Beyond Chat: Building Real Business Value with Microsoft 365 Copilot Agents featuring Steve Corey [MVP]

1
00:00:00,000 --> 00:00:04,800
Yeah, welcome back to the Amc.65 as Am podcast.

2
00:00:04,800 --> 00:00:07,560
We are diving into deep Microsoft technology

3
00:00:07,560 --> 00:00:10,640
with the architect's MEPs and industry experts

4
00:00:10,640 --> 00:00:12,200
shaping the future of the enterprise.

5
00:00:12,200 --> 00:00:15,000
RT today's guest has spent more than two decades

6
00:00:15,000 --> 00:00:17,600
helping organization gets the most out of the Microsoft

7
00:00:17,600 --> 00:00:20,520
technologies from the early days of SharePoint development

8
00:00:20,520 --> 00:00:22,840
to today's AI power workplaces.

9
00:00:22,840 --> 00:00:25,600
He has seen multiple levels of transformation

10
00:00:25,600 --> 00:00:31,760
and now he's believing the biggest and one has only just begun,

11
00:00:31,760 --> 00:00:32,560
I think.

12
00:00:32,560 --> 00:00:36,960
And yeah, his current passion is Microsoft 365 co-pilot

13
00:00:36,960 --> 00:00:40,840
agent development through consulting, architectural work

14
00:00:40,840 --> 00:00:42,760
and educational content on YouTube.

15
00:00:42,760 --> 00:00:45,720
He helping organization understands the future

16
00:00:45,720 --> 00:00:48,760
or isn't just chatting with the AI,

17
00:00:48,760 --> 00:00:53,440
it's building intelligent agents and actually works

18
00:00:53,440 --> 00:00:54,520
alongside employees.

19
00:00:54,520 --> 00:00:58,800
Today we are exploring agent 365, Microsoft growing agents

20
00:00:58,800 --> 00:01:00,520
ecosystem.

21
00:01:00,520 --> 00:01:03,160
We talk about implementation strategies, governance,

22
00:01:03,160 --> 00:01:07,120
architecture and enterprise AI is heading over the next

23
00:01:07,120 --> 00:01:07,640
several years.

24
00:01:07,640 --> 00:01:11,320
So Steve, welcome to the Amc.65 podcast.

25
00:01:11,320 --> 00:01:12,200
Thank you so much.

26
00:01:12,200 --> 00:01:14,120
I'm happy to be here.

27
00:01:14,120 --> 00:01:16,120
Yeah, thank you.

28
00:01:16,120 --> 00:01:19,840
Can you tell us a lot of people who don't know you a little bit

29
00:01:19,840 --> 00:01:27,840
about yourself and how was your journey into Microsoft

30
00:01:27,840 --> 00:01:29,960
ecosystem?

31
00:01:29,960 --> 00:01:34,480
Yeah, so yeah, I started with Microsoft technologies

32
00:01:34,480 --> 00:01:37,080
and probably the late '90s, like straight out of college,

33
00:01:37,080 --> 00:01:41,120
just learning, you know, into four server and all that kind

34
00:01:41,120 --> 00:01:41,600
of stuff.

35
00:01:41,600 --> 00:01:49,200
And I bounced around between kind of the networking

36
00:01:49,200 --> 00:01:53,040
side of things, an application administration and development

37
00:01:53,040 --> 00:01:55,920
just throughout my entire career because I kind of like all

38
00:01:55,920 --> 00:02:00,280
the aspects of the Microsoft technologies or really

39
00:02:00,280 --> 00:02:01,920
any technologies.

40
00:02:01,920 --> 00:02:05,160
But I definitely gravitated a lot more towards Microsoft

41
00:02:05,160 --> 00:02:10,120
and really, really liked SharePoint when I first saw it.

42
00:02:10,120 --> 00:02:13,600
And I think it was SharePoint portal server 2003.

43
00:02:13,600 --> 00:02:18,120
It was really, really early in the life of SharePoint.

44
00:02:18,120 --> 00:02:20,320
And actually started as a developer.

45
00:02:20,320 --> 00:02:23,800
So, you know, did a lot of development and really

46
00:02:23,800 --> 00:02:27,480
got a passion for all of these collaboration tools.

47
00:02:27,480 --> 00:02:29,880
Of course, you know, several years ago, AI started

48
00:02:29,880 --> 00:02:32,560
to come out and be kind of a thing for enterprises

49
00:02:32,560 --> 00:02:36,400
and started kind of like with everyone else did.

50
00:02:36,400 --> 00:02:39,120
Just you're going to chat, GPT, and you're just talking

51
00:02:39,120 --> 00:02:40,800
to what you're asking questions.

52
00:02:40,800 --> 00:02:42,640
And you don't really know how to do it.

53
00:02:42,640 --> 00:02:43,800
You don't know what a prompt is.

54
00:02:43,800 --> 00:02:49,200
You're just using it the way you would think to use it.

55
00:02:49,200 --> 00:02:52,040
Of course, over time, you start to learn more about prompts

56
00:02:52,040 --> 00:02:53,640
and a lot of that stuff.

57
00:02:53,640 --> 00:02:55,800
And then you start to hear about some of these agents.

58
00:02:55,800 --> 00:02:59,800
And now you can build custom versions of AI

59
00:02:59,800 --> 00:03:03,840
and they could be like really, really good at doing,

60
00:03:03,840 --> 00:03:05,320
you know, particular things.

61
00:03:05,320 --> 00:03:09,560
And then you start to find some of these use cases

62
00:03:09,560 --> 00:03:12,760
to build out an agent to start to save you time.

63
00:03:12,760 --> 00:03:14,280
And that's--

64
00:03:14,280 --> 00:03:17,960
I use agents more than a lot of other types of AI.

65
00:03:17,960 --> 00:03:21,280
Oh, certainly more than a co-pilot chat, I would say,

66
00:03:21,280 --> 00:03:24,160
because there's a lot of processes

67
00:03:24,160 --> 00:03:27,760
that I'll do both as a content creator, as a consultant,

68
00:03:27,760 --> 00:03:30,280
you know, across a number of different roles

69
00:03:30,280 --> 00:03:34,640
that I will always be using agents for something.

70
00:03:34,640 --> 00:03:37,280
It helps me save it a lot of time.

71
00:03:37,280 --> 00:03:41,000
And I've really gotten into the agent ecosystem--

72
00:03:41,000 --> 00:03:44,720
no code, no code, pro code, all of that stuff.

73
00:03:44,720 --> 00:03:47,920
I kind of live and breathe that every single day

74
00:03:47,920 --> 00:03:50,040
at my job at Quizative.

75
00:03:50,040 --> 00:03:52,320
Yeah, also.

76
00:03:52,320 --> 00:03:56,480
What would you say made the agent develop so much more

77
00:03:56,480 --> 00:04:02,320
interesting than, yeah, try to add additional co-pilot prompting?

78
00:04:02,320 --> 00:04:05,240
Because it scratched my age for development,

79
00:04:05,240 --> 00:04:09,560
I was a developer, but well before college,

80
00:04:09,560 --> 00:04:14,360
back in the BBS days, and I was writing, let's see, Pascal,

81
00:04:14,360 --> 00:04:19,040
and a couple of the languages for bulletin board systems.

82
00:04:19,040 --> 00:04:21,800
If you remember those from the dial-up days,

83
00:04:21,800 --> 00:04:24,080
this was pre-internet.

84
00:04:24,080 --> 00:04:26,840
And definitely, I've always liked writing code.

85
00:04:26,840 --> 00:04:28,880
I've liked learning different languages,

86
00:04:28,880 --> 00:04:31,160
and because you're creating something from scratch.

