Copilot Studio, AI Agents, RAG, and the Future of Business Automation with Nilüfer Doğan [MVP]

1
00:00:00,000 --> 00:00:05,760
Yeah, welcome back to the M365 of M podcast.

2
00:00:05,760 --> 00:00:10,840
We explore here are people, technologies and ideas shaping the future of the Microsoft

3
00:00:10,840 --> 00:00:14,440
365 power cut form, Azure and AI world.

4
00:00:14,440 --> 00:00:18,840
Today we are diving deep into one of the hottest topics in the Microsoft ecosystem.

5
00:00:18,840 --> 00:00:22,720
Cooperated studio, AI agents and intelligent business automation.

6
00:00:22,720 --> 00:00:29,280
My guest is Niliou Fadogan, a platform developer at the International Atomic Energy Agency.

7
00:00:29,280 --> 00:00:35,440
Where she designs AI power solutions using co-pile studio power apps, power automate AI

8
00:00:35,440 --> 00:00:42,840
build and AI services through her career at organizations like Siemens, Bosch, the

9
00:00:42,840 --> 00:00:47,480
Abyss, A home, applicants and the EA.

10
00:00:47,480 --> 00:00:54,040
She has helped build enterprise grad applications, automatic complex business processes and

11
00:00:54,040 --> 00:00:57,200
power citizen development through mentoring and workshops.

12
00:00:57,200 --> 00:01:03,800
In this episode, we explore how co-pile studio is evolving far beyond traditional chat box,

13
00:01:03,800 --> 00:01:10,280
how Azure AI found the Azure Open AI and AI search are enabled new generation of intelligent

14
00:01:10,280 --> 00:01:16,640
agents and what organizations need to know to move for AI experiments to real business

15
00:01:16,640 --> 00:01:17,640
value.

16
00:01:17,640 --> 00:01:22,080
Whatever you are a power platform developer, solution architect IT leader or simple careers

17
00:01:22,080 --> 00:01:26,600
about the future of AI power business applications, this conversation, this pack was a

18
00:01:26,600 --> 00:01:28,680
practical insight, a real world lesson.

19
00:01:28,680 --> 00:01:34,040
So grab your favorite, Bruce, Sit back and enjoy my conversation with Niliou Fadogan.

20
00:01:34,040 --> 00:01:38,080
Welcome Niliou Fadogan, thank you for being here.

21
00:01:38,080 --> 00:01:40,440
Thank you, Miko, for the support.

22
00:01:40,440 --> 00:01:44,320
Thanks for having me and I'm really glad to be here.

23
00:01:44,320 --> 00:01:47,280
And yeah, thank you.

24
00:01:47,280 --> 00:01:53,040
Yeah, you work for the EA EA.

25
00:01:53,040 --> 00:02:01,480
And tell a little bit about what you're doing there that sounds really rocker sides.

26
00:02:01,480 --> 00:02:06,960
You are the explain what I'm doing there, what kind of technologies I'm using there, but

27
00:02:06,960 --> 00:02:08,640
let me explain a bit.

28
00:02:08,640 --> 00:02:15,440
I work there as a 365 platform specialist, so my specialty and my background totally

29
00:02:15,440 --> 00:02:17,600
aligns what I'm doing right now.

30
00:02:17,600 --> 00:02:26,600
I'm mostly creating power platform and co-operative solution like creating agents and applications

31
00:02:26,600 --> 00:02:28,480
for internal processes.

32
00:02:28,480 --> 00:02:37,600
So this is, I work for the team which just aims digital transformation within the organization.

33
00:02:37,600 --> 00:02:42,280
So that's, that's it, I guess, yeah.

34
00:02:42,280 --> 00:02:47,560
Yeah, and yeah, I think I see your academic background is

35
00:02:47,560 --> 00:02:56,560
yeah, in economics, I think that's not the, not the mobile way to power platform in the

36
00:02:56,560 --> 00:03:02,280
ID-bound, how did you transition into power platform in the ID-bound?

37
00:03:02,280 --> 00:03:06,760
When I told people that I graduated from economics, this is the first question that comes

38
00:03:06,760 --> 00:03:09,000
mind, how do you switch there?

39
00:03:09,000 --> 00:03:10,200
Why you switch there?

40
00:03:10,200 --> 00:03:17,200
And actually, before I graduate, I realized that I don't want to work in the economics field.

41
00:03:17,200 --> 00:03:21,880
Then I started to get some courses about data science.

42
00:03:21,880 --> 00:03:28,720
Then I, at that time, data science became a bit famous and everybody wanted to work in

43
00:03:28,720 --> 00:03:30,440
the data science field.

44
00:03:30,440 --> 00:03:35,440
And that's why I get some courses and I try to find an intern about data science.

45
00:03:35,440 --> 00:03:44,760
And I think my career change just began with this idea and I was a bit lucky because I

46
00:03:44,760 --> 00:03:47,960
get an internship in the BSH in that way.

47
00:03:47,960 --> 00:03:51,040
And I started the data science project student.

48
00:03:51,040 --> 00:03:56,960
I worked there for nine months and after that, I realized another technology is arising and

49
00:03:56,960 --> 00:03:58,800
I talk with my manager.

50
00:03:58,800 --> 00:04:03,880
And of course, I wasn't sure what this field and what can I do there.

51
00:04:03,880 --> 00:04:07,320
But my manager offered me, do you want to work on this?

52
00:04:07,320 --> 00:04:13,800
And I started about power platform and power apps, power automates and what we can do there

53
00:04:13,800 --> 00:04:17,200
and do our interest with these technologies.

54
00:04:17,200 --> 00:04:23,280
And then I realized that I really loved what power apps and power automates do at that

55
00:04:23,280 --> 00:04:27,160
time because I was very interested with UI.

56
00:04:27,160 --> 00:04:35,280
And instead of just doing hard-cooked things, I was getting, I was into just doing something

57
00:04:35,280 --> 00:04:38,360
simply and writing little code.

58
00:04:38,360 --> 00:04:41,360
So this is low-code, no-code platform.

59
00:04:41,360 --> 00:04:47,680
This was totally aligned my interest and then I switched to actually there.

60
00:04:47,680 --> 00:04:53,760
So my start point was being a data science project student and switching to another team

61
00:04:53,760 --> 00:04:55,840
which is power platform.

