Building Enterprise AI Agents with Copilot Studio, Power Platform & AI Governance with Sailaja Mantripragada [MVP/MCT]

1
00:00:00,000 --> 00:00:04,800
>> Yeah, welcome back to another edition of the MC65 podcast,

2
00:00:04,800 --> 00:00:08,480
where we explore Microsoft 365 Power Platform,

3
00:00:08,480 --> 00:00:10,800
AI, Co-Pilot, and the Future of Work,

4
00:00:10,800 --> 00:00:13,680
which is brought from around the world.

5
00:00:13,680 --> 00:00:17,200
Today I join in for Celia, Matri Pagra,

6
00:00:17,200 --> 00:00:22,000
sometimes the people love me because I pronounce it wrong.

7
00:00:22,000 --> 00:00:25,880
She is a Microsoft Application MVP,

8
00:00:25,880 --> 00:00:28,380
Microsoft Certificate Trainer,

9
00:00:28,380 --> 00:00:33,380
a personal cloud architect and found off the low-code power

10
00:00:33,380 --> 00:00:35,740
with more than 20 years of experience

11
00:00:35,740 --> 00:00:37,660
in enterprise technology in Celia,

12
00:00:37,660 --> 00:00:39,980
specialized in Microsoft Power Platform,

13
00:00:39,980 --> 00:00:40,960
Co-Pilot Studio,

14
00:00:40,960 --> 00:00:43,900
Agentic AI, Architecture, AI Governance,

15
00:00:43,900 --> 00:00:45,900
and Enterprise Scale Automation.

16
00:00:45,900 --> 00:00:47,900
She is helping organization move

17
00:00:47,900 --> 00:00:50,300
beyond simple chatboards and into

18
00:00:50,300 --> 00:00:53,220
a new area of intelligent agents

19
00:00:53,220 --> 00:00:55,460
that can perform deep search,

20
00:00:55,460 --> 00:00:58,460
generate complex document,

21
00:00:58,460 --> 00:01:00,540
business or classification,

22
00:01:00,540 --> 00:01:04,380
process and security index with enterprise data.

23
00:01:04,380 --> 00:01:06,740
Well, welcome.

24
00:01:06,740 --> 00:01:10,580
We discuss what separates real enterprise from demos,

25
00:01:10,580 --> 00:01:13,580
how I, Genic AI has changed software architecture,

26
00:01:13,580 --> 00:01:16,300
where Co-Pilot Studio is heading,

27
00:01:16,300 --> 00:01:18,660
and what organization need to deploy

28
00:01:18,660 --> 00:01:22,500
or to prepare for the next wave of AI automation?

29
00:01:22,500 --> 00:01:26,140
Celia, welcome to the MC65 podcast.

30
00:01:26,140 --> 00:01:28,620
Can you ask tell a little bit

31
00:01:28,620 --> 00:01:32,900
how you're drawn into technology?

32
00:01:32,900 --> 00:01:36,380
Thank you so much for having me, Merkel.

33
00:01:36,380 --> 00:01:39,220
It's so pleasure to be in this podcast.

34
00:01:39,220 --> 00:01:43,900
My name is Sila Ja, Matri Pagra,

35
00:01:43,900 --> 00:01:50,340
and I am a Power Platform Architect and

36
00:01:50,340 --> 00:01:53,580
a Co-Pilot Architect.

37
00:01:53,580 --> 00:01:58,220
Building, I'm a practitioner who's moved into governance,

38
00:01:58,220 --> 00:02:01,340
because I see the importance of governance.

39
00:02:01,340 --> 00:02:06,540
So to talk about my, where I come from,

40
00:02:06,540 --> 00:02:10,860
I've been in the Microsoft Space for the last 20 years,

41
00:02:10,860 --> 00:02:14,500
and I did my master's in computer applications,

42
00:02:14,500 --> 00:02:20,980
and my entire journey, entire professional career,

43
00:02:20,980 --> 00:02:23,980
has been with Microsoft.

44
00:02:23,980 --> 00:02:30,980
So, yeah, I worked for 15 years as a SharePoint subject matter expert,

45
00:02:30,980 --> 00:02:35,100
and then moved into Power Platform because

46
00:02:35,100 --> 00:02:38,740
I was doing a lot of business process re-engineering,

47
00:02:38,740 --> 00:02:42,420
and with After Power Platform,

48
00:02:42,420 --> 00:02:47,580
of course, Co-Pilot just turned on all the lights,

49
00:02:47,580 --> 00:02:53,900
and I am, you know, the thread that has been on

50
00:02:53,900 --> 00:02:58,260
from SharePoint to now is governance,

51
00:02:58,260 --> 00:03:03,660
and it's important enterprise scalability.

52
00:03:03,660 --> 00:03:09,540
Oh. And this is awesome, 20 years.

53
00:03:09,540 --> 00:03:15,340
When you look back over the 20 years in this industry, especially the last 15 years,

54
00:03:15,340 --> 00:03:23,140
and Microsoft ecosystem, what was the biggest technology you have seen?

55
00:03:23,140 --> 00:03:31,860
So, I've seen that Microsoft has moved,

56
00:03:31,860 --> 00:03:37,660
so there was one part where Microsoft technologies were really

57
00:03:37,660 --> 00:03:39,340
proprietary, right?

58
00:03:39,340 --> 00:03:44,300
So, earlier days of SharePoint were very,

59
00:03:44,300 --> 00:03:52,860
you know, ASP.NET, and you could only code in the set Microsoft platform,

60
00:03:52,860 --> 00:03:55,700
and once Microsoft turned open source,

61
00:03:55,700 --> 00:04:01,380
I believe that's a biggest change once they allowed JavaScript,

62
00:04:01,380 --> 00:04:08,220
to take query into the ecosystem, I think that's one big change.

63
00:04:08,220 --> 00:04:14,420
And from there on, I believe Microsoft has been instrumental in

64
00:04:14,420 --> 00:04:17,340
democratizing technology.

65
00:04:17,340 --> 00:04:22,460
First, we democratized information using SharePoint,

66
00:04:22,460 --> 00:04:31,340
and then we democratized apps and application development using Power Platform,

67
00:04:31,340 --> 00:04:36,940
and now we are democratizing the intelligence by making co-pilot agents

68
00:04:36,940 --> 00:04:39,660
accessible to citizen developers.

69
00:04:39,660 --> 00:04:46,060
So, that's the pattern I look at, and as everybody says,

70
00:04:46,060 --> 00:04:51,180
AI is the biggest technology for SharePoint.

71
00:04:51,180 --> 00:04:54,940
Yeah, I think this is really awesome change,

72
00:04:54,940 --> 00:04:59,100
especially all the co-pilot and AI foundry and stuff,

73
00:04:59,100 --> 00:05:09,500
but what was the, I think, the first part or first idea that I track you to

74
00:05:09,500 --> 00:05:13,020
to low-code development?

75
00:05:13,020 --> 00:05:15,620
Oh, this is an interesting story for me.

76
00:05:15,620 --> 00:05:20,820
I started as a SharePoint developer in 2008,

77
00:05:20,820 --> 00:05:25,740
and I've worked with companies like Blue Cross Blue Shield,

78
00:05:25,740 --> 00:05:30,900
Securities and Exchange Commission, and AARP,

79
00:05:30,900 --> 00:05:33,740
like some of the big giant companies,

80
00:05:33,740 --> 00:05:36,740
and I was a daughter of developer until then,

81
00:05:36,740 --> 00:05:42,220
before I started with SharePoint, I was a role-pored developer.

82
00:05:42,220 --> 00:05:46,380
What happened with me was the first three projects,

83
00:05:46,380 --> 00:05:52,740
where, you know, apparently, everybody who's a SharePoint developer

84
00:05:52,740 --> 00:05:58,380
was building custom web parts using .NET, right?

85
00:05:58,380 --> 00:06:03,380
At that time, these three projects that I worked with,

86
00:06:03,380 --> 00:06:07,940
would span, like, first four years of my career,

87
00:06:07,940 --> 00:06:12,580
and each of these places, I was stored not to code.

88
00:06:12,580 --> 00:06:13,740
Oh, Blue?

89
00:06:13,740 --> 00:06:18,340
They said, "Use what we have, use out of the box,

90
00:06:18,340 --> 00:06:22,500
and third party tools, right?"

91
00:06:22,500 --> 00:06:26,860
So I would leverage into path, and, you know,

92
00:06:26,860 --> 00:06:29,580
out of the box SharePoint opportunities,

93
00:06:29,580 --> 00:06:34,100
and maybe in some, NINTEX workflows and stuff like that,

94
00:06:34,100 --> 00:06:36,900
and build for their companies.

95
00:06:36,900 --> 00:06:40,860
So in my fourth endeavor, right?

96
00:06:40,860 --> 00:06:44,020
Once I completed these three projects,

97
00:06:44,020 --> 00:06:49,740
I tried to analyze a pattern because I was having the formal of,

98
00:06:49,740 --> 00:06:53,940
you know, the fear of missing out, working with word, right?

99
00:06:53,940 --> 00:06:57,500
I didn't want to be obsolete, but then I observed that

100
00:06:57,500 --> 00:07:02,140
if three giant companies don't want me to code,

101
00:07:02,140 --> 00:07:05,140
that's because they didn't want an overhead

102
00:07:05,140 --> 00:07:09,300
of maintaining another developer,

103
00:07:09,300 --> 00:07:13,940
because their goal was on business outcomes.