87
00:04:31,160 --> 00:04:33,440
So I'm not that much of a creative person.

88
00:04:33,440 --> 00:04:35,480
I'm certainly not an artistic person,

89
00:04:35,480 --> 00:04:37,640
but I know how to think through things logically.

90
00:04:37,640 --> 00:04:43,520
I know how to order micro tasks into a larger process

91
00:04:43,520 --> 00:04:46,880
to accomplish a big goal.

92
00:04:46,880 --> 00:04:50,040
And so agents really fit right into that.

93
00:04:50,040 --> 00:04:56,040
And so that's, I guess that's a long-winded version of the answer.

94
00:04:56,040 --> 00:04:57,280
Yeah.

95
00:04:57,280 --> 00:04:58,360
Yeah.

96
00:04:58,360 --> 00:05:05,760
So for people who most are familiar with OpenAI or co-pilot,

97
00:05:05,760 --> 00:05:08,400
yeah, they think it's a prompting tool.

98
00:05:08,400 --> 00:05:14,040
But what is Microsoft 365 co-pilot agent?

99
00:05:14,040 --> 00:05:18,560
So if you think about what M365 co-pilot is,

100
00:05:18,560 --> 00:05:20,720
it is just a personal assistant.

101
00:05:20,720 --> 00:05:24,240
It is there to help you find information faster across all

102
00:05:24,240 --> 00:05:25,440
of M365.

103
00:05:25,440 --> 00:05:29,520
It's there to help you in general ways.

104
00:05:29,520 --> 00:05:34,640
If you've heard of the phrase a Jack of All Trades Master of None,

105
00:05:34,640 --> 00:05:40,240
well, the, I would say M365 co-pilot is the Jack of All Trades.

106
00:05:40,240 --> 00:05:43,400
They can do a lot of things sort of well.

107
00:05:43,400 --> 00:05:45,200
And that goes the same for chat GPT.

108
00:05:45,200 --> 00:05:47,720
It goes the same for blogs chat.

109
00:05:47,720 --> 00:05:50,080
It's fairly good at a lot of things.

110
00:05:50,080 --> 00:05:52,920
It's never really great at any one of those things,

111
00:05:52,920 --> 00:05:54,800
though, because it's very broad.

112
00:05:54,800 --> 00:05:57,880
The agent is the opposite of that.

113
00:05:57,880 --> 00:06:01,760
It's not designed to do everything under the sun.

114
00:06:01,760 --> 00:06:03,600
It's designed to do one thing,

115
00:06:03,600 --> 00:06:06,160
and it's designed to do it exceptionally well,

116
00:06:06,160 --> 00:06:09,920
because you're going to teach this agent what it needs to know.

117
00:06:09,920 --> 00:06:11,160
You're going to be providing,

118
00:06:11,160 --> 00:06:14,520
if you think crops can get long, instructions,

119
00:06:14,520 --> 00:06:17,040
or agents can get a whole lot longer.

120
00:06:17,040 --> 00:06:19,120
And usually in the Microsoft world,

121
00:06:19,120 --> 00:06:21,840
you have about 8,000 characters.

122
00:06:21,840 --> 00:06:26,320
And I usually recommend about using almost all of those things,

123
00:06:26,320 --> 00:06:28,840
because you're going to teach it everything it needs to know

124
00:06:28,840 --> 00:06:32,000
about how to do this one job it needs to do.

125
00:06:32,000 --> 00:06:35,520
It should not do anything else, but that one process,

126
00:06:35,520 --> 00:06:40,600
that one task, that way you put blinders on that agent.

127
00:06:40,600 --> 00:06:44,600
It's not thinking about anything else, except for that one thing.

128
00:06:44,600 --> 00:06:49,000
So that's kind of the opposite of what co-pilot is,

129
00:06:49,000 --> 00:06:53,560
and why those two complement each other very well.

130
00:06:53,560 --> 00:06:56,120
And when we think about,

131
00:06:56,120 --> 00:06:59,520
how do you think about is an agent more like a software?

132
00:06:59,520 --> 00:07:02,920
It's more like an automation, it's more like an employee,

133
00:07:02,920 --> 00:07:06,880
or something, entry-lead different.

134
00:07:06,880 --> 00:07:09,560
It is depending on the,

135
00:07:09,560 --> 00:07:11,000
like it be any of those things,

136
00:07:11,000 --> 00:07:14,440
and that's part of the evolution that we've seen with agents.

137
00:07:14,440 --> 00:07:18,040
First, they started out with simple pasts.

138
00:07:18,040 --> 00:07:24,520
Maybe it was creating a particular type of document,

139
00:07:24,520 --> 00:07:27,240
and it was a little bit more of an automation,

140
00:07:27,240 --> 00:07:32,240
or maybe a blend of automation and assistant,

141
00:07:32,240 --> 00:07:36,640
but it was, it was therefore that one thing.

142
00:07:36,640 --> 00:07:40,680
We later saw agents evolve it more into automations,

143
00:07:40,680 --> 00:07:42,840
and we see that was certainly with a pro code side,

144
00:07:42,840 --> 00:07:45,040
but also with something like co-pilot studio,

145
00:07:45,040 --> 00:07:49,040
where now you can have an agent that can trigger a flow,

146
00:07:49,040 --> 00:07:51,920
or a flow could trigger an agent.

147
00:07:51,920 --> 00:07:54,920
And, and, and,

148
00:07:54,920 --> 00:07:59,480
then, so now it's able to start to interact with other systems,

149
00:07:59,480 --> 00:08:05,160
write files, and, and, and do more system integration level tasks.

150
00:08:05,160 --> 00:08:09,720
And, more recently, I think it was November of last year,

151
00:08:09,720 --> 00:08:13,800
at Ignite, Microsoft introduced agent 365

152
00:08:13,800 --> 00:08:16,440
in this notion of AI teammates.

153
00:08:16,440 --> 00:08:18,640
So now, instead of agents,

154
00:08:18,640 --> 00:08:22,640
which had been previously more like a piece of software,

155
00:08:22,640 --> 00:08:25,520
an agent, you, you deploy an agent to an organization,

156
00:08:25,520 --> 00:08:27,280
and all users just simply use it.

157
00:08:27,280 --> 00:08:28,120
It's there.

158
00:08:28,120 --> 00:08:31,720
It's kind of like any other app that might show up inside teams,

159
00:08:31,720 --> 00:08:33,640
everyone uses the same copy.

160
00:08:33,640 --> 00:08:37,680
AI teammates kind of change that now they're acting like people.

161
00:08:37,680 --> 00:08:41,600
They don't require someone to go talk to it.

162
00:08:41,600 --> 00:08:44,200
It's not just sitting there waiting for someone to talk to it.

163
00:08:44,200 --> 00:08:46,040
It can do things on its own.

164
00:08:46,040 --> 00:08:49,560
It can react to changes in data into,

165
00:08:49,560 --> 00:08:57,640
in, and other systems. And so they act more like an actual AI being.

166
00:08:57,640 --> 00:08:59,600
But it's still an agent on the back end.

167
00:08:59,600 --> 00:09:05,440
There's just a lot more sophistication that has been brought into this,

168
00:09:05,440 --> 00:09:09,560
this agent technology.

169
00:09:09,560 --> 00:09:15,440
And, I think what, what, when we build an agent or, or re-ashitecture an agent,

170
00:09:15,440 --> 00:09:20,760
what are the main components to make up an agent?

171
00:09:20,760 --> 00:09:22,040
So that's a great question.

172
00:09:22,040 --> 00:09:27,680
It's the, it's actually three things, three core things that really make up an agent.

