54:54.180 --> 00:54:56.587
And you can train your own model.

00:54:56.670 --> 00:55:00.727
You can take a base model, which is a
pretrained model an open-weight model,

00:55:00.810 --> 00:55:03.847
and train that with these
traces, and then put it back

00:55:03.930 --> 00:55:08.257
in Copilot in Auto,
and it'll then route to your model.

00:55:08.340 --> 00:55:14.677
So I think that this idea
and this consideration

00:55:14.760 --> 00:55:20.107
around sovereignty of a firm because,
ultimately, what is the future of a firm?

00:55:20.190 --> 00:55:24.067
The future of a firm is its
ability to continuously,

00:55:24.150 --> 00:55:28.267
monotonically get better
at its knowledge creation,

00:55:28.350 --> 00:55:32.497
and it's able to capture
its tacit knowledge

00:55:32.580 --> 00:55:36.757
that is not only inside
of context or in skills,

00:55:36.840 --> 00:55:39.817
but even in the weights
that you control as your own IP.

00:55:39.900 --> 00:55:42.457
And so that's what we're
building the agent platform for,

00:55:42.540 --> 00:55:43.717
and you'll continue to see us,

00:55:43.800 --> 00:55:46.597
even going into our developer
conference in early part

00:55:46.680 --> 00:55:50.107
of June, talk about this in
great detail because we think

00:55:50.190 --> 00:55:52.867
that this is what, as I said, any startup,

00:55:52.950 --> 00:55:55.297
any software company,
and, for that matter,

00:55:55.380 --> 00:55:59.227
any part of the economy that now
needs to be thinking of all

00:55:59.310 --> 00:56:02.137
of these pretty strategically.

00:56:02.220 --> 00:56:06.367
And sort of speaking of that in Australia,
we have fantastic momentum in terms

00:56:06.450 --> 00:56:08.977
of people building these agentic systems.

00:56:09.060 --> 00:56:13.237
I had a chance yesterday
to meet with many of them.

00:56:13.320 --> 00:56:19.927
AEMO has taken Fabric and Foundry and
built out a fantastic system to be able

00:56:20.010 --> 00:56:23.137
to improve the signal-to-noise
ratio around alerts, right?

00:56:23.220 --> 00:56:28.417
Because how do you really have
AI help them manage, in fact,

00:56:28.500 --> 00:56:31.207
some of the most critical
infrastructure around the grid

00:56:31.290 --> 00:56:34.957
in a much more streamlined
way and an effective way?

00:56:35.040 --> 00:56:38.137
Had a chance to sort of meet
the team at the Commonwealth Bank,

00:56:38.220 --> 00:56:42.997
where they have built and extended
both their customer-facing bot

00:56:43.080 --> 00:56:46.027
with all the new power of these models,

00:56:46.110 --> 00:56:50.377
but also for their agents
in their call centers to be able

00:56:50.460 --> 00:56:52.987
to take the escalations
and have the tools.

00:56:53.070 --> 00:56:56.647
The productivity gains that
they're seeing is tremendous.

00:56:56.730 --> 00:57:00.277
I had a chance to meet with
even one student and as well

00:57:00.360 --> 00:57:04.597
as a software company
called Cogniti that's working

00:57:04.680 --> 00:57:09.127
with the University of Sydney
in enabling, in fact, teachers, right?

00:57:09.210 --> 00:57:11.917
We know that a lot of the
students are using these tools

00:57:12.000 --> 00:57:16.117
to learn, but teachers being
able to create new course material

00:57:16.200 --> 00:57:22.627
and change the way they conduct their
sessions and so on were fantastic to see.

00:57:22.710 --> 00:57:26.290
And of course, I had a chance
to meet with Cricket Australia.

00:57:27.330 --> 00:57:29.797
And look, I mean, it's always awesome.

00:57:29.880 --> 00:57:31.717
I think Australia has been collecting --

00:57:31.800 --> 00:57:34.057
and Cricket Australia -- has
been collecting stats from,

00:57:34.140 --> 00:57:36.547
I forget, 1896 or something like that.

00:57:36.630 --> 00:57:38.677
In fact, the last time I was here,
somebody was telling me,

00:57:38.760 --> 00:57:41.557
hey, if you ever wanted to check
the scorecards of Steve Smith

00:57:41.640 --> 00:57:44.197
when he was a 15-year-old,
there's no problem.

00:57:44.280 --> 00:57:48.907
You have that all data.
And I'm a big cricket stats buff.

00:57:48.990 --> 00:57:52.687
And one of the things, in fact,
one of the queries I have,

00:57:52.770 --> 00:57:54.427
which I'm going to do this weekend, is --

00:57:54.510 --> 00:57:58.957
and in fact, I was asking them
whether they have basically an

00:57:59.040 --> 00:58:01.960
MCP server for all of their data.

00:58:03.390 --> 00:58:08.827
And it's all in Azure Blob Storage,
so I must assume it is there.

00:58:08.910 --> 00:58:12.397
But one of my things I want to check
out is the conversion ratios.

00:58:12.480 --> 00:58:16.897
If truly Steve Smith's
under-15 scores are there,

00:58:16.980 --> 00:58:20.407
I want to see them next
Vaibhav Suryavanshi's scores

00:58:20.490 --> 00:58:25.687
and see how good is this guy
going to be when he grows up?

00:58:25.770 --> 00:58:32.317
And so that type of ability
to use data in a rich way

00:58:32.400 --> 00:58:35.887
and then translate it
in their case to the outcome

00:58:35.970 --> 00:58:40.147
that they desire, which is fan
engagement, is fantastic to see.