104
00:07:13,940 --> 00:07:17,060
So then in my fourth project itself,

105
00:07:17,060 --> 00:07:24,900
so we're talking 2014, I already started using the term

106
00:07:24,900 --> 00:07:28,140
"low code specialist" in my resume,

107
00:07:28,140 --> 00:07:29,980
even before Microsoft coined it.

108
00:07:29,980 --> 00:07:34,100
Yeah.

109
00:07:34,100 --> 00:07:35,980
Yeah, this is great.

110
00:07:35,980 --> 00:07:38,900
Low code is so interesting.

111
00:07:38,900 --> 00:07:42,540
Now we have also just these other words like,

112
00:07:42,540 --> 00:07:45,260
"No code, my code," and so on.

113
00:07:45,260 --> 00:07:54,060
But let me a little bit look in your CV,

114
00:07:54,060 --> 00:07:58,700
and it was a solution actually, and now you are a strategist,

115
00:07:58,700 --> 00:08:03,460
an advisor, how has your change,

116
00:08:03,460 --> 00:08:06,940
your role change over the time?

117
00:08:06,940 --> 00:08:09,740
And what can you say?

118
00:08:09,740 --> 00:08:13,620
What did you think a little bit about your future?

119
00:08:13,620 --> 00:08:16,820
How will all this change?

120
00:08:16,820 --> 00:08:26,820
So what I've seen in my experience over the two decades is that,

121
00:08:26,820 --> 00:08:31,900
because like I said, I moved to low code very quickly,

122
00:08:31,900 --> 00:08:35,020
so I was building a lot of info path,

123
00:08:35,020 --> 00:08:39,140
but then I was also part of big migrations.

124
00:08:39,140 --> 00:08:45,580
So migrating from SharePoint 2010 to 2013 to SharePoint Online,

125
00:08:45,580 --> 00:08:52,300
during this process, my job was to evaluate the existing apps

126
00:08:52,300 --> 00:08:56,580
and workflows and even the power platform, right?

127
00:08:56,580 --> 00:08:59,940
So whenever we did this move,

128
00:08:59,940 --> 00:09:06,860
I observed that citizen development has been a major win

129
00:09:06,860 --> 00:09:10,580
for what we're doing with low code,

130
00:09:10,580 --> 00:09:15,660
but also a big challenge because now, once you say,

131
00:09:15,660 --> 00:09:21,140
for example, you've provided access to everybody

132
00:09:21,140 --> 00:09:25,180
in the organization to create a SharePoint size.

133
00:09:25,180 --> 00:09:30,580
After two years, you're ending up with thousands of sites

134
00:09:30,580 --> 00:09:34,820
that have been orphaned, people created it out of excitement,

135
00:09:34,820 --> 00:09:37,900
and they've left it, right?

136
00:09:37,900 --> 00:09:40,580
So things like that, same with power apps,

137
00:09:40,580 --> 00:09:44,260
same with hope-ilet agents now.

138
00:09:44,260 --> 00:09:50,740
So as I lived through these different experiences,

139
00:09:50,740 --> 00:09:55,140
I believe governance is one thing that becomes...

140
00:09:55,140 --> 00:09:56,700
So when I say governance, right?

141
00:09:56,700 --> 00:09:57,220
We have to...

142
00:09:57,220 --> 00:10:00,220
Governance has like four or five years.

143
00:10:00,220 --> 00:10:07,820
I am talking about agent governance or solution governance.

144
00:10:07,820 --> 00:10:14,300
So when we actually start building a agent,

145
00:10:14,300 --> 00:10:16,460
what are the things we keep in mind

146
00:10:16,460 --> 00:10:19,140
and the best practices when we do that governance

147
00:10:19,140 --> 00:10:21,820
is the major focus for me.

148
00:10:21,820 --> 00:10:28,820
So I believe the future is in trainings.

149
00:10:28,820 --> 00:10:31,500
It is in developers to do it, right?

150
00:10:31,500 --> 00:10:34,740
Because they are domain experts, right?

151
00:10:34,740 --> 00:10:37,980
So they have a lot of power

152
00:10:37,980 --> 00:10:39,780
because they know the pain points.

153
00:10:39,780 --> 00:10:41,900
They have the...

154
00:10:41,900 --> 00:10:44,660
They're ready to roll.

155
00:10:44,660 --> 00:10:49,260
So training is a primary factor, I believe,

156
00:10:49,260 --> 00:10:51,660
and there's a lot more.

157
00:10:51,660 --> 00:10:53,900
I think one more thing is,

158
00:10:55,580 --> 00:11:00,580
I could see organizations building AI...

159
00:11:00,580 --> 00:11:06,580
But center of excellence.

160
00:11:06,580 --> 00:11:09,380
Like we used to have center of excellence otherwise,

161
00:11:09,380 --> 00:11:11,580
right, power platforms, CEOs,

162
00:11:11,580 --> 00:11:14,700
because you can't expect all these...

163
00:11:14,700 --> 00:11:20,300
Citizen developers to know everything about everything.

164
00:11:22,780 --> 00:11:27,980
Every organization would need an entire AI center of excellence

165
00:11:27,980 --> 00:11:33,380
with AI architects who are in charge to review

166
00:11:33,380 --> 00:11:35,180
what the teens are building.

167
00:11:35,180 --> 00:11:40,700
Yeah, that's a really interesting topic.

168
00:11:40,700 --> 00:11:43,180
What did you think that we...

169
00:11:43,180 --> 00:11:46,860
What is the work of, I call it,

170
00:11:46,860 --> 00:11:51,740
pro developers and we have the citizen developers

171
00:11:51,740 --> 00:11:54,940
and we have this year...

172
00:11:54,940 --> 00:11:56,500
I don't know.

173
00:11:56,500 --> 00:11:58,340
They built their applications.

174
00:11:58,340 --> 00:11:59,900
We have the pro developers.

175
00:11:59,900 --> 00:12:05,580
How did you see both roles in this AI area?

176
00:12:05,580 --> 00:12:13,180
Both are really important.

177
00:12:13,180 --> 00:12:19,180
So when an organization looks at AI, for example, co-pilot,

178
00:12:19,700 --> 00:12:24,700
it's important to distinguish between...

179
00:12:24,700 --> 00:12:27,140
I'm thinking loud here.

180
00:12:27,140 --> 00:12:33,540
So, yeah, it's important to distinguish between the four different types of AI

181
00:12:33,540 --> 00:12:36,460
that Microsoft provides for us,

182
00:12:36,460 --> 00:12:38,980
starting with just...

183
00:12:38,980 --> 00:12:41,740
Are we saying we just want information,

184
00:12:41,740 --> 00:12:46,420
just give a prompt and get some information?

185
00:12:46,420 --> 00:12:50,020
So we're talking about a regular co-pilot

186
00:12:50,020 --> 00:12:53,300
and then there's M365 co-pilot,

187
00:12:53,300 --> 00:12:55,940
there is co-pilot studio,

188
00:12:55,940 --> 00:13:01,420
there's a gently A&I and AI foundry, right?

189
00:13:01,420 --> 00:13:03,020
So it is very important.

190
00:13:03,020 --> 00:13:06,420
The reason why I say AI center of excellence

191
00:13:06,420 --> 00:13:12,900
is somebody on the top needs to separate these use cases

192
00:13:12,900 --> 00:13:16,180
based on the complexity

193
00:13:16,180 --> 00:13:19,700
and the detailness of each of the use cases

194
00:13:19,700 --> 00:13:23,780
as to which co-pilot do you need, right?

195
00:13:23,780 --> 00:13:29,380
Sometimes you don't even need an agent.

196
00:13:29,380 --> 00:13:31,660
You could do it using a power automate flow.

197
00:13:31,660 --> 00:13:36,860
So what I've seen is a lot of comparison

198
00:13:36,860 --> 00:13:42,100
has been going on saying power platform versus co-pilot,

199
00:13:42,100 --> 00:13:49,100
whereas it should be power platform enabled by co-pilot.

200
00:13:49,100 --> 00:13:56,740
I think there was a really funny example.

201
00:13:56,740 --> 00:13:58,660
I think it was Amazon.

202
00:13:58,660 --> 00:14:01,620
They have the people paid for...

203
00:14:01,620 --> 00:14:06,740
or getting leather or something or budgets

204
00:14:06,740 --> 00:14:09,140
for using AI tokens, that I think.

205
00:14:09,140 --> 00:14:12,860
This was not a good idea, they also say it was a good idea

206
00:14:12,860 --> 00:14:17,620
because there are a lot of good machine learning

207
00:14:17,620 --> 00:14:19,940
with staff also and Microsoft.

208
00:14:19,940 --> 00:14:25,460
I say you don't need all the time using AI

209
00:14:25,460 --> 00:14:27,500
for a simple problem.

210
00:14:27,500 --> 00:14:29,340
That's beautiful.

211
00:14:29,340 --> 00:14:30,700
Yeah.

212
00:14:30,700 --> 00:14:31,700
Yeah.

213
00:14:31,700 --> 00:14:36,700
I've heard people saying,

214
00:14:36,700 --> 00:14:42,740
in general discussions, I heard some organizations

215
00:14:42,740 --> 00:14:47,100
are deciding whether to keep employees or not

216
00:14:47,100 --> 00:14:51,140
based on how many tokens they're burning a month.