62
00:04:55,840 --> 00:04:59,040
And yeah, it started in that way.

63
00:04:59,040 --> 00:05:02,160
Yeah, awesome.

64
00:05:02,160 --> 00:05:07,840
How has your economics background influenced the way you approach technology and automation

65
00:05:07,840 --> 00:05:09,440
projects?

66
00:05:09,440 --> 00:05:18,480
Actually, while I'm studying in the university, I got lots of statistics courses and at that

67
00:05:18,480 --> 00:05:28,600
time I realized that I really love analyzing something and also just approaching to this

68
00:05:28,600 --> 00:05:35,400
some issues and when I face something during my work, I can use that background, I think.

69
00:05:35,400 --> 00:05:44,760
But I cannot say of course 100%, I'm using economics background, but of course a bit of learning

70
00:05:44,760 --> 00:05:57,040
process of courses and attitudes and tradition in my university just shape myself and my personality

71
00:05:57,040 --> 00:06:05,360
and also my way of looking to the issues that I am facing during the work and while solving

72
00:06:05,360 --> 00:06:11,960
the problems, then I may tell in that way it helped me.

73
00:06:11,960 --> 00:06:19,520
And can you tell us how did typical day look like for you as platform development, working

74
00:06:19,520 --> 00:06:23,760
with AI solutions?

75
00:06:23,760 --> 00:06:33,800
Actually, I am experiencing different kind of days of power platform development days.

76
00:06:33,800 --> 00:06:43,080
For now, actually, I am just working by myself, creating applications and sometimes helping

77
00:06:43,080 --> 00:06:51,440
other people who are not coming from the technical background, but they want to develop some

78
00:06:51,440 --> 00:06:53,840
agents or some applications.

79
00:06:53,840 --> 00:06:59,920
I also help them and we make call them like a student developer and mostly helping the student

80
00:06:59,920 --> 00:07:05,400
developers and also working on my own projects and my own tasks.

81
00:07:05,400 --> 00:07:08,760
I may tell in that way.

82
00:07:08,760 --> 00:07:15,600
Yeah, so that's a little bit deep dive into the co-pilot studio topic.

83
00:07:15,600 --> 00:07:22,040
So many people still think co-pilot series, adjustable in chatbos, what's the biggest

84
00:07:22,040 --> 00:07:26,240
misconception about the platform you see?

85
00:07:26,240 --> 00:07:33,360
I think misconception is people think like co-pilot studio is a very intelligent something

86
00:07:33,360 --> 00:07:40,400
and they can write the instructions and it should work for them and that's enough.

87
00:07:40,400 --> 00:07:42,520
But they are thinking in that way.

88
00:07:42,520 --> 00:07:47,360
But I think this is the misconception that people think they can solve everything with

89
00:07:47,360 --> 00:07:48,360
agents.

90
00:07:48,360 --> 00:07:55,000
Sometimes, some demand comes from the client side and when we talk, the pain points I understand

91
00:07:55,000 --> 00:08:03,520
that is not an agent that cannot be solved via agents that can be solved via applications

92
00:08:03,520 --> 00:08:07,600
or powerpates or just with power automate.

93
00:08:07,600 --> 00:08:14,560
So I think misconception starts from, first of all, deciding which product, for which purpose

94
00:08:14,560 --> 00:08:20,880
they couldn't sometimes understand the people can not sometimes understand this and so navigating

95
00:08:20,880 --> 00:08:23,000
through wrong direction.

96
00:08:23,000 --> 00:08:29,440
And another thing is people think that they are really intelligent and just writing simple

97
00:08:29,440 --> 00:08:31,360
instruction will be enough.

98
00:08:31,360 --> 00:08:32,920
But it's not like that.

99
00:08:32,920 --> 00:08:39,200
We have to construct the knowledge part and also tools and also automation.

100
00:08:39,200 --> 00:08:42,560
All together became a good agent.

101
00:08:42,560 --> 00:08:49,520
So sometimes they are thinking, okay, I just write the instructions by helping, by getting

102
00:08:49,520 --> 00:08:55,720
help via some chat tools, some AI tools and that's enough.

103
00:08:55,720 --> 00:08:58,080
But it's not working at the end of the day.

104
00:08:58,080 --> 00:09:03,360
So I think these two parts are the misconception about co-pies studio.

105
00:09:03,360 --> 00:09:11,080
Yeah, I'm a little bit overwhelmed with the updates on co-pies studio.

106
00:09:11,080 --> 00:09:15,480
It's an earlier feel every three days.

107
00:09:15,480 --> 00:09:17,240
They wrote something new out.

108
00:09:17,240 --> 00:09:24,200
Can you explain how co-pies studio evolved over the last month?

109
00:09:24,200 --> 00:09:28,440
Actually over the last month, there were lots of updates.

110
00:09:28,440 --> 00:09:35,760
So I also struggled to just keep up what's going on, what kind of new features arise and which

111
00:09:35,760 --> 00:09:38,840
one would be valuable for me to learn.

112
00:09:38,840 --> 00:09:42,120
So I'm still just trying to catch up.

113
00:09:42,120 --> 00:09:49,520
So there were some big events like Microsoft Build and the output from this event was very

114
00:09:49,520 --> 00:09:54,440
important and we saw lots of important features there.

115
00:09:54,440 --> 00:10:02,320
I think it's going to be like in the previous time, we were just telling everything else and

116
00:10:02,320 --> 00:10:09,240
just doing the whole configurations and adjustments inside the co-pies studio.

117
00:10:09,240 --> 00:10:14,200
That was creating, that became the agent at the end of the day.

118
00:10:14,200 --> 00:10:21,400
But now I'm seeing that it will be enough to write the instructions in a good way.

119
00:10:21,400 --> 00:10:27,960
And if you write the better instructions, if you structure it very well, it will be better.

120
00:10:27,960 --> 00:10:34,440
For example, they changed the UI of the co-pies studio and for now they just kind of switched

121
00:10:34,440 --> 00:10:39,760
them, topic and also the topic with the skills.

122
00:10:39,760 --> 00:10:47,480
So in skills, you should write the good instructions to be sure that agent will behave very well.

123
00:10:47,480 --> 00:10:53,600
So I think it's going to be a bit just describing what you want very well.

