1
00:00:02,620 --> 00:00:04,640
[music]

2
00:00:05,240 --> 00:00:06,837
[applause]

3
00:00:06,920 --> 00:00:09,437
>> Good morning. It's

4
00:00:09,520 --> 00:00:12,757
It's uh wonderful uh to be back in India

5
00:00:12,840 --> 00:00:15,957
and here in Mumbai and have a chance uh

6
00:00:16,040 --> 00:00:19,997
this morning to talk about uh AI and

7
00:00:20,080 --> 00:00:23,717
mostly to celebrate uh all the AI and

8
00:00:23,800 --> 00:00:26,797
the AI solutions uh that are being built

9
00:00:26,880 --> 00:00:29,997
uh by people in this country and this

10
00:00:30,080 --> 00:00:32,197
state and everywhere. And it's just

11
00:00:32,280 --> 00:00:35,757
exciting to be able to come back and see

12
00:00:35,840 --> 00:00:38,797
uh the progress even year on year uh

13
00:00:38,880 --> 00:00:43,157
that um I you know see
uh you know in action.

14
00:00:43,240 --> 00:00:47,757
To us, it's always about empowering
people and organizations. So, when we

15
00:00:47,840 --> 00:00:50,757
talk about even new technology um

16
00:00:50,840 --> 00:00:52,837
you know platforms, it comes down to how

17
00:00:52,920 --> 00:00:55,957
is it being really used to empower an

18
00:00:56,040 --> 00:00:58,677
institution, an organization.

19
00:00:58,760 --> 00:01:03,117
Um I'm going to talk about in fact Crime
OS that the Chief Minister was here, you

20
00:01:03,200 --> 00:01:05,517
know, really sponsored and made happen

21
00:01:05,600 --> 00:01:08,477
uh in this state so that even uh every

22
00:01:08,560 --> 00:01:10,717
citizen of this uh state can benefit.

23
00:01:10,800 --> 00:01:12,717
And that's the type of innovation we

24
00:01:12,800 --> 00:01:16,517
want to see um in any place.

25
00:01:16,600 --> 00:01:19,317
And but the key thing for us is to be

26
00:01:19,400 --> 00:01:23,837
grounded in the fact that the
frontier of what is possible

27
00:01:23,920 --> 00:01:27,597
is changing, right? I mean, that's
the exciting part for any of us is when we

28
00:01:27,680 --> 00:01:29,877
talk about technology, it allows us to

29
00:01:29,960 --> 00:01:33,397
reimagine what we can now do uh whether

30
00:01:33,480 --> 00:01:38,637
it's in our state, whether it is in our
organization, whether it's in our lives,

31
00:01:38,720 --> 00:01:41,357
right? So, the ability
to move the frontier.

32
00:01:41,440 --> 00:01:46,157
So, we always start there, right?
With the customer experience. Now, you can

33
00:01:46,240 --> 00:01:49,877
think about and reimagine it. Whether
it's the employee experience, whether

34
00:01:49,960 --> 00:01:52,557
it's the operational efficiency, uh

35
00:01:52,640 --> 00:01:57,277
whether it's the rate of innovation.
If you're a drug company, can you really

36
00:01:57,360 --> 00:01:59,277
get through clinical trials faster? Can

37
00:01:59,360 --> 00:02:02,237
we get more new drugs out there? Uh that

38
00:02:02,320 --> 00:02:06,757
to me is ultimately moving of the
frontier that we are all chasing.

39
00:02:06,840 --> 00:02:08,717
And to do that though,

40
00:02:08,800 --> 00:02:11,877
you kind of have to embrace a new

41
00:02:11,960 --> 00:02:16,397
approach even, right?
You What worked previously

42
00:02:16,480 --> 00:02:20,837
in some sense, you have to do this hard
job of unlearning and learning what's

43
00:02:20,920 --> 00:02:23,157
possible, right? Sometimes we sort of

44
00:02:23,240 --> 00:02:25,757
pattern match too much uh from what

45
00:02:25,840 --> 00:02:30,197
worked in the previous era uh may not
necessarily work. In fact, there's a new

46
00:02:30,280 --> 00:02:34,677
frontier. So, when I think about that,
it starts with that mindset, freeing

47
00:02:34,760 --> 00:02:38,357
ourselves uh that there's a new approach.
For example, in when it comes

48
00:02:38,440 --> 00:02:42,477
to AI, right? If you look at even
a building of an AI solution,

49
00:02:42,560 --> 00:02:47,637
you don't start with a spec.
You actually start with the test, right?

50
00:02:47,720 --> 00:02:50,517
I mean, when people they have fancy
new words talk people talk about

51
00:02:50,600 --> 00:02:55,437
evaluations, right? They you create
create these rubrics to do evals on a

52
00:02:55,520 --> 00:02:59,877
particular outcome, then you create
a learning system, then you hill climb on

53
00:02:59,960 --> 00:03:02,037
that learning system.

54
00:03:02,120 --> 00:03:03,757
Guess what? That's kind of literally

55
00:03:03,840 --> 00:03:07,357
inverting the process of building a an

56
00:03:07,440 --> 00:03:09,597
an information system. You start from

57
00:03:09,680 --> 00:03:12,757
the outcome, you create a robust eval,

58
00:03:12,840 --> 00:03:17,117
and then you spec versus
going left to right.

59
00:03:17,200 --> 00:03:19,677
But the key thing is that means you got

60
00:03:19,760 --> 00:03:23,357
to pick up new skills, all of us. Uh and

61
00:03:23,440 --> 00:03:27,757
and in order to pick up the new skills,
you kind of need a new tool chain that

62
00:03:27,840 --> 00:03:31,317
you are using every day. Uh and

63
00:03:31,400 --> 00:03:33,877
and to be able to then have all of your

64
00:03:33,960 --> 00:03:38,117
data that is being brought uh to harness

65
00:03:38,200 --> 00:03:42,757
this new tool chain, this new skill set
with the new mindset. That's really uh

66
00:03:42,840 --> 00:03:44,557
the way we want to approach it.

67
00:03:44,640 --> 00:03:48,877
So, to facilitate this, right?
This can't just be an abstract framework.

68
00:03:48,960 --> 00:03:52,557
This has got to be an everyday
practice for all of us.

69
00:03:52,640 --> 00:03:58,117
Um you know, it's an information
worker today or a software developer.

70
00:03:58,200 --> 00:04:00,277
Um all of us are going to be using

71
00:04:00,360 --> 00:04:02,957
essentially the same approach. To

72
00:04:03,040 --> 00:04:05,277
realize that, we've built a tech stack.

73
00:04:05,360 --> 00:04:09,917
And I want to just literally walk up
this stack or walk down this uh tech

74
00:04:10,000 --> 00:04:13,397
stack. So, those starting
with the experience layer

75
00:04:13,480 --> 00:04:15,837
where we have taken this idea that, you

76
00:04:15,920 --> 00:04:19,637
know, let's bring this AI copilot

77
00:04:19,720 --> 00:04:24,557
right where you work, right? You may
be you may be in Outlook, you may be in

78
00:04:24,640 --> 00:04:26,517
Word, you may be in Excel, you may be in

79
00:04:26,600 --> 00:04:29,037
Teams, you may whatever you are doing

80
00:04:29,120 --> 00:04:31,677
your work, how do we bring

81
00:04:31,760 --> 00:04:36,237
the copilot there?
And think of copilot though

82
00:04:36,320 --> 00:04:38,157
as not just a request response, right?

83
00:04:38,240 --> 00:04:41,117
That's what we've been doing now nearly
for 2 years, right? You you know, we

84
00:04:41,200 --> 00:04:45,757
know now that hey, we can ask it any
question, it'll come back with a better

85
00:04:45,840 --> 00:04:50,237
answer than anything that we have
previously experienced uh in let's call

86
00:04:50,320 --> 00:04:52,037
it an information retrieval search type

87
00:04:52,120 --> 00:04:55,317
experience. But now, we have agents. In

88
00:04:55,400 --> 00:04:59,557
fact, think of copilot as your browser

89
00:04:59,640 --> 00:05:01,917
for the agentic web.

90
00:05:02,000 --> 00:05:07,397
In fact, right inside a copilot, you
have very powerful agents built in. Like

91
00:05:07,480 --> 00:05:10,837
one such uh agent is your researcher.

92
00:05:10,920 --> 00:05:13,957
And researcher with computer use

93
00:05:14,040 --> 00:05:17,877
is a real breakthrough.
It's literally like having 24/7

94
00:05:17,960 --> 00:05:20,557
someone very sophisticated who is your

95
00:05:20,640 --> 00:05:23,477
research assistant who can go take any

96
00:05:23,560 --> 00:05:28,077
topic and then because they have
a computer, right? So, they just are not

97
00:05:28,160 --> 00:05:32,117
like think of it as not a single shot
I go and bring back information after

98
00:05:32,200 --> 00:05:34,797
thinking. They can the researcher can

99
00:05:34,880 --> 00:05:37,637
use a computer to do research. Right?

100
00:05:37,720 --> 00:05:41,677
So, it's not only you have a computer,
your researcher has a computer. So,

101
00:05:41,760 --> 00:05:43,877
that's the way to conceptualize. Same

102
00:05:43,960 --> 00:05:46,717
thing with analyst. So, I can load a

103
00:05:46,800 --> 00:05:51,197
whole bunch of Excel spreadsheets, you
give it to an analyst, and it's kind of

104
00:05:51,280 --> 00:05:53,997
like giving it to your data science
department, right? Or the data

105
00:05:54,080 --> 00:05:57,997
scientist. So, it's again having a 24/7

106
00:05:58,080 --> 00:06:02,197
data analyst who can
then go give you insights.

107
00:06:02,280 --> 00:06:04,997
And lastly, it's like one of the things

108
00:06:05,080 --> 00:06:07,437
that I'm very very excited about is what

109
00:06:07,520 --> 00:06:09,597
happened in software development is

110
00:06:09,680 --> 00:06:13,117
going to happen with Excel and Excel

111
00:06:13,200 --> 00:06:17,397
agent, right? So, I in fact had a chance
to participate recently in the Excel

112
00:06:17,480 --> 00:06:19,837
world championship. By the way, ESPN

113
00:06:19,920 --> 00:06:22,117
even broadcast Excel championships. We

114
00:06:22,200 --> 00:06:24,837
have an honest-to-god uh great way to

115
00:06:24,920 --> 00:06:27,077
sort of celebrate the Excel prowess. And

116
00:06:27,160 --> 00:06:29,917
so, I was able to participate in it you

117
00:06:30,000 --> 00:06:34,237
the digital challenge using this Excel
agent. It is so cool to be able to see

118
00:06:34,320 --> 00:06:36,317
and I worked on Excel all my life

119
00:06:36,400 --> 00:06:38,597
essentially. And to be able to sort of

120
00:06:38,680 --> 00:06:42,317
see how the agent mode understands

121
00:06:42,400 --> 00:06:44,997
deeply the entire semantics of a

122
00:06:45,080 --> 00:06:46,517
spreadsheet, right? So, it's not just a

123
00:06:46,600 --> 00:06:49,557
one shot I created a great model. But

124
00:06:49,640 --> 00:06:52,157
it's the ability to iterate on a pretty

125
00:06:52,240 --> 00:06:56,877
sophisticated model continuously using
this agent mode just like how we do in

126
00:06:56,960 --> 00:07:00,797
software engineering. So, it's just
exciting to see this. And also, it's

127
00:07:00,880 --> 00:07:03,557
really to me at the end of the day, all

128
00:07:03,640 --> 00:07:06,277
these tools are like the new Word,

129
00:07:06,360 --> 00:07:10,957
Excel, PowerPoint, Office, right? That's
kind of they're as simple as that. Uh

130
00:07:11,040 --> 00:07:12,797
and they're in the hands of everyone

131
00:07:12,880 --> 00:07:15,717
here. And I see it being used like that,

132
00:07:15,800 --> 00:07:18,197
right? So, no one is sort of you know

133
00:07:18,280 --> 00:07:20,117
praying at the altar of these tools.