00:58:40.230 --> 00:58:44.797
So let's roll the video. TODD GREENBERG:
Cricket is our national sport.

00:58:44.880 --> 00:58:47.977
It's an Australian pastime.
It brings communities together.

00:58:48.060 --> 00:58:50.977
So fan engagement is at the very
core of everything we think

00:58:51.060 --> 00:58:52.417
about here at Cricket Australia.

00:58:52.500 --> 00:58:55.717
We're engaging with more than a million
fans on the Cricket Australia Live app.

00:58:55.800 --> 00:58:58.447
The challenge for us is to
personalize their experience.

00:58:58.530 --> 00:58:59.707
LEIGH DAVIES: When I'm
watching a cricket match,

00:58:59.790 --> 00:59:02.047
I'm paying attention to the
battle between bat and ball.

00:59:02.130 --> 00:59:04.387
It's fascinating. It keeps me involved.

00:59:04.470 --> 00:59:06.487
CASSIE GRAY: It's a pretty nuanced sport.

00:59:06.570 --> 00:59:10.237
It can speed up or surprise
you at any point in the game,

00:59:10.320 --> 00:59:13.807
and I love to follow along
and learn more about the players.

00:59:13.890 --> 00:59:16.687
TODD GREENBERG: AI Insights
provides a level of opportunity

00:59:16.770 --> 00:59:18.397
for all fans, whether that's engaging

00:59:18.480 --> 00:59:20.647
with the rich statistics
and history of the sport

00:59:20.730 --> 00:59:23.707
or it's just learning
the simple rules and regulations.

00:59:23.790 --> 00:59:25.147
BALAMURUGAN P M: Ease of use and accuracy

00:59:25.230 --> 00:59:27.487
to check the player stats,
to check the match stats,

00:59:27.570 --> 00:59:30.967
was non-negotiable for us
and Azure Open AI just hit the mark.

00:59:31.050 --> 00:59:35.977
Millions of data points that sits
in our Azure Cosmos TV are being served

00:59:36.060 --> 00:59:39.280
to our fans in subseconds.

00:59:40.950 --> 00:59:42.457
SPEAKER 8: Oh, that's a good strike.

00:59:42.540 --> 00:59:45.037
He's a strong boy is Ryan Harris.

00:59:45.120 --> 00:59:47.107
CASSIE GRAY: I like that I
can switch between profiles.

00:59:47.190 --> 00:59:49.387
So most of the time I'm a newbie.

00:59:49.470 --> 00:59:52.057
Sometimes I'll switch to stats guru.

00:59:52.140 --> 00:59:54.907
I like to deep dive,
especially on great players,

00:59:54.990 --> 00:59:57.877
what their stats are,
what they're known for.

00:59:57.960 --> 00:59:59.557
LEIGH DAVIES: I could
really see myself finding

00:59:59.640 --> 01:00:01.687
out information about test history.

01:00:01.770 --> 01:00:05.797
I noticed that the AI Insights
goes as far back as the 1800s.

01:00:05.880 --> 01:00:07.417
BALAMURUGAN P M:
With our Azure infrastructure set

01:00:07.500 --> 01:00:11.677
up in a way that is scalable,
it gives us a whole new avenue

01:00:11.760 --> 01:00:14.167
to interact with our fans
like never before.

01:00:14.250 --> 01:00:18.067
The frequency on which they
use the app, that has increased

01:00:18.150 --> 01:00:21.067
by 30% from the past season
to the current season.

01:00:21.150 --> 01:00:23.077
LEIGH DAVIES: Cricket is a
game that's in your blood.

01:00:23.160 --> 01:00:24.907
We're all from different parts
of life, but we're connected

01:00:24.990 --> 01:00:26.497
to cricket and the game that we love.

01:00:26.580 --> 01:00:28.567
TODD GREENBERG:
AI Insights gives us the ability

01:00:28.650 --> 01:00:30.727
to supercharge someone's
knowledge of the game,

01:00:30.810 --> 01:00:34.437
make them feel part of a community
and part of the sport.

01:00:34.520 --> 01:00:39.437
[Music] SATYA NADELLA: All right, so cool.

01:00:39.520 --> 01:00:43.937
[Applause]

01:00:44.020 --> 01:00:48.047
For a stats-rich sport,
this is just game changer.

01:00:48.130 --> 01:00:53.887
In fact, they were showing me
how you're following a match,

01:00:53.970 --> 01:00:57.937
and it just keeps coming ball by ball
with this unbelievable statistics.

01:00:58.020 --> 01:01:02.770
And one of the things that they had
was these conversion ratios, right?

01:01:04.170 --> 01:01:06.037
I guess, I don't know,
at least I'm obsessed about it,

01:01:06.120 --> 01:01:09.457
which is what's the greatest ever
test team from different eras?

01:01:09.540 --> 01:01:13.807
And you can't compare. One of the hard
statistical problems is how do you compare

01:01:13.890 --> 01:01:18.037
people across generations?
And it turns out there is causality,

01:01:18.120 --> 01:01:19.447
and it's called conversion ratio.

01:01:19.530 --> 01:01:23.317
So if you sort of really like -- at least,
that's one of the predictable things.

01:01:23.400 --> 01:01:26.590
And so I'm going to have
a lot of fun this weekend.