124
00:10:53,600 --> 00:10:59,320
Means you will have the good quality agents at the end.

125
00:10:59,320 --> 00:11:04,000
I think it goes in that way.

126
00:11:04,000 --> 00:11:12,080
And why are especially enterprises suddenly paying so much attention accurately to AI agents?

127
00:11:12,080 --> 00:11:19,400
My first and sincere answer would be of course there are AI trends and people really understand

128
00:11:19,400 --> 00:11:25,680
that there is a train and this train is going and we have to just jump into the train because

129
00:11:25,680 --> 00:11:26,680
we should catch it.

130
00:11:26,680 --> 00:11:34,800
But this is the first thing I think that they are just trying to get into this field and

131
00:11:34,800 --> 00:11:36,840
want to have agents.

132
00:11:36,840 --> 00:11:42,680
But another thing, people understand that they have some manual things and the world we

133
00:11:42,680 --> 00:11:53,960
are living is just the world we may automate, make an automation and also we can get rid

134
00:11:53,960 --> 00:11:56,640
of these manual works.

135
00:11:56,640 --> 00:12:02,960
So that's why after they try some AI tools, they understand that we need agents.

136
00:12:02,960 --> 00:12:04,760
We can create this internally.

137
00:12:04,760 --> 00:12:11,840
So this is the beginning part for I think.

138
00:12:11,840 --> 00:12:22,480
And what will you say is the difference between an AI agent to a traditional AI chatbot system?

139
00:12:22,480 --> 00:12:31,000
Actually, between AI agents and chatbots there is one that prevents and the main differences

140
00:12:31,000 --> 00:12:38,240
in AI agents, it can decide based on what we give to that agent.

141
00:12:38,240 --> 00:12:46,720
But for chatbots it was like you have to sell every step and decision, decision you have to

142
00:12:46,720 --> 00:12:53,640
just configure the whole decision in that way chatbots act like what you define.

143
00:12:53,640 --> 00:13:00,240
So you were telling all things to the chatbots and just doing manually in the background

144
00:13:00,240 --> 00:13:01,240
all things.

145
00:13:01,240 --> 00:13:06,280
But for AI agents, you are describing what you want and if you describe it well then it

146
00:13:06,280 --> 00:13:11,600
behaves in a way you like you want.

147
00:13:11,600 --> 00:13:18,600
And what did you think should organization choose co-pilot studio over custom AI development

148
00:13:18,600 --> 00:13:22,520
or the M-Mobile?

149
00:13:22,520 --> 00:13:30,280
It depends but can you say when you use co-pilot studio and then a custom AI development?

150
00:13:30,280 --> 00:13:36,480
Can you elaborate on the question I didn't get what you mean with the custom agents?

151
00:13:36,480 --> 00:13:39,440
Yeah, I think we have this also.

152
00:13:39,440 --> 00:13:50,800
So we have Azure AI especially where the difference and when it choose what?

153
00:13:50,800 --> 00:14:00,400
I think it's where I use only cases and when you as an M365 specialist I may tell that

154
00:14:00,400 --> 00:14:08,440
when I talk with the client for a project at the beginning and when I analyze the project

155
00:14:08,440 --> 00:14:16,680
I see some functionalities they want and sometimes I know from my experiences like this can be

156
00:14:16,680 --> 00:14:23,200
solved just via co-pilot studio but sometimes I understand this should be tried and if we

157
00:14:23,200 --> 00:14:27,200
couldn't solve this we may move on to Azure site.

158
00:14:27,200 --> 00:14:36,200
So for me actually I start with the co-pilot studio and try to find the work around solution

159
00:14:36,200 --> 00:14:44,720
sometimes and in case I cannot solve it I switch to Azure and another thing is I think

160
00:14:44,720 --> 00:14:54,800
the another main criteria that defines where to go for any scenario is if they are working

161
00:14:54,800 --> 00:15:03,240
with a huge amount of data then that should be first of all cleared and should be structured

162
00:15:03,240 --> 00:15:10,200
in another place so in that kind of case is mostly we have switched to Azure platforms and

163
00:15:10,200 --> 00:15:13,480
use the Azure AI tools.

164
00:15:13,480 --> 00:15:23,040
Okay, can you tell us a real case scenario for co-pilot studio you have developed?

165
00:15:23,040 --> 00:15:27,160
Yes, I can but what kind of scenarios you need?

166
00:15:27,160 --> 00:15:34,160
I have lots of scenarios and I couldn't think which one I may explain.

167
00:15:34,160 --> 00:15:37,160
Yeah.

168
00:15:37,160 --> 00:15:44,280
Only when you are allowed to tell about it.

169
00:15:44,280 --> 00:15:48,280
Okay, okay.

170
00:15:48,280 --> 00:15:54,440
I may tell one agent to, yeah, okay, let me tell one agent.

171
00:15:54,440 --> 00:16:02,560
I created a Confluence agent and these agents were helping for the people who are searching

172
00:16:02,560 --> 00:16:11,160
some knowledge inside the Confluence and it needs both indexation of the documents which

173
00:16:11,160 --> 00:16:18,640
are listed inside the on-prem Confluence so that's why I had to use the tool from Azure

174
00:16:18,640 --> 00:16:25,960
sites which means that I have to make indexation and rock inside the Azure and make it as

175
00:16:25,960 --> 00:16:30,200
a knowledge source and use it inside the co-pilot studio.

176
00:16:30,200 --> 00:16:36,040
Another thing in this project was defining the permission level of the people and the group

177
00:16:36,040 --> 00:16:44,160
so by getting retrieving the data from Confluence also I get the permission levels and permission

178
00:16:44,160 --> 00:16:52,040
groups and then I put as a variable insight indexation in that way when people ask some

179
00:16:52,040 --> 00:16:58,520
question if they don't have access in the Confluence they couldn't get the answer about permission

180
00:16:58,520 --> 00:17:05,760
level issues sold and also they can easily access the information without just scrolling

181
00:17:05,760 --> 00:17:10,480
the page and trying to find it within lots of documents.

182
00:17:10,480 --> 00:17:18,160
So this was one of the real case and I think it was nice because it was the first project

183
00:17:18,160 --> 00:17:25,560
for me that I retrieved the data from on-premise system and make it structured way and also make

184
00:17:25,560 --> 00:17:28,320
rough from there.