134
00:07:20,200 --> 00:07:23,117
They're just using them. Uh and and to

135
00:07:23,200 --> 00:07:27,877
be to see the proliferation,
the diffusion is probably faster than

136
00:07:27,960 --> 00:07:32,077
anything I've seen, right? I've come
to back to India, I've come back to Mumbai

137
00:07:32,160 --> 00:07:34,437
throughout every generational shift in

138
00:07:34,520 --> 00:07:37,597
technology starting in the early '90s.

139
00:07:37,680 --> 00:07:41,317
And but when I see now the rate of pace
of change, there's no difference quite

140
00:07:41,400 --> 00:07:45,997
frankly. I could be in any city anywhere
in the world uh and to be able to see uh

141
00:07:46,080 --> 00:07:50,277
all of this uh applications just being
broadly diffused. To show you some of

142
00:07:50,360 --> 00:07:56,840
this in action, let me invite up on stage
my colleague Zoe. Zoe, take it away.

143
00:07:59,460 --> 00:08:03,317
>> [applause] >> Thanks,
Satya. I'm Zoe, a product

144
00:08:03,400 --> 00:08:08,117
marketer on the Microsoft 365 Copilot
team. And like all of you, I start my

145
00:08:08,200 --> 00:08:10,277
day commuting in and preparing for the

146
00:08:10,360 --> 00:08:15,960
day ahead. For that, I'm going
to use Copilot on my mobile.

147
00:08:16,840 --> 00:08:21,160
Hey Copilot, what's
on my agenda for today?

148
00:08:23,400 --> 00:08:25,917
You've got a few things lined up today.

149
00:08:26,000 --> 00:08:28,797
First, there's an AI Tour India meeting

150
00:08:28,880 --> 00:08:31,357
that started earlier this morning. Then

151
00:08:31,440 --> 00:08:33,757
at 11:00 a.m., there's a Contoso sales

152
00:08:33,840 --> 00:08:35,997
call. Later in the afternoon, you've got

153
00:08:36,080 --> 00:08:39,237
a team sync at 2:00 p.m.

154
00:08:39,320 --> 00:08:44,480
Thanks, Copilot.
And has my manager sent me any emails?

155
00:08:45,600 --> 00:08:49,837
Yes, you've got a few emails from Nicole.
Uh just yesterday, she reminded

156
00:08:49,920 --> 00:08:54,557
the team about a Friday mixer happening
in the common area. She also mentioned

157
00:08:54,640 --> 00:08:59,077
that finalizing the monthly business
review should be a high priority. She

158
00:08:59,160 --> 00:09:01,877
also Thanks, Copilot.

159
00:09:01,960 --> 00:09:03,757
Copilot doesn't just have my work

160
00:09:03,840 --> 00:09:06,677
context, it knows who I am, what I work

161
00:09:06,760 --> 00:09:09,757
on, and who I work with. It's why I use

162
00:09:09,840 --> 00:09:13,117
Copilot. It's a true AI assistant. Now,

163
00:09:13,200 --> 00:09:17,797
I wanted to take a look at my manager's
ask around that MBR updates for my

164
00:09:17,880 --> 00:09:19,037
V-team.

165
00:09:19,120 --> 00:09:22,237
And you'll see I'm actually using GPT

166
00:09:22,320 --> 00:09:25,117
5.2, which we released yesterday and

167
00:09:25,200 --> 00:09:27,077
it's immediately available to all

168
00:09:27,160 --> 00:09:30,477
Microsoft 365 customers. Do you Yeah,

169
00:09:30,560 --> 00:09:34,117
it's exciting. I know. >> [applause]

170
00:09:34,200 --> 00:09:37,237
>> What I love about GPT 5.2 is it's

171
00:09:37,320 --> 00:09:39,877
allowing Copilot to be even better at

172
00:09:39,960 --> 00:09:42,797
finding the most relevant resources. So,

173
00:09:42,880 --> 00:09:45,797
my responses are very contextually

174
00:09:45,880 --> 00:09:50,517
relevant and I'm able to take quick
action. And it makes a very visual

175
00:09:50,600 --> 00:09:54,277
output. Now,
only Copilot is able to combine

176
00:09:54,360 --> 00:09:59,317
this constant model innovation with your
work data to provide the most relevant

177
00:09:59,400 --> 00:10:00,997
response.

178
00:10:01,080 --> 00:10:03,597
Now, I want to go and update my existing

179
00:10:03,680 --> 00:10:06,557
monthly business review document, but I

180
00:10:06,640 --> 00:10:09,597
want to change it with my November data.

181
00:10:09,680 --> 00:10:13,597
For that, I'm going
to use agent mode in Word.

182
00:10:13,680 --> 00:10:16,597
Agent mode allows the agent to edit, to

183
00:10:16,680 --> 00:10:19,437
draft, to format the document, all with

184
00:10:19,520 --> 00:10:24,077
natural language, but it still
keeps me in complete control.

185
00:10:24,160 --> 00:10:25,957
And I can see it's starting to look

186
00:10:26,040 --> 00:10:28,477
across all of that work data, right? My

187
00:10:28,560 --> 00:10:31,277
emails, my meetings, my team chats, my

188
00:10:31,360 --> 00:10:34,717
files. Now, that normally takes me hours

189
00:10:34,800 --> 00:10:36,677
to do. I have to find the right

190
00:10:36,760 --> 00:10:39,277
information, I have to sift through it,

191
00:10:39,360 --> 00:10:43,957
but now you'll see the agent is actually
actively applying it on the left-hand

192
00:10:44,040 --> 00:10:47,157
side. And you'll see that it's actually

193
00:10:47,240 --> 00:10:51,117
leaving parts alone that haven't changed
month over month over month, like our

194
00:10:51,200 --> 00:10:54,077
mission, but rather using this to update

195
00:10:54,160 --> 00:10:56,677
everything and keep me on par with

196
00:10:56,760 --> 00:10:59,597
exactly the November updates.

197
00:10:59,680 --> 00:11:04,157
Now, last, I want to make sure I
understand our team budget for the next

198
00:11:04,240 --> 00:11:08,197
fiscal year. I'm going to again go
to Excel, and I'm going to turn on agent

199
00:11:08,280 --> 00:11:10,997
mode and ask it to get me a little bit

200
00:11:11,080 --> 00:11:13,037
more aware of what's been spent, what's

201
00:11:13,120 --> 00:11:16,117
uncommitted, what's at risk, and to bid

202
00:11:16,200 --> 00:11:18,197
build a visual dashboard so it's easy

203
00:11:18,280 --> 00:11:20,837
for me to understand my data.

204
00:11:20,920 --> 00:11:25,877
Agent mode in Excel natively speaks
Excel just like Satya has said. So, it's

205
00:11:25,960 --> 00:11:29,597
not going to create a random number that I
don't know where it came from. It will

206
00:11:29,680 --> 00:11:32,037
actually build formulas right in my

207
00:11:32,120 --> 00:11:33,717
spreadsheet, right where my budget

208
00:11:33,800 --> 00:11:36,757
already exists. And it will even create

209
00:11:36,840 --> 00:11:40,757
native objects like charts and tables,
things that I can continue to

210
00:11:40,840 --> 00:11:44,317
manipulate. It also walks
through everything it's

211
00:11:44,400 --> 00:11:48,477
doing on the right-hand side. So,
that way, if say my manager Nicole

212
00:11:48,560 --> 00:11:50,637
comes to me and asks, "How did I come to

213
00:11:50,720 --> 00:11:53,037
my conclusion?" Well, Copilot actually

214
00:11:53,120 --> 00:11:56,077
helps me understand what it did. That

215
00:11:56,160 --> 00:11:59,037
way, I can continue to work with it.

216
00:11:59,120 --> 00:12:02,597
I can also ask agent mode to do simple
things like conditional formatting,

217
00:12:02,680 --> 00:12:04,557
which you'll see, highlighting things in

218
00:12:04,640 --> 00:12:07,277
red that maybe are at risk. All of these

219
00:12:07,360 --> 00:12:09,837
things to continue to work with Copilot

220
00:12:09,920 --> 00:12:12,277
and agent mode in order to get a better

221
00:12:12,360 --> 00:12:14,637
understanding. And you can see it's

222
00:12:14,720 --> 00:12:19,317
created this dashboard live. I can If I
click in, I can see these rich formulas

223
00:12:19,400 --> 00:12:22,997
it's actually built to give me an answer.
The conditional formatting to

224
00:12:23,080 --> 00:12:25,717
highlight what's at risk in red. And a

225
00:12:25,800 --> 00:12:28,597
native graph that actually I'm able to

226
00:12:28,680 --> 00:12:31,437
manipulate even further, should I want.

227
00:12:31,520 --> 00:12:33,597
Now, Copilot, it hasn't just made me

228
00:12:33,680 --> 00:12:35,597
better and faster at my job, it's helped

229
00:12:35,680 --> 00:12:38,797
me make more impact. And I can't imagine

230
00:12:38,880 --> 00:12:43,280
working without it. Back to you, Satya.

231
00:12:43,740 --> 00:12:45,760
>> [applause]

232
00:12:47,400 --> 00:12:51,637
>> Thank you so much, Zoe. Whenever I
see that Excel demo, I like to think about

233
00:12:51,720 --> 00:12:55,717
it that, you know, when I had hair,
I loved Excel, and when I don't have hair,

234
00:12:55,800 --> 00:13:00,597
I love Excel. Uh, I'll tell you,
it's one of those generational things that

235
00:13:00,680 --> 00:13:02,637
just gets better and better in spite of

236
00:13:02,720 --> 00:13:05,517
my follicle challenges. Um,

237
00:13:05,600 --> 00:13:09,757
you know, to me, uh, if you
sort of take that experience

238
00:13:09,840 --> 00:13:14,117
layer, one of the things
that's super important

239
00:13:14,200 --> 00:13:18,677
in the age of AI is to think about
your data as one of the most strategic

240
00:13:18,760 --> 00:13:22,797
assets. But you got to bring that data

241
00:13:22,880 --> 00:13:26,917
contextually to the AI.
In fact, people talk about it

242
00:13:27,000 --> 00:13:29,677
as context engineering. Uh, and I think

243
00:13:29,760 --> 00:13:31,797
it's a good term for it because in fact,

244
00:13:31,880 --> 00:13:35,277
a lot of what needs to be done,

245
00:13:35,360 --> 00:13:37,517
both at this engineering level and even,

246
00:13:37,600 --> 00:13:39,197
I would say, business decision-making

247
00:13:39,280 --> 00:13:42,677
level, is to think about not the silos

248
00:13:42,760 --> 00:13:45,997
of the previous generation, but the

249
00:13:46,080 --> 00:13:49,597
contextual way you can feed intelligence

250
00:13:49,680 --> 00:13:53,877
with data. So, to that end,
for example, underneath

251
00:13:53,960 --> 00:13:55,717
Zoe's demo,

252
00:13:55,800 --> 00:13:59,397
is perhaps the most important database

253
00:13:59,480 --> 00:14:01,717
in any company that uses Microsoft 365

254
00:14:01,800 --> 00:14:03,917
today, right? If you think about it, the

255
00:14:04,000 --> 00:14:06,957
tacit knowledge inside any corporation,

256
00:14:07,040 --> 00:14:10,037
any organization, is the knowledge of

257
00:14:10,120 --> 00:14:12,397
people, their relationships with other

258
00:14:12,480 --> 00:14:15,197
people, their work artifacts, whether

259
00:14:15,280 --> 00:14:18,237
they're spreadsheets, documents, their

260
00:14:18,320 --> 00:14:21,517
conversations in email, meetings in

261
00:14:21,600 --> 00:14:24,957
Teams. To be able to relate all that,

262
00:14:25,040 --> 00:14:29,557
because that's where the tacit knowledge
of an organization is. Uh, and it's in

263
00:14:29,640 --> 00:14:33,477
that database. And so, fundamentally,
what we have transformed that is into

264
00:14:33,560 --> 00:14:36,437
work IQ, right? So, when we say work IQ,

265
00:14:36,520 --> 00:14:41,517
it's the stateful part underneath M365

266
00:14:41,600 --> 00:14:44,117
that now can be brought to bear in the

267
00:14:44,200 --> 00:14:46,877
context of any AI solution you want to

268
00:14:46,960 --> 00:14:49,197
build. Not just Copilot, any agent you

269
00:14:49,280 --> 00:14:51,997
build out there, you can now use work

270
00:14:52,080 --> 00:14:56,277
IQ. Think about that.
The second thing we've done is, in fact,

271
00:14:56,360 --> 00:14:58,557
one of the other ubiquitous tools all

272
00:14:58,640 --> 00:15:01,797
over is Excel and Power BI.