173
00:09:27,680 --> 00:09:29,120
One is the instructions.

174
00:09:29,120 --> 00:09:30,520
I touched on that.

175
00:09:30,520 --> 00:09:37,400
And that is essentially the very, very long prompt that teaches this agent,

176
00:09:37,400 --> 00:09:39,560
which is a blank slate to start with.

177
00:09:39,560 --> 00:09:43,200
It teaches it what it needs to know about its job.

178
00:09:43,200 --> 00:09:47,680
You tell it what its role is, how it should be responding to people,

179
00:09:47,680 --> 00:09:53,720
what tasks it's going to be asked to help with or what processes,

180
00:09:53,720 --> 00:09:55,320
all the steps of that process.

181
00:09:55,320 --> 00:09:56,640
You're going to go into great detail.

182
00:09:56,640 --> 00:10:00,360
I usually tell people, pre an agent, like it's a five-year-old,

183
00:10:00,360 --> 00:10:05,280
you're going to spell out every single thing that it needs to know on how to do its job.

184
00:10:05,280 --> 00:10:07,040
And if you do that, do it well.

185
00:10:07,040 --> 00:10:12,760
That agent's going to know and respond exactly the way you want it to.

186
00:10:12,760 --> 00:10:16,840
So the instructions are definitely the most important thing,

187
00:10:16,840 --> 00:10:19,160
it's make or break for the agent.

188
00:10:19,160 --> 00:10:24,400
Then you've got the knowledge, what knowledge does this agent have access to,

189
00:10:24,400 --> 00:10:26,640
if any, to do its job.

190
00:10:26,640 --> 00:10:30,840
Not every agent actually needs to be grounded on organizational knowledge.

191
00:10:30,840 --> 00:10:36,080
A lot of times, just the, the, the large language model itself,

192
00:10:36,080 --> 00:10:39,400
it will, will provide enough knowledge for this.

193
00:10:39,400 --> 00:10:41,560
And then there's the tool set.

194
00:10:41,560 --> 00:10:45,160
So what tools does this agent have access to?

195
00:10:45,160 --> 00:10:48,160
Does it have access to find M365 data?

196
00:10:48,160 --> 00:10:50,200
Does it have access to find?

197
00:10:50,200 --> 00:10:52,920
Maybe it's just SharePoint, maybe it's just OneDrive.

198
00:10:52,920 --> 00:10:55,920
Maybe there's a service now integration.

199
00:10:55,920 --> 00:10:59,560
And this agent is going to be able to connect into service now

200
00:10:59,560 --> 00:11:05,360
and pull ticket information in the case of maybe a helpdesk agent.

201
00:11:05,360 --> 00:11:10,160
So other, this is more into the system integration side,

202
00:11:10,160 --> 00:11:14,960
but it's giving the agent the different tools in its tool belt

203
00:11:14,960 --> 00:11:19,120
so that when it needs to, based on your instructions,

204
00:11:19,120 --> 00:11:25,640
when it does have the ability to talk to other systems

205
00:11:25,640 --> 00:11:30,360
to call specialized tools, whatever the need is,

206
00:11:30,360 --> 00:11:35,240
and it can use those tools to accomplish its job,

207
00:11:35,240 --> 00:11:38,400
which again is you're defining in the agent instructions.

208
00:11:38,400 --> 00:11:41,960
But really those three things, instructions, knowledge and tools

209
00:11:41,960 --> 00:11:45,000
are the core things that make up an agent.

210
00:11:45,000 --> 00:11:51,200
Yeah, I think a little bit, yeah, before agents come,

211
00:11:51,200 --> 00:11:55,800
we have, if we would like to work with different tools,

212
00:11:55,800 --> 00:12:00,000
we have, yeah, work with Microsoft Graph.

213
00:12:00,000 --> 00:12:01,800
Do this also play a role?

214
00:12:02,800 --> 00:12:12,600
So the Microsoft Graph is also an important part.

215
00:12:12,600 --> 00:12:18,800
Graph is that so yes, Graph would really fall under the knowledge

216
00:12:18,800 --> 00:12:23,200
and potentially tools because there's so much there to graph.

217
00:12:23,200 --> 00:12:27,200
Behind the scenes, when co-pilot is going to go get knowledge

218
00:12:27,200 --> 00:12:31,200
from Microsoft 365, it's going to be calling Graph

219
00:12:31,200 --> 00:12:33,000
or if it's getting knowledge from Azure,

220
00:12:33,000 --> 00:12:37,600
almost any entry point into the Microsoft Knowledge Network

221
00:12:37,600 --> 00:12:41,000
is always going through Graph.

222
00:12:41,000 --> 00:12:47,000
So that any, like behind the scenes, you'll be using Graph,

223
00:12:47,000 --> 00:12:51,000
unless you're getting into more advanced agents,

224
00:12:51,000 --> 00:12:54,000
specifically do something like Crocode,

225
00:12:54,000 --> 00:12:58,400
you don't have to worry too much about the Graph entry point

226
00:12:58,400 --> 00:13:02,800
because you're presented with friendlier terminology

227
00:13:02,800 --> 00:13:06,200
like SharePoint or OneDrive or Teams or things like that,

228
00:13:06,200 --> 00:13:08,400
but it's always using Graph behind the scenes.

229
00:13:08,400 --> 00:13:10,400
In addition, suppose you've got an agent

230
00:13:10,400 --> 00:13:16,400
that is doing some work inside of Azure,

231
00:13:16,400 --> 00:13:22,800
then it could be calling Graph to provision resources inside Azure.

232
00:13:22,800 --> 00:13:26,400
So you can have tools that way as well.

233
00:13:26,400 --> 00:13:28,700
And there's a number of different ways to define tools,

234
00:13:28,700 --> 00:13:32,200
but yes, Graph, if you're having an agent interact

235
00:13:32,200 --> 00:13:34,800
with Microsoft 365 in any way,

236
00:13:34,800 --> 00:13:40,000
it's always going to be leveraging Graph as its doorway

237
00:13:40,000 --> 00:13:42,800
into that world.

238
00:13:42,800 --> 00:13:44,600
And you said knowledge.

239
00:13:44,600 --> 00:13:49,600
And I think about knowledge, how important is

240
00:13:49,600 --> 00:13:53,200
for the agents that we have structured data?

241
00:13:56,200 --> 00:14:01,200
So it depends if you want, I mean,

242
00:14:01,200 --> 00:14:05,200
what I usually, the way I approach this is,

243
00:14:05,200 --> 00:14:09,200
if an agent has a need for structured data,

244
00:14:09,200 --> 00:14:12,600
then I will incorporate that.

245
00:14:12,600 --> 00:14:16,200
But one of the interesting things about,

246
00:14:16,200 --> 00:14:19,800
and this is getting more into the data science side, I believe,

247
00:14:19,800 --> 00:14:22,600
and that's definitely not my expertise,

248
00:14:22,600 --> 00:14:27,600
but the ability to index content has changed

249
00:14:27,600 --> 00:14:30,600
with the evolution of AI, because in the past,

250
00:14:30,600 --> 00:14:33,600
we would have a traditional like Google type search,

251
00:14:33,600 --> 00:14:35,600
where it was based off of keywords.

252
00:14:35,600 --> 00:14:40,600
Well, with AI, we had a semantic index that came about

253
00:14:40,600 --> 00:14:45,600
where now all of this data that is unstructured

254
00:14:45,600 --> 00:14:47,600
along with any metadata,

255
00:14:47,600 --> 00:14:50,600
so things like your custom columns and such,

256
00:14:50,600 --> 00:14:53,600
SharePoint and things like that, all of that stuff,

257
00:14:53,600 --> 00:14:57,600
it's indexed into one large index that was specifically tuned

258
00:14:57,600 --> 00:14:59,600
for AI.