217
00:14:51,140 --> 00:14:52,140
Yeah.

218
00:14:52,140 --> 00:14:55,860
Great idea.

219
00:14:55,860 --> 00:15:00,460
I don't know where it's taking them, but yeah.

220
00:15:00,460 --> 00:15:06,100
Yeah, I think when we work for OpenAO

221
00:15:06,100 --> 00:15:09,140
and Tropic, it's a good idea about, I don't know,

222
00:15:09,140 --> 00:15:12,700
if it's really good for companies.

223
00:15:12,700 --> 00:15:15,860
But we have this discoverments topics.

224
00:15:15,860 --> 00:15:22,860
And it's one thing I often discussed with was my clients

225
00:15:22,860 --> 00:15:25,060
and other people.

226
00:15:25,060 --> 00:15:30,100
And how should we balance innovation and control?

227
00:15:30,100 --> 00:15:32,300
Have you there any tips?

228
00:15:35,100 --> 00:15:41,700
So guardrails is definitely important,

229
00:15:41,700 --> 00:15:45,900
but it shouldn't definitely, you know,

230
00:15:45,900 --> 00:15:50,500
I always say governance should not be the cost.

231
00:15:50,500 --> 00:15:53,180
It should be the enabler, right?

232
00:15:53,180 --> 00:15:57,420
So I compare it with the traffic system,

233
00:15:57,420 --> 00:16:02,140
you know, just because you have a Lamborghini or you can afford

234
00:16:02,140 --> 00:16:07,140
a Lamborghini, you don't hand it to your 16 year old new drivers

235
00:16:07,140 --> 00:16:08,340
on, right?

236
00:16:08,340 --> 00:16:11,380
You ensure he gets the drivers manual,

237
00:16:11,380 --> 00:16:15,620
does the new religions and the road rules

238
00:16:15,620 --> 00:16:17,860
and then let him touch the steering.

239
00:16:17,860 --> 00:16:27,540
So my first trip would be to start with, you know,

240
00:16:27,540 --> 00:16:29,500
start with a small work force.

241
00:16:29,500 --> 00:16:34,420
So if you're asking the cities and developer to go through

242
00:16:34,420 --> 00:16:35,580
co-pilot, right?

243
00:16:35,580 --> 00:16:40,420
Like you just say, okay, take one or two workflows.

244
00:16:40,420 --> 00:16:44,820
Let us do the due religions on what data sources you're

245
00:16:44,820 --> 00:16:48,340
going to touch for them, just these two workflows.

246
00:16:48,340 --> 00:16:51,340
Example, they say, I'm going to build a co-pilot agent

247
00:16:51,340 --> 00:16:57,500
on the HR document library because I want to identify

248
00:16:57,500 --> 00:17:01,780
all the opportunities for this quarter, right?

249
00:17:01,780 --> 00:17:02,780
Something like that.

250
00:17:02,780 --> 00:17:08,420
And the governance team's goal should be to ensure

251
00:17:08,420 --> 00:17:11,900
that those three document libraries are not

252
00:17:11,900 --> 00:17:15,460
side that this workflow is going to touch.

253
00:17:15,460 --> 00:17:19,220
It is clean, the permissions are good.

254
00:17:19,220 --> 00:17:25,340
So I believe because companies are drowning in data right now.

255
00:17:25,340 --> 00:17:29,820
And if we try to boil the ocean, can't go too far.

256
00:17:29,820 --> 00:17:33,380
So that's one thing I'm thinking of.

257
00:17:33,380 --> 00:17:38,700
I'm actually building, I've built this entire

258
00:17:38,700 --> 00:17:44,020
triage system for my customers.

259
00:17:44,020 --> 00:17:48,660
If you're a vanilla co-pilot implementation company,

260
00:17:48,660 --> 00:17:50,700
you're just getting started.

261
00:17:50,700 --> 00:17:55,260
I would suggest create a governance portal first.

262
00:17:55,260 --> 00:18:02,100
And ensure all the guidance is in there in terms of acceptable AI

263
00:18:02,100 --> 00:18:05,540
usage, policy, and things like that.

264
00:18:05,540 --> 00:18:09,660
And I would start with a triage form, which

265
00:18:09,660 --> 00:18:13,260
will say, why are you building this agent?

266
00:18:13,260 --> 00:18:15,500
Who is the business of who?

267
00:18:15,500 --> 00:18:17,380
What is the business purpose?

268
00:18:17,380 --> 00:18:21,660
There are like a 20 different fields that you can fill in

269
00:18:21,660 --> 00:18:24,020
that you can use as audit trail later on.

270
00:18:24,620 --> 00:18:28,260
This is interesting.

271
00:18:28,260 --> 00:18:29,140
And there are two topics.

272
00:18:29,140 --> 00:18:36,060
I really, you say, but let us start with the HR example.

273
00:18:36,060 --> 00:18:39,660
So we have a lot of personal data.

274
00:18:39,660 --> 00:18:44,940
So how can we use this secure and compliant

275
00:18:44,940 --> 00:18:46,220
from your perspective?

276
00:18:51,340 --> 00:18:55,420
So there are some governance tools, Microsoft provides

277
00:18:55,420 --> 00:19:01,260
to check for DLP policies and you provide the details

278
00:19:01,260 --> 00:19:05,820
and it gives you the kindliness of your data.

279
00:19:05,820 --> 00:19:13,180
But I am actually building a tool for this.

280
00:19:13,180 --> 00:19:17,580
I call it the Justing Time AI governance framework,

281
00:19:17,580 --> 00:19:22,260
where once we know these are the data sources,

282
00:19:22,260 --> 00:19:27,140
you're going to touch maybe a few data worth tables,

283
00:19:27,140 --> 00:19:30,940
a couple of SharePoint sites, on click of a button,

284
00:19:30,940 --> 00:19:37,140
the tool uses Graph API to go scan all your data sources.

285
00:19:37,140 --> 00:19:41,820
And it gives you a remediation plan saying, yeah, you're

286
00:19:41,820 --> 00:19:43,620
good to go.

287
00:19:43,620 --> 00:19:50,100
So you check for the documents and see if there's PII,

288
00:19:50,100 --> 00:19:54,300
if the documents are labeled correctly, just check per view

289
00:19:54,300 --> 00:19:59,340
and see if the DLP policies are correctly in place,

290
00:19:59,340 --> 00:20:03,820
if the connectors are blocked and things like that.

291
00:20:03,820 --> 00:20:06,180
I believe the Justing Time governance

292
00:20:06,180 --> 00:20:08,900
will be hearing useful.

293
00:20:08,900 --> 00:20:14,780
Because and then intent versus outcome,

294
00:20:14,780 --> 00:20:18,740
then it's to be a comparison when the developer said,

295
00:20:18,740 --> 00:20:22,020
this is what I'm building, this is a goal,

296
00:20:22,020 --> 00:20:27,860
there should be a mechanism to evaluate what has been built.

297
00:20:27,860 --> 00:20:30,500
Is it matching the intent?

298
00:20:30,500 --> 00:20:34,340
Because that's where compliance meets.

299
00:20:37,340 --> 00:20:42,540
Is this tool available, actually?

300
00:20:42,540 --> 00:20:46,380
It's in progress to, I mean, I have deployed it in a couple

301
00:20:46,380 --> 00:20:51,140
of customers who, you know, where I consult and directly

302
00:20:51,140 --> 00:20:53,540
built it in the tenants.

303
00:20:53,540 --> 00:20:57,060
The goal is to make this available so that it can be deployed

304
00:20:57,060 --> 00:20:59,940
in any tenant that way.

305
00:20:59,940 --> 00:21:05,780
So it's not readily available right now, but we are close to that.

306
00:21:05,780 --> 00:21:08,420
This is cool, then we need another session, I think.

307
00:21:08,420 --> 00:21:09,860
This is really interesting.

308
00:21:09,860 --> 00:21:12,380
I really interesting in your tool.

309
00:21:12,380 --> 00:21:13,220
Thank you.

310
00:21:13,220 --> 00:21:20,420
So we have this topic, I say, I govlin's or I read this,

311
00:21:20,420 --> 00:21:26,100
or this has so many, many, many, many words,

312
00:21:26,100 --> 00:21:31,100
but I say I govlin's, this becoming, I think, a major topic,

313
00:21:31,100 --> 00:21:35,060
why is it so important, actually?

314
00:21:36,060 --> 00:21:41,380
Okay, so governance has always been there, right?

315
00:21:41,380 --> 00:21:48,140
Why it has suddenly become so important is because

316
00:21:48,140 --> 00:21:54,100
we've created many, many, many messy rooms in this house

317
00:21:54,100 --> 00:21:56,060
called an organization.

318
00:21:56,060 --> 00:22:00,500
We've made them messy, not intentionally,

319
00:22:00,500 --> 00:22:05,500
but somebody had created a table, a share point site,

320
00:22:05,500 --> 00:22:10,140
five years ago, set permissions to a certain degree

321
00:22:10,140 --> 00:22:14,500
and probably even left the organization or the department.

322
00:22:14,500 --> 00:22:19,500
All these siloed sites, I call them messy rooms,

323
00:22:19,500 --> 00:22:22,180
with lights turned off.