185
00:17:28,320 --> 00:17:32,600
Yeah, that's interesting.

186
00:17:32,600 --> 00:17:39,120
What did you think are the biggest strengths on Limit and the co-pilot studio today?

187
00:17:39,120 --> 00:17:48,480
Actually for me, recent change sometimes can be confusing and I couldn't decide where

188
00:17:48,480 --> 00:17:58,720
to solve the case because I think there is no way not only one way for some cases so sometimes

189
00:17:58,720 --> 00:18:03,040
it can be hard to just decide which way I want to solve this.

190
00:18:03,040 --> 00:18:10,160
I want to use the tools, I want to adjust the connector and use inside the topic and do

191
00:18:10,160 --> 00:18:17,120
I need any other agents for that kind of questions it can be sometimes confusing based on the

192
00:18:17,120 --> 00:18:18,120
cases.

193
00:18:18,120 --> 00:18:26,560
So there is no only one way for some cases so you can solve it with three different way.

194
00:18:26,560 --> 00:18:36,240
So that kind of case can be a bit complicated and decision making can be hard but besides

195
00:18:36,240 --> 00:18:44,600
that I'm just thinking for example for me UI change is a bit hard for me because I am also

196
00:18:44,600 --> 00:18:52,200
surprised when UI of power automate change and I have to switch back to the previous version

197
00:18:52,200 --> 00:19:03,760
always so UI change also makes a bit it can be a bit challenging for me and also I think

198
00:19:03,760 --> 00:19:11,800
sometimes especially as I saw in the season developer they cannot address what is pre-view

199
00:19:11,800 --> 00:19:19,360
feature, what is experimental feature so they create project and in some way they publish

200
00:19:19,360 --> 00:19:29,360
it in the prod environment and that can be sometimes difficult to just handle those things.

201
00:19:29,360 --> 00:19:33,360
But that's all I guess that comes my mind.

202
00:19:33,360 --> 00:19:36,280
Yeah awesome.

203
00:19:36,280 --> 00:19:42,680
I'm a little bit thinking about or I read a little bit and look what you do and there is

204
00:19:42,680 --> 00:19:45,040
something I found really interesting.

205
00:19:45,040 --> 00:19:52,040
You spoke about the combination of co-pilot studio with Azure, AI Foundry and AI search

206
00:19:52,040 --> 00:19:56,520
what makes this combination so powerful.

207
00:19:56,520 --> 00:20:08,880
Actually before Azure AI search integration comes to co-pilot studio and also some knowledge

208
00:20:08,880 --> 00:20:17,000
source wasn't developed well at that time so making RAC for big amount of data was quite

209
00:20:17,000 --> 00:20:18,000
hard.

210
00:20:18,000 --> 00:20:27,480
For example people have 1000 or 2000, 3000 documents and making indexation and getting

211
00:20:27,480 --> 00:20:31,000
quick response from the agent was quite hard.

212
00:20:31,000 --> 00:20:43,000
So those combination were bringing RAC solution and while it is RAC solution it was both getting

213
00:20:43,000 --> 00:20:48,920
much better answer and in a more quick way.

214
00:20:48,920 --> 00:20:59,000
So this combination brings this but for now also it is valid for big amount of data making

215
00:20:59,000 --> 00:21:06,000
and the exchange in some inside Azure is still available.

216
00:21:06,000 --> 00:21:12,320
That's really really cool and interesting and can you a little bit explain how you architecture

217
00:21:12,320 --> 00:21:15,600
a modern enterprise AI agent?

218
00:21:15,600 --> 00:21:18,720
Yes I can.

219
00:21:18,720 --> 00:21:25,880
For me actually I am not a solution architect but since as a developer we have to make it

220
00:21:25,880 --> 00:21:33,200
sometimes by ourselves then I follow a couple of things like first of all I analyze the

221
00:21:33,200 --> 00:21:38,800
project that I get and with steps I will follow which technologies I will use.

222
00:21:38,800 --> 00:21:47,680
If it needs some POC process then first I tried POC then later on decide which tool I will

223
00:21:47,680 --> 00:21:55,680
use, which way I will go and then later on I create one template for myself and just

224
00:21:55,680 --> 00:22:02,400
visualizing everything where the data source will be and how I will configure this, how

225
00:22:02,400 --> 00:22:05,520
I will create the permission layer.

226
00:22:05,520 --> 00:22:12,160
I decide all of those and then I start a create project.

227
00:22:12,160 --> 00:22:16,360
Yeah I can tell.

228
00:22:16,360 --> 00:22:22,240
And we have also I don't know if everyone knows this.

229
00:22:22,240 --> 00:22:30,760
Can you a little bit explain what Azure AI search is and what role it play in the agent

230
00:22:30,760 --> 00:22:32,080
solutions?

231
00:22:32,080 --> 00:22:41,400
Azure AI search is a place, a tool that you can index your documents and indexation means

232
00:22:41,400 --> 00:22:49,840
that indexation and vectorization is done by Azure AI search at the same time.

233
00:22:49,840 --> 00:22:57,520
So that means that if you have church documents in normal indexation doesn't exist if you are

234
00:22:57,520 --> 00:23:06,240
using it directly then it means that it just read the whole page to find the relevant

235
00:23:06,240 --> 00:23:08,240
information.

236
00:23:08,240 --> 00:23:10,680
The post that you are immersed in in the second page.

237
00:23:10,680 --> 00:23:16,600
So when you make an indexation and when you use Azure AI search it just go to the second

238
00:23:16,600 --> 00:23:24,160
page and find this information or if this information is simply just mentioned in the third page

239
00:23:24,160 --> 00:23:32,840
is just look second and third page and do not skim the whole page, each page and it doesn't

240
00:23:32,840 --> 00:23:34,080
spend a lot of time.

241
00:23:34,080 --> 00:23:42,560
So it's a both increase the quality of the answers and also you can get easily answer within

242
00:23:42,560 --> 00:23:52,000
a couple of not minutes but less than that so yeah that's nice.

243
00:23:52,000 --> 00:24:03,920
And what role did retrieval augmented to generation play in this kind of your own?

244
00:24:03,920 --> 00:24:12,000
Yeah, retrieval augmentation actually retrieval augmentation and indexation are indexation

245
00:24:12,000 --> 00:24:17,080
and vectorization are whole process that works together actually.