259
00:14:59,600 --> 00:15:03,600
So you may start with structured data in,

260
00:15:03,600 --> 00:15:05,600
maybe it's an Azure SQL,

261
00:15:05,600 --> 00:15:07,600
or maybe it's in SharePoint,

262
00:15:07,600 --> 00:15:14,600
but they'll usually get indexed into some sort of an AI-friendly index

263
00:15:14,600 --> 00:15:17,600
so that AI can find this quickly,

264
00:15:17,600 --> 00:15:20,600
because you certainly don't want,

265
00:15:20,600 --> 00:15:25,600
you typically don't want AI executing SQL statements

266
00:15:25,600 --> 00:15:27,600
on your SQL server.

267
00:15:27,600 --> 00:15:30,600
That's usually not desirable.

268
00:15:30,600 --> 00:15:35,600
The DBAs would just freak out if there was such,

269
00:15:35,600 --> 00:15:37,600
that got a behavior,

270
00:15:37,600 --> 00:15:39,600
and a large language model could just execute

271
00:15:39,600 --> 00:15:41,600
whatever SQL it wanted.

272
00:15:41,600 --> 00:15:43,600
Usually there is an index involved,

273
00:15:43,600 --> 00:15:48,600
so co-pilot specifically can leverage your structured data,

274
00:15:48,600 --> 00:15:50,600
it's gotten a lot better.

275
00:15:50,600 --> 00:15:52,600
Now, when it first came out,

276
00:15:52,600 --> 00:15:56,600
basically it's ability to use metadata columns

277
00:15:56,600 --> 00:15:59,600
was like non-existent in SharePoint,

278
00:15:59,600 --> 00:16:02,600
and that was the first thing that a lot of people noticed.

279
00:16:02,600 --> 00:16:05,600
That was around the time that I became a Microsoft MVP as well,

280
00:16:05,600 --> 00:16:09,600
and I heard a lot of MVs complaining about this as well,

281
00:16:09,600 --> 00:16:11,600
because Microsoft teaches us,

282
00:16:11,600 --> 00:16:13,600
they've always taught us,

283
00:16:13,600 --> 00:16:15,600
use metadata columns in SharePoint.

284
00:16:15,600 --> 00:16:18,600
This is not just an online file share.

285
00:16:18,600 --> 00:16:23,600
This is not what we used to do in the Maps Drive days.

286
00:16:23,600 --> 00:16:25,600
You need metadata columns,

287
00:16:25,600 --> 00:16:27,600
you need to enrich your data,

288
00:16:27,600 --> 00:16:29,600
it's not just the document.

289
00:16:29,600 --> 00:16:32,600
It's now the metadata that goes along with that.

290
00:16:32,600 --> 00:16:34,600
But yeah, co-pilot came out,

291
00:16:34,600 --> 00:16:36,600
and it couldn't even read any of that stuff.

292
00:16:36,600 --> 00:16:39,600
It has gotten better significantly better in the last,

293
00:16:39,600 --> 00:16:41,600
12 months, I would say,

294
00:16:41,600 --> 00:16:43,600
with being able to read some of that data,

295
00:16:43,600 --> 00:16:47,600
all the metadata that accompanies the actual data,

296
00:16:47,600 --> 00:16:51,600
the actual documents, for instance.

297
00:16:51,600 --> 00:16:53,600
And the agent,

298
00:16:53,600 --> 00:16:55,600
is it more like,

299
00:16:55,600 --> 00:16:57,600
I think, a pro-code,

300
00:16:57,600 --> 00:17:00,600
visual, basic studios thing,

301
00:17:00,600 --> 00:17:04,600
or it's also in citizen development environments,

302
00:17:04,600 --> 00:17:06,600
like power platforms,

303
00:17:06,600 --> 00:17:09,600
I don't call pages, or power automated.

304
00:17:09,600 --> 00:17:15,600
That is the interesting thing about how Microsoft introduced agents.

305
00:17:15,600 --> 00:17:18,600
They made them accessible for everyone,

306
00:17:18,600 --> 00:17:21,600
regardless of their skill level.

307
00:17:21,600 --> 00:17:24,600
On one end of the spectrum you've got,

308
00:17:24,600 --> 00:17:29,600
SharePoint agents, probably the newest type of agent that was introduced,

309
00:17:29,600 --> 00:17:31,600
is SharePoint agents,

310
00:17:31,600 --> 00:17:34,600
where all you had to do is provide agent instructions.

311
00:17:34,600 --> 00:17:37,600
Like, at a core level, that's it.

312
00:17:37,600 --> 00:17:39,600
You, maybe if you want to go a little bit more,

313
00:17:39,600 --> 00:17:41,600
get a little bit fancier,

314
00:17:41,600 --> 00:17:43,600
you could say you just want an agent that,

315
00:17:43,600 --> 00:17:44,600
with these instructions,

316
00:17:44,600 --> 00:17:46,600
it's grounded just on this library.

317
00:17:46,600 --> 00:17:48,600
You've got agent builder,

318
00:17:48,600 --> 00:17:49,600
which gets a little bit fancier.

319
00:17:49,600 --> 00:17:52,600
You've started, you start to get some tools now.

320
00:17:52,600 --> 00:17:54,600
You got, you got co-pilot studio,

321
00:17:54,600 --> 00:17:59,600
which has definitely had some very significant changes recently.

322
00:17:59,600 --> 00:18:01,600
But that had more complexity,

323
00:18:01,600 --> 00:18:03,600
and now you're inside Power Platform,

324
00:18:03,600 --> 00:18:07,600
which has its own set of challenges that you deal with.

325
00:18:07,600 --> 00:18:09,600
Then you've got Pro Code on the,

326
00:18:09,600 --> 00:18:13,600
the, the, highest difficulty, I would say.

327
00:18:13,600 --> 00:18:17,600
And you, you have to do everything yourself.

328
00:18:17,600 --> 00:18:20,600
So, and, and I know that recently,

329
00:18:20,600 --> 00:18:23,600
I've heard Microsoft say that, you know,

330
00:18:23,600 --> 00:18:26,600
you should be able to create the agent you want

331
00:18:26,600 --> 00:18:29,600
with the technology that you want.

332
00:18:29,600 --> 00:18:32,600
They, they wanted, they, they wanted accessible to everyone.

333
00:18:32,600 --> 00:18:37,600
They want to enable people to create their own agents,

334
00:18:37,600 --> 00:18:41,600
even if they don't have the same skill set as someone else.

335
00:18:41,600 --> 00:18:42,600
Certainly, as a pro developer,

336
00:18:42,600 --> 00:18:46,600
they don't want you to have to have a pro developer skill set.

337
00:18:46,600 --> 00:18:51,600
So, there's, they do their best to,

338
00:18:51,600 --> 00:18:56,600
have, to bring parity to all of those different experiences.

339
00:18:56,600 --> 00:19:02,600
So, you're not limited while also not getting overwhelmed

340
00:19:02,600 --> 00:19:03,600
with the things.

341
00:19:03,600 --> 00:19:06,600
If someone is brand new to AI in general,

342
00:19:06,600 --> 00:19:08,600
they're just starting to learn things.

343
00:19:08,600 --> 00:19:10,600
They don't need to be in co-pilot studio,

344
00:19:10,600 --> 00:19:12,600
where they've got all these settings,

345
00:19:12,600 --> 00:19:14,600
all these things that, there's too much there

346
00:19:14,600 --> 00:19:16,600
that would overwhelm them.