324
00:22:22,180 --> 00:22:26,460
I, if I'm a contractor in your organization

325
00:22:26,460 --> 00:22:29,820
and you have set the HR share point site

326
00:22:29,820 --> 00:22:34,300
on the change to everyone, at the rate everyone,

327
00:22:34,300 --> 00:22:38,420
you didn't know that because somebody else has set it,

328
00:22:38,420 --> 00:22:41,460
I don't know it because I'm a new contractor.

329
00:22:41,460 --> 00:22:46,140
So I wouldn't go and look up your HR documents

330
00:22:46,140 --> 00:22:51,140
because I'm unaware, but with a rise of co-pilot

331
00:22:51,140 --> 00:22:55,300
and citizen developers building agents,

332
00:22:56,300 --> 00:22:59,820
anybody can give a prompt and say,

333
00:22:59,820 --> 00:23:04,420
go look at these three share point sites and libraries

334
00:23:04,420 --> 00:23:07,620
and bring me all the details.

335
00:23:07,620 --> 00:23:10,620
And now that's a light bulb moment

336
00:23:10,620 --> 00:23:15,300
where your messy room is up now, right?

337
00:23:15,300 --> 00:23:19,020
That's why governance has become important.

338
00:23:19,020 --> 00:23:22,580
Earlier, it would be more like a check box

339
00:23:22,580 --> 00:23:25,820
that you would check for compliance and for audit.

340
00:23:25,820 --> 00:23:30,020
It was always an afterthought.

341
00:23:30,020 --> 00:23:32,940
I can't generalize too much.

342
00:23:32,940 --> 00:23:36,540
There would be organizations that have been meticulous

343
00:23:36,540 --> 00:23:41,540
from the get go maybe, but a large portion of the enterprises

344
00:23:41,540 --> 00:23:46,620
I have seen have been struggling with this

345
00:23:46,620 --> 00:23:50,780
because sometimes I used to call myself a king of us manager

346
00:23:50,780 --> 00:23:54,620
because once you're a senior SME,

347
00:23:54,620 --> 00:23:58,540
people bring you into projects where they say

348
00:23:58,540 --> 00:24:01,060
we have four thousand sites or you know,

349
00:24:01,060 --> 00:24:04,260
two thousand power platform solutions

350
00:24:04,260 --> 00:24:06,300
that people have built.

351
00:24:06,300 --> 00:24:09,780
I need you to help me govern them now

352
00:24:09,780 --> 00:24:12,340
because it was an afterthought.

353
00:24:12,340 --> 00:24:17,260
Yeah, I often see there are a lot of people

354
00:24:17,260 --> 00:24:21,340
or leaders in companies that are happy.

355
00:24:21,340 --> 00:24:23,660
Oh, my view is green.

356
00:24:23,660 --> 00:24:25,620
We are happy.

357
00:24:25,620 --> 00:24:32,100
But what are misconceptions do leaders

358
00:24:32,100 --> 00:24:36,380
or stakeholders often have a special when we talk

359
00:24:36,380 --> 00:24:40,300
about AI governance from your perspective or your project?

360
00:24:40,300 --> 00:24:46,860
A few places that I went to,

361
00:24:47,620 --> 00:24:50,980
it's sad, but there are many people

362
00:24:50,980 --> 00:24:54,860
that are not aware what governance means at all.

363
00:24:54,860 --> 00:25:01,260
So, misconceptions would be governance is a security team's project.

364
00:25:01,260 --> 00:25:08,820
That's the first one because people,

365
00:25:08,820 --> 00:25:10,380
it's about governing, right?

366
00:25:10,380 --> 00:25:14,660
So they think that it is, there's a GRC team,

367
00:25:14,660 --> 00:25:16,420
governance risk and compliance.

368
00:25:17,260 --> 00:25:22,260
Our team is responsible for AI governance.

369
00:25:22,260 --> 00:25:28,780
That's one misconception and then probably,

370
00:25:28,780 --> 00:25:34,260
another thing is people have heard the word

371
00:25:34,260 --> 00:25:37,540
the term responsible AI.

372
00:25:37,540 --> 00:25:42,060
So they also think that governance lives in the LLMs.

373
00:25:44,460 --> 00:25:46,220
wouldn't technology people,

374
00:25:46,220 --> 00:25:53,620
when people who take classes or go to YouTube to understand AI

375
00:25:53,620 --> 00:25:56,540
will attend the first class, right?

376
00:25:56,540 --> 00:25:58,780
It could be from anybody, we don't know.

377
00:25:58,780 --> 00:26:02,940
So they would just do their first video

378
00:26:02,940 --> 00:26:07,100
and I've known that responsible AI,

379
00:26:07,100 --> 00:26:12,100
I mean ethics and compliance and observability,

380
00:26:12,900 --> 00:26:16,980
these are the words they'll hear if they open a YouTube video

381
00:26:16,980 --> 00:26:18,780
on large language models.

382
00:26:18,780 --> 00:26:21,620
So they will think it's open AI's job

383
00:26:21,620 --> 00:26:24,820
or anthropic will take care of governance.

384
00:26:24,820 --> 00:26:27,540
So that's another one.

385
00:26:27,540 --> 00:26:33,100
- Yeah, one thing I think really interesting is

386
00:26:33,100 --> 00:26:34,740
we have this GRC,

387
00:26:34,740 --> 00:26:37,460
governance, risk and compliance topics

388
00:26:37,460 --> 00:26:39,700
and everything, it's all the same.

389
00:26:39,700 --> 00:26:40,860
But yeah,

390
00:26:41,860 --> 00:26:45,860
it's so interesting,

391
00:26:45,860 --> 00:26:50,860
but you have these awesome skills also in power platform.

392
00:26:50,860 --> 00:26:55,220
Let's a little bit talk about this.

393
00:26:55,220 --> 00:27:00,220
So power platform has evolved dramatically over the last few years.

394
00:27:00,220 --> 00:27:05,380
What changes excited you most?

395
00:27:08,300 --> 00:27:12,060
- I have, so within power platform,

396
00:27:12,060 --> 00:27:16,220
my expertise in power apps,

397
00:27:16,220 --> 00:27:20,620
so I've built a lot of Canvas apps and power automate.

398
00:27:20,620 --> 00:27:23,980
These are my favorite tools.

399
00:27:23,980 --> 00:27:28,340
I, obviously, like everything that I don't want to,

400
00:27:28,340 --> 00:27:32,140
I believe it's important to specialize

401
00:27:32,140 --> 00:27:34,700
and go deep in a few things,

402
00:27:34,700 --> 00:27:39,700
while learning what all the other tools can do for you.

403
00:27:39,700 --> 00:27:42,900
So I had focused on Canvas apps,

404
00:27:42,900 --> 00:27:45,660
more YouTube apps and power automate.

405
00:27:45,660 --> 00:27:52,980
And tremendous changes in the last couple of years.

406
00:27:52,980 --> 00:27:57,980
And the ability to use AI to build Canvas apps

407
00:27:57,980 --> 00:28:01,980
and

408
00:28:01,980 --> 00:28:09,420
just trying to think of the list of

409
00:28:09,420 --> 00:28:12,220
our new content came in.

410
00:28:12,220 --> 00:28:15,180
It's a never ending day to day list.

411
00:28:15,180 --> 00:28:19,380
Like yesterday, you thought of something and today it's changed, right?

412
00:28:19,380 --> 00:28:24,580
- Yeah, that's really interesting.

413
00:28:24,580 --> 00:28:27,100
Have you tried out this new, I don't know,

414
00:28:27,100 --> 00:28:30,260
it's code apps, I think, in power apps?

415
00:28:31,580 --> 00:28:36,580
- Yes, I've built a couple of apps using code apps,

416
00:28:36,580 --> 00:28:40,820
but I am still stuck at vibe coding

417
00:28:40,820 --> 00:28:44,740
because more than code apps,

418
00:28:44,740 --> 00:28:48,620
I've tried more of AI-poundry.

419
00:28:48,620 --> 00:28:53,620
So I like how the fusion development

420
00:28:53,620 --> 00:28:56,980
is what I'm a fan of.

421
00:28:56,980 --> 00:29:00,700
Like, take your power apps

422
00:29:00,700 --> 00:29:05,700
and supercharge them as your functions, for example.

423
00:29:05,700 --> 00:29:07,660
So when you do those things,

424
00:29:07,660 --> 00:29:12,660
it's a lot of win and now with AI and it's easier

425
00:29:12,660 --> 00:29:16,900
to build your function.

426
00:29:16,900 --> 00:29:21,940
And then, so yeah, building Canvas apps

427
00:29:21,940 --> 00:29:27,620
with AI capabilities, that's what I mean.

428
00:29:28,980 --> 00:29:31,700
Did you think the power platform,

429
00:29:31,700 --> 00:29:33,300
I think a lot of executive stuff,

430
00:29:33,300 --> 00:29:35,580
so it's only for citizenship,

431
00:29:35,580 --> 00:29:39,780
but it's also for pro-re-development,

432
00:29:39,780 --> 00:29:42,260
and will you say it's enterprise-ready?

433
00:29:42,260 --> 00:29:46,900
- It is pretty enterprise-ready right now.

434
00:29:46,900 --> 00:29:51,420
The ALM has evolved so much as your DevOps

435
00:29:51,420 --> 00:29:56,260
and the ALM strategy is this application lifecycle management

436
00:29:56,260 --> 00:30:01,260
for everything power platform.