273
00:15:01,880 --> 00:15:05,157
So, that means you've taken real great

274
00:15:05,240 --> 00:15:07,757
pain over decades building these

275
00:15:07,840 --> 00:15:10,917
semantic models to do analysis.

276
00:15:11,000 --> 00:15:15,197
And now, with Fabric and Fabric IQ,

277
00:15:15,280 --> 00:15:17,637
you can take all of that data, along

278
00:15:17,720 --> 00:15:22,757
with all the operational data,
and make it available to your AI.

279
00:15:22,840 --> 00:15:26,197
And then the last one sort
of really makes it all come

280
00:15:26,280 --> 00:15:31,117
together. So, what we've done with
Foundry IQ is said what people describe

281
00:15:31,200 --> 00:15:34,117
as retrieval augmented generation,
right? So, one of the biggest things

282
00:15:34,200 --> 00:15:35,357
that we have done in the last couple of

283
00:15:35,440 --> 00:15:37,437
years as we build these AI solutions is

284
00:15:37,520 --> 00:15:41,197
to do retrieval of data. And we've now

285
00:15:41,280 --> 00:15:45,237
built the next generation retrieval
system, essentially. It does more than

286
00:15:45,320 --> 00:15:48,037
retrieval. It in fact plans

287
00:15:48,120 --> 00:15:51,117
and executes and reasons that plan all

288
00:15:51,200 --> 00:15:54,517
that plan on how to retrieve relevant

289
00:15:54,600 --> 00:15:59,477
data from work IQ, Fabric IQ, and any

290
00:15:59,560 --> 00:16:01,757
other blob storage, unstructured

291
00:16:01,840 --> 00:16:04,397
storage, or what have you, and bring it

292
00:16:04,480 --> 00:16:06,917
in the context of your AI application.

293
00:16:07,000 --> 00:16:09,117
So, that's what this IQ layer is. In

294
00:16:09,200 --> 00:16:10,877
fact, it's probably one of the more

295
00:16:10,960 --> 00:16:14,797
salient, yet important pieces on how

296
00:16:14,880 --> 00:16:17,157
anyone who builds any AI solution going

297
00:16:17,240 --> 00:16:19,517
forward will be more successful, right?

298
00:16:19,600 --> 00:16:22,277
In fact, direct correlation between the

299
00:16:22,360 --> 00:16:25,037
evaluations or the sort of performance

300
00:16:25,120 --> 00:16:30,037
of your models will be how good is your
data being brought. Just because in some

301
00:16:30,120 --> 00:16:32,717
sense, models are becoming, quite
frankly, a commodity, right? So, there

302
00:16:32,800 --> 00:16:36,317
is lots and lots of models out there
which are very, very capable. The

303
00:16:36,400 --> 00:16:40,477
question is, how do you marshal that
capability? I think it'll come down to

304
00:16:40,560 --> 00:16:44,237
having a great experience layer where
people can discover the agents you're

305
00:16:44,320 --> 00:16:46,917
building, and having a fantastic sort of

306
00:16:47,000 --> 00:16:49,837
capability around context engineering.

307
00:16:49,920 --> 00:16:53,037
So, the next layer for us

308
00:16:53,120 --> 00:16:57,677
uh, is really about now that we have
an experience layer, we have the data set

309
00:16:57,760 --> 00:17:01,637
up, let's build agents, right?
And the beauty of building agents, quite

310
00:17:01,720 --> 00:17:05,717
frankly, is like building spreadsheets
or writing a document. It's not more

311
00:17:05,800 --> 00:17:07,717
mystical than that. In fact, agent

312
00:17:07,800 --> 00:17:11,077
builder that's built right into Copilot

313
00:17:11,160 --> 00:17:13,837
is like saying, "Hey, I doc new," right?

314
00:17:13,920 --> 00:17:16,797
Or spreadsheet new. Uh, and so we start

315
00:17:16,880 --> 00:17:21,157
there. In fact, now we have an app
builder. So, in fact, Bill's big thing

316
00:17:21,240 --> 00:17:23,757
always, for the from ever since I joined

317
00:17:23,840 --> 00:17:28,677
Microsoft in the early '90s, was, "Why
the heck do I have a difference between

318
00:17:28,760 --> 00:17:32,517
a document, a website, and an
application?" Like, in you know, the his

319
00:17:32,600 --> 00:17:36,357
concept was there should just be one
thing, and it should I should be able to

320
00:17:36,440 --> 00:17:40,197
transform. And guess what? We finally
have it. Uh, which is app builder can

321
00:17:40,280 --> 00:17:44,037
you can go in and just specify the app
you want to build, and it'll build it.

322
00:17:44,120 --> 00:17:47,117
It's kind of like a document, it's kind
of like a website, it's kind of like an

323
00:17:47,200 --> 00:17:49,357
app, right? So, that's the power. Then

324
00:17:49,440 --> 00:17:54,477
Copilot Studio is another way to think
of it as build out of your agent. So,

325
00:17:54,560 --> 00:17:58,157
again, as a natural you're using
a natural language prompt, you can go

326
00:17:58,240 --> 00:17:59,957
build an agent, you can ground it in

327
00:18:00,040 --> 00:18:02,237
some data, uh, and you there you have

328
00:18:02,320 --> 00:18:05,277
it. You have an agent. Um, and so we're

329
00:18:05,360 --> 00:18:08,237
building out Foundry, in fact, without

330
00:18:08,320 --> 00:18:12,517
any cliffs to be able to then say,
"Okay, we now have agents. Can I even

331
00:18:12,600 --> 00:18:17,437
orchestrate multiple agents and build
fully agentic systems?" Microsoft

332
00:18:17,520 --> 00:18:20,917
Foundry allows you to do that as well,
because the multi-agent frameworks are

333
00:18:21,000 --> 00:18:25,357
there. All of the tools, uh,
that you need in order to be able to build

334
00:18:25,440 --> 00:18:27,717
multi-agentic systems are there.

335
00:18:27,800 --> 00:18:32,477
We also now, of course,
have all these models. We have 11,000

336
00:18:32,560 --> 00:18:36,557
models. You say, "Why Why do I need 11,000

337
00:18:36,640 --> 00:18:40,237
models?" In fact,
even for me this week, it is a

338
00:18:40,320 --> 00:18:43,077
great reminder of why having this

339
00:18:43,160 --> 00:18:45,797
diversity of models is going to really

340
00:18:45,880 --> 00:18:48,477
make this world a better place. In fact,

341
00:18:48,560 --> 00:18:51,317
we launched, uh, and we announced, uh,

342
00:18:51,400 --> 00:18:54,477
in Foundry, uh, as an open weight, uh,

343
00:18:54,560 --> 00:18:57,877
model, uh, a model called Giga Time. It

344
00:18:57,960 --> 00:19:02,317
There was an article published in Cell,
and quite frankly, it gave me goosebumps

345
00:19:02,400 --> 00:19:05,837
when I sort of read about
it and learned about it,

346
00:19:05,920 --> 00:19:09,557
uh, because it's work Microsoft did
with both the University of Washington as

347
00:19:09,640 --> 00:19:12,557
well as Providence Hospitals. Um, and

348
00:19:12,640 --> 00:19:15,117
what it does is, if you think about the

349
00:19:15,200 --> 00:19:18,477
challenge of someone with cancer,

350
00:19:18,560 --> 00:19:21,237
um, today, when you sort of try to

351
00:19:21,320 --> 00:19:23,117
understand whether immunotherapy will

352
00:19:23,200 --> 00:19:26,597
work, you kind of have to model the

353
00:19:26,680 --> 00:19:30,477
immune system, essentially.
And so, there's a test for that.

354
00:19:30,560 --> 00:19:33,117
The problem with that test is,

355
00:19:33,200 --> 00:19:35,917
uh, it's a complex test, it takes a long

356
00:19:36,000 --> 00:19:38,917
time, and it costs a lot of money.

357
00:19:39,000 --> 00:19:42,397
And so, what this particular model does

358
00:19:42,480 --> 00:19:46,637
is, it does a simulation of the test.

359
00:19:46,720 --> 00:19:49,117
So, you can literally take a simple

360
00:19:49,200 --> 00:19:54,597
pathology of the tumor, right? So,
literally like a microscope image of a a

361
00:19:54,680 --> 00:19:59,157
tumor. And then use
essentially GenAI to fill

362
00:19:59,240 --> 00:20:03,117
in. So, it will do the simulation

363
00:20:03,200 --> 00:20:05,797
of what is the

364
00:20:05,880 --> 00:20:08,437
basically spatial proteomics of the

365
00:20:08,520 --> 00:20:12,397
immune system and say, you now can have

366
00:20:12,480 --> 00:20:16,517
a test. So, I was thinking about it,
right? With that model,

367
00:20:16,600 --> 00:20:21,157
that means some entrepreneur
in Mumbai is going to pick it up

368
00:20:21,240 --> 00:20:22,437
and say, you know what? I'm going to

369
00:20:22,520 --> 00:20:27,077
build an entire sort of simulation lab

370
00:20:27,160 --> 00:20:28,837
in silico

371
00:20:28,920 --> 00:20:31,397
and make it available to every city in

372
00:20:31,480 --> 00:20:34,277
Maharashtra and in this country, right?

373
00:20:34,360 --> 00:20:37,397
So, every hospital can start doing what

374
00:20:37,480 --> 00:20:41,597
now today is done in the few tier one
hospitals and cost a lot of money. I

375
00:20:41,680 --> 00:20:44,997
mean, that's in some sense, when you're
playing the game, sometimes you forget

376
00:20:45,080 --> 00:20:48,797
why you're playing it. This is why we
are playing it. We are playing it so

377
00:20:48,880 --> 00:20:51,637
that we can actually use this innovation

378
00:20:51,720 --> 00:20:53,877
to make a real difference uh in the

379
00:20:53,960 --> 00:20:56,597
world. So, it's exciting to see uh how

380
00:20:56,680 --> 00:20:59,357
this is all coming together. Now, we're

381
00:20:59,440 --> 00:21:01,357
bringing, as I said, all this together

382
00:21:01,440 --> 00:21:03,797
in this tool chain. This tool chain is

383
00:21:03,880 --> 00:21:06,757
going to become what you will use. Uh in

384
00:21:06,840 --> 00:21:11,597
fact, GitHub is where you would start
interfacing with all of this app stack,

385
00:21:11,680 --> 00:21:13,037
right? So, if you think of the new app

386
00:21:13,120 --> 00:21:15,877
server as Foundry, uh the tools are

387
00:21:15,960 --> 00:21:19,077
Visual Studio and VS Code and GitHub.

388
00:21:19,160 --> 00:21:23,757
And and it Talking about GitHub,
it's really exciting to see the growth

389
00:21:23,840 --> 00:21:28,517
of GitHub in India. It's I mean, it's
tremendous. By 2030, India is expected

390
00:21:28,600 --> 00:21:34,350
to be the number one country in terms
of GitHub participation. [applause]

391
00:21:35,040 --> 00:21:36,837
You know, it's it's tremendous to see

392
00:21:36,920 --> 00:21:38,877
not only just the participation, but the

393
00:21:38,960 --> 00:21:42,317
but the ambition level of the projects

394
00:21:42,400 --> 00:21:46,157
uh coming out of India are just
continuing to improve and, you know, get

395
00:21:46,240 --> 00:21:48,717
better. And so,
it's fantastic to see that.