347
00:19:16,600 --> 00:19:19,600
So, trying to find that balance is, I think,

348
00:19:19,600 --> 00:19:22,600
what Microsoft is currently faced with,

349
00:19:22,600 --> 00:19:28,600
with how to, how to enable someone without overwhelming them.

350
00:19:28,600 --> 00:19:31,600
And you have, before you say,

351
00:19:31,600 --> 00:19:35,600
there, there's integration for, for, for, service now.

352
00:19:35,600 --> 00:19:40,600
Is it possible to have, is it only with working with integrations?

353
00:19:40,600 --> 00:19:43,600
Or is it only in the Microsoft ecosystem?

354
00:19:43,600 --> 00:19:46,600
Or can I use also, I don't know,

355
00:19:46,600 --> 00:19:48,600
other things like, I don't know,

356
00:19:48,600 --> 00:19:53,600
Google neps are, are something to work in, in this agent.

357
00:19:53,600 --> 00:20:00,600
So, the, the agent ecosystem in general in the world,

358
00:20:00,600 --> 00:20:04,600
it has settled on a number of different protocols

359
00:20:04,600 --> 00:20:10,600
and, and ways to communicate that you can have an agent talk to,

360
00:20:10,600 --> 00:20:13,600
a, so let's say you have an agent in the co-pilot world,

361
00:20:13,600 --> 00:20:17,600
you can have a talk to a Google vertex agent.

362
00:20:17,600 --> 00:20:20,600
Or an Amazon bedrock agent.

363
00:20:20,600 --> 00:20:23,600
Because they'll all be implementing the same languages

364
00:20:23,600 --> 00:20:26,600
or protocols, things like, agent to agent,

365
00:20:26,600 --> 00:20:30,600
or MCP, model context protocol.

366
00:20:30,600 --> 00:20:33,600
They'll, they'll be talking to, those same languages.

367
00:20:33,600 --> 00:20:38,600
So, that's, that's where I'm seeing a lot of customers

368
00:20:38,600 --> 00:20:42,600
preferring certain technologies, a lot of customers I talk to,

369
00:20:42,600 --> 00:20:45,600
prefer the anthropic ecosystem.

370
00:20:45,600 --> 00:20:48,600
They want to use the anthropic API, they want to use the

371
00:20:48,600 --> 00:20:51,600
Clawed models, or they want to use the Google,

372
00:20:51,600 --> 00:20:55,600
the Google infrastructure, because of their,

373
00:20:55,600 --> 00:20:57,600
the agent building experience there,

374
00:20:57,600 --> 00:21:01,600
but they also still want the Microsoft grounding

375
00:21:01,600 --> 00:21:03,600
and other things like that.

376
00:21:03,600 --> 00:21:07,600
So, you can use any technology you really want to,

377
00:21:07,600 --> 00:21:11,600
and you would still be leveraging a lot of the common things

378
00:21:11,600 --> 00:21:13,600
that we've talked about so far, things like graph.

379
00:21:13,600 --> 00:21:16,600
If you want to get the graph data,

380
00:21:16,600 --> 00:21:18,600
if you want to leverage these semantic index

381
00:21:18,600 --> 00:21:21,600
that is provided from a through graph,

382
00:21:21,600 --> 00:21:24,600
you can do all that so you could have the agent live

383
00:21:24,600 --> 00:21:29,600
where you want grounded and using the data that you want,

384
00:21:29,600 --> 00:21:34,600
even if it's not on the Microsoft platform.

385
00:21:34,600 --> 00:21:37,600
Okay, that's really interesting.

386
00:21:37,600 --> 00:21:40,600
That's, that's a lot of capability about things.

387
00:21:40,600 --> 00:21:46,600
So, when, when people think about building an agent ecosystem,

388
00:21:46,600 --> 00:21:52,600
what, should they do before they start with building?

389
00:21:52,600 --> 00:21:55,600
Yeah, that, that is the million dollar question,

390
00:21:55,600 --> 00:22:02,600
because that is where most companies will immediately start

391
00:22:02,600 --> 00:22:04,600
incorrectly.

392
00:22:04,600 --> 00:22:07,600
They will start with just turning on AI,

393
00:22:07,600 --> 00:22:10,600
and they've already made the first mistake,

394
00:22:10,600 --> 00:22:13,600
because I, I talked about the semantic index.

395
00:22:13,600 --> 00:22:17,600
Most, most of these platform providers now understand

396
00:22:17,600 --> 00:22:22,600
semantic indexes, they are incredibly well at finding data.

397
00:22:22,600 --> 00:22:25,600
That's why everyone started to love chat G-P-T,

398
00:22:25,600 --> 00:22:27,600
when it, when it first came out,

399
00:22:27,600 --> 00:22:30,600
and I, I usually go back to that example,

400
00:22:30,600 --> 00:22:32,600
because for me personally, that was my first,

401
00:22:32,600 --> 00:22:36,600
I'm really hearing about an AI chat interface,

402
00:22:36,600 --> 00:22:39,600
and being able to talk about, talk to,

403
00:22:39,600 --> 00:22:41,600
about things, find information,

404
00:22:41,600 --> 00:22:43,600
because it could find anything,

405
00:22:43,600 --> 00:22:45,600
I didn't have to worry about what keywords to use.

406
00:22:45,600 --> 00:22:48,600
But the problem with business data now is,

407
00:22:48,600 --> 00:22:52,600
if you turn on AI, even if it's just, you know,

408
00:22:52,600 --> 00:22:56,600
in 365 co-pilot, users are going to be able to find data

409
00:22:56,600 --> 00:23:00,600
incredibly well, including the data they're not supposed to have access to,

410
00:23:00,600 --> 00:23:05,600
but they do, because people have had such bad practices,

411
00:23:05,600 --> 00:23:08,600
historically within the M365,

412
00:23:08,600 --> 00:23:11,600
world or other environments, with sharing information.

413
00:23:11,600 --> 00:23:13,600
A lot of people would just share information

414
00:23:13,600 --> 00:23:15,600
with everywhere in the company,

415
00:23:15,600 --> 00:23:18,600
because it just made, it made it easier.

416
00:23:18,600 --> 00:23:20,600
They, they'd have to deal with permission issues anymore,

417
00:23:20,600 --> 00:23:22,600
because if everything was, was wide open,

418
00:23:22,600 --> 00:23:23,600
then great.

419
00:23:23,600 --> 00:23:28,600
If no one knew that this one particular SharePoint site had,

420
00:23:28,600 --> 00:23:32,600
HR data, with a salary spreadsheet,

421
00:23:32,600 --> 00:23:35,600
no one knew about it, well, no one could find it, right?

422
00:23:35,600 --> 00:23:37,600
Only the people who, who were supposed to have access

423
00:23:37,600 --> 00:23:41,600
would have that link, and easy, no problem.

424
00:23:41,600 --> 00:23:44,600
And with Microsoft Graph Search, the old,

425
00:23:44,600 --> 00:23:47,600
like the original kind of SharePoint Search,

426
00:23:47,600 --> 00:23:50,600
that was, it wasn't too much of a challenge.

427
00:23:50,600 --> 00:23:54,600
Like, if someone worked hard, they might be able to find that,

428
00:23:54,600 --> 00:23:58,600
but with AI, again, everything became easy.

429
00:23:58,600 --> 00:24:00,600
You didn't have to know the name of the file.

430
00:24:00,600 --> 00:24:02,600
You didn't have to know any keywords.

431
00:24:02,600 --> 00:24:05,600
If you just described, I want the spreadsheet

432
00:24:05,600 --> 00:24:07,600
with all of the salaries for everyone.

433
00:24:07,600 --> 00:24:09,600
AI just simply found it.