246
00:24:17,080 --> 00:24:22,800
I may sell let me explain you in a more better way.

247
00:24:22,800 --> 00:24:29,480
We have documents lots of documents we are putting in a vector database or one database

248
00:24:29,480 --> 00:24:37,800
then this is just divided into sections this is the chunks and then via large language

249
00:24:37,800 --> 00:24:43,960
models which communicates and also create the response via this we get the response and

250
00:24:43,960 --> 00:24:52,240
hold this process vectorization chunking and indexation and creating response via LLM

251
00:24:52,240 --> 00:24:53,800
is rock.

252
00:24:53,800 --> 00:24:59,400
So this is the whole process I talk about actually.

253
00:24:59,400 --> 00:25:05,640
And what did you think what misconceptions to organization commonly make when they implementing

254
00:25:05,640 --> 00:25:10,200
REC?

255
00:25:10,200 --> 00:25:20,440
Sometimes okay I think yeah even they have little amount of documents sometimes they try

256
00:25:20,440 --> 00:25:28,720
it but sometimes it's not necessary to use it in according to my opinion you can just

257
00:25:28,720 --> 00:25:34,520
solve it easily by just setting it a knowledge inside the global files studio.

258
00:25:34,520 --> 00:25:44,520
Another thing is they are thinking if they upload a excel file or CSV file it will be indexed

259
00:25:44,520 --> 00:25:53,280
properly but the thing is they have to select the right database for indexation.

260
00:25:53,280 --> 00:25:57,920
So selecting the right database based on the data is also important.

261
00:25:57,920 --> 00:26:04,920
For example if you have excel like data, tabular data then those indexation groups different

262
00:26:04,920 --> 00:26:11,680
compared to text based documents so this should be also regarded while making rock.

263
00:26:11,680 --> 00:26:17,560
Yeah I think Microsoft REC would recommend it to use Microsoft fabric.

264
00:26:17,560 --> 00:26:19,320
Yeah yeah yeah.

265
00:26:19,320 --> 00:26:20,320
Industrial.

266
00:26:20,320 --> 00:26:29,160
One of your topics you speak about is this business value.

267
00:26:29,160 --> 00:26:38,320
How do you evaluate AI agents that are actually delivering a business value?

268
00:26:38,320 --> 00:26:43,120
How we can do it or how do you do it?

269
00:26:43,120 --> 00:26:50,280
Actually this is hard topic and still there are some discussions continue.

270
00:26:50,280 --> 00:26:55,200
There are a couple of ways to do this.

271
00:26:55,200 --> 00:27:03,560
I used some automation via Power Automate that I was getting the conversation history and

272
00:27:03,560 --> 00:27:12,200
in that way I was telling what kind of question I asked and in normal how many how much time

273
00:27:12,200 --> 00:27:18,560
do people spend without the agents and it was kind of manual process but there are some

274
00:27:18,560 --> 00:27:25,680
tools like Copial Studio Kit to see the whole matrix and put the key PIs there and even

275
00:27:25,680 --> 00:27:34,800
you can just tell for example this would cost 5 euro for the process but we gain 10 Rs

276
00:27:34,800 --> 00:27:39,640
so we just gained 50 euros per day.

277
00:27:39,640 --> 00:27:45,840
You can just put that kind of key PIs and in that way you can adjust the return of investment

278
00:27:45,840 --> 00:27:53,080
and you can see the real value in that way so I think Copial Studio Kit is very valuable

279
00:27:53,080 --> 00:27:59,760
for this via a commutation key PIs and also for other purpose for governing purpose also

280
00:27:59,760 --> 00:28:00,960
it's nice.

281
00:28:00,960 --> 00:28:09,200
So yeah I would say Copial Studio Kit and some manual, not manual but some automation can

282
00:28:09,200 --> 00:28:11,600
be done to detect it.

283
00:28:11,600 --> 00:28:18,840
Also there are some tools inside the Copial Studio that you can put some matrix in a different

284
00:28:18,840 --> 00:28:28,480
way and it can detect it for you so this is the more easiest way for people without just

285
00:28:28,480 --> 00:28:32,040
setting up the Copial Studio Kit they can also use this.

286
00:28:32,040 --> 00:28:34,920
Yeah there are some ways.

287
00:28:34,920 --> 00:28:45,240
And when you start with a proof of concept, what challenge and counter or did you see

288
00:28:45,240 --> 00:28:50,720
when you go to production?

289
00:28:50,720 --> 00:28:59,600
Some proof of concepts just end up like we don't need this and some of them end up like

290
00:28:59,600 --> 00:29:08,440
okay we need this let's just move to create the real products and so it depends on the cases

291
00:29:08,440 --> 00:29:12,280
I guess.

292
00:29:12,280 --> 00:29:21,240
And what is the role of governance and security especially by deploying AI agents at scale?

293
00:29:21,240 --> 00:29:29,440
It is also a hard topic for every company because it's very important for governing and

294
00:29:29,440 --> 00:29:30,440
security part.

295
00:29:30,440 --> 00:29:40,160
Yeah this really needs a bold attention and bold effort to work on.

296
00:29:40,160 --> 00:29:48,000
Also Copial Studio Kit has some tools, some subproducts inside of it and also agent 365

297
00:29:48,000 --> 00:29:56,600
has lots of capabilities to govern those agents and detect who created which agents and do

298
00:29:56,600 --> 00:30:03,760
these agents use knowledge or AI tools or how many AI credits they are consuming.

299
00:30:03,760 --> 00:30:11,160
We can detect this via those tools and for security parts via admin center also it can

300
00:30:11,160 --> 00:30:20,560
be just bands or just sewage of some features and in that way it can be controlled but yeah

301
00:30:20,560 --> 00:30:29,440
for some tools that's not possible for example for co-pilot and 365 co-pilot agents it cannot

302
00:30:29,440 --> 00:30:41,280
be, it cannot be just prohibited or just cannot removed or controlling them is a bit still

303
00:30:41,280 --> 00:30:48,120
the issue on the desk so I may just tell in that way it's a bit complicated actually and

304
00:30:48,120 --> 00:30:56,360
it needs more attention and more effort to work on and to understand because I think there

305
00:30:56,360 --> 00:31:05,120
are some misconception also about these people do not give importance to the governance

306
00:31:05,120 --> 00:31:13,240
and also security they are just using everything and they are not looking where they are giving

307
00:31:13,240 --> 00:31:19,920
their data what they are doing so these are really important things.