437
00:30:01,260 --> 00:30:05,580
It lives in solutions and environment variables.

438
00:30:05,580 --> 00:30:08,860
And so the lifecycle, as long as you're doing

439
00:30:08,860 --> 00:30:11,140
the lifecycle management right,

440
00:30:11,140 --> 00:30:13,860
it is definitely enterprise-ready.

441
00:30:13,860 --> 00:30:17,500
Again, it was done to

442
00:30:17,500 --> 00:30:22,980
we as developers building expertise

443
00:30:23,820 --> 00:30:28,820
in each of the features that are available.

444
00:30:28,820 --> 00:30:34,820
So we have ALM features

445
00:30:34,820 --> 00:30:39,340
simply sitting in as your DevOps

446
00:30:39,340 --> 00:30:43,060
and you need to understand solutions,

447
00:30:43,060 --> 00:30:46,900
you need to understand why you need to create

448
00:30:46,900 --> 00:30:50,700
a solution where workflows spread.

449
00:30:50,700 --> 00:30:53,780
There is so much in the building of it

450
00:30:53,780 --> 00:31:00,900
that is involved in ensuring that you're creating enterprise-ready

451
00:31:00,900 --> 00:31:04,140
applications.

452
00:31:04,140 --> 00:31:09,140
So it is enterprise-ready if you are building your capabilities

453
00:31:09,140 --> 00:31:11,940
right within your TX.

454
00:31:11,940 --> 00:31:20,420
- What did you think did we draw in a world

455
00:31:20,420 --> 00:31:22,620
or a future where let's where language

456
00:31:22,620 --> 00:31:25,540
becomes the primary developed enterprise?

457
00:31:25,540 --> 00:31:30,500
- It will, it is already becoming,

458
00:31:30,500 --> 00:31:37,220
yeah, I have a huge take on that natural language.

459
00:31:37,220 --> 00:31:40,060
So low-code power was built because,

460
00:31:40,060 --> 00:31:46,500
I mean, it was built even slightly before

461
00:31:46,500 --> 00:31:48,700
natural language took over,

462
00:31:48,700 --> 00:31:52,860
but because we want to do,

463
00:31:52,860 --> 00:31:55,500
we want to help organizations,

464
00:31:55,500 --> 00:31:59,060
harness natural language, programming

465
00:31:59,060 --> 00:32:03,940
and application building in the right way.

466
00:32:03,940 --> 00:32:08,100
And I see totally that development

467
00:32:08,100 --> 00:32:12,460
is becoming a natural language framework, right?

468
00:32:12,460 --> 00:32:14,980
That is how it's gonna be in the future

469
00:32:14,980 --> 00:32:17,220
even for pro-code developers.

470
00:32:18,220 --> 00:32:21,220
So people who understand,

471
00:32:21,220 --> 00:32:24,060
you see how we started with prompt engineering

472
00:32:24,060 --> 00:32:27,100
that said let's build good prompts

473
00:32:27,100 --> 00:32:31,140
and then make a library of the reusable prompts.

474
00:32:31,140 --> 00:32:35,260
Then they realized from engineering means

475
00:32:35,260 --> 00:32:40,260
you're assigning tasks to AI.

476
00:32:40,260 --> 00:32:43,060
You're saying do this for me

477
00:32:43,060 --> 00:32:45,820
or you're asking it some questions.

478
00:32:45,820 --> 00:32:48,620
Then we realized that this is moving

479
00:32:48,620 --> 00:32:53,620
from prompt engineering to context engineering, right?

480
00:32:53,620 --> 00:32:58,620
And context differs, so you can create a prompt library

481
00:32:58,620 --> 00:33:01,300
but can you create a context library

482
00:33:01,300 --> 00:33:05,900
is a debatable question, right?

483
00:33:05,900 --> 00:33:10,900
So I believe now that organizations

484
00:33:10,900 --> 00:33:15,460
need to create global prompt libraries

485
00:33:16,180 --> 00:33:21,020
of approved prompts wherein everybody,

486
00:33:21,020 --> 00:33:25,100
you don't have to be an expert at prompting,

487
00:33:25,100 --> 00:33:29,380
you could use existing prompts from your library,

488
00:33:29,380 --> 00:33:31,700
but when it comes to context,

489
00:33:31,700 --> 00:33:34,620
it should be a department level venture,

490
00:33:34,620 --> 00:33:37,700
each department, for example HR, right?

491
00:33:37,700 --> 00:33:40,980
In the HR department,

492
00:33:40,980 --> 00:33:44,900
some citizen developer is building an agent,

493
00:33:44,900 --> 00:33:47,780
and they will need a prompt,

494
00:33:47,780 --> 00:33:50,300
they will also need to provide context.

495
00:33:50,300 --> 00:33:54,260
And if they're new to the organization

496
00:33:54,260 --> 00:33:57,620
or not everybody has complete context

497
00:33:57,620 --> 00:33:59,500
of your department, right?

498
00:33:59,500 --> 00:34:03,660
So I believe at the department level,

499
00:34:03,660 --> 00:34:07,380
there needs to be approved context,

500
00:34:07,380 --> 00:34:10,700
list of context built

501
00:34:12,020 --> 00:34:16,620
and somebody needs to take ownership of keeping them current.

502
00:34:16,620 --> 00:34:21,220
For example, today you have a certain strategy

503
00:34:21,220 --> 00:34:26,220
then tomorrow as a department, you have pivoted, right?

504
00:34:26,220 --> 00:34:29,980
You would want the agent to gain

505
00:34:29,980 --> 00:34:33,900
to get the latest context of that department.

506
00:34:33,900 --> 00:34:39,860
So a global prompt library and local department

507
00:34:39,860 --> 00:34:41,860
wise complex libraries.

508
00:34:41,860 --> 00:34:49,380
And I'll talk about frameworks going forward,

509
00:34:49,380 --> 00:34:52,780
but that framework libraries is also something

510
00:34:52,780 --> 00:34:57,260
I'm thinking about because with natural language,

511
00:34:57,260 --> 00:34:59,660
everybody has a different way of prompting,

512
00:34:59,660 --> 00:35:02,700
different way of talking to doing I,

513
00:35:02,700 --> 00:35:06,580
but as an organization, you want to stand for the same thing,

514
00:35:07,700 --> 00:35:10,260
right, that's where governance happens.

515
00:35:10,260 --> 00:35:15,980
You say we have this context libraries.

516
00:35:15,980 --> 00:35:23,500
How did they compare to co-pilot skills or skills in AI?

517
00:35:23,500 --> 00:35:29,700
I'd say co-pilot skills is a technology piece

518
00:35:29,700 --> 00:35:35,540
and prompting or context engineering is the domain part.

519
00:35:35,540 --> 00:35:42,540
So any agent that you build should not come from the thought

520
00:35:42,540 --> 00:35:47,540
that I now have mastered creating co-pilot agents.

521
00:35:47,540 --> 00:35:48,980
So let me create the next one.

522
00:35:48,980 --> 00:35:55,740
You mastered it and now the goal is to understand

523
00:35:55,740 --> 00:35:58,740
a company or a department's business

524
00:35:58,740 --> 00:36:02,540
and their current way of working

525
00:36:03,540 --> 00:36:08,540
and identify use cases that can be automated

526
00:36:08,540 --> 00:36:12,020
and AI sprinkled.

527
00:36:12,020 --> 00:36:16,500
So the business is achieved better outcomes.

528
00:36:16,500 --> 00:36:19,740
Unless that part is figured,

529
00:36:19,740 --> 00:36:21,860
you shouldn't start building agents

530
00:36:21,860 --> 00:36:23,580
unless you're doing it for practice.

531
00:36:23,580 --> 00:36:28,540
Yeah, let's talk about agents.

532
00:36:28,540 --> 00:36:31,220
How did you define a Genetic AI?

533
00:36:33,020 --> 00:36:37,540
A Genetic AI again is, you have seen people understanding

534
00:36:37,540 --> 00:36:42,540
and using AI agents and a Genetic AI interchangeably.

535
00:36:42,540 --> 00:36:47,620
And that is a huge piece.

536
00:36:47,620 --> 00:36:52,620
So we are moving from the step one was asking AI

537
00:36:52,620 --> 00:36:59,060
and it meticulously giving you answers to anything you ask

538
00:37:00,660 --> 00:37:05,220
and then we've moved from there to deliverables.

539
00:37:05,220 --> 00:37:10,220
I just did a presentation on the concept of prompt

540
00:37:10,220 --> 00:37:12,740
to deliverable.

541
00:37:12,740 --> 00:37:15,980
So the goal should now be,

542
00:37:15,980 --> 00:37:20,980
I will just give you one prompt and the agent,

543
00:37:20,980 --> 00:37:23,860
I'm coming to agent the AI.

544
00:37:23,860 --> 00:37:26,740
I just wanted to talk about the agent first.

545
00:37:26,740 --> 00:37:30,980
So we evolved correctly into agent the AI.

546
00:37:30,980 --> 00:37:34,020
So we give a prompt.

547
00:37:34,020 --> 00:37:38,340
So I did one for company profiling.