434
00:24:09,600 --> 00:24:12,600
So the biggest thing that organizations have to do well before

435
00:24:12,600 --> 00:24:17,600
they want to turn on AI is, they've got to govern their data better.

436
00:24:17,600 --> 00:24:22,600
They have to look at all of the over-promission files,

437
00:24:22,600 --> 00:24:24,600
sites, libraries, everything.

438
00:24:24,600 --> 00:24:27,600
They've got to make sure that everything is secured correctly.

439
00:24:27,600 --> 00:24:31,600
They may have to make sure that if there's a confidential file,

440
00:24:31,600 --> 00:24:34,600
that they're leveraging something like Microsoft Perview

441
00:24:34,600 --> 00:24:36,600
to label that file as confidential

442
00:24:36,600 --> 00:24:39,600
and have additional protections on that,

443
00:24:39,600 --> 00:24:42,600
things like making sure it can't be shared outside the organization,

444
00:24:42,600 --> 00:24:45,600
it can't be printed, whatever those protections are.

445
00:24:45,600 --> 00:24:48,600
You need to make sure that that data is secured

446
00:24:48,600 --> 00:24:51,600
with all of the proper methods,

447
00:24:51,600 --> 00:24:54,600
because, at least with M365.

448
00:24:54,600 --> 00:24:55,600
M365.

449
00:24:55,600 --> 00:24:57,600
It's aware of those protections,

450
00:24:57,600 --> 00:24:59,600
and it's going to honor those protections.

451
00:24:59,600 --> 00:25:00,600
I always tell people,

452
00:25:00,600 --> 00:25:02,600
"Her view will always win out,

453
00:25:02,600 --> 00:25:04,600
but that's only if you're leveraging it."

454
00:25:04,600 --> 00:25:10,600
So, starting by securing your whole environment,

455
00:25:10,600 --> 00:25:12,600
that's always going to be the first step to AI,

456
00:25:12,600 --> 00:25:15,600
because whatever weaknesses you have,

457
00:25:15,600 --> 00:25:18,600
AI will expose it to all of your users,

458
00:25:18,600 --> 00:25:22,600
because it will make everything so easy to find.

459
00:25:22,600 --> 00:25:25,600
For good or for bad?

460
00:25:25,600 --> 00:25:31,600
Can you tell us how?

461
00:25:31,600 --> 00:25:35,600
I don't know, a real project without,

462
00:25:35,600 --> 00:25:39,600
for a personal or without saying the name of the company,

463
00:25:39,600 --> 00:25:41,600
but how did you build,

464
00:25:41,600 --> 00:25:44,600
can you check or through step by step?

465
00:25:44,600 --> 00:25:46,600
A little bit.

466
00:25:46,600 --> 00:25:49,600
So, like an example agent?

467
00:25:49,600 --> 00:25:52,600
Yes, so, there's one that I'm building right now.

468
00:25:52,600 --> 00:25:56,600
This is usually one of the absolute best use cases.

469
00:25:56,600 --> 00:25:59,600
It is an intranet agent.

470
00:25:59,600 --> 00:26:02,600
Most companies, I would say, have intranets,

471
00:26:02,600 --> 00:26:04,600
not all of them, but most of them.

472
00:26:04,600 --> 00:26:07,600
And so, you've got a whole bunch of different shipwants sites

473
00:26:07,600 --> 00:26:10,600
organized into a hub network,

474
00:26:10,600 --> 00:26:13,600
and you've got information scattered all throughout there.

475
00:26:13,600 --> 00:26:15,600
And if you look at typical customer,

476
00:26:15,600 --> 00:26:18,600
it usually is hard for anyone to find anything

477
00:26:18,600 --> 00:26:23,600
on there, because someone may be looking in one site for a file,

478
00:26:23,600 --> 00:26:25,600
but it's in a whole different site.

479
00:26:25,600 --> 00:26:27,600
So, a lot of people have to dig around,

480
00:26:27,600 --> 00:26:31,600
constantly find, start the end of emailing someone.

481
00:26:31,600 --> 00:26:34,600
And then, hour later, they finally find the page

482
00:26:34,600 --> 00:26:36,600
of the file they're looking for.

483
00:26:36,600 --> 00:26:39,600
A agent can't solve that because I could have an agent

484
00:26:39,600 --> 00:26:41,600
that is an expert at that intranet.

485
00:26:41,600 --> 00:26:45,600
It's got instructions on exactly what it's supposed to do.

486
00:26:45,600 --> 00:26:49,600
It's given only the knowledge it needs to be able to do that.

487
00:26:49,600 --> 00:26:53,600
In this case, it's grounded on all of the intranet sites.

488
00:26:53,600 --> 00:26:56,600
And it doesn't even need tools in this case.

489
00:26:56,600 --> 00:26:59,600
It's just there to help people find the information

490
00:26:59,600 --> 00:27:00,600
they're looking for.

491
00:27:00,600 --> 00:27:03,600
So, if someone wants to start with a chat first,

492
00:27:03,600 --> 00:27:06,600
experienced on their intranet, they can.

493
00:27:06,600 --> 00:27:09,600
If they want to just keep browsing for things

494
00:27:09,600 --> 00:27:12,600
because maybe they know exactly where to find the information,

495
00:27:12,600 --> 00:27:15,600
then they start that way.

496
00:27:15,600 --> 00:27:17,600
So, the one I'm thinking of in particular,

497
00:27:17,600 --> 00:27:19,600
I'm currently working with them.

498
00:27:19,600 --> 00:27:26,600
And they're not a big fan of the co-pilot sidecar in SharePoint.

499
00:27:26,600 --> 00:27:30,600
It's not quite that obvious that it's there.

500
00:27:30,600 --> 00:27:33,600
A lot of users who are startly new to co-pilot,

501
00:27:33,600 --> 00:27:34,600
but definitely new to agents.

502
00:27:34,600 --> 00:27:36,600
They don't know that it's there.

503
00:27:36,600 --> 00:27:39,600
And so, they, but they wanted a different way.

504
00:27:39,600 --> 00:27:42,600
Now, if you look on a lot of internet sites,

505
00:27:42,600 --> 00:27:44,600
there's so many websites out there right now

506
00:27:44,600 --> 00:27:48,600
that have that little chat window on the bottom right corner

507
00:27:48,600 --> 00:27:51,600
and pretty much everyone recognizes that as,

508
00:27:51,600 --> 00:27:54,600
"Okay, there's an AI assistant here.

509
00:27:54,600 --> 00:27:56,600
So, I can just chat with us instead."

510
00:27:56,600 --> 00:28:00,600
So, this company wanted that experience.

511
00:28:00,600 --> 00:28:03,600
So, that's not provided.

512
00:28:03,600 --> 00:28:05,600
Like, there is no mechanism.

513
00:28:05,600 --> 00:28:08,600
Now, Microsoft actually recently rolled out

514
00:28:08,600 --> 00:28:11,600
the key to the name of the company.

515
00:28:11,600 --> 00:28:14,600
Well, they keep changing the name.

516
00:28:14,600 --> 00:28:15,600
What is the knowledge agent?

517
00:28:15,600 --> 00:28:18,600
Now, I think the current name is AI and SharePoint.

518
00:28:18,600 --> 00:28:20,600
They might rename it again.

519
00:28:20,600 --> 00:28:23,600
It's hard to keep up with the rebrands with Microsoft,

520
00:28:23,600 --> 00:28:25,600
but they wanted something like that.

521
00:28:25,600 --> 00:28:29,600
So, I used SharePoint framework and created a little chat window.

522
00:28:29,600 --> 00:28:31,600
They could just click on this icon.