308
00:31:19,920 --> 00:31:28,840
You are also a platform developer and my question is we have co-pilot studio and AI Foundry

309
00:31:28,840 --> 00:31:35,960
is there any reason for to use longer power apps or power automate or let's say more

310
00:31:35,960 --> 00:31:42,720
diplomatic how are power apps and power automate changing in the age of AI?

311
00:31:42,720 --> 00:31:53,360
Some people believe that power apps will not exist in the next feature maybe but I am not

312
00:31:53,360 --> 00:31:59,720
sure actually they are evolving of course they are evolving to a more intelligent way for

313
00:31:59,720 --> 00:32:05,320
example there are web apps and my web apps which can create the applications easily

314
00:32:05,320 --> 00:32:15,840
and we can customize this so I think the way they are evolving is not a bad place actually

315
00:32:15,840 --> 00:32:24,880
I would say that not a bad way it's actually in a good way because if we write our description

316
00:32:24,880 --> 00:32:30,920
if we write what we want very well then we can create something within a minute then we

317
00:32:30,920 --> 00:32:37,840
have time to customize it and work on it and add more important features and work on more

318
00:32:37,840 --> 00:32:45,200
security and governance also so it will give us time for those kind of things and those

319
00:32:45,200 --> 00:32:50,360
are really important I guess so it's evolving in that way.

320
00:32:50,360 --> 00:33:00,720
Yeah and what are some practical use cases where AI agent and power automate work together?

321
00:33:00,720 --> 00:33:08,600
Absolutely to be honest I didn't been in a project where I combined those two very effectively

322
00:33:08,600 --> 00:33:15,000
I've been in some projects but I don't think that they were the best case to combine

323
00:33:15,000 --> 00:33:26,560
those two but there are some options like sometimes people need to feed the user interface

324
00:33:26,560 --> 00:33:33,920
with the widescreen and interact with it so sometimes navigating from agent to applications

325
00:33:33,920 --> 00:33:41,920
could be possible if they need more interaction in the user interface or if they want to

326
00:33:41,920 --> 00:33:49,760
see the table some statistics and some values in that way it can be integrated but as I said

327
00:33:49,760 --> 00:33:54,280
I didn't remove that much.

328
00:33:54,280 --> 00:34:08,320
Okay I think a little bit about what are good candidates for AI powered automation how

329
00:34:08,320 --> 00:34:17,680
can we identify processes that are good for that or how do we, how do you do it?

330
00:34:17,680 --> 00:34:26,920
How do we, I didn't get the question, how do we define the AI projects I didn't get?

331
00:34:26,920 --> 00:34:32,240
How did you identify process that are good candidates for AI powered automation that's

332
00:34:32,240 --> 00:34:39,880
my question.

333
00:34:39,880 --> 00:34:49,240
I don't know I didn't think any case for this I couldn't get it sorry.

334
00:34:49,240 --> 00:34:55,480
Then another question, where did you see the human in the loop or human approval, what

335
00:34:55,480 --> 00:35:01,720
role did this still play in AI driven workflows?

336
00:35:01,720 --> 00:35:09,720
I think for serious decision of course agents shouldn't take the decision and sometimes

337
00:35:09,720 --> 00:35:18,960
of course for basic tasks like creating tickets or some basic approvals these are very straightforward

338
00:35:18,960 --> 00:35:28,440
and human in the loop scenarios are valid for those but there are some scenarios which

339
00:35:28,440 --> 00:35:36,360
are really important to be to involve the human for example expense reports and those

340
00:35:36,360 --> 00:35:45,000
kind of cases and also if we need to escalate to the sea level then for that kind of cases

341
00:35:45,000 --> 00:35:50,040
we need another approval before we address it to the sea level so for that kind of cases

342
00:35:50,040 --> 00:35:59,120
human in the loop became really important so I can differentiate in that way.

343
00:35:59,120 --> 00:36:05,640
Okay let's a little bit look at the tutorial power platform.

344
00:36:05,640 --> 00:36:13,200
What will you say is the most underrated yeah feature in the power platform for you?

345
00:36:13,200 --> 00:36:21,880
Wow that's good question I didn't think about this before.

346
00:36:21,880 --> 00:36:24,040
I'm thinking.

347
00:36:24,040 --> 00:36:32,240
I think the response to design make me always like that's good at the end of the design

348
00:36:32,240 --> 00:36:40,000
because you can use it either in the tablets or in the laptop or in the phone.

349
00:36:40,000 --> 00:36:49,840
I really love this feature so I think responsive design makes me always excited.

350
00:36:49,840 --> 00:36:59,880
You are an MVP so you have this you get all the information more earlier than the novel

351
00:36:59,880 --> 00:37:08,920
humans like me so what did you think is coming or what's the future of AI inside the power

352
00:37:08,920 --> 00:37:12,120
platform ecosystem?

353
00:37:12,120 --> 00:37:18,320
I cannot declare the whole information of course and also I am not getting a little

354
00:37:18,320 --> 00:37:27,000
further but yeah in my opinion beside being an MVP I'm not sure that it became like we

355
00:37:27,000 --> 00:37:37,680
have to be a person who can understand from security governance and whole concepts together

356
00:37:37,680 --> 00:37:47,720
because as a developer if we skip some parts then it became really hard to understand

357
00:37:47,720 --> 00:37:54,040
what we are doing and okay we developed this but will this be sustainable will people

358
00:37:54,040 --> 00:38:00,760
satisfy from this so to understand those also we need the security and governance parts

359
00:38:00,760 --> 00:38:14,600
a lot and also we should understand that we have to also we have to have lots of information

360
00:38:14,600 --> 00:38:21,680
about the system for example when we get the case we should understand where we should

361
00:38:21,680 --> 00:38:28,480
address this and which tool we can solve this because there are some options and it's

362
00:38:28,480 --> 00:38:38,480
either decreased cost or increased the cost so seeing the whole perspective became important

363
00:38:38,480 --> 00:38:45,600
in this time I met all so that's I guess.