01:01:28.590 --> 01:01:32.317
So the last layer of the
tech stack I want to talk

01:01:32.400 --> 01:01:36.847
about is the cloud and AI
token factory, right?

01:01:36.930 --> 01:01:40.957
So this is the rich
infrastructure that you get

01:01:41.040 --> 01:01:47.947
to build to power all of these agent
systems and applications going forward.

01:01:48.030 --> 01:01:52.897
And it comes down to getting
this one super important formula

01:01:52.980 --> 01:01:59.197
right, which is performance is really
about tokens per dollar per watt.

01:01:59.280 --> 01:02:03.397
And you really want to be best-in-class
at it as a hyperscaler, and you want

01:02:03.480 --> 01:02:06.307
to be best-in-class at it as a country.

01:02:06.390 --> 01:02:08.617
And that, I think,
is what's going to really create

01:02:08.700 --> 01:02:12.787
that abundance of tokens
that then allows the rest

01:02:12.870 --> 01:02:15.937
of what we talked about to come alive.

01:02:16.020 --> 01:02:22.657
And so we are building out Azure
as the world's computer with all

01:02:22.740 --> 01:02:25.267
of the regions around the world.

01:02:25.350 --> 01:02:30.277
We have 70 plus data center
regions around the world

01:02:30.360 --> 01:02:32.467
with one fungible fleet.

01:02:32.550 --> 01:02:38.137
And of course, when it comes
to Australia, we have regions

01:02:38.220 --> 01:02:41.077
in Sydney, in Melbourne, and Canberra.

01:02:41.160 --> 01:02:45.217
And it's growing.
It's 100% renewable energy-powered.

01:02:45.300 --> 01:02:48.877
We're very proud of it.
We're very committed to it.

01:02:48.960 --> 01:02:54.227
We're also, yeah. [Applause]

01:02:54.310 --> 01:02:58.547
And one of the ways we
continue to improve performance

01:02:58.630 --> 01:03:03.107
and power performance is by
innovating on the systems level

01:03:03.190 --> 01:03:05.407
and even all the way
to the silicon level, right?

01:03:05.490 --> 01:03:08.197
So we obviously partner
with NVIDIA and AMD,

01:03:08.280 --> 01:03:11.287
but we're thrilled about
the progress Maia is making.

01:03:11.370 --> 01:03:18.127
And Maia 200 now has 30 plus percent over
any leading ASIC out there in the market.

01:03:18.210 --> 01:03:20.917
And we'll continue to innovate
and co-innovate, in fact,

01:03:21.000 --> 01:03:26.977
between those agent run times
as well as the model architecture,

01:03:27.060 --> 01:03:32.047
plus the AI system itself because there
are very efficient ways for us to be able

01:03:32.130 --> 01:03:36.187
to really drive that tokens per
dollar per watt going forward,

01:03:36.270 --> 01:03:39.697
and we want to be able
to have that ability.

01:03:39.780 --> 01:03:46.747
We also are very focused
on our sovereignty considerations,

01:03:46.830 --> 01:03:52.777
and so this is where we are making
sure that you have all of the controls.

01:03:52.860 --> 01:03:57.037
Not only do we have the regions
in country, but we conform

01:03:57.120 --> 01:03:58.837
to all the regulatory requirements.

01:03:58.920 --> 01:04:02.227
You have the ability
to manage your own keys.

01:04:02.310 --> 01:04:08.197
You can encrypt data. You can use
confidential computing such that the data

01:04:08.280 --> 01:04:15.157
in use is also encrypted. So there's tons
of sovereignty control that is built-in.

01:04:15.240 --> 01:04:18.247
And the one other consideration
I would also submit is

01:04:18.330 --> 01:04:21.277
that you want to think about
cyber resilience as well

01:04:21.360 --> 01:04:22.777
as sovereignty together, right?

01:04:22.860 --> 01:04:27.097
Because you want the global
intelligence of signals

01:04:27.180 --> 01:04:33.127
when it comes to your overall
resilience in addition to sovereignty.

01:04:33.210 --> 01:04:35.497
You don't want to sacrifice
one for the other.

01:04:35.580 --> 01:04:40.057
And that's, again, part of the design
of what we have put together.

01:04:40.140 --> 01:04:43.597
And with all of that,
one of the things I'm absolutely thrilled

01:04:43.680 --> 01:04:47.047
to announce is our doubling
down our commitment.

01:04:47.130 --> 01:04:50.317
We are making our biggest
commitment in Australia

01:04:50.400 --> 01:04:53.437
of Australian dollars of 25 billion.

01:04:53.520 --> 01:04:57.127
[Applause]

01:04:57.210 --> 01:05:04.057
So this will allow us to expand
this ability to generate the tokens

01:05:04.140 --> 01:05:06.097
and provide the cloud infrastructure.

01:05:06.180 --> 01:05:09.367
I think one of the things that gets
lost is we just think about the tokens.

01:05:09.450 --> 01:05:12.607
But when I talked about
those hosted agents,

01:05:12.690 --> 01:05:17.557
hosted agents need classic
cores, classic storage.

01:05:17.640 --> 01:05:22.657
Same thing. Every Cowork session
needs storage and compute.

01:05:22.740 --> 01:05:24.217
And so one of the things that we are going

01:05:24.300 --> 01:05:30.877
to see is the explosion of both
cloud and AI accelerators,

01:05:30.960 --> 01:05:33.787
and that's one of the things
that we are very excited about,

01:05:33.870 --> 01:05:37.687
making sure we continue
to invest and bring all of that.