548
00:37:38,340 --> 00:37:42,940
If you're in the sales team and monthly,

549
00:37:42,940 --> 00:37:46,780
weekly, you have to vet customers

550
00:37:46,780 --> 00:37:50,380
and come up with an entire customer profile

551
00:37:50,380 --> 00:37:52,820
for like 50 customers, 100 customers.

552
00:37:52,820 --> 00:37:56,420
That is a manual process, right?

553
00:37:56,420 --> 00:38:00,020
So the goal should be just say give me the company profile

554
00:38:00,020 --> 00:38:05,020
for say M365 show and it needs to give you a detailed

555
00:38:05,020 --> 00:38:11,620
templated, predefined,

556
00:38:11,620 --> 00:38:16,460
output, a deliverable that you can directly use

557
00:38:16,460 --> 00:38:17,500
for your sales call.

558
00:38:17,500 --> 00:38:18,580
That's it.

559
00:38:18,580 --> 00:38:22,300
Everything else should happen in the co-pilot studio

560
00:38:22,300 --> 00:38:23,060
orchestration.

561
00:38:23,060 --> 00:38:26,100
Now this is a simpler example.

562
00:38:26,100 --> 00:38:30,700
So this could be achieved using a co-pilot agent.

563
00:38:30,700 --> 00:38:35,700
So basically co-pilot agent is your task takeer.

564
00:38:35,700 --> 00:38:40,060
It's like a developer that you assign a task and say,

565
00:38:40,060 --> 00:38:45,060
I want to accomplish this task and the agent will do it for you.

566
00:38:45,060 --> 00:38:49,340
And agentic AI is a concept.

567
00:38:49,340 --> 00:38:55,140
It's a system of agents that are able to talk to each other

568
00:38:55,700 --> 00:39:00,700
and can retrieve context and make decisions on the fly.

569
00:39:00,700 --> 00:39:09,340
How did you see in this discussion between agents?

570
00:39:09,340 --> 00:39:12,140
What role did MCP play?

571
00:39:12,140 --> 00:39:16,140
Oh, MCP does play a huge role.

572
00:39:16,140 --> 00:39:20,660
MCP is your unifying layer because

573
00:39:22,140 --> 00:39:27,140
it's similar to APIs for your development world.

574
00:39:27,140 --> 00:39:32,420
So I'll give a two simple example.

575
00:39:32,420 --> 00:39:35,460
If you wanted to find out what the weather is

576
00:39:35,460 --> 00:39:39,740
and show it on your website, you would just pick the API

577
00:39:39,740 --> 00:39:44,740
for the weather channel or that particular company's API

578
00:39:44,740 --> 00:39:49,940
and code it into your website to

579
00:39:51,020 --> 00:39:53,300
get the details from that company.

580
00:39:53,300 --> 00:39:59,140
Now that's the same role, similar role and MCP plays.

581
00:39:59,140 --> 00:40:04,820
It's called a model context protocol,

582
00:40:04,820 --> 00:40:09,820
which is a unified layer for you to access any data source.

583
00:40:09,820 --> 00:40:16,620
For example, you would use like a MCP,

584
00:40:17,580 --> 00:40:22,580
a SharePoint MCP that within your canvas app

585
00:40:22,580 --> 00:40:28,140
so that everything you get now, it is grounded.

586
00:40:28,140 --> 00:40:29,940
It helps in grounding.

587
00:40:29,940 --> 00:40:33,780
We haven't covered the concept of grounding within AI,

588
00:40:33,780 --> 00:40:36,940
but now that you're here.

589
00:40:36,940 --> 00:40:41,940
So we've heard that AI hallucinates often, right?

590
00:40:43,900 --> 00:40:48,900
That's because there is no ground,

591
00:40:48,900 --> 00:40:54,860
your information or your agent has not been grounded

592
00:40:54,860 --> 00:40:58,340
in correct information.

593
00:40:58,340 --> 00:41:01,900
What MCP does is ensures that

594
00:41:01,900 --> 00:41:06,140
you are venue access SharePoint

595
00:41:06,140 --> 00:41:10,860
instead of doing it directly when you do it through an MCP,

596
00:41:10,860 --> 00:41:15,860
it allows you to access data from your data source

597
00:41:15,860 --> 00:41:21,820
in a unified standardized method.

598
00:41:21,820 --> 00:41:28,220
- Yeah, I'm more of the, I found the dude,

599
00:41:28,220 --> 00:41:30,820
and okay, Microsoft will hate me.

600
00:41:30,820 --> 00:41:33,700
I never had to look into a co-pilot studio,

601
00:41:33,700 --> 00:41:38,700
but it's becoming such a strategic platform for enterprises.

602
00:41:40,020 --> 00:41:43,300
Why, I don't understand, can you explain this?

603
00:41:43,300 --> 00:41:48,060
- Sure.

604
00:41:48,060 --> 00:41:53,060
So, co-pilot studio is your, as the name suggests actually,

605
00:41:53,060 --> 00:41:59,100
it is that studio, it is your orchestration engine,

606
00:41:59,100 --> 00:42:03,460
so there are a bunch of things

607
00:42:03,460 --> 00:42:07,020
that you want your agents to do,

608
00:42:07,020 --> 00:42:12,020
whether it is a single agent or now agent AI

609
00:42:12,020 --> 00:42:17,020
where you're saying you would provide an epic of details

610
00:42:17,020 --> 00:42:22,020
and your first agent gets triggered

611
00:42:22,020 --> 00:42:26,860
and it automatically calls other agents

612
00:42:26,860 --> 00:42:31,860
and sometimes even creates some other artifacts for you

613
00:42:31,860 --> 00:42:34,220
to talk to each other.

614
00:42:34,220 --> 00:42:38,020
So, it is like a bunch of developers supposing you have

615
00:42:38,020 --> 00:42:42,620
a 15 developer team, right?

616
00:42:42,620 --> 00:42:47,620
Now, you can completely replace that team

617
00:42:47,620 --> 00:42:51,620
with agent AI architecture.

618
00:42:51,620 --> 00:42:58,460
And co-pilot studio is the engine for it.

619
00:42:58,460 --> 00:43:02,660
It won't, you do not want to give it the full load,

620
00:43:02,660 --> 00:43:06,420
you will want to utilize the other parts

621
00:43:06,420 --> 00:43:12,900
that, for example, the knowledge sources, the tools,

622
00:43:12,900 --> 00:43:17,220
you know, where you add the MCP server to it,

623
00:43:17,220 --> 00:43:20,220
you're making this co-pilot agent ProCode,

624
00:43:20,220 --> 00:43:23,380
the co-pilot studio agent ProCode,

625
00:43:23,380 --> 00:43:28,380
but the studio agent itself is the orchestration engine.

626
00:43:28,380 --> 00:43:30,580
You can call agent flows into it,

627
00:43:30,580 --> 00:43:35,580
you can call, you know, you can provide MCP servers to it

628
00:43:35,580 --> 00:43:39,980
and that is where everything will live.

629
00:43:39,980 --> 00:43:44,980
So, going forward, co-pilot studio governance,

630
00:43:44,980 --> 00:43:52,500
again, becomes very, very important

631
00:43:52,500 --> 00:43:54,980
because now you're not even saying

632
00:43:54,980 --> 00:43:57,660
there are five agents that I need to govern.

633
00:43:57,660 --> 00:44:02,660
You're saying the agents will take decisions on the fly

634
00:44:02,660 --> 00:44:06,780
and so I'll give you an example.

635
00:44:06,780 --> 00:44:13,460
For example, we're going long division to a student, right?

636
00:44:13,460 --> 00:44:18,260
A school student, you're teaching long division.

637
00:44:18,260 --> 00:44:23,260
So they start, you know, some number of 5,200 divided

638
00:44:24,060 --> 00:44:29,060
by 24, each us would take not just the answer,

639
00:44:29,060 --> 00:44:31,660
they'll say show your work, correct?

640
00:44:31,660 --> 00:44:33,980
They'll say show how you did this

641
00:44:33,980 --> 00:44:37,300
and how you arrived at the answer.

642
00:44:37,300 --> 00:44:41,540
That is what we will be doing going forward

643
00:44:41,540 --> 00:44:45,100
with co-pilot, you know, agent AI,

644
00:44:45,100 --> 00:44:50,100
because the final outcome might be correct,

645
00:44:50,100 --> 00:44:53,740
but in the second step,

646
00:44:53,740 --> 00:44:56,700
they could have in a long decision.

647
00:44:56,700 --> 00:45:03,500
So now the goal is to govern the entire step-by-step

648
00:45:03,500 --> 00:45:07,860
governance is needed to make sure every step

649
00:45:07,860 --> 00:45:10,060
that agent has taken correct decisions.

650
00:45:10,060 --> 00:45:15,060
- This is really cool.

651
00:45:15,060 --> 00:45:18,980
Can you walk us without exposing a client

652
00:45:18,980 --> 00:45:20,980
through a real world scenario

653
00:45:20,980 --> 00:45:25,980
where a agent flows, creates a significant business value?

654
00:45:25,980 --> 00:45:29,220
- Yeah, sure.

655
00:45:29,220 --> 00:45:34,220
So I help an R&D tax credits boutique firm.

656
00:45:34,220 --> 00:45:43,100
And IRS process for calculating R&D tax credits

657
00:45:43,100 --> 00:45:49,940
and evaluating it is a very, very cumbersome process.