364
00:38:45,600 --> 00:38:51,360
I think in your role you have x yeah you have a must or you can connect with a lot of

365
00:38:51,360 --> 00:38:58,000
experience mentoring citizens in the developer what separates a successful citizen developer

366
00:38:58,000 --> 00:39:03,600
program from their failed ones.

367
00:39:03,600 --> 00:39:10,800
I think common things amongst these developers is not only this in developer but the developer

368
00:39:10,800 --> 00:39:19,400
some new developers are also doing this thing sometimes they are just using the AI and for

369
00:39:19,400 --> 00:39:26,200
example they are asking to close they are asking to chat GPT and then they do what they see

370
00:39:26,200 --> 00:39:34,280
but they don't understand what they are doing or what they are allowing so if they do this

371
00:39:34,280 --> 00:39:41,680
they are not being the good ones of course because at the end when they get some issue they

372
00:39:41,680 --> 00:39:47,680
couldn't understand the concept they cannot understand where to start and why this is

373
00:39:47,680 --> 00:39:57,920
resulted from where this is resulted and that's why I think asking and being very reluctant

374
00:39:57,920 --> 00:40:05,760
to do AI response of course I don't say that I also use the clause and other code tools

375
00:40:05,760 --> 00:40:13,080
for this but I try to understand how we did this and when I get an error I am not reluctant

376
00:40:13,080 --> 00:40:20,640
to do I am not a stick to the answer that I ideas to me so this is the important part I

377
00:40:20,640 --> 00:40:28,040
think for also citizen developer too but some of them just try to search a bit the forums

378
00:40:28,040 --> 00:40:38,400
and check the documentation and try to understand what's happening then they became more better

379
00:40:38,400 --> 00:40:42,160
compared to other cases.

380
00:40:42,160 --> 00:40:51,800
In the companies we have a say two to M&A fighters and the one side of the ring is innovation

381
00:40:51,800 --> 00:41:02,760
and the other side is governance which one wins the game the metal decides.

382
00:41:02,760 --> 00:41:10,000
That's also a good question I think it depends on the company I am sure that in some companies

383
00:41:10,000 --> 00:41:18,840
like governance wins and in some companies innovation wins if they are very aggressive about just

384
00:41:18,840 --> 00:41:27,280
expanding their earnings and expanding their project and cycle then in that way of course

385
00:41:27,280 --> 00:41:36,440
innovation wins but for some of them they became more protected and they became unwilling

386
00:41:36,440 --> 00:41:43,280
to share all of their data and first of all they try to understand why we are allowing

387
00:41:43,280 --> 00:41:52,000
this and why we are doing this and for those of course the other side wins.

388
00:41:52,000 --> 00:42:01,280
Okay yeah I think a lot of IT departments actually say a little bit worried about citizen

389
00:42:01,280 --> 00:42:08,480
developer because they are creating risk.

390
00:42:08,480 --> 00:42:11,400
What advice can you give?

391
00:42:11,400 --> 00:42:19,480
I think try to call for for for for colleagues to your case I mean so that I saw also those

392
00:42:19,480 --> 00:42:27,160
kind of cases people the organization try to give access to whole people because they

393
00:42:27,160 --> 00:42:35,400
want to give people playground and they can be able to they grow the developer on agents

394
00:42:35,400 --> 00:42:42,160
that's understandable but that's quite chaotic too because to manage all these both need

395
00:42:42,160 --> 00:42:48,680
governance and security so for that purpose we have environments tenants in the power

396
00:42:48,680 --> 00:42:54,880
under power platform so they have to manage those environments very well for example they

397
00:42:54,880 --> 00:42:59,840
should have the test environments and they should have sandbox environment if they want

398
00:42:59,840 --> 00:43:06,360
to try something maybe they can put to whole people in those environments but they if they

399
00:43:06,360 --> 00:43:13,520
see real cases and real scenarios they may move to the environment so they have to be

400
00:43:13,520 --> 00:43:19,600
stick to application lifecycle management very well and also they should just restrict some

401
00:43:19,600 --> 00:43:27,520
features and check the other people is in the background and based on the people based on

402
00:43:27,520 --> 00:43:33,920
the departments maybe or the cases based on the cases they should just allow people to

403
00:43:33,920 --> 00:43:40,520
use those and in that way I think it could be more manageable.

404
00:43:40,520 --> 00:43:47,320
Yeah the these things we are moving towards a future where everyone becomes a developer?

405
00:43:47,320 --> 00:43:55,800
I don't think so everybody thinks like that but I don't think in that way because everybody

406
00:43:55,800 --> 00:44:03,160
can be a developer everybody can create an agent because yeah this became really easy

407
00:44:03,160 --> 00:44:10,440
you can just write what you want and within a minute there are some AI tools that can create

408
00:44:10,440 --> 00:44:15,920
application for but that's not a development development doesn't mean you tell something

409
00:44:15,920 --> 00:44:21,840
and it creates something it means you have to follow also what you created and what's

410
00:44:21,840 --> 00:44:26,760
the results of this and what you turn off investment you are getting and hold this process

411
00:44:26,760 --> 00:44:32,920
also included in the development so I don't believe that's the everybody will be a developer

412
00:44:32,920 --> 00:44:40,400
but I think to be a developer became easier but I don't think everybody will be a developer

413
00:44:40,400 --> 00:44:41,400
easily.

414
00:44:41,400 --> 00:44:50,200
Yeah okay cool let's jump in the rapid fire round I ask you short questions and you say first

415
00:44:50,200 --> 00:44:56,960
what come in your your mind so my first question is what's your favorite Microsoft product?

416
00:44:56,960 --> 00:45:08,160
Okay I think Copa Listerio unfortunately I couldn't answer it very quickly but yeah Copa

417
00:45:08,160 --> 00:45:14,640
Listerio what is the one AI trend that's overhyped?

418
00:45:14,640 --> 00:45:17,520
Sorry can you repeat it again?

419
00:45:17,520 --> 00:45:21,880
Yeah one AI trend that is overhyped.

420
00:45:21,880 --> 00:45:33,440
One AI trends I couldn't think about this but

421
00:45:33,440 --> 00:45:42,600
have you one that's under unhiked?

422
00:45:42,600 --> 00:45:51,720
I think creating agents within a minute on not underhyped all things about AI is hyped

423
00:45:51,720 --> 00:46:01,600
I guess so I cannot think it is under underhyped so I couldn't see what this is.