396
00:21:48,800 --> 00:21:53,557
Now, we're building on GitHub, by the way,
as Agent HQ. So, what I mean by

397
00:21:53,640 --> 00:21:55,437
that is you go to GitHub and you will

398
00:21:55,520 --> 00:21:58,117
have access to all the coding models

399
00:21:58,200 --> 00:22:03,197
that you can use in the context of your
repo. Uh and so, that's the idea. It's

400
00:22:03,280 --> 00:22:05,117
not about sort of having one model, but

401
00:22:05,200 --> 00:22:10,397
you have your code repo, just like Work
IQ, think of this as Code IQ, right? So,

402
00:22:10,480 --> 00:22:14,277
that's kind of really it. Like,
your code is your most important resource.

403
00:22:14,360 --> 00:22:18,757
Yeah, you want that. And then you want
many many models over the years to sort

404
00:22:18,840 --> 00:22:23,357
of keep working and creating more
of that repo uh and the repo features. And

405
00:22:23,440 --> 00:22:27,157
so, that's what And by the way, the form
factors, right? If people say, oh, wow,

406
00:22:27,240 --> 00:22:30,037
will I only use autonomous agents? Will

407
00:22:30,120 --> 00:22:35,237
I use only an IDE with agent mode? Will
I use a command line? Guess what? I use

408
00:22:35,320 --> 00:22:36,997
all three all day long, right? I mean,

409
00:22:37,080 --> 00:22:39,597
that's it. Like, the form factors

410
00:22:39,680 --> 00:22:42,477
uh of how I interface with AI are going

411
00:22:42,560 --> 00:22:45,477
to remain essentially always uh some

412
00:22:45,560 --> 00:22:50,277
IDE, some uh CLI interface, uh and a a

413
00:22:50,360 --> 00:22:53,637
new chat interface, if you will. And
that I think is the way you'll interact

414
00:22:53,720 --> 00:22:58,277
with it. But to show you all of this in
its full glory, I wanted to invite up on

415
00:22:58,360 --> 00:23:02,637
stage my colleague Karan.
Karan, take it away.

416
00:23:02,720 --> 00:23:06,117
>> [applause] >> Thanks, Satya.

417
00:23:06,200 --> 00:23:08,677
A few weeks ago at GitHub Universe, we

418
00:23:08,760 --> 00:23:11,757
announced Agent HQ, which is an agentic

419
00:23:11,840 --> 00:23:16,557
platform that unites every agent, making
them native to the GitHub flow you're

420
00:23:16,640 --> 00:23:18,477
already familiar with. So, let me show

421
00:23:18,560 --> 00:23:21,117
you a demo of what that looks like.

422
00:23:21,200 --> 00:23:26,117
All right. So, I have a demo app here
that's a storefront for some of the

423
00:23:26,200 --> 00:23:30,477
nostalgic toys, which might have
been common in many of our households.

424
00:23:30,560 --> 00:23:33,117
However, as you can see, the app doesn't

425
00:23:33,200 --> 00:23:35,757
really have any search or filter options

426
00:23:35,840 --> 00:23:38,357
and I want to add that functionality.

427
00:23:38,440 --> 00:23:41,077
So, I have my code repo here on GitHub.

428
00:23:41,160 --> 00:23:43,477
With the new agents panel, I can now

429
00:23:43,560 --> 00:23:46,277
kick off an agentic task from anywhere

430
00:23:46,360 --> 00:23:49,117
on GitHub by just describing the task

431
00:23:49,200 --> 00:23:53,917
and delegating it to the Copilot coding
agent. So, Copilot starts working on it

432
00:23:54,000 --> 00:23:55,557
in the background.

433
00:23:55,640 --> 00:23:57,717
I can assign a task to Copilot from

434
00:23:57,800 --> 00:24:01,277
anywhere I work on github.com, in VS

435
00:24:01,360 --> 00:24:05,717
Code, in the CLI, or even on my mobile.

436
00:24:05,800 --> 00:24:07,517
All right. While working across multiple

437
00:24:07,600 --> 00:24:10,757
repositories and tasks, I have delegated

438
00:24:10,840 --> 00:24:13,597
many of these tasks to Copilot and I

439
00:24:13,680 --> 00:24:15,757
want to know what's going on with it.

440
00:24:15,840 --> 00:24:18,677
So, now I can see all of my agent

441
00:24:18,760 --> 00:24:21,077
sessions in a single mission control

442
00:24:21,160 --> 00:24:25,717
view, both ongoing as well
as completed ones over here.

443
00:24:25,800 --> 00:24:27,597
Not just that, I can also start off a

444
00:24:27,680 --> 00:24:30,837
task from right here while using custom

445
00:24:30,920 --> 00:24:33,797
agents that I have created or have been

446
00:24:33,880 --> 00:24:36,637
built by partners. So, these custom

447
00:24:36,720 --> 00:24:39,757
agents are domain experts, which extend

448
00:24:39,840 --> 00:24:44,637
the Copilot coding agent across
all of your tools and workflows.

449
00:24:44,720 --> 00:24:48,837
All right. So, let's dive into the uh
session that we just kicked off and see

450
00:24:48,920 --> 00:24:51,437
what Copilot is working on. So, I can

451
00:24:51,520 --> 00:24:53,717
see the progress of Copilot here on the

452
00:24:53,800 --> 00:24:57,477
left and also the draft PR on the right.

453
00:24:57,560 --> 00:25:00,317
So, I'm a developer, so which means I am

454
00:25:00,400 --> 00:25:04,597
still in control. So, I can even steer a

455
00:25:04,680 --> 00:25:08,397
ongoing session midway to have Copilot

456
00:25:08,480 --> 00:25:12,037
work in a slightly different direction.

457
00:25:12,120 --> 00:25:15,917
So, this will take a few minutes,
so I'll let Copilot continue working on

458
00:25:16,000 --> 00:25:17,797
that in the background, but I want to

459
00:25:17,880 --> 00:25:20,477
show you one of the uh agent sessions I

460
00:25:20,560 --> 00:25:23,157
kicked off earlier for the same feature.

461
00:25:23,240 --> 00:25:25,077
So, I can see it's completed and then

462
00:25:25,160 --> 00:25:28,157
there's a summary as well. And oh, nice

463
00:25:28,240 --> 00:25:30,317
screenshots as well, thanks to the

464
00:25:30,400 --> 00:25:33,437
Playwright uh MCP. I can also take a

465
00:25:33,520 --> 00:25:36,477
look at uh the files it has changed over

466
00:25:36,560 --> 00:25:40,037
here and also ask it for any change, if

467
00:25:40,120 --> 00:25:42,237
I need, right over here.

468
00:25:42,320 --> 00:25:44,517
All right. Amidst all of this, if I want

469
00:25:44,600 --> 00:25:47,317
to have uh you know, a chat with Copilot

470
00:25:47,400 --> 00:25:49,957
to ask it something or so, I can jump

471
00:25:50,040 --> 00:25:52,717
into Copilot chat right here as well.

472
00:25:52,800 --> 00:25:57,037
So, Copilot provides me a choice
of various different models from different

473
00:25:57,120 --> 00:26:01,077
providers, including GPT-5.2,

474
00:26:01,160 --> 00:26:05,997
which just launched last night and already
available on GitHub Copilot. See

475
00:26:06,080 --> 00:26:08,517
how fast it is. And

476
00:26:08,600 --> 00:26:11,077
yeah. All right. Not just that, I can

477
00:26:11,160 --> 00:26:14,037
also bring in my own models from

478
00:26:14,120 --> 00:26:17,637
Microsoft Foundry. So, I have a few

479
00:26:17,720 --> 00:26:20,037
models that I've deployed on Microsoft

480
00:26:20,120 --> 00:26:22,797
Foundry, including a couple of them that

481
00:26:22,880 --> 00:26:25,597
I had fine-tuned myself and I have

482
00:26:25,680 --> 00:26:27,837
brought in all of these models to

483
00:26:27,920 --> 00:26:30,557
Copilot by connecting them to my

484
00:26:30,640 --> 00:26:33,117
deployments in Foundry.

485
00:26:33,200 --> 00:26:36,597
So, now I can actually use all of these

486
00:26:36,680 --> 00:26:39,837
models directly within Copilot, be it on

487
00:26:39,920 --> 00:26:41,917
GitHub or VS Code.

488
00:26:42,000 --> 00:26:44,517
Now, I was just doing something and I

489
00:26:44,600 --> 00:26:46,877
fine-tuned one of the model

490
00:26:46,960 --> 00:26:50,077
uh just for fun to see if I can converse

491
00:26:50,160 --> 00:26:53,077
almost like a developer, you know, in

492
00:26:53,160 --> 00:26:57,957
Mumbai. So, I just asked it something
simple saying, what is Kubernetes? Your

493
00:26:58,040 --> 00:27:02,477
pods, nodes, kuch samajh
mein nahi aata hai. And this

494
00:27:02,560 --> 00:27:06,757
is coming and it says, simple language
mein samjho. This is what containers

495
00:27:06,840 --> 00:27:11,797
are. This is what Kubernetes is.
And it's so awesome, right? Because this

496
00:27:11,880 --> 00:27:14,437
Yeah. This

497
00:27:14,520 --> 00:27:17,517
is coming from my own model that's

498
00:27:17,600 --> 00:27:19,517
deployed on Foundry, which I can use in

499
00:27:19,600 --> 00:27:22,277
Copilot, right? That's pretty awesome.

500
00:27:22,360 --> 00:27:24,557
All right. So, speaking of deployments

501
00:27:24,640 --> 00:27:29,837
and everything, um I had also created
a PR to implement a new product detail

502
00:27:29,920 --> 00:27:33,237
page uh you know,
sometime earlier and I can

503
00:27:33,320 --> 00:27:35,957
see that Copilot has already reviewed my

504
00:27:36,040 --> 00:27:39,597
PR and also left a few comments on it.

505
00:27:39,680 --> 00:27:41,517
So, let's take a look at some of the

506
00:27:41,600 --> 00:27:44,597
suggestions it has left, like this one

507
00:27:44,680 --> 00:27:47,237
here, which speaks about uh DB

508
00:27:47,320 --> 00:27:49,717
connection pool exhaustion or, you know,

509
00:27:49,800 --> 00:27:52,477
this one uh right here, which speaks

510
00:27:52,560 --> 00:27:54,717
about type safety.

511
00:27:54,800 --> 00:27:58,197
So, this is the new agentic Copilot code

512
00:27:58,280 --> 00:28:01,757
review, which combines LLM detections,

513
00:28:01,840 --> 00:28:06,277
tool calling, and also deterministic
scans that help you with smarter

514
00:28:06,360 --> 00:28:09,837
reviews. Not just that, I can actually ask

515
00:28:09,920 --> 00:28:12,837
Copilot to help me implement all of

516
00:28:12,920 --> 00:28:18,077
these suggestions in a single PR, which
will again kick off an agentic task for

517
00:28:18,160 --> 00:28:21,397
me. All right. You all know how important

518
00:28:21,480 --> 00:28:24,477
security is, right? So, thankfully, I'm

519
00:28:24,560 --> 00:28:26,917
also a security admin for this account

520
00:28:27,000 --> 00:28:30,197
on GitHub. So, with the new integration

521
00:28:30,280 --> 00:28:33,277
between Microsoft Defender for Cloud and

522
00:28:33,360 --> 00:28:35,917
GitHub Advanced Security, I can now

523
00:28:36,000 --> 00:28:38,037
specifically filter

524
00:28:38,120 --> 00:28:41,597
for runtime risk issues in production

525
00:28:41,680 --> 00:28:44,277
that have been identified by Defender.