01:05:37.770 --> 01:05:41.137
We are also thrilled
to expand our partnership

01:05:41.220 --> 01:05:45.437
with the Signals Directorate and Cyber
Shield because, again, going back to --

01:05:45.520 --> 01:05:51.197
[Applause] -- because one of the
challenges we all collectively face is how

01:05:51.280 --> 01:05:56.227
do we protect ourselves, defend
ourselves and our organizations,

01:05:56.310 --> 01:06:00.157
against cyber threat actors?
And so that's, again, a signals game.

01:06:00.240 --> 01:06:04.717
That's an intelligence game. And we want
to make sure that we are able to extend

01:06:04.800 --> 01:06:08.227
that Cyber Shield arrangement
such that the most vulnerable,

01:06:08.310 --> 01:06:11.617
right -- that is that small
business or the consumer --

01:06:11.700 --> 01:06:16.657
is protected when it
comes to any cyber attack.

01:06:16.740 --> 01:06:20.047
And lastly, we are
really thrilled to double

01:06:20.130 --> 01:06:23.077
down on the skilling
investments we're making.

01:06:23.160 --> 01:06:28.837
We're going to train three plus
million Australians around AI skills.

01:06:28.920 --> 01:06:34.687
And again, there's going to be
a real change in the way we work,

01:06:34.770 --> 01:06:40.477
the tools we use to work. And so to be
able to get ahead of all of the jobs

01:06:40.560 --> 01:06:43.327
and the economic opportunity
that gets created,

01:06:43.410 --> 01:06:45.577
I think it's going to be very,
very important for all of us,

01:06:45.660 --> 01:06:50.107
including Microsoft, to do
the work to bring the skilling

01:06:50.190 --> 01:06:52.447
and the upskilling that is required.

01:06:52.530 --> 01:06:54.517
So I kind of want
to close where I started,

01:06:54.600 --> 01:06:58.327
which is our mission always remains

01:06:58.410 --> 01:07:02.617
to empower every person --
in this case, in Australia

01:07:02.700 --> 01:07:06.967
and every organization --
to be able to achieve more,

01:07:07.050 --> 01:07:09.937
and I want to leave you with, I think,

01:07:10.020 --> 01:07:14.467
a video that perhaps best
captures what this mission means

01:07:14.550 --> 01:07:18.607
to us and what it means to us
collectively to strive for.

01:07:18.690 --> 01:07:19.837
Thank you all very, very much.

01:07:19.920 --> 01:07:21.460
Have a restful day.

01:07:22.520 --> 01:07:27.817
[Music]

01:07:27.900 --> 01:07:32.107
GRANT MAHER: Deadly Coders,
we're there to ignite that spark

01:07:32.190 --> 01:07:33.967
of interest in STEM,
particularly in coding

01:07:34.050 --> 01:07:37.417
and robotics for Indigenous
kids throughout Australia.

01:07:37.500 --> 01:07:39.367
LEVI DUGGER: They need an opportunity,
like everyone else,

01:07:39.450 --> 01:07:42.607
to be able to move with the
ever-changing world and keep

01:07:42.690 --> 01:07:48.340
up with the demands of potential
future careers and educational pathways.

01:07:49.410 --> 01:07:51.997
JACK REIS: Deadly Coders
are here to help bring skills

01:07:52.080 --> 01:07:56.107
such as coding, education
around artificial intelligence,

01:07:56.190 --> 01:08:01.207
and we are doing that in partnership
with organizations such as Microsoft.

01:08:01.290 --> 01:08:04.297
>> Deadly Coders' Minecraft
Education program help break

01:08:04.380 --> 01:08:07.957
down those stereotypes that mob
usually see for themselves.

01:08:08.040 --> 01:08:10.687
The program makes AI and coding fun.

01:08:10.770 --> 01:08:14.317
It gamifies the whole process,
which actually then creates

01:08:14.400 --> 01:08:17.977
that energy and that spark.
EYLANIA NAAWI: It's a new way of learning,

01:08:18.060 --> 01:08:21.877
and I think kids really need
to know how to use it correctly,

01:08:21.960 --> 01:08:25.387
how to use it safely,
how to use it culturally,

01:08:25.470 --> 01:08:30.627
and how we can use it alongside
culture, not replacing culture.

01:08:30.710 --> 01:08:32.547
JACK REIS: When you are
able to impact the lives

01:08:32.630 --> 01:08:37.917
of young kids who typically would not have
access to technology, to be able to bring

01:08:38.000 --> 01:08:41.037
that to their world is priceless.

01:08:41.120 --> 01:08:44.277
EYLANIA NAAWI: Making sure that
younger mob had the opportunity

01:08:44.360 --> 01:08:46.437
to do whatever they wanted.

01:08:46.520 --> 01:08:53.520
[Music]

01:08:55.980 --> 01:08:58.777
SPEAKER 9: Please welcome
the President of Microsoft Australia

01:08:58.860 --> 01:09:01.437
and New Zealand, Jane Livesey.

01:09:01.520 --> 01:09:08.520
[Music]

01:09:08.780 --> 01:09:11.137
JANE LIVESEY: Welcome to the AI tour.

01:09:11.220 --> 01:09:14.317
It's wonderful to see so
many creators and innovators,

01:09:14.400 --> 01:09:19.327
all who are helping shape Australia's
next chapter here with us today.

01:09:19.410 --> 01:09:21.907
Thank you for taking the time.