658
00:45:50,660 --> 00:45:55,660
So reuse power platform and co-pilot agents together

659
00:45:55,660 --> 00:46:04,020
to make it easy for their customers.

660
00:46:04,020 --> 00:46:09,020
So for example, a huge giant company,

661
00:46:09,020 --> 00:46:12,460
we will think of Microsoft because it's easy.

662
00:46:12,460 --> 00:46:16,140
Imagine Microsoft is doing its R&D tax credits.

663
00:46:16,140 --> 00:46:19,700
So the IRS looks for,

664
00:46:20,220 --> 00:46:24,060
proof that for each department

665
00:46:24,060 --> 00:46:27,060
that you want to claim R&D tax credits,

666
00:46:27,060 --> 00:46:31,620
you want to show, you need to show the experimentation

667
00:46:31,620 --> 00:46:35,940
gone and the documentation you created, the patents,

668
00:46:35,940 --> 00:46:42,540
there's like tons of work, it's called Form 6.765.

669
00:46:42,540 --> 00:46:46,940
It is a firm that companies need to fill, it's huge.

670
00:46:46,940 --> 00:46:50,860
So we are automating that for their customers.

671
00:46:50,860 --> 00:46:53,700
So I use Canvas apps.

672
00:46:53,700 --> 00:46:59,900
So I start with a Teams channel in the customers environment

673
00:46:59,900 --> 00:47:04,420
because the goal is to make sure data is extremely secure

674
00:47:04,420 --> 00:47:09,140
and it is in the customer's tenant.

675
00:47:09,140 --> 00:47:11,180
Everything should be within their own tenant

676
00:47:11,180 --> 00:47:12,940
so nothing goes out.

677
00:47:12,940 --> 00:47:15,740
So we start with a Teams channel

678
00:47:15,740 --> 00:47:18,060
and use the underlying SharePoint.

679
00:47:18,060 --> 00:47:24,140
How every Teams comes with its underlying SharePoint.

680
00:47:24,140 --> 00:47:29,140
So not SharePoints, I had created lists and libraries in there

681
00:47:29,140 --> 00:47:33,660
and use that as the data source

682
00:47:33,660 --> 00:47:37,780
and create Canvas app that goes through the entire process

683
00:47:37,780 --> 00:47:42,780
of qualitative and quantitative proof

684
00:47:43,220 --> 00:47:47,460
of a company doing research and development.

685
00:47:47,460 --> 00:47:51,220
Sorry.

686
00:47:51,220 --> 00:47:58,340
How I do that is the company identifies subject matter experts

687
00:47:58,340 --> 00:48:03,220
in the managers in their organization, right?

688
00:48:03,220 --> 00:48:08,220
They fill out this long Canvas app and attached documents

689
00:48:09,300 --> 00:48:13,620
and then on top of this, I use SharePoint MCP

690
00:48:13,620 --> 00:48:20,300
to query and provide reports on the data that has been,

691
00:48:20,300 --> 00:48:27,900
you know, the work that has been done for experimentation.

692
00:48:27,900 --> 00:48:32,500
And we could also, you know, for the same company,

693
00:48:32,500 --> 00:48:36,300
use the Prompto-Deliverable Google that I told you

694
00:48:36,300 --> 00:48:38,700
for customer research, right?

695
00:48:38,700 --> 00:48:43,700
So I built agents that do customer research

696
00:48:43,700 --> 00:48:47,420
that also do, like, you've done a sales call,

697
00:48:47,420 --> 00:48:52,500
you can take the transcript and completely build an agent

698
00:48:52,500 --> 00:48:56,700
on top of it that will give you ready to use deliverables

699
00:48:56,700 --> 00:48:59,700
after your sales call.

700
00:48:59,700 --> 00:49:02,700
So, you know, there's a huge workflow.

701
00:49:02,700 --> 00:49:05,700
Yeah, this is a really huge workflow.

702
00:49:05,700 --> 00:49:11,700
When we talk about an agent, there is a lot of basmurts

703
00:49:11,700 --> 00:49:15,700
but something is really interesting and it's called reasoning.

704
00:49:15,700 --> 00:49:20,700
Can you explain what is reasoning and what role it play

705
00:49:20,700 --> 00:49:23,700
in AI agents?

706
00:49:23,700 --> 00:49:27,700
Reasoning.

707
00:49:27,700 --> 00:49:30,700
Yeah, so for example, like I mentioned earlier,

708
00:49:33,700 --> 00:49:38,700
you might just think about humans, how we do reasoning, right?

709
00:49:38,700 --> 00:49:39,700
Yeah.

710
00:49:39,700 --> 00:49:45,700
If you're given something to do or if you're thinking about something,

711
00:49:45,700 --> 00:49:50,700
your reason based on your knowledge, right?

712
00:49:50,700 --> 00:49:52,700
So you have based on what you know,

713
00:49:52,700 --> 00:49:54,700
supposing I want to invest in a stock.

714
00:49:54,700 --> 00:49:58,700
I will reason based on how much money I have,

715
00:49:58,700 --> 00:50:01,700
my risk appetite,

716
00:50:01,700 --> 00:50:03,700
do I want to invest,

717
00:50:03,700 --> 00:50:05,700
do I want some cash in hand,

718
00:50:05,700 --> 00:50:08,700
which companies are doing well?

719
00:50:08,700 --> 00:50:11,700
So there are a bunch of things that you need to know

720
00:50:11,700 --> 00:50:15,700
in order to be able to invest in a stock.

721
00:50:15,700 --> 00:50:19,700
So compare this with reasoning in agents, right?

722
00:50:19,700 --> 00:50:22,700
It is a similar concept.

723
00:50:22,700 --> 00:50:27,700
Basically, if you're assigning some job to an agent

724
00:50:27,700 --> 00:50:29,700
or a set of agents,

725
00:50:29,700 --> 00:50:34,700
you want them to reason before they make decisions.

726
00:50:34,700 --> 00:50:38,700
That's where I was talking about step-step governance

727
00:50:38,700 --> 00:50:41,700
to make sure that every step

728
00:50:41,700 --> 00:50:44,700
if the agent has taken the right decision,

729
00:50:44,700 --> 00:50:47,700
that stems from reasoning.

730
00:50:47,700 --> 00:50:50,700
Yeah, awesome.

731
00:50:50,700 --> 00:50:53,700
I think a little bit about,

732
00:50:53,700 --> 00:50:58,700
or how did you think about orchestration versus autonomy?

733
00:50:58,700 --> 00:51:01,700
What did you think about this?

734
00:51:01,700 --> 00:51:02,700
Is it the first?

735
00:51:02,700 --> 00:51:03,700
I don't know.

736
00:51:03,700 --> 00:51:10,700
How will you make your decision?

737
00:51:10,700 --> 00:51:16,700
Autonomy, yeah, I believe it's definitely not

738
00:51:16,700 --> 00:51:24,700
completely recommended to give an agent AI system autonomous.

739
00:51:24,700 --> 00:51:27,700
Definitely there needs to be humans in the loop.

740
00:51:27,700 --> 00:51:32,700
A very, very important concept as much as we say

741
00:51:32,700 --> 00:51:36,700
that AI can do everything.

742
00:51:36,700 --> 00:51:41,700
We have the capability to create a agent AI.

743
00:51:41,700 --> 00:51:44,700
Some decisions, you know,

744
00:51:44,700 --> 00:51:48,700
there are two things attached here.

745
00:51:48,700 --> 00:51:52,700
One is business decisions where some approvals

746
00:51:52,700 --> 00:51:56,700
need to come from senior level executives.

747
00:51:56,700 --> 00:52:01,700
You do not want agents to make decisions

748
00:52:01,700 --> 00:52:05,700
that involve security, risk, compliance,

749
00:52:05,700 --> 00:52:07,700
and things like that.

750
00:52:07,700 --> 00:52:13,700
Even if they do, we need human oversight on any autonomous agents.

751
00:52:13,700 --> 00:52:17,700
So it's a combination of understanding

752
00:52:17,700 --> 00:52:19,700
what you need to leave to an agent

753
00:52:19,700 --> 00:52:25,700
versus what needs to be controlled by a human.

754
00:52:25,700 --> 00:52:31,700
Even technically there are some places where,

755
00:52:31,700 --> 00:52:38,700
for example, you've asked to build a huge application.

756
00:52:38,700 --> 00:52:42,700
You could even talk about Claude or ClaudeCore,

757
00:52:42,700 --> 00:52:51,700
for example, and it built you millions of lines of code.

758
00:52:51,700 --> 00:52:56,700
How do you review all this?

759
00:52:56,700 --> 00:53:01,700
There's a lot involved in this because imagine your current team leaves.

760
00:53:01,700 --> 00:53:04,700
What happens to all that code that was built?

761
00:53:04,700 --> 00:53:08,700
How would another income and understand what was built?

762
00:53:08,700 --> 00:53:14,700
So there's that part, scalability, reliability,

763
00:53:14,700 --> 00:53:18,700
keep it going, first of all, that's one part.

764
00:53:18,700 --> 00:53:21,700
And the other part is business outcomes.

765
00:53:21,700 --> 00:53:26,700
Do we, once a workflow reaches a point,

766
00:53:26,700 --> 00:53:30,700
do we, you need to route it to a human

767
00:53:30,700 --> 00:53:35,700
and then some review be done and then get back to the agent.

768
00:53:35,700 --> 00:53:38,700
That would be done.