526
00:28:44,360 --> 00:28:47,077
All right, right here. So, that's really

527
00:28:47,160 --> 00:28:49,917
nice and a campaign to address all of

528
00:28:50,000 --> 00:28:53,717
these alerts was created and oops, I see

529
00:28:53,800 --> 00:28:56,797
some of them in my repo as well, but not

530
00:28:56,880 --> 00:28:59,597
to worry because Copilot can also help

531
00:28:59,680 --> 00:29:02,877
me fix these security issues. So, I can

532
00:29:02,960 --> 00:29:06,517
just select all of these and then say,

533
00:29:06,600 --> 00:29:09,037
assign to Copilot. And then Copilot will

534
00:29:09,120 --> 00:29:12,397
go off and then create a PR to address

535
00:29:12,480 --> 00:29:15,157
all of these security issues.

536
00:29:15,240 --> 00:29:17,597
All right. You know, so now we have

537
00:29:17,680 --> 00:29:19,757
multiple different agents working on

538
00:29:19,840 --> 00:29:22,517
different tasks in the background, but I

539
00:29:22,600 --> 00:29:25,157
want to write some code myself as well.

540
00:29:25,240 --> 00:29:27,757
So, here is uh the agent session for the

541
00:29:27,840 --> 00:29:30,037
search and filter function that we were

542
00:29:30,120 --> 00:29:32,917
looking at earlier and um I want to make

543
00:29:33,000 --> 00:29:36,237
some changes to it. So, I can continue

544
00:29:36,320 --> 00:29:38,717
iterating with this by opening this

545
00:29:38,800 --> 00:29:42,597
directly in VS Code, which will open up

546
00:29:42,680 --> 00:29:45,197
the session for me. So, I can either

547
00:29:45,280 --> 00:29:48,237
check out the branch or even apply the

548
00:29:48,320 --> 00:29:51,197
changes directly in my current branch

549
00:29:51,280 --> 00:29:53,637
and I can continue working. So, I will

550
00:29:53,720 --> 00:29:56,997
open up the familiar chat interface here

551
00:29:57,080 --> 00:30:01,757
in VS code and I'll
switch to my agent mode

552
00:30:01,840 --> 00:30:03,757
probably use one of the models I

553
00:30:03,840 --> 00:30:06,917
deployed from foundry and I will ask it

554
00:30:07,000 --> 00:30:09,837
to make one small change, which is I

555
00:30:09,920 --> 00:30:14,437
also want an auto complete
feature for the search bar as well.

556
00:30:14,520 --> 00:30:16,357
So, this can continue working in the

557
00:30:16,440 --> 00:30:19,437
background and you would have seen that

558
00:30:19,520 --> 00:30:23,717
just like on github.com, I also have a

559
00:30:23,800 --> 00:30:27,477
mission control view in VS code as well,

560
00:30:27,560 --> 00:30:30,077
where I can see all of my local agents,

561
00:30:30,160 --> 00:30:35,357
my cloud agents, my background
agents, which is running as well.

562
00:30:35,440 --> 00:30:39,197
All right. So, I want
to implement a new feature.

563
00:30:39,280 --> 00:30:41,477
The toys are great, but I want to make a

564
00:30:41,560 --> 00:30:44,237
bundle of it and almost gift them as a

565
00:30:44,320 --> 00:30:46,437
play box. I know many of you would love

566
00:30:46,520 --> 00:30:48,917
that as well, right? But, I need to plan

567
00:30:49,000 --> 00:30:51,837
how to go about implementing it. So,

568
00:30:51,920 --> 00:30:54,997
with the new plan agent in VS code, I

569
00:30:55,080 --> 00:30:57,917
can ask Copilot to create an

570
00:30:58,000 --> 00:31:00,357
implementation plan and also keep

571
00:31:00,440 --> 00:31:03,797
iterating with it until I am confident

572
00:31:03,880 --> 00:31:07,477
and ready with it.
So, I had created one of

573
00:31:07,560 --> 00:31:11,957
you know, this plan earlier as well,
which I'll just open up right over here.

574
00:31:12,040 --> 00:31:14,357
So, you can see it gives me a detailed

575
00:31:14,440 --> 00:31:16,517
plan for all the steps that needs to be

576
00:31:16,600 --> 00:31:19,717
accomplished. So, whenever I'm ready, I

577
00:31:19,800 --> 00:31:23,837
can kind of continue coding this or I

578
00:31:23,920 --> 00:31:26,877
can hand it off to an agent as well to

579
00:31:26,960 --> 00:31:30,037
work on it in the background. So, I can

580
00:31:30,120 --> 00:31:33,277
hand it off to the CLI agent and I will

581
00:31:33,360 --> 00:31:35,957
say include the changes. That's fine and

582
00:31:36,040 --> 00:31:39,477
this will start working on this agent in

583
00:31:39,560 --> 00:31:43,277
the background. All right.
So, some of this will take a

584
00:31:43,360 --> 00:31:48,037
bit of time to complete. So, meanwhile,
if when all of this task would have been

585
00:31:48,120 --> 00:31:51,557
completed, let's see how
our app would look like.

586
00:31:51,640 --> 00:31:54,917
So, it would look like something like
that with all of the filters that are

587
00:31:55,000 --> 00:32:00,157
there and also my search and auto
complete feature and I can go start

588
00:32:00,240 --> 00:32:04,327
playing my brick game. All right.

589
00:32:04,410 --> 00:32:07,717
>> [applause] >> So,

590
00:32:07,800 --> 00:32:10,117
in the past few minutes, we went from so

591
00:32:10,200 --> 00:32:13,797
many different ideas, planned them and

592
00:32:13,880 --> 00:32:16,757
also coded them using Copilot coding

593
00:32:16,840 --> 00:32:20,317
agents in the cloud on github.com, in VS

594
00:32:20,400 --> 00:32:23,677
code, in background using CLI agents,

595
00:32:23,760 --> 00:32:26,357
had that code reviewed with the Copilot

596
00:32:26,440 --> 00:32:29,557
code review agent and also implemented

597
00:32:29,640 --> 00:32:33,157
the security fixes using Copilot auto

598
00:32:33,240 --> 00:32:36,597
fix agent. All the while having

599
00:32:36,680 --> 00:32:38,917
ability to look at all of my agent

600
00:32:39,000 --> 00:32:41,677
sessions, steer them in the single

601
00:32:41,760 --> 00:32:44,917
mission control view on GitHub or even

602
00:32:45,000 --> 00:32:48,517
in VS code. And that is Agent HQ. Thank

603
00:32:48,600 --> 00:32:52,280
you. Back to you, Satya. >> [applause]

604
00:32:54,280 --> 00:32:58,237
>> Thank you so much, Karan. And so,
you know, when you look at all of these

605
00:32:58,320 --> 00:33:00,557
tools that Karan showed, you know, what

606
00:33:00,640 --> 00:33:02,477
does it do? It inspires you. So,

607
00:33:02,560 --> 00:33:05,277
recently we had Thanksgiving in the

608
00:33:05,360 --> 00:33:07,837
United States. So, we had some time off.

609
00:33:07,920 --> 00:33:12,637
And so, what does one do when you have
time off? You build. And so, I built an

610
00:33:12,720 --> 00:33:15,717
app. And so,
this is by the way, this is my

611
00:33:15,800 --> 00:33:19,157
regular laptop you are seeing. So, this

612
00:33:19,240 --> 00:33:22,917
is um in fact, my Azure subscription.

613
00:33:23,000 --> 00:33:24,317
And

614
00:33:24,400 --> 00:33:27,597
this is in fact my app that's deployed

615
00:33:27,680 --> 00:33:31,517
right now, I think in in Canada Central.

616
00:33:31,600 --> 00:33:33,717
This is my repo. This is my GitHub repo.

617
00:33:33,800 --> 00:33:37,197
So, this is my code.

618
00:33:37,280 --> 00:33:39,157
In fact, the way my standard sort of way

619
00:33:39,240 --> 00:33:43,197
I work is I have a Windows 365

620
00:33:43,280 --> 00:33:46,957
setup. So, it travels with me everywhere.
And then, of course, I use

621
00:33:47,040 --> 00:33:50,677
GitHub. And then I use code spaces.
So, so this is I think it'll show up with

622
00:33:50,760 --> 00:33:54,197
the code space here. So, it's kind
of like turtles all the way down, right?

623
00:33:54,280 --> 00:33:57,837
So, I have my code space running
on GitHub, running on Windows 365. So,

624
00:33:57,920 --> 00:34:00,837
that's kind of my virtual
environment that I travel with.

625
00:34:00,920 --> 00:34:04,677
And of course, like as Karan
was showing, every morning I get up

626
00:34:04,760 --> 00:34:08,197
and my favorite place to go is here,
right? Which is I go to my repo and I go

627
00:34:08,280 --> 00:34:12,317
to my you know, Copilot
and start assigning it

628
00:34:12,400 --> 00:34:17,637
tasks. Like every day there's some
multiple agents I kick off. Like this

629
00:34:17,720 --> 00:34:21,317
morning I said, "Please upgrade this
entire app to GPT-52." Right? So, and

630
00:34:21,400 --> 00:34:24,277
it's actually done and when I'm done
with my keynote and I'm on a flight

631
00:34:24,360 --> 00:34:26,437
back, I'll kind of get there

632
00:34:26,520 --> 00:34:28,877
on my flight Wi-Fi and be able to

633
00:34:28,960 --> 00:34:31,957
actually look at the PR and push it.

634
00:34:32,040 --> 00:34:35,477
If you will. In fact, I'll test it
once with my code spaces and then I'll

635
00:34:35,560 --> 00:34:38,237
commit. But, that's in fact, it shows

636
00:34:38,320 --> 00:34:39,437
you

637
00:34:39,520 --> 00:34:43,517
the power of how you can literally start

638
00:34:43,600 --> 00:34:46,077
really taking simple tasks to big

639
00:34:46,160 --> 00:34:48,717
updates and start powering them powering

640
00:34:48,800 --> 00:34:51,917
through them. So, what did I build? Um

641
00:34:52,000 --> 00:34:53,677
So, what I built

642
00:34:53,760 --> 00:34:58,677
is essentially a new deep research tool.

643
00:34:58,760 --> 00:35:02,957
You know, one of my life's ambition
is to figure out how to get a job in the

644
00:35:03,040 --> 00:35:06,837
Copilot team. And so,
I'm preparing for it. I'm sort

645
00:35:06,920 --> 00:35:11,357
of desperately trying how
to get competent enough to get

646
00:35:11,440 --> 00:35:15,717
hired. And so, therefore, I said, "Okay,
let me build a new more sophisticated

647
00:35:15,800 --> 00:35:18,957
deep research application."
And so, the what the thing

648
00:35:19,040 --> 00:35:22,837
that I did is you know,
I said, "Look, there are all

649
00:35:22,920 --> 00:35:25,717
these models available. So, why don't I

650
00:35:25,800 --> 00:35:29,917
create decision frameworks that work
on top of the model?" And one of those

651
00:35:30,000 --> 00:35:32,357
decision frameworks is Andrej Karpathy,

652
00:35:32,440 --> 00:35:37,077
who was a famous AI researcher,
came up with this LLM council, he called

653
00:35:37,160 --> 00:35:41,077
it. Right? So, the idea is you
will have like like a selection

654
00:35:41,160 --> 00:35:42,597
committee, you will have multiple

655
00:35:42,680 --> 00:35:44,837
members of the committee, right? So, in

656
00:35:44,920 --> 00:35:48,517
this case, I can have GPT-51 or 52

657
00:35:48,600 --> 00:35:53,077
today, Claude, Gemini, Lama, you know,
Grok, whatever, right? Pick your

658
00:35:53,160 --> 00:35:57,677
favorite selection committee. Then you
elect even a chairman. Like you can say,

659
00:35:57,760 --> 00:36:01,797
"Who is the chairman?
Is it Claude or Gemini or

660
00:36:01,880 --> 00:36:03,797
you know, GPT or what have you?" Pick

661
00:36:03,880 --> 00:36:06,557
your chair and then go ahead and issue a

662
00:36:06,640 --> 00:36:09,197
query and ask it to do stuff.