01:09:21.990 --> 01:09:25.327
This is my first AI tour since
taking on the leadership role

01:09:25.410 --> 01:09:27.487
at the end of last year.

01:09:27.570 --> 01:09:31.867
After three decades in business,
I've been asked why Microsoft?

01:09:31.950 --> 01:09:34.357
And for me, the answer
has always been the same.

01:09:34.440 --> 01:09:38.797
It's two words: trust and safety.

01:09:38.880 --> 01:09:42.937
So much of our world is changing
from healthcare, education,

01:09:43.020 --> 01:09:45.697
business and communications --

01:09:45.780 --> 01:09:49.357
there couldn't be a more
important time to be Microsoft.

01:09:49.440 --> 01:09:52.267
We have the privilege and the obligation

01:09:52.350 --> 01:09:55.297
to help Australia navigate
this change with trust

01:09:55.380 --> 01:09:58.867
and safety embedded in everything we do.

01:09:58.950 --> 01:10:03.817
You just heard Satya outline
Microsoft's vision for an AI era

01:10:03.900 --> 01:10:06.847
where intelligence moves
from experimentation

01:10:06.930 --> 01:10:11.347
into everyday work embedded
in the world's most trusted cloud

01:10:11.430 --> 01:10:15.777
and productivity platforms.
AI is no longer abstract.

01:10:15.860 --> 01:10:20.647
It's here, and it's changing
how organizations operate.

01:10:20.730 --> 01:10:23.707
Across every sector,
we're seeing the appetite

01:10:23.790 --> 01:10:27.427
to adopt AI early and responsibly.

01:10:27.510 --> 01:10:31.297
It's not just about doing things faster.

01:10:31.380 --> 01:10:34.557
It's about unleashing human ambition,

01:10:34.640 --> 01:10:38.517
and that's what we call
frontier transformation.

01:10:38.600 --> 01:10:42.267
It's no longer about
pilots or proof of concepts.

01:10:42.350 --> 01:10:46.017
It's about empowering people,
reinventing business process,

01:10:46.100 --> 01:10:48.717
and unlocking the value of data.

01:10:48.800 --> 01:10:55.167
And this moment isn't just about
technology; it's about shared prosperity.

01:10:55.250 --> 01:10:57.657
There couldn't be a more important time

01:10:57.740 --> 01:11:02.457
to unlock the value that AI
brings to all of Australia.

01:11:02.540 --> 01:11:05.937
Through our local ecosystem,
Microsoft contributes more

01:11:06.020 --> 01:11:11.307
than $36 billion to Australia's
GDP, and we support more

01:11:11.390 --> 01:11:15.547
than 150,000 jobs nationwide.

01:11:15.630 --> 01:11:20.467
But the opportunity ahead
of us is even more enormous,

01:11:20.550 --> 01:11:22.807
but none of this will happen on its own.

01:11:22.890 --> 01:11:28.357
It takes leadership -- leadership that
is bold and curious and people-centered --

01:11:28.440 --> 01:11:34.567
and partnership across industry,
technology, and across all

01:11:34.650 --> 01:11:39.067
of Australia to ensure that we
adopt AI in a way that is safe,

01:11:39.150 --> 01:11:43.567
inclusive, and benefits all Australians.

01:11:43.650 --> 01:11:46.267
And that's why this conversation matters.

01:11:46.350 --> 01:11:49.597
I first want to thank
our partners and our sponsors

01:11:49.680 --> 01:11:52.437
for bringing these moments to life.

01:11:52.520 --> 01:11:57.217
[Applause]

01:11:57.300 --> 01:11:59.997
There are two organizations
that are at the heart

01:12:00.080 --> 01:12:04.407
of Australia's national fabric:
Telstra and Westpac,

01:12:04.490 --> 01:12:06.907
and they're putting this into practice.

01:12:06.990 --> 01:12:09.457
They're not just talking
about what's possible.

01:12:09.540 --> 01:12:15.007
They're doing it. So please join me
in welcoming Vicki Brady, CEO of Telstra,

01:12:15.090 --> 01:12:19.987
and Anthony Miller, CEO of Westpac,
to talk about how AI is transforming their

01:12:20.070 --> 01:12:23.887
organizations, their people,
and Australia's future.

01:12:23.970 --> 01:12:26.437
Thank you. Let's get started.

01:12:26.520 --> 01:12:33.437
[Applause]

01:12:33.520 --> 01:12:38.407
[Music]

01:12:38.490 --> 01:12:41.797
I couldn't be more excited to have you
both here for this conversation today.

01:12:41.880 --> 01:12:45.667
I've been looking forward
to it as part of the AI tour.

01:12:45.750 --> 01:12:50.257
Both Telstra and Westpac are operating
in an environment that's rapidly changing.

01:12:50.340 --> 01:12:52.987
That's something that
we're all experiencing.

01:12:53.070 --> 01:12:56.797
How are you navigating this?
Maybe, Anthony, I'll start with you?

01:12:56.880 --> 01:13:00.067
ANTHONY MILLER: So certainly,
it's one of the most challenging

01:13:00.150 --> 01:13:04.327
and invigorating
environments for a long time.

01:13:04.410 --> 01:13:07.537
The way we're thinking about
particularly the AI opportunity is

01:13:07.620 --> 01:13:12.487
that it's just this fantastic way
in which people can do their job better.

01:13:12.570 --> 01:13:14.857
They can do their job faster.
They can do it more safely.

01:13:14.940 --> 01:13:16.657
They can do it more consistently.