523
00:28:31,600 --> 00:28:33,600
It would pop up the chat interface.

524
00:28:33,600 --> 00:28:34,600
They can talk with it.

525
00:28:34,600 --> 00:28:36,600
They can close it when they're done with it.

526
00:28:36,600 --> 00:28:41,600
They wanted, they didn't want to go into the Procode side

527
00:28:41,600 --> 00:28:43,600
because you have way different costs.

528
00:28:43,600 --> 00:28:45,600
You have more infrastructure.

529
00:28:45,600 --> 00:28:46,600
It's a more complex solution.

530
00:28:46,600 --> 00:28:50,600
And is it necessarily even needed

531
00:28:50,600 --> 00:28:53,600
for something as simple of a use case as this?

532
00:28:53,600 --> 00:28:56,600
So, we used Copilot Studio to build the agent.

533
00:28:56,600 --> 00:29:02,600
And we do have Procode for the SharePoint East

534
00:29:02,600 --> 00:29:05,600
because for that level of customization,

535
00:29:05,600 --> 00:29:08,600
we did have to create TypeScript for that.

536
00:29:08,600 --> 00:29:12,600
So, you have a currently testing that right now.

537
00:29:12,600 --> 00:29:16,600
We're making some tweaks to styling because if we're,

538
00:29:16,600 --> 00:29:19,600
you know, we're presenting a custom chat window.

539
00:29:19,600 --> 00:29:22,600
So, they get to tweak this all they want.

540
00:29:22,600 --> 00:29:24,600
If they want it wider, if they want it taller,

541
00:29:24,600 --> 00:29:27,600
or all these different behaviors, you can tweak.

542
00:29:27,600 --> 00:29:31,600
And it's still just accessing the same back end agent.

543
00:29:31,600 --> 00:29:35,600
So, that's one of the easiest and best use cases

544
00:29:35,600 --> 00:29:41,600
to start to realize value in an organization leveraging AI is.

545
00:29:41,600 --> 00:29:45,600
Something that most users will be going to at least once or twice a week.

546
00:29:45,600 --> 00:29:50,600
And that's an internet and just letting AI help users find content faster

547
00:29:50,600 --> 00:29:53,600
and get back to work.

548
00:29:53,600 --> 00:30:00,600
And when you have built an MVP and many other Vibram product,

549
00:30:00,600 --> 00:30:09,600
and you will roll this out in an company.

550
00:30:09,600 --> 00:30:13,600
What exactly have you do for it?

551
00:30:13,600 --> 00:30:19,600
And what are, yeah, say, misconceptions company have on the rollout?

552
00:30:19,600 --> 00:30:23,600
So, this is really where organization change management comes in

553
00:30:23,600 --> 00:30:27,600
where this isn't just something you'll just turn on

554
00:30:27,600 --> 00:30:29,600
and people will just notice.

555
00:30:29,600 --> 00:30:31,600
A lot of people, they'll never notice it.

556
00:30:31,600 --> 00:30:33,600
A lot of people in this case, like the internet case,

557
00:30:33,600 --> 00:30:36,600
maybe they've given up on ever finding anything on the internet

558
00:30:36,600 --> 00:30:38,600
because it's always been too difficult.

559
00:30:38,600 --> 00:30:43,600
So, getting the message out for, like assuming you've gone through a pilot test

560
00:30:43,600 --> 00:30:48,600
and the people in that pilot program have, like, they love it.

561
00:30:48,600 --> 00:30:50,600
They're done with the iterations.

562
00:30:50,600 --> 00:30:52,600
They're done with any sort of changes.

563
00:30:52,600 --> 00:30:53,600
It is ready.

564
00:30:53,600 --> 00:30:55,600
It's gone at Scott public.

565
00:30:55,600 --> 00:30:59,600
That's when you really want to start a communication campaign,

566
00:30:59,600 --> 00:31:02,600
an awareness campaign to make sure people understand

567
00:31:02,600 --> 00:31:07,600
this thing is now here and that, you know, this can help them find

568
00:31:07,600 --> 00:31:09,600
that document on the internet.

569
00:31:09,600 --> 00:31:13,600
It can help them find that policy that HR put out on travel

570
00:31:13,600 --> 00:31:16,600
or vacation time, whatever that is.

571
00:31:16,600 --> 00:31:23,600
And have a way to get some feedback from those users.

572
00:31:23,600 --> 00:31:26,600
Have maybe the using the even engage.

573
00:31:26,600 --> 00:31:32,600
And there could be a community for feedback on this particular agent

574
00:31:32,600 --> 00:31:35,600
or for AI feedback or co-pilot feedback.

575
00:31:35,600 --> 00:31:39,600
Having a communication loop, though, is important.

576
00:31:39,600 --> 00:31:43,600
Having an awareness campaign is definitely important

577
00:31:43,600 --> 00:31:48,600
because if you roll out AI, you're going to be spending money for this.

578
00:31:48,600 --> 00:31:53,600
Whether that's the time for someone to develop this agent

579
00:31:53,600 --> 00:31:57,600
or ongoing costs in Azure.

580
00:31:57,600 --> 00:31:59,600
If it's something like a pro code agent,

581
00:31:59,600 --> 00:32:00,600
you get Azure resources.

582
00:32:00,600 --> 00:32:01,600
They're always running.

583
00:32:01,600 --> 00:32:06,600
You're going to spend money implementing AI of any kind.

584
00:32:06,600 --> 00:32:10,600
And if your users don't ever find out that it's even there,

585
00:32:10,600 --> 00:32:12,600
then you've just wasted money.

586
00:32:12,600 --> 00:32:14,600
And I'm being by the time they realize it,

587
00:32:14,600 --> 00:32:16,600
you've already got to make updates

588
00:32:16,600 --> 00:32:19,600
because the large language models have changed or whatever.

589
00:32:19,600 --> 00:32:20,600
You've got to restart testing.

590
00:32:20,600 --> 00:32:26,600
You're going to have an ongoing cost there to even maintain that AI.

591
00:32:26,600 --> 00:32:29,600
And so, yeah, if the users don't even know if it's there,

592
00:32:29,600 --> 00:32:36,600
you're already failed, in my opinion.

593
00:32:36,600 --> 00:32:40,600
And I think a little bit where I have a little bit problem

594
00:32:40,600 --> 00:32:43,600
with is I think it was a really good explain.

595
00:32:43,600 --> 00:32:50,600
But we have so much tools now for me to really to build something like

596
00:32:50,600 --> 00:32:53,600
we want to work with AI in Microsoft.

597
00:32:53,600 --> 00:32:54,600
We have the co-pilot studio.

598
00:32:54,600 --> 00:32:56,600
We have the co-pilot agent builder.

599
00:32:56,600 --> 00:33:00,600
And also we have Azure AI Foundry when you,

600
00:33:00,600 --> 00:33:06,600
when should I use what?

601
00:33:06,600 --> 00:33:08,600
So, there is a stuffy, a really good question.

602
00:33:08,600 --> 00:33:12,600
I will always tell people start with this simplest thing.

603
00:33:12,600 --> 00:33:17,600
First thing, first, if you're not experienced enough building agents

604
00:33:17,600 --> 00:33:22,600
to know which tools don't even have the capabilities

605
00:33:22,600 --> 00:33:24,600
that your agent would need,

606
00:33:24,600 --> 00:33:26,600
always start with a SharePoint agent.

607
00:33:26,600 --> 00:33:28,600
See if you can get that to work.

608
00:33:28,600 --> 00:33:31,600
If you can't start graduating up,

609
00:33:31,600 --> 00:33:34,600
there are going to be some things you'll learn quickly,

610
00:33:34,600 --> 00:33:36,600
like if you need to integrate with systems,

611
00:33:36,600 --> 00:33:40,600
with external systems, then a lot of these tools will be immediately out.