663
00:36:09,280 --> 00:36:12,077
And then another one I did was

664
00:36:12,160 --> 00:36:14,437
DXO. This is a pretty cool decision

665
00:36:14,520 --> 00:36:19,717
framework. This is comes out of health
care. In fact, we first implemented In

666
00:36:19,800 --> 00:36:24,317
fact, we've implemented it in health
care, where you have just different

667
00:36:24,400 --> 00:36:26,437
roles. So, for example, one role is a

668
00:36:26,520 --> 00:36:29,197
lead researcher, who does Let think of

669
00:36:29,280 --> 00:36:32,597
it as exhaustive breadth first research.

670
00:36:32,680 --> 00:36:36,117
There's a critical reviewer. And the
And by the way, you can assign different

671
00:36:36,200 --> 00:36:39,197
models to these different roles. So,
I have a lead researcher that happens to

672
00:36:39,280 --> 00:36:42,917
be Claude, a critical reviewer that
happens to be a GPT. And what does a

673
00:36:43,000 --> 00:36:45,677
critical reviewer do? They

674
00:36:45,760 --> 00:36:47,917
That particular role finds method

675
00:36:48,000 --> 00:36:51,677
problems, right? Say, recency bias or

676
00:36:51,760 --> 00:36:56,397
period bias, whatever are the issues
which are cognitive biases.

677
00:36:56,480 --> 00:37:00,357
Then you can have domain expert,
data analyst, so different roles. And In

678
00:37:00,440 --> 00:37:05,157
fact, when we applied DXO in health care,
it turned out that having multiple

679
00:37:05,240 --> 00:37:09,797
roles work together in a multi-agent
system outperformed any single model.

680
00:37:09,880 --> 00:37:13,037
Right? It's intuitive, right? Why not
have all the smart people working with

681
00:37:13,120 --> 00:37:16,037
you on a particular decision versus one.

682
00:37:16,120 --> 00:37:18,997
And so, that was DXO. Then I implemented

683
00:37:19,080 --> 00:37:21,477
even another one called ensemble.

684
00:37:21,560 --> 00:37:23,637
And the idea with ensemble is you sort

685
00:37:23,720 --> 00:37:26,277
of fire off parallel queries,

686
00:37:26,360 --> 00:37:28,637
anonymize the responses, so there is no

687
00:37:28,720 --> 00:37:31,157
bias even in selection

688
00:37:31,240 --> 00:37:33,797
and then synthesize, right? In fact,

689
00:37:33,880 --> 00:37:35,797
last night I said, "Well, let's even

690
00:37:35,880 --> 00:37:38,837
implement a regular old debate." So,

691
00:37:38,920 --> 00:37:42,967
this is a new feature I added
and thank God it's showed up here.

692
00:37:43,050 --> 00:37:46,077
>> [laughter] >> And basically,
think of it as chain of

693
00:37:46,160 --> 00:37:50,037
debate. In fact, I I think of all
of these as instead of thinking of them as

694
00:37:50,120 --> 00:37:53,477
chain of thought, thinking of all
of this as chain of debate. And I said,

695
00:37:53,560 --> 00:37:56,917
"Let's actually

696
00:37:57,000 --> 00:38:00,597
have a full-on debate." So, the bottom
line here is you can have different

697
00:38:00,680 --> 00:38:02,917
formats of you know, even debates. You

698
00:38:03,000 --> 00:38:07,757
can have pros and cons, SWOT. You can
have risk versus impact, counterfactual

699
00:38:07,840 --> 00:38:11,877
versus devil's advocate, right? So,
you can even pick your form framework. You

700
00:38:11,960 --> 00:38:15,957
can have critiques. So, each sort of each
person makes an argument, everybody

701
00:38:16,040 --> 00:38:20,037
else critiques. You can have multiple
rounds of it. So, I was just kind of I

702
00:38:20,120 --> 00:38:23,117
had a lot of time, I guess,
in my hotel room yesterday. But,

703
00:38:23,200 --> 00:38:26,917
so I went on and added features.
Like So, I said, "Okay, let's have" Oh, by

704
00:38:27,000 --> 00:38:30,277
the way, I even added because I was
spending so many tokens, I said, "Man, I

705
00:38:30,360 --> 00:38:33,837
might as well actually add cost control
to my thing." And so, I said, "Do I need

706
00:38:33,920 --> 00:38:36,437
fast lower cost or deep higher cost?"

707
00:38:36,520 --> 00:38:40,117
I'm picking balanced here.
I can pick the different research

708
00:38:40,200 --> 00:38:44,477
analyst, system architect, risk officer,
three types of roles, and then have

709
00:38:44,560 --> 00:38:47,077
multiple rounds.
And so, you get the point.

710
00:38:47,160 --> 00:38:49,797
So, of course, you build all this,

711
00:38:49,880 --> 00:38:52,557
what does one good old South Asian do?

712
00:38:52,640 --> 00:38:55,877
the best ever Indian test cricket team.

713
00:38:55,960 --> 00:38:59,997
I think we need one. And so,
if I go to my history, you'll

714
00:39:00,080 --> 00:39:03,317
see it. Like I've been at it.
I've been working at it. In fact, I have a

715
00:39:03,400 --> 00:39:08,917
fantastic MLB lineup as well.
But, since we are here in Mumbai,

716
00:39:09,000 --> 00:39:10,117
I'll show you some of the stuff, right?

717
00:39:10,200 --> 00:39:14,597
So, I'll show you the council one.
I'll show you

718
00:39:14,680 --> 00:39:17,557
what it came back with.
In fact, let me kind of

719
00:39:17,640 --> 00:39:21,797
increase the font here. So, this is the
chairman's synthesis. It shows me, you

720
00:39:21,880 --> 00:39:23,917
know, obviously Sunil opens with Sehwag,

721
00:39:24,000 --> 00:39:27,277
Rahul, Sachin, Virat's A man, it picked

722
00:39:27,360 --> 00:39:31,157
So, this is the synthesis and it shows
VVS at six and seven and then it goes

723
00:39:31,240 --> 00:39:32,837
on. Here's the interesting thing, right?

724
00:39:32,920 --> 00:39:37,437
The chairman's sort of report says
that there was unanimous essentially

725
00:39:37,520 --> 00:39:40,997
consensus, right? Gavaskar, Sehwag,
Dravid, Tendulkar, Kohli, Kapil Dev,

726
00:39:41,080 --> 00:39:43,277
Ashwin, Bumrah. But, the big debate, of

727
00:39:43,360 --> 00:39:48,797
course, was, you know, do you need an
additional batsman or not? And thank God

728
00:39:48,880 --> 00:39:51,677
they selected, at least being
a Hyderabadi, I'm really proud that they

729
00:39:51,760 --> 00:39:53,197
made

730
00:39:53,280 --> 00:39:57,077
VVS Laxman in there. Um and here's the

731
00:39:57,160 --> 00:39:59,677
interesting thing. So, he waited 5-1 and

732
00:39:59,760 --> 00:40:01,597
Claude's inclusion. I mean, think about

733
00:40:01,680 --> 00:40:04,757
this. These two models bid for my man

734
00:40:04,840 --> 00:40:07,637
Laxman, right? Because uh why? Of his

735
00:40:07,720 --> 00:40:09,917
crisis management. And sort of made the

736
00:40:10,000 --> 00:40:14,917
point about why. Uh and then the other
debate was who's the captain? Um in

737
00:40:15,000 --> 00:40:18,357
fact, I think Oh, by the way, the Kumble
versus Zaheer was also a very good one,

738
00:40:18,440 --> 00:40:20,997
right? Which is uh it sort of chose Anil

739
00:40:21,080 --> 00:40:25,477
versus uh Zaheer. Zaheer is going to be
12th man. Um you know, depends, right? I

740
00:40:25,560 --> 00:40:28,637
I guess that's where I think there's
more work to be done. I'll pick Zaheer

741
00:40:28,720 --> 00:40:32,437
some days and sometimes um uh
depending on where where you're

742
00:40:32,520 --> 00:40:34,637
playing. Captaincy debate, you know, it

743
00:40:34,720 --> 00:40:36,797
picked uh Kohli. So, this

744
00:40:36,880 --> 00:40:41,157
the key thing is I see the annotations

745
00:40:41,240 --> 00:40:43,797
uh of every the chain of debate and it

746
00:40:43,880 --> 00:40:46,597
gives me insight. Um in fact, if I go

747
00:40:46,680 --> 00:40:50,317
back um to show you one other case, let

748
00:40:50,400 --> 00:40:53,157
me go back to my history again.

749
00:40:53,240 --> 00:40:56,837
Uh let's see. Let me go to the

750
00:40:56,920 --> 00:41:01,037
uh the DXO view. I think this is
interesting because the DXO view is

751
00:41:01,120 --> 00:41:03,477
slightly different, right? Uh it says

752
00:41:03,560 --> 00:41:06,717
here is what the lead researcher did. Um

753
00:41:06,800 --> 00:41:09,117
then it says here's what the critical

754
00:41:09,200 --> 00:41:11,357
reviewer did. But here's the cool thing.

755
00:41:11,440 --> 00:41:14,557
Here Look at the way it found the

756
00:41:14,640 --> 00:41:17,717
biases, the error bias. Right? So, the

757
00:41:17,800 --> 00:41:20,597
critical reviewer's job was to say,

758
00:41:20,680 --> 00:41:25,797
"Okay, when I'm selecting something,
are you mixing up your stats?" Right? I

759
00:41:25,880 --> 00:41:28,597
mean, like I always say, like, you know,
sometimes you could say, "Wow, the

760
00:41:28,680 --> 00:41:32,277
fitness levels of the modern teams are
so great and wonderful and what have

761
00:41:32,360 --> 00:41:36,717
you?" But man, like to watch Gavaskar
go and bat without a helmet or you know,

762
00:41:36,800 --> 00:41:40,397
even a I don't know
um Hazare go and uh you

763
00:41:40,480 --> 00:41:45,317
know, uncovered wicket in play uh
in England. Like, how do you really

764
00:41:45,400 --> 00:41:49,557
equate for that, right? So, that's the
error bias. And so, to me, to be able to

765
00:41:49,640 --> 00:41:54,917
kind of have all these models debate
that issue uh is what this is all about.

766
00:41:55,000 --> 00:41:56,397
So, anyway, so you get the point, right?

767
00:41:56,480 --> 00:41:57,357
Which is

768
00:41:57,440 --> 00:42:02,157
I You can use these decision frameworks

769
00:42:02,240 --> 00:42:06,997
as the new form of meta cognition,
right? So, if you have all this

770
00:42:07,080 --> 00:42:08,557
abundance

771
00:42:08,640 --> 00:42:12,077
of models, the ability to literally do

772
00:42:12,160 --> 00:42:14,877
what I did, which is just build your own

773
00:42:14,960 --> 00:42:17,037
multi-agent system, right? That's the

774
00:42:17,120 --> 00:42:22,117
new commodity. I mean, the reality is,
you know, this is lit- all of us are

775
00:42:22,200 --> 00:42:26,277
going to be doing this work just like
how we do spreadsheets or documents.

776
00:42:26,360 --> 00:42:30,397
And if we have that power, then the
question is what are we going to do with

777
00:42:30,480 --> 00:42:34,317
it? And one of the things I think
is exercise better judgment, better

778
00:42:34,400 --> 00:42:37,317
decision. And in high-stakes situations,

779
00:42:37,400 --> 00:42:39,717
in health care, uh in financial

780
00:42:39,800 --> 00:42:43,477
services, in insurance, in supply chain.

781
00:42:43,560 --> 00:42:47,077
And that's the beauty of it, right?
To be able to think about these as meta

782
00:42:47,160 --> 00:42:49,997
cognition frameworks for us to apply

783
00:42:50,080 --> 00:42:54,237
better judgment uh in a world where
we have all of this. One of the other

784
00:42:54,320 --> 00:42:58,077
things I should mention is I built all
this. I even made it in fact a co-pilot.