01:13:16.740 --> 01:13:20.257
And so the way we think about
AI is how do we get everybody set

01:13:20.340 --> 01:13:22.177
up to use it, and how are we investing

01:13:22.260 --> 01:13:25.147
in making sure they can use
it and know how to use it

01:13:25.230 --> 01:13:26.737
and continue to improve how they use it?

01:13:26.820 --> 01:13:29.227
So funny enough,
from my perspective, when we think

01:13:29.310 --> 01:13:33.967
about AI, it's a technology, yes,
but it, really, it's a people issue.

01:13:34.050 --> 01:13:38.227
It's a people challenge. And so how
well are we investing in our people?

01:13:38.310 --> 01:13:39.637
How well are we training them?

01:13:39.720 --> 01:13:42.667
How well are we leading them? How well
are we inspiring them and motivating them

01:13:42.750 --> 01:13:45.727
to use this tool to improve how
they do things, day in, day out?

01:13:45.810 --> 01:13:48.297
That's the way we're
thinking about it at Westpac.

01:13:48.380 --> 01:13:50.457
JANE LIVESEY: Vicki?
VICKI BRADY: Yeah, so many parallels.

01:13:50.540 --> 01:13:54.567
I guess I've been trying
to think about it in three ways.

01:13:54.650 --> 01:14:00.177
So first off, as a telco,
lots of increasing demands,

01:14:00.260 --> 01:14:02.997
expectation, just demand on our network.

01:14:03.080 --> 01:14:06.927
So to continue to be
a leader in connectivity,

01:14:07.010 --> 01:14:09.747
we've got to be a leader
in how we're embedding

01:14:09.830 --> 01:14:12.937
and applying AI inside the business.

01:14:13.020 --> 01:14:18.847
And that's a whole
of business transformation,

01:14:18.930 --> 01:14:24.747
the way we engage with customers, the way
we empower our teams, even the way we run

01:14:24.830 --> 01:14:26.187
and operate our network.

01:14:26.270 --> 01:14:31.027
So the thing on my mind in that part
is always are we going fast enough?

01:14:31.110 --> 01:14:34.207
I mean, the world is just changing
so fast, as Anthony just said.

01:14:34.290 --> 01:14:35.707
It's sort of unprecedented.

01:14:35.790 --> 01:14:40.987
I know that gets used a lot,
and I just think to be competitive,

01:14:41.070 --> 01:14:43.867
we've got to be moving
at a scale and a pace.

01:14:43.950 --> 01:14:46.267
The second way is very much the skills.

01:14:46.350 --> 01:14:50.227
Are we giving sufficient opportunity
to our teams to grow those skills?

01:14:50.310 --> 01:14:52.327
We use a data and AI academy.

01:14:52.410 --> 01:14:55.537
We've just put in specific
learning pathways for jobs.

01:14:55.620 --> 01:14:58.387
We're working with Microsoft
at the moment on a pilot

01:14:58.470 --> 01:15:01.447
for an AI learning agent,
which would be great.

01:15:01.530 --> 01:15:06.247
And then the final way we think about
it is really the national conversation

01:15:06.330 --> 01:15:11.827
around how are the AI benefits
going to be shared across Australia?

01:15:11.910 --> 01:15:15.967
Because AI is really
a whole-of-society transition.

01:15:16.050 --> 01:15:18.307
It's not just affecting business.

01:15:18.390 --> 01:15:19.717
It's affecting the way we work.

01:15:19.800 --> 01:15:21.877
It affects the economy.

01:15:21.960 --> 01:15:25.327
So we're thinking about things
that are important around trust,

01:15:25.410 --> 01:15:29.707
around inclusion, around skills,
and we do think about some

01:15:29.790 --> 01:15:31.537
of those groups that are
going to need more support:

01:15:31.620 --> 01:15:34.147
small businesses, regional communities,

01:15:34.230 --> 01:15:36.697
people facing barriers
to digital inclusion.

01:15:36.780 --> 01:15:41.077
And again, Jane will announce something
today around small business, trying

01:15:41.160 --> 01:15:43.267
to help small businesses
get greater access.

01:15:43.350 --> 01:15:44.887
But I think businesses need to be

01:15:44.970 --> 01:15:48.267
in that conversation alongside
government and industry,

01:15:48.350 --> 01:15:51.247
and be helping contribute
to those solutions.

01:15:51.330 --> 01:15:53.527
JANE LIVESEY: If we come
back into your own businesses,

01:15:53.610 --> 01:15:55.627
you've both been quite ambitious in terms

01:15:55.710 --> 01:15:58.837
of your AI transformations,
can you share maybe a little bit

01:15:58.920 --> 01:16:02.017
about what you've learned and some
of the outcomes you've achieved?

01:16:02.100 --> 01:16:02.947
VICKI BRADY: Yeah, absolutely.

01:16:03.030 --> 01:16:05.977
I might start with
challenges and learnings.

01:16:06.060 --> 01:16:10.927
Lots of those. Top of the list would
be just how important getting the

01:16:11.010 --> 01:16:14.107
foundational tech and data is.

01:16:14.190 --> 01:16:19.177
Just as an example, a few years ago, we
had 80 platforms where our data was held.

01:16:19.260 --> 01:16:21.187
Our target is to get to three.

01:16:21.270 --> 01:16:25.717
We're at 20 right now, so making progress,
but you've got to be relentless, I think,

01:16:25.800 --> 01:16:30.087
relentless on removing complexity
and getting much simpler.