769
00:53:38,700 --> 00:53:40,700
Awesome.

770
00:53:40,700 --> 00:53:41,700
I think a little bit,

771
00:53:41,700 --> 00:53:46,700
what your tips or ideas or,

772
00:53:46,700 --> 00:53:51,700
let me change this question.

773
00:53:51,700 --> 00:53:56,700
How do you prevent agents from becoming expensive, unreliable,

774
00:53:56,700 --> 00:53:59,700
or difficult to govern?

775
00:53:59,700 --> 00:54:01,700
How do you do this?

776
00:54:01,700 --> 00:54:07,700
AI governance from the one that is the whole code.

777
00:54:07,700 --> 00:54:09,700
I say very fast,

778
00:54:09,700 --> 00:54:15,700
so that the tagline of local power is very fast,

779
00:54:15,700 --> 00:54:18,700
and the first govern faster.

780
00:54:18,700 --> 00:54:22,700
So with all these capabilities in hand,

781
00:54:22,700 --> 00:54:27,700
completely start with governance at every level.

782
00:54:27,700 --> 00:54:32,700
Having governance team or somebody who's overseeing governance,

783
00:54:32,700 --> 00:54:37,700
a governance portal, training practitioners and developers

784
00:54:37,700 --> 00:54:40,700
to build with a governance mindset.

785
00:54:40,700 --> 00:54:44,700
So governance layered at various levels is how you accomplish.

786
00:54:44,700 --> 00:54:49,700
You know, preventing that swarm of agents

787
00:54:49,700 --> 00:54:52,700
that becomes uncontrollable.

788
00:54:52,700 --> 00:54:54,700
Awesome.

789
00:54:54,700 --> 00:54:57,700
Yeah, oh, we run a little bit off time.

790
00:54:57,700 --> 00:55:00,700
So we have a joint in the rapid fire round.

791
00:55:00,700 --> 00:55:07,700
So I ask a question and you tell me you work first, come in your mind.

792
00:55:07,700 --> 00:55:09,700
So exciting.

793
00:55:09,700 --> 00:55:13,700
Yeah, one Microsoft technology, more people should pay attention to.

794
00:55:13,700 --> 00:55:18,700
Co-pilot studio.

795
00:55:18,700 --> 00:55:25,700
What's the most underrated platform feature?

796
00:55:25,700 --> 00:55:27,700
Platform feature?

797
00:55:27,700 --> 00:55:30,700
Yeah.

798
00:55:30,700 --> 00:55:35,700
Deal.

799
00:55:35,700 --> 00:55:41,700
When Microsoft gives you all resources you need, money, personal,

800
00:55:41,700 --> 00:55:46,700
and what will you develop?

801
00:55:46,700 --> 00:55:51,700
Wow, that's not a rapid fire question.

802
00:55:51,700 --> 00:55:56,700
Let me take.

803
00:55:56,700 --> 00:56:01,700
I will develop an affordable,

804
00:56:01,700 --> 00:56:06,700
agent AI governance framework for every organization.

805
00:56:06,700 --> 00:56:13,700
A lot of AI trends are interesting, something is overhyped.

806
00:56:13,700 --> 00:56:26,700
Did you see the what trend will you see have have a future and whatnot?

807
00:56:26,700 --> 00:56:31,700
Vibe coding is overhyped.

808
00:56:31,700 --> 00:56:37,700
One pro-tib-habit you can live about.

809
00:56:37,700 --> 00:56:40,700
Can you repeat?

810
00:56:40,700 --> 00:56:44,700
What pro-tib-t-habit you have you can live about?

811
00:56:44,700 --> 00:56:51,700
I have built something for myself called life operating system in my SharePoint tenant.

812
00:56:51,700 --> 00:57:00,700
It's personal, so I am a mom of two kids and I have like 2000 things that I am working on at the same time.

813
00:57:00,700 --> 00:57:11,700
So I've built an entire system with now with co-pilot in it that I can't live without anymore.

814
00:57:11,700 --> 00:57:18,700
What's the best career advice you ever received?

815
00:57:18,700 --> 00:57:21,700
It's not enough that you do great work.

816
00:57:21,700 --> 00:57:27,700
It's important to show your work and show the world what you're doing.

817
00:57:27,700 --> 00:57:32,700
That they're one book or conference or learning resource.

818
00:57:32,700 --> 00:57:41,700
Everyone working with the A-should-read or listen to.

819
00:57:41,700 --> 00:57:45,700
That's interesting, I didn't think about that.

820
00:57:45,700 --> 00:57:53,700
I can suggest a book off of my mind right now, but show up in multiple places.

821
00:57:53,700 --> 00:58:03,700
What changed for me is the first power platform conference that I sponsored myself to.

822
00:58:03,700 --> 00:58:13,700
Once you're visible and once you see more people doing the same thing as you are doing, the confidence doubles.

823
00:58:13,700 --> 00:58:17,700
It's useful in multiple levels.

824
00:58:17,700 --> 00:58:27,700
For your life or career perspective, what's more important than the MCT, the trainer or the MVP?

825
00:58:27,700 --> 00:58:29,700
For me both.

826
00:58:29,700 --> 00:58:41,700
Because like I said, the goal with the company is to enable organizations to adopt AI securely responsibly

827
00:58:41,700 --> 00:58:47,700
and train a major role in that.

828
00:58:47,700 --> 00:58:56,700
When I share, when I can only share when I learn, so it also propels me to learn well.

829
00:58:56,700 --> 00:59:10,700
Learning sharing is really important in today's day and age where you can take a break for one day without looking at LinkedIn.

830
00:59:10,700 --> 00:59:14,700
I don't know what you already missed.

831
00:59:14,700 --> 00:59:16,700
Yeah, yeah, I'm working.

832
00:59:16,700 --> 00:59:18,700
What are important for me?

833
00:59:18,700 --> 00:59:22,700
Yeah, I think my whole life is something to me.

834
00:59:22,700 --> 00:59:28,700
But what is something you will learn next?

835
00:59:28,700 --> 00:59:36,700
I am this week I will be taking the agent AI business solutions architect exam.

836
00:59:36,700 --> 00:59:46,700
But yeah, on a bigger picture, I am also working on getting my AI GP.

837
00:59:46,700 --> 00:59:54,700
It's AI governance professional exam that is provided by IAPP.

838
00:59:54,700 --> 00:59:56,700
It's an organization.

839
00:59:56,700 --> 01:00:07,700
So basically that certification is primarily the goal of lawyers and security teams.

840
01:00:07,700 --> 01:00:23,700
And I would love to understand the GDPR, EUH Act, ISO, everything that relates to governance from a legal standpoint.

841
01:00:23,700 --> 01:00:28,700
Because that's where I see my future career going.

842
01:00:28,700 --> 01:00:36,700
What's the best food you ever eating in your whole life?

843
01:00:36,700 --> 01:00:38,700
Food?

844
01:00:38,700 --> 01:00:45,700
I come from Hyderabad in India and I love Biryani.

845
01:00:45,700 --> 01:00:52,700
I know vegetarian, it's diluted but veg tam biryani.

846
01:00:52,700 --> 01:00:55,700
Yeah, so we are over now.

847
01:00:55,700 --> 01:00:57,700
So well, thank you for the time.

848
01:00:57,700 --> 01:01:11,700
So my last question is when you meet your 20 year younger yourself, what tips will you give GIFR?

849
01:01:11,700 --> 01:01:22,700
I would tell her to show up in communities earlier.

850
01:01:22,700 --> 01:01:31,700
You know, just share your work without being, leave your imposter syndrome.

851
01:01:31,700 --> 01:01:33,700
That's my advice.

852
01:01:33,700 --> 01:01:38,700
Oh, yeah, we have had also nice conversation about the enforcers in the room.

853
01:01:38,700 --> 01:01:42,700
Yeah, so thank you for for this fantastic conversation.

854
01:01:42,700 --> 01:01:44,700
I really love it.

855
01:01:44,700 --> 01:01:51,700
And I think when you're too ready, we have to do a live stream on the MC65 show.

856
01:01:51,700 --> 01:01:54,700
I really, really, really interested to see it.

857
01:01:54,700 --> 01:02:06,700
And yeah, I think thank you so much for sharing your insights about Power Platform, Co-Pilot Studio, AgenteGi, Enterprise Automation.

858
01:02:06,700 --> 01:02:09,700
And especially governance.

859
01:02:09,700 --> 01:02:12,700
Thank you so much, Valko.

860
01:02:12,700 --> 01:02:16,700
I think all the links are in the show notes.

861
01:02:16,700 --> 01:02:23,700
And yeah, so you until next time keep learning, keep building, keep exploring what's possible with Microsoft technology in AI.

862
01:02:23,700 --> 01:02:26,700
And yeah, thank you for all the listeners.

863
01:02:26,700 --> 01:02:32,700
And yeah, I give you the last words, the shout out to the listeners.

864
01:02:32,700 --> 01:02:36,700
You can say, "What do you want?"

865
01:02:36,700 --> 01:02:41,700
To all my listeners, very fast, go on faster.

866
01:02:41,700 --> 01:02:42,700
It's really good.

867
01:02:42,700 --> 01:02:44,700
Thank you, bye.

868
01:02:44,700 --> 01:02:46,700
Bye.

869
01:02:46,700 --> 01:02:49,080