785
00:42:58,160 --> 00:43:00,277
So, this is all in my Azure private

786
00:43:00,360 --> 00:43:04,077
tenant. Um and I So, I then did a you

787
00:43:04,160 --> 00:43:07,237
know, basically an uh an open API uh to

788
00:43:07,320 --> 00:43:09,357
it. Uh and so then I made it into a

789
00:43:09,440 --> 00:43:11,677
co-pilot agent. And then I said, "Great,

790
00:43:11,760 --> 00:43:14,317
let me go deploy." And guess what? You

791
00:43:14,400 --> 00:43:19,517
know, I was caught. So, in other words,
the Microsoft DLP data you know, loss

792
00:43:19,600 --> 00:43:23,437
prevention came in and said,
"You can't deploy it uh because

793
00:43:23,520 --> 00:43:25,997
uh you this has to be deployed inside

794
00:43:26,080 --> 00:43:27,837
the Microsoft tenant. You just can't

795
00:43:27,920 --> 00:43:30,517
point it to an external agent uh because

796
00:43:30,600 --> 00:43:33,037
that's a massive exfil problem." And

797
00:43:33,120 --> 00:43:35,757
that is thanks to this other part, which

798
00:43:35,840 --> 00:43:38,157
I think is the runtime of agents we have

799
00:43:38,240 --> 00:43:41,197
built called Agent 365. And Agent 365

800
00:43:41,280 --> 00:43:45,357
takes our identity
management, our Defender,

801
00:43:45,440 --> 00:43:50,077
our things like Purview and data loss
prevention, all of that and brings it to

802
00:43:50,160 --> 00:43:53,637
the agent world. Just like what we
have done for end-user computing, if you

803
00:43:53,720 --> 00:43:55,357
will, we're now doing it for agents.

804
00:43:55,440 --> 00:43:58,717
Because if people are going to build
powerful agents like the one I built,

805
00:43:58,800 --> 00:44:02,917
you want full compliance, you want full
visibility, you want full governance of

806
00:44:03,000 --> 00:44:06,637
it. And that's what the Agent 365 does,
right? Because you can't just run around

807
00:44:06,720 --> 00:44:11,037
and say, "I have thousands and hundreds
of thousands of agents without that

808
00:44:11,120 --> 00:44:13,437
security and governance and visibility."

809
00:44:13,520 --> 00:44:15,077
And that's really

810
00:44:15,160 --> 00:44:18,277
uh what uh Agent 365 delivers.

811
00:44:18,360 --> 00:44:20,677
You know, the the beauty of all this is

812
00:44:20,760 --> 00:44:23,557
there is significant momentum,

813
00:44:23,640 --> 00:44:26,717
significant ambition level to translate

814
00:44:26,800 --> 00:44:30,957
this ultimately into people doing
unbelievable things. So, this morning I

815
00:44:31,040 --> 00:44:32,797
had a chance to meet with, for example,

816
00:44:32,880 --> 00:44:35,117
the folks from Adani Cement uh who are

817
00:44:35,200 --> 00:44:38,037
really doing this invoice to payment uh

818
00:44:38,120 --> 00:44:40,277
and reduce something that took 20 plus

819
00:44:40,360 --> 00:44:45,077
days to 4 hours. Uh I had a chance
to meet the team from Aditya Birla where

820
00:44:45,160 --> 00:44:48,957
they took their mobile app that's
doing complex products and sort of said,

821
00:44:49,040 --> 00:44:52,917
"Okay, here's a conversational interface
that just makes it simpler for them to

822
00:44:53,000 --> 00:44:56,917
be able to interact." I had a chance
to meet with the folks at LTI Mindtree.

823
00:44:57,000 --> 00:44:58,717
They're doing a complete wall-to-wall

824
00:44:58,800 --> 00:45:01,197
transformation using things from

825
00:45:01,280 --> 00:45:04,557
co-pilot to co-pilot studio to a foundry

826
00:45:04,640 --> 00:45:07,117
to do every business process and relook

827
00:45:07,200 --> 00:45:09,277
at it from an AI-first perspective.

828
00:45:09,360 --> 00:45:11,957
Mankind, it's a beautiful use case of

829
00:45:12,040 --> 00:45:15,237
really putting that a power of AI and

830
00:45:15,320 --> 00:45:20,077
the expertise in the hands of the people
who are use the med- medical reps who

831
00:45:20,160 --> 00:45:23,517
are visiting doctors
in rural areas to be able to

832
00:45:23,600 --> 00:45:25,357
make sure that you can the medical rep

833
00:45:25,440 --> 00:45:28,397
is confident in answering the questions

834
00:45:28,480 --> 00:45:31,117
and then improving care outcomes.

835
00:45:31,200 --> 00:45:35,837
Yes Bank has done a fantastic job
of really the invoice turnaround times,

836
00:45:35,920 --> 00:45:38,597
right? So, they're also doing
a wall-to-wall transformation with 40 plus

837
00:45:38,680 --> 00:45:41,677
use cases. But really seeing the ROI in

838
00:45:41,760 --> 00:45:46,157
each one of these examples. Now,
one of the chances I had again was

839
00:45:46,240 --> 00:45:47,757
to meet

840
00:45:47,840 --> 00:45:52,037
uh with the team that built this new

841
00:45:52,120 --> 00:45:56,077
agentic system called the Maha Crime OS.

842
00:45:56,160 --> 00:46:00,117
Uh so, this is was sponsored by uh
the Chief Minister of Maharashtra who is

843
00:46:00,200 --> 00:46:03,960
here with us. >> [applause] >> And

844
00:46:06,480 --> 00:46:11,197
And it was really great to meet
uh both the SP as well as

845
00:46:11,280 --> 00:46:15,557
uh the people in the team uh
and in fact, the investigative team.

846
00:46:15,640 --> 00:46:17,957
Uh and the pride they had

847
00:46:18,040 --> 00:46:21,157
in in being able to use this system

848
00:46:21,240 --> 00:46:24,997
to deliver sort of justice to a citizen

849
00:46:25,080 --> 00:46:27,837
in Nagpur who was

850
00:46:27,920 --> 00:46:30,197
uh unfortunately a victim of a crime.

851
00:46:30,280 --> 00:46:35,037
But the ability for them to use this
agentic system to speed up the time to

852
00:46:35,120 --> 00:46:40,640
justice was just fantastic to see.
So, let's go ahead and play the video.

853
00:46:40,810 --> 00:46:42,830
>> [music]

854
00:46:45,200 --> 00:46:51,207
>> Today, policing is all about cyber
crime. It is all about financial frauds.

855
00:46:51,290 --> 00:46:54,557
>> [music] >> And it is global.

856
00:46:54,640 --> 00:46:57,077
Nagpur rural receives about 30 cyber

857
00:46:57,160 --> 00:47:00,077
crime reports every single day. So, you

858
00:47:00,160 --> 00:47:04,237
can imagine the kind of influx of cyber
crime reporting that that's taking

859
00:47:04,320 --> 00:47:08,380
place. >> [music]

860
00:47:12,910 --> 00:47:14,930
[music]

861
00:47:17,490 --> 00:47:19,510
[music]

862
00:47:29,280 --> 00:47:31,300
[music]

863
00:47:32,280 --> 00:47:34,877
>> Maha Crime OS serves
as an investigation

864
00:47:34,960 --> 00:47:38,117
co-pilot. You're able to have an agent

865
00:47:38,200 --> 00:47:42,677
that is able to detect the patterns,
able to make sense of it, and have it

866
00:47:42,760 --> 00:47:46,797
speak to the direction
of the investigation itself.

867
00:47:46,880 --> 00:47:49,677
The Maharashtra government through
Marvels has collaborated with both

868
00:47:49,760 --> 00:47:51,957
Microsoft as well as Cyber Eye in terms

869
00:47:52,040 --> 00:47:55,877
of adapting Crime OS for the needs of

870
00:47:55,960 --> 00:47:57,997
Maharashtra. Ethical [music] and

871
00:47:58,080 --> 00:48:01,197
responsible AI for public good is our

872
00:48:01,280 --> 00:48:04,597
motto. And Microsoft has provided us a

873
00:48:04,680 --> 00:48:09,360
very safe and advanced platform.

874
00:48:13,820 --> 00:48:15,840
>> [music]

875
00:48:19,060 --> 00:48:21,080
[music]

876
00:48:23,000 --> 00:48:25,917
>> Maha Crime OS AI is built using you

877
00:48:26,000 --> 00:48:31,117
know, multiple technologies. Underlying
this is Azure and AI services through

878
00:48:31,200 --> 00:48:34,697
Microsoft Foundry. We're also using
Defender for for cloud and Microsoft

879
00:48:34,780 --> 00:48:37,437
[music] Fabric.

880
00:48:37,520 --> 00:48:41,907
With Maha Crime OS AI,
the turnaround times of investigations are

881
00:48:41,990 --> 00:48:44,957
>> [music] >> decreased
by 80%. Almost 100% of

882
00:48:45,040 --> 00:48:49,317
investigations are getting digitally
registered and investigators are acting

883
00:48:49,400 --> 00:48:53,180
upon. >> [music]

884
00:48:54,320 --> 00:48:55,997
>> So, that makes me really proud that

885
00:48:56,080 --> 00:48:58,957
we're able to adapt emerging technology

886
00:48:59,040 --> 00:49:03,517
to fulfill the needs of several
millions of of people across

887
00:49:03,600 --> 00:49:05,997
>> [music] >> the state
and make government function

888
00:49:06,080 --> 00:49:08,517
much more effectively and transparently

889
00:49:08,600 --> 00:49:11,797
for them. If AI is properly used in

890
00:49:11,880 --> 00:49:17,640
governance, we can make
the lives of people easier.

891
00:49:25,220 --> 00:49:27,240
>> [music]

892
00:49:33,730 --> 00:49:35,750
[music]

893
00:49:40,000 --> 00:49:43,317
>> It's [applause] It's a great
example, quite frankly, of

894
00:49:43,400 --> 00:49:46,437
using the power of all this technology

895
00:49:46,520 --> 00:49:48,997
to help citizens of this country.

896
00:49:49,080 --> 00:49:53,157
And more importantly, it also sort
of is a great example of how quickly this

897
00:49:53,240 --> 00:49:56,517
technology is diffusing, right? I mean,
if you would have thought about any

898
00:49:56,600 --> 00:50:00,717
previous era of tech and the
rate at which it would come to

899
00:50:00,800 --> 00:50:04,877
sort of impact in a context like this
versus where we are. And that I think is

900
00:50:04,960 --> 00:50:08,397
the opportunity in front of us.
So, it's fantastic to

901
00:50:08,480 --> 00:50:10,637
see. So, the last piece I want to talk

902
00:50:10,720 --> 00:50:13,117
about is powering all of this, of

903
00:50:13,200 --> 00:50:16,077
course, ultimately is the

904
00:50:16,160 --> 00:50:18,317
infrastructure, right? You have got to

905
00:50:18,400 --> 00:50:21,557
run essentially a token factory.

906
00:50:21,640 --> 00:50:25,157
And when it comes to token factory, one
of the most important things is this one

907
00:50:25,240 --> 00:50:28,797
simple equation, yet very
very important, which is

908
00:50:28,880 --> 00:50:31,677
performance per tokens

909
00:50:31,760 --> 00:50:36,717
is all about tokens
per dollar or per rupee

910
00:50:36,800 --> 00:50:41,477
per watt. Right? You have to optimize.