424
00:46:01,600 --> 00:46:07,920
And ever open AI or open source models?

425
00:46:07,920 --> 00:46:16,520
It depends on the case so I cannot tell only ones actually but to democratization of course

426
00:46:16,520 --> 00:46:18,760
open source I will say.

427
00:46:18,760 --> 00:46:24,920
Okay where are to get the best Turkish food in Vienna?

428
00:46:24,920 --> 00:46:28,400
Is this real question?

429
00:46:28,400 --> 00:46:36,000
Okay maybe I will say next dollar.

430
00:46:36,000 --> 00:46:43,240
What's the best career advice you ever received?

431
00:46:43,240 --> 00:46:56,800
I think when I get first in my first interview at the BSAJ chart the later I asked the

432
00:46:56,800 --> 00:47:03,640
later I don't want to work in finance but I want to work in data science but I don't have

433
00:47:03,640 --> 00:47:09,720
any experience on this and later told me that if you want this you can do this and I

434
00:47:09,720 --> 00:47:19,960
was like yeah why not then I became I believe myself more after this interview because it

435
00:47:19,960 --> 00:47:22,000
was inspiring and so forth.

436
00:47:22,000 --> 00:47:28,280
So not only one sentence but I think the whole interview was impressed me so I may tell

437
00:47:28,280 --> 00:47:40,040
that that was also my career so I would say this interview but I don't have specific order.

438
00:47:40,040 --> 00:47:50,960
Does there have you ever any certificate that's the most well you ever wrote?

439
00:47:50,960 --> 00:47:59,800
And I get 400 for now I didn't get there is recently an insert to get unfortunately but when

440
00:47:59,800 --> 00:48:07,840
I get Pia 400 it was very valuable for me I learned a lot while preparing for this one

441
00:48:07,840 --> 00:48:10,960
so I would say that one.

442
00:48:10,960 --> 00:48:20,520
Yeah and what's the biggest benefit you get from the MVP award?

443
00:48:20,520 --> 00:48:28,600
I couldn't attend in person event but I think even I attend online events like the MVP

444
00:48:28,600 --> 00:48:39,080
summits and the Microsoft build those events were very valuable for me because there are

445
00:48:39,080 --> 00:48:47,360
really lots of topics going on there and even you can't attend all of them I have to watch

446
00:48:47,360 --> 00:48:54,200
the record later and I learned a lot of things from those so yeah it's very valuable and I

447
00:48:54,200 --> 00:49:01,640
really I want to say the Microsoft events that I can attend.

448
00:49:01,640 --> 00:49:09,320
What's your favorite book podcast or YouTube channel or resource every professional

449
00:49:09,320 --> 00:49:22,480
show or reading? Okay I may tell as a YouTube channel I just watch lots of people so I cannot

450
00:49:22,480 --> 00:49:29,840
say only one people but I may tell for example I'm not saying just because he is my friend but

451
00:49:29,840 --> 00:49:37,820
I really love Raph Sun's channel and I watch Raph Sun's video a lot and learn from his video

452
00:49:37,820 --> 00:49:46,360
a lot and of course when I see some videos shared in the link and I also share I also watch

453
00:49:46,360 --> 00:49:54,980
love Matthew Matthew's also such a blog post I really benefit from it and of course a power

454
00:49:54,980 --> 00:50:04,460
pilot from Veechley I follow these a lot and I read it a lot so I benefit from this too.

455
00:50:04,460 --> 00:50:11,360
Yeah thank you for for for the quick fire rounds so then it's my last questions if listen

456
00:50:11,360 --> 00:50:21,620
I want to start building intelligence agents tomorrow where should they begin?

457
00:50:21,620 --> 00:50:29,620
Can you ask the question again? Yeah I will catch it story.

458
00:50:29,620 --> 00:50:37,220
The listeners today want to start building intelligent agents tomorrow where should they start?

459
00:50:37,220 --> 00:50:48,580
Okay. They I think the YouTube is really valuable it has really valuable sources and it has

460
00:50:48,580 --> 00:50:58,380
lots of real cases so I think they should find those YouTube channels and try to build their

461
00:50:58,380 --> 00:51:05,020
own agents by just following the steps on those videos and in that way they don't have

462
00:51:05,020 --> 00:51:10,500
to spend a lot of money to learn something or just get a course they can just follow the

463
00:51:10,500 --> 00:51:16,660
steps that's shown in the YouTube videos and this is the easiest way I think because when

464
00:51:16,660 --> 00:51:23,780
I start also a power platform developer position I didn't know anything and I was following

465
00:51:23,780 --> 00:51:29,700
the steps that is shown in the YouTube videos and it really helps me a lot because especially

466
00:51:29,700 --> 00:51:36,940
some some channels are explaining very well why we are doing this and what's the idea behind

467
00:51:36,940 --> 00:51:42,540
this so in that way they can understand the concept well so I think they can start from

468
00:51:42,540 --> 00:51:49,140
here and of course I can get some help from for example clothes but they should be

469
00:51:49,140 --> 00:51:54,780
evaluated what they are doing and they should try to understand why we are doing this in

470
00:51:54,780 --> 00:51:59,100
that way they can learn it quickly I'm sure.

471
00:51:59,100 --> 00:52:00,100
Awesome.

472
00:52:00,100 --> 00:52:07,900
Yeah so then whoa yeah thank you so much that was fantastic conversation thank you and

473
00:52:07,900 --> 00:52:13,660
and there we have so many covered so many topics around co-pilot through your architecture

474
00:52:13,660 --> 00:52:21,460
or AI services governance and citizens but yeah and yeah I think it's it's really really interesting

475
00:52:21,460 --> 00:52:31,140
so I thank you so much and yeah for the listeners you find all links and enforce a familiar

476
00:52:31,140 --> 00:52:38,380
for in the show notes and yeah I say thank you so much for for being here and spent nearly

477
00:52:38,380 --> 00:52:45,820
one hour with me so thank you thank you so much thank you because thank you so much for

478
00:52:45,820 --> 00:52:51,940
inviting me and in the moment me in this series I really enjoyed it was really nice thank you.

479
00:52:51,940 --> 00:52:52,500
Bye!

480
00:52:52,820 --> 00:52:54,060
Bye.