01:16:30.170 --> 01:16:33.277
We had more than 400 software partners.

01:16:33.360 --> 01:16:36.217
We consolidated down
to two to move faster.

01:16:36.300 --> 01:16:39.757
And then it is skills and access.

01:16:39.840 --> 01:16:43.897
And we've got a goal this year to get 85%

01:16:43.980 --> 01:16:49.777
of our teams are using AI tools
at least weekly, and we're at 75% now.

01:16:49.860 --> 01:16:53.047
So that's been important, foundationally.

01:16:53.130 --> 01:16:55.987
If I think about outcomes,
I mean, everyone will talk

01:16:56.070 --> 01:17:01.147
about getting better customer outcomes
alongside being more efficient.

01:17:01.230 --> 01:17:05.797
But if I drill down and just
use one example, last November,

01:17:05.880 --> 01:17:10.897
on our website, we launched
a generative-AI-powered assistant.

01:17:10.980 --> 01:17:16.597
First time we had GenAI in that tool,
and we, in March, took that into our app.

01:17:16.680 --> 01:17:19.597
So on our website,
it does pretty simple stuff,

01:17:19.680 --> 01:17:22.537
be able to check plans, activate a SIM.

01:17:22.620 --> 01:17:24.697
In the app, it can do
more sophisticated things

01:17:24.780 --> 01:17:27.427
because we know it's who
it is, which customer,

01:17:27.510 --> 01:17:31.857
so it can make changes to plans,
accounts, troubleshoot.

01:17:31.940 --> 01:17:36.597
It's already, of all of that
volume coming in on the website

01:17:36.680 --> 01:17:39.837
and our app,
it is dealing with 30% of that

01:17:39.920 --> 01:17:44.737
without our teams needing to get involved,
and we continue to build that capability.

01:17:44.820 --> 01:17:47.977
Of course, it hands off where
it's particularly sensitive

01:17:48.060 --> 01:17:51.877
or there might be indicators
of vulnerable circumstances,

01:17:51.960 --> 01:17:56.257
but it's allowing our teams
to spend more time with customers

01:17:56.340 --> 01:17:59.887
on those more complex issues.
So that's one example.

01:17:59.970 --> 01:18:01.177
ANTHONY MILLER: Well, I mean, Vicki,

01:18:01.260 --> 01:18:04.837
I agree with almost
everything you said there.

01:18:04.920 --> 01:18:06.727
That's the learnings we've had.

01:18:06.810 --> 01:18:10.837
I think the challenge that or the learning

01:18:10.920 --> 01:18:14.557
that I've really found
interesting with this tool,

01:18:14.640 --> 01:18:18.187
this technology, is the challenge

01:18:18.270 --> 01:18:21.577
that people need to be open-minded.

01:18:21.660 --> 01:18:25.447
I run -- and it is a very regulated
industry -- banking as appropriately

01:18:25.530 --> 01:18:30.037
so banking license is a real privilege,
and we need to meet a certain standard.

01:18:30.120 --> 01:18:32.767
And so therefore there's a real
culture, or that's the way we do it.

01:18:32.850 --> 01:18:38.167
We must do it this way. And so the tool
asks you and gives you an opportunity

01:18:38.250 --> 01:18:40.387
to think about doing things differently.

01:18:40.470 --> 01:18:43.987
And so therefore, it's a very
big challenge for us to sort

01:18:44.070 --> 01:18:46.567
of create an environment where
people feel they can step back

01:18:46.650 --> 01:18:49.207
and go is there a different
way that I can do my job?

01:18:49.290 --> 01:18:51.247
Is there a different way
that we can deliver this service

01:18:51.330 --> 01:18:53.647
or product to the customer
and do it in a way

01:18:53.730 --> 01:18:56.737
that meets the standards
rightly expected of us?

01:18:56.820 --> 01:19:01.747
And that is not a skill
that people naturally challenge

01:19:01.830 --> 01:19:04.567
to acquire as they
progress in an organization.

01:19:04.650 --> 01:19:07.987
Equally, someone has a better
idea than you or a better way

01:19:08.070 --> 01:19:10.897
of doing things is not way
people have classically been

01:19:10.980 --> 01:19:13.867
rewarded to acknowledge
someone else has a better idea.

01:19:13.950 --> 01:19:16.837
And so therefore this challenge
that sits in front of us is

01:19:16.920 --> 01:19:20.557
that it asks us to think about things
differently and be more open-minded.

01:19:20.640 --> 01:19:22.897
And so that's the learning we've taken.

01:19:22.980 --> 01:19:24.277
And so hence my observation

01:19:24.360 --> 01:19:27.997
around it being perhaps not
withstanding it's a technology.

01:19:28.080 --> 01:19:31.267
It's one of the great people
issues or opportunities of all time.

01:19:31.350 --> 01:19:33.757
We need to ask people to think
about things differently.

01:19:33.840 --> 01:19:37.267
And thankfully, Jane, too,
your organization, with the Copilot,

01:19:37.350 --> 01:19:41.457
we've given every single member
of the organization the tool,

01:19:41.540 --> 01:19:43.077
every single member of the company.

01:19:43.160 --> 01:19:47.007
Thirty-five thousand employees
is challenged, encouraged,

01:19:47.090 --> 01:19:49.407
and motivated to try
and experiment and learn,

01:19:49.490 --> 01:19:52.437
and then layering on top of that,
we have more ambition,

01:19:52.520 --> 01:19:56.287
more goals, and more
opportunities as we go forward.