911
00:50:41,560 --> 00:50:43,237
And And this, by the way, is going to be

912
00:50:43,320 --> 00:50:45,837
the most important thing, right? Whether

913
00:50:45,920 --> 00:50:49,197
any country, any community, any state

914
00:50:49,280 --> 00:50:51,517
that is able to sort of really have the

915
00:50:51,600 --> 00:50:56,357
infrastructure that delivers
this efficient frontier of the token

916
00:50:56,440 --> 00:50:58,397
factory is just going to have more

917
00:50:58,480 --> 00:51:01,037
abundance of tokens that can then be

918
00:51:01,120 --> 00:51:05,717
deployed to drive the productivity,
which directly is going to be correlated

919
00:51:05,800 --> 00:51:08,757
to the GDP growth in the economy. And so,

920
00:51:08,840 --> 00:51:11,157
to that end, we are all in, right? We

921
00:51:11,240 --> 00:51:13,477
are building out Azure as the world's

922
00:51:13,560 --> 00:51:16,317
computer. We are investing in building

923
00:51:16,400 --> 00:51:19,157
out Azure in India. We now have multiple

924
00:51:19,240 --> 00:51:23,557
regions. In fact, we right in in in
Maharashtra, we have a Central India

925
00:51:23,640 --> 00:51:25,157
region in Pune.

926
00:51:25,240 --> 00:51:28,237
We have West India in Mumbai. We have

927
00:51:28,320 --> 00:51:30,837
South India in Chennai. We have

928
00:51:30,920 --> 00:51:34,477
a DC with in partnership with Jio.

929
00:51:34,560 --> 00:51:39,277
We are also very very excited about sort
of our new South Central region that is

930
00:51:39,360 --> 00:51:44,077
going to come out in Hyderabad next year,
which is going to be on the next big

931
00:51:44,160 --> 00:51:47,997
data center region.
And we keep expanding. And to that end,

932
00:51:48,080 --> 00:51:52,077
I'm really By the way, the the region in
Hyderabad is going to be 100% renewable,

933
00:51:52,160 --> 00:51:57,037
right? So, that's the other cool thing is
we are really focusing on sustainability

934
00:51:57,120 --> 00:51:59,997
as well. >> [applause]

935
00:52:00,080 --> 00:52:05,117
>> And And to us,
we are also building all of

936
00:52:05,200 --> 00:52:09,637
this with the making
sure that there's real

937
00:52:09,720 --> 00:52:13,717
sovereignty. Because after all, at the
end of the day, this is not just about

938
00:52:13,800 --> 00:52:18,757
bringing the capabilities, but it's also
ensuring that there is real sovereignty

939
00:52:18,840 --> 00:52:22,477
controls. And so, for that, we have
a full set of options, right? You can use

940
00:52:22,560 --> 00:52:26,277
public cloud. You can use public cloud
with sovereign controls. That means you

941
00:52:26,360 --> 00:52:30,517
have you know, you're the key management
locally. You can use confidential

942
00:52:30,600 --> 00:52:34,357
computing in concert with it. So, that
means when the data is being processed,

943
00:52:34,440 --> 00:52:37,357
even in use, at rest, in flight, it's

944
00:52:37,440 --> 00:52:39,877
encrypted and the keys are with you.

945
00:52:39,960 --> 00:52:42,757
We also have the private cloud option.

946
00:52:42,840 --> 00:52:44,637
We also obviously have a region that's

947
00:52:44,720 --> 00:52:47,197
operated by local partners like Jio or

948
00:52:47,280 --> 00:52:50,397
NIC and others. And so, we have a full

949
00:52:50,480 --> 00:52:55,157
set of options. So, think of it
as a portfolio that you can then

950
00:52:55,240 --> 00:52:57,317
use that portfolio for the different

951
00:52:57,400 --> 00:52:59,557
workloads you have and have the

952
00:52:59,640 --> 00:53:02,677
assurance of your own sovereignty and

953
00:53:02,760 --> 00:53:06,637
control over it. We are also doing,
in fact, sovereign data control. So,

954
00:53:06,720 --> 00:53:10,717
Copilot and all of the things we talked
about are all being locally processed.

955
00:53:10,800 --> 00:53:15,597
So, none of the data is going
outside of the the data boundary of

956
00:53:15,680 --> 00:53:18,637
India. And so, we're
very excited about sort of

957
00:53:18,720 --> 00:53:21,397
all of these options being available.

958
00:53:21,480 --> 00:53:23,877
Now, one other point I would also make,

959
00:53:23,960 --> 00:53:25,997
since we talked about sort of crime OS

960
00:53:26,080 --> 00:53:28,517
and and what have you, is

961
00:53:28,600 --> 00:53:31,877
one of the biggest challenges is for any

962
00:53:31,960 --> 00:53:36,237
online infrastructure
is cyber and cyber risk

963
00:53:36,320 --> 00:53:40,917
and cyber security. And it's
an intelligence game, right?

964
00:53:41,000 --> 00:53:44,597
And ultimately, whoever has the best
intelligence is going to be able to

965
00:53:44,680 --> 00:53:48,677
defend you the best. And so,
Microsoft sees trillions of

966
00:53:48,760 --> 00:53:52,437
signals every day. We see every
nation-state and every cybercriminal

967
00:53:52,520 --> 00:53:56,597
everywhere in the world.
And we are able to harness that to

968
00:53:56,680 --> 00:53:59,877
protect. In fact, in the demo current
showed was pretty cool, right? Which is

969
00:53:59,960 --> 00:54:03,917
he basically was able to take
your production sort of instance, have

970
00:54:04,000 --> 00:54:08,237
Defender check in real time what are all
the vulnerabilities, and then have you

971
00:54:08,320 --> 00:54:10,037
even go back and fix those

972
00:54:10,120 --> 00:54:11,997
vulnerabilities. And that's not a static

973
00:54:12,080 --> 00:54:15,397
list. That is a dynamic list that is

974
00:54:15,480 --> 00:54:17,757
sort of being monitoring essentially

975
00:54:17,840 --> 00:54:19,757
what is the cyber activity and the cyber

976
00:54:19,840 --> 00:54:22,837
attack sort of graph. And so, to me,

977
00:54:22,920 --> 00:54:25,717
really thinking about sovereignty

978
00:54:25,800 --> 00:54:28,917
and cyber resilience are two sides of

979
00:54:29,000 --> 00:54:33,077
risk, right? So, when you think about
even you managing your risk, you want to

980
00:54:33,160 --> 00:54:35,797
think about it. Because if you get
disconnected, you have a sovereign

981
00:54:35,880 --> 00:54:40,157
everything, except it doesn't
have the best intelligence,

982
00:54:40,240 --> 00:54:41,957
it's sort of you're going to be very

983
00:54:42,040 --> 00:54:44,157
very vulnerable. And so, therefore, it's

984
00:54:44,240 --> 00:54:47,397
a much more nuanced discussion that

985
00:54:47,480 --> 00:54:51,437
needs real thought. And then, at the end
end of the day, it's not a one-time

986
00:54:51,520 --> 00:54:53,997
decision. It's a continuous monitoring

987
00:54:54,080 --> 00:54:58,677
of that portfolio of risk and opportunity.
So, to that end, we are

988
00:54:58,760 --> 00:55:03,077
very very committed to this country. And
therefore, one of the things that I was

989
00:55:03,160 --> 00:55:06,757
excited to have a chance is meet
the Prime Minister earlier this week and

990
00:55:06,840 --> 00:55:12,040
even announce the biggest announcement.

991
00:55:13,030 --> 00:55:16,437
>> [applause] >> It's It's
our biggest investment in

992
00:55:16,520 --> 00:55:20,077
Asia, 17 and a half billion,
which builds on top of the 3 billion we

993
00:55:20,160 --> 00:55:22,717
announced earlier in the year, in fact.

994
00:55:22,800 --> 00:55:24,837
And so, we are obviously very excited to

995
00:55:24,920 --> 00:55:27,437
bring all of this infrastructure to

996
00:55:27,520 --> 00:55:30,797
empower ultimately all the citizens of

997
00:55:30,880 --> 00:55:34,557
this country, all the developers of this
country. And so, that's where I want to

998
00:55:34,640 --> 00:55:37,277
close out with, which is to us,

999
00:55:37,360 --> 00:55:41,917
all this is about empowering every
person and every organization in India

1000
00:55:42,000 --> 00:55:43,677
to achieve more.

1001
00:55:43,760 --> 00:55:46,437
To us, you know, whether it is

1002
00:55:46,520 --> 00:55:48,997
you know, things like gigatime,

1003
00:55:49,080 --> 00:55:52,237
which allow us to think about

1004
00:55:52,320 --> 00:55:57,677
making what is personalized care or even

1005
00:55:57,760 --> 00:56:00,477
precision oncology possible in every

1006
00:56:00,560 --> 00:56:03,397
tier two city in this country because

1007
00:56:03,480 --> 00:56:06,077
they can an entrepreneur here will take

1008
00:56:06,160 --> 00:56:11,197
an advance like that and parlay it
into health outcomes that matter.

1009
00:56:11,280 --> 00:56:14,717
Or the Maha crime OS that's going to

1010
00:56:14,800 --> 00:56:16,477
deliver

1011
00:56:16,560 --> 00:56:19,717
the justice to someone who was a victim

1012
00:56:19,800 --> 00:56:22,957
of some crime in Nagpur. To us,

1013
00:56:23,040 --> 00:56:26,277
ultimately, that's why we do what we do.

1014
00:56:26,360 --> 00:56:28,677
All of us working together to realize

1015
00:56:28,760 --> 00:56:33,357
the opportunity ahead of us in this age
of AI is an exciting time. I hope you

1016
00:56:33,440 --> 00:56:37,957
all have high ambition for what you can
do. We sure do for you all. Thank you

1017
00:56:38,040 --> 00:56:41,510
all very very much. >> [applause]

1018
00:56:45,800 --> 00:56:49,757
>> India has more than 400 million
unorganized workers. They're earning on

1019
00:56:49,840 --> 00:56:53,517
the day-to-day wages and not
getting any [music] benefits.

1020
00:56:53,600 --> 00:56:56,797
We need to shift the people from the
informal to the formal sector. [music]

1021
00:56:56,880 --> 00:56:59,957
That would help us to develop our economy.
Through the initiative of

1022
00:57:00,040 --> 00:57:04,517
e-Shram, the national unorganized
workers database was developed and we

1023
00:57:04,600 --> 00:57:08,237
are able to provide those workers
[music] with social security benefits,

1024
00:57:08,320 --> 00:57:10,397
also giving them access to the national

1025
00:57:10,480 --> 00:57:12,997
career service portal. NCS is a one-stop

1026
00:57:13,080 --> 00:57:17,677
solution for job seekers, employers,
training providers, and counselors. We

1027
00:57:17,760 --> 00:57:21,157
have AI-based matching between
[music] the job seekers' skills and the

1028
00:57:21,240 --> 00:57:23,957
employers. So, any informal worker can

1029
00:57:24,040 --> 00:57:28,677
go to the NCS [music] portal and match
with the right employer. NCS has an AI

1030
00:57:28,760 --> 00:57:32,237
resume builder. And if a person
is lacking certain skills, the [music] AI

1031
00:57:32,320 --> 00:57:36,717
tool helps him to develop those
skills by giving them recommendations.

1032
00:57:36,800 --> 00:57:39,997
We were expecting millions
of unorganized workers to register in this

1033
00:57:40,080 --> 00:57:44,677
system. [music] So, we wanted to have
scalability, reliability, and security.

1034
00:57:44,760 --> 00:57:48,357
Microsoft Azure met all these
qualifications. In addition [music] to

1035
00:57:48,440 --> 00:57:54,640
that, Azure Open AI has become
a game-changer for unorganized workers.

1036
00:58:00,050 --> 00:58:02,070
>> [music]

1037
00:58:11,000 --> 00:58:13,237
>> Around 150 million workers are now on

1038
00:58:13,320 --> 00:58:15,557
the NCS portal. We are also [music] now

1039
00:58:15,640 --> 00:58:20,277
trying to bring in more and more
schemes to benefit these workers. AI is

1040
00:58:20,360 --> 00:58:23,997
supporting us to provide opportunities
for the individuals and youth of the

1041
00:58:24,080 --> 00:58:26,310
country. [music]