01:19:56.370 --> 01:20:00.607
JANE LIVESEY: Amazing. I think as leaders,
we're all operating in the heart

01:20:00.690 --> 01:20:02.047
of Australia at the moment.

01:20:02.130 --> 01:20:04.297
What are some of the things
that you think we should be thinking

01:20:04.380 --> 01:20:08.167
about and your hopes
as leaders for Australia?

01:20:08.250 --> 01:20:10.537
VICKI BRADY: It's a big
question, isn't it?

01:20:10.620 --> 01:20:13.717
I mean, the first thing I'd
say is I think we've, course,

01:20:13.800 --> 01:20:16.867
we've got to consider
the risks and challenges,

01:20:16.950 --> 01:20:20.527
but I really hope we grab the opportunity

01:20:20.610 --> 01:20:23.857
because I just see it
as to remain competitive,

01:20:23.940 --> 01:20:28.087
to remain relevant as companies,
but as a country,

01:20:28.170 --> 01:20:31.567
grabbing that opportunity right
now and moving fast enough.

01:20:31.650 --> 01:20:38.557
And I think, for me, that means choices
we make today as businesses, as a country,

01:20:38.640 --> 01:20:40.447
around some of those foundational pieces

01:20:40.530 --> 01:20:44.637
about investing in skills
in Australia, building trust

01:20:44.720 --> 01:20:47.487
through the risk
and governance frameworks,

01:20:47.570 --> 01:20:51.207
inclusion so people can
share in those benefits,

01:20:51.290 --> 01:20:53.397
and getting the digital
infrastructure right

01:20:53.480 --> 01:20:56.227
so we unlock those productivity benefits.

01:20:56.310 --> 01:20:59.767
I just think it's a huge
opportunity, and those things

01:20:59.850 --> 01:21:03.307
to position us well are
necessary and they're urgent.

01:21:03.390 --> 01:21:04.690
Yeah.

01:21:05.940 --> 01:21:07.147
ANTHONY MILLER: I think, I agree.

01:21:07.230 --> 01:21:11.857
I mean, the real challenge
or what I really want to see is

01:21:11.940 --> 01:21:14.137
like let's have some ambition.

01:21:14.220 --> 01:21:17.077
The current discourse
publicly is so negative.

01:21:17.160 --> 01:21:23.137
It's so informed by what's not
going well or what is a challenge,

01:21:23.220 --> 01:21:28.117
as opposed to there are opportunities this
country has today which are extraordinary,

01:21:28.200 --> 01:21:30.307
and I think this technology
is a contributor

01:21:30.390 --> 01:21:34.207
to what we should be aiming for,
which is to really take Australia forward.

01:21:34.290 --> 01:21:37.537
And so without wanting to be
too dramatic about it, I mean,

01:21:37.620 --> 01:21:41.107
one of the challenges that faces
the country is lifting productivity.

01:21:41.190 --> 01:21:44.887
And classically, to grow productivity,
it's about investment in things

01:21:44.970 --> 01:21:47.347
such as technology and innovation.

01:21:47.430 --> 01:21:49.267
And normally, that's
technology and innovation,

01:21:49.350 --> 01:21:52.447
it only drives productivity
when it's diffused right

01:21:52.530 --> 01:21:54.967
across the community
and right across society.

01:21:55.050 --> 01:21:57.757
And so think electricity,
it's only when everyone's using it

01:21:57.840 --> 01:22:01.027
that we truly realized its benefits.

01:22:01.110 --> 01:22:03.217
And funny enough, this technology has

01:22:03.300 --> 01:22:07.387
so quickly diffused and so
quickly being adopted and used

01:22:07.470 --> 01:22:11.557
by everyone that it sets us up quicker
than we've ever had the opportunity to,

01:22:11.640 --> 01:22:15.727
so how are we embracing it? And so
we need an ambition for the country.

01:22:15.810 --> 01:22:17.707
We need a public discourse that's focused

01:22:17.790 --> 01:22:20.167
on how do we improve
the country using this?

01:22:20.250 --> 01:22:22.897
And it's not the panacea
for everything we need to do, but,

01:22:22.980 --> 01:22:24.277
gee, it's got a big role to play.

01:22:24.360 --> 01:22:27.697
And I think things like
regulators, governments,

01:22:27.780 --> 01:22:32.047
businesses, have a real
obligation to encourage

01:22:32.130 --> 01:22:35.797
and motivate and invest
in and challenge people to use it,

01:22:35.880 --> 01:22:39.217
to improve how they do their job, deliver
better outcomes for their customers.

01:22:39.300 --> 01:22:41.947
And I think that's that
catalyst will then drive some

01:22:42.030 --> 01:22:43.417
of what I think we can realize from it.

01:22:43.500 --> 01:22:47.617
And as I say, I just wish we were much
more positive and ambitious around it.

01:22:47.700 --> 01:22:48.187
JANE LIVESEY: Amazing.

01:22:48.270 --> 01:22:51.667
So Vicki, Anthony, I can't thank
you enough for your partnership

01:22:51.750 --> 01:22:54.037
and for sharing your insights today.

01:22:54.120 --> 01:22:58.177
We're really excited about the opportunity
that AI brings to all Australians.

01:22:58.260 --> 01:23:01.807
Thank you all for joining us today, and we
hope you enjoy the rest of the AI tour.

01:23:01.890 --> 01:23:03.437
Thank you. ANTHONY MILLER: Thank you.

01:23:03.520 --> 01:23:10.520
[Music]