612
00:33:40,600 --> 00:33:47,600
But the, you always want to go with the simplest technology

613
00:33:47,600 --> 00:33:51,600
that you can because if you start with Foundry

614
00:33:51,600 --> 00:33:53,600
on every single agent,

615
00:33:53,600 --> 00:33:56,600
you're going to have a tremendous amount of technical debt

616
00:33:56,600 --> 00:33:59,600
and it'll be much more expensive to maintain.

617
00:33:59,600 --> 00:34:05,600
You'll be leveraging less of your M365 co-pilot benefits as well

618
00:34:05,600 --> 00:34:10,600
because you don't use the M365 large language model

619
00:34:10,600 --> 00:34:12,600
if you're building in Foundry,

620
00:34:12,600 --> 00:34:14,600
you're holding models yourself.

621
00:34:14,600 --> 00:34:17,600
You're calling custom deployed models.

622
00:34:17,600 --> 00:34:23,600
So, like everything is much harder when you start inside Foundry

623
00:34:23,600 --> 00:34:26,600
and it's not necessary.

624
00:34:26,600 --> 00:34:31,600
There's a lot more decisions.

625
00:34:31,600 --> 00:34:34,600
There's a lot of technical decisions you would be making

626
00:34:34,600 --> 00:34:36,600
in something like Foundry.

627
00:34:36,600 --> 00:34:39,600
Like do you just leverage the model

628
00:34:39,600 --> 00:34:43,600
and build using something like the M365 Agents SDK

629
00:34:43,600 --> 00:34:47,600
or the Agents 365 SDK or the Teams SDK?

630
00:34:47,600 --> 00:34:52,600
There's three different ones that would result in deploying

631
00:34:52,600 --> 00:34:57,600
an application into Azure App Service

632
00:34:57,600 --> 00:35:00,600
and just leveraging a model in Foundry.

633
00:35:00,600 --> 00:35:02,600
Or there's the Foundry Agent Service,

634
00:35:02,600 --> 00:35:05,600
something I hear a lot more people talking about.

635
00:35:05,600 --> 00:35:09,600
And that has some simplistic advantages.

636
00:35:09,600 --> 00:35:15,600
Honestly, that's depending on if you're creating a simpler agent

637
00:35:15,600 --> 00:35:17,600
inside Foundry,

638
00:35:17,600 --> 00:35:20,600
it'll look and feel a lot more like Agent Builder actually.

639
00:35:20,600 --> 00:35:25,600
Or if you're building a Code First model or Code First Agent,

640
00:35:25,600 --> 00:35:30,600
then there's a lot more complexity to it there.

641
00:35:30,600 --> 00:35:35,600
And then you give up some crucial things that they can't integrate

642
00:35:35,600 --> 00:35:40,600
quite as well with an environment like Teams, for instance,

643
00:35:40,600 --> 00:35:46,600
compared to what more of a co-pilot specific toolset

644
00:35:46,600 --> 00:35:49,600
would let you do.

645
00:35:49,600 --> 00:35:51,600
Yeah, awesome.

646
00:35:51,600 --> 00:35:54,600
I think, what I think is also, I think...

647
00:35:54,600 --> 00:35:56,600
I don't know.

648
00:35:56,600 --> 00:36:01,600
There's a lot of these horror stories that the agent delete all,

649
00:36:01,600 --> 00:36:06,600
I don't know, a content from a project or something.

650
00:36:06,600 --> 00:36:12,600
How do you test an agent that you can feel?

651
00:36:12,600 --> 00:36:16,600
I don't know, that you are in a safe environment

652
00:36:16,600 --> 00:36:20,600
that's just not happening at something bad.

653
00:36:20,600 --> 00:36:25,600
I hear the same concern out of a lot of customers that I talk to.

654
00:36:25,600 --> 00:36:28,600
They hear all these same horror stories.

655
00:36:28,600 --> 00:36:31,600
An agent has just deleted all of their data.

656
00:36:31,600 --> 00:36:36,600
Or it's gone and done things that it's not supposed to do.

657
00:36:36,600 --> 00:36:40,600
And it usually comes down to...

658
00:36:40,600 --> 00:36:44,600
And the other thing is I hear them a lot more now

659
00:36:44,600 --> 00:36:48,600
than I have in maybe two years ago.

660
00:36:48,600 --> 00:36:51,600
And for one of those is there's an agent harness

661
00:36:51,600 --> 00:36:53,600
and then there's an agent.

662
00:36:53,600 --> 00:36:56,600
And when something kind of goes rogue

663
00:36:56,600 --> 00:36:58,600
and it's just deleting all these other things,

664
00:36:58,600 --> 00:37:01,600
a lot of times that's not a regular agent.

665
00:37:01,600 --> 00:37:04,600
That's something similar to OpenClaw

666
00:37:04,600 --> 00:37:09,600
where it's an environment that is in call one

667
00:37:09,600 --> 00:37:13,600
or can create on the fly one or more agents

668
00:37:13,600 --> 00:37:15,600
and start doing a lot of things.

669
00:37:15,600 --> 00:37:17,600
If you've heard of Clawed Code,

670
00:37:17,600 --> 00:37:19,600
that's more of an agent harness.

671
00:37:19,600 --> 00:37:22,600
Clawed Codework, opilot codework, OpenClaw.

672
00:37:22,600 --> 00:37:24,600
And a number of these different things

673
00:37:24,600 --> 00:37:27,600
that are designed to be incredibly powerful.

674
00:37:27,600 --> 00:37:31,600
Agents are not designed to do that

675
00:37:31,600 --> 00:37:34,600
because that's more of like an individual component

676
00:37:34,600 --> 00:37:37,600
versus an entire ecosystem or framework.

677
00:37:37,600 --> 00:37:41,600
But in general, you want to work

678
00:37:41,600 --> 00:37:44,600
on the least privilege principle.

679
00:37:44,600 --> 00:37:46,600
So whatever the agent needs,

680
00:37:46,600 --> 00:37:49,600
it should have the ability to do just that and nothing more.

681
00:37:49,600 --> 00:37:51,600
If it's going to be reading SharePoint files,

682
00:37:51,600 --> 00:37:53,600
there is no reason whatsoever.

683
00:37:53,600 --> 00:37:57,600
It should have right access to SharePoint files.

684
00:37:57,600 --> 00:38:00,600
Now, a lot of this you may not be able to control

685
00:38:00,600 --> 00:38:03,600
or you wouldn't previously

686
00:38:03,600 --> 00:38:05,600
because a lot of these agents,

687
00:38:05,600 --> 00:38:06,600
certainly in the Microsoft world,

688
00:38:06,600 --> 00:38:08,600
would operate based off of whatever privileges

689
00:38:08,600 --> 00:38:10,600
the end user had.

690
00:38:10,600 --> 00:38:13,600
So if the end user had right access to SharePoint,

691
00:38:13,600 --> 00:38:15,600
then so did the agent.

692
00:38:15,600 --> 00:38:17,600
But if they didn't, then it couldn't.

693
00:38:17,600 --> 00:38:20,600
If the user didn't have access to read HR files,

694
00:38:20,600 --> 00:38:22,600
and I protected library,

695
00:38:22,600 --> 00:38:25,600
well, the agent wouldn't have it even if the agent was told

696
00:38:25,600 --> 00:38:28,600
to use that library because it would still use the end user

697
00:38:28,600 --> 00:38:29,600
permission.

698
00:38:29,600 --> 00:38:34,600
Agents 365 kind of came along the same late last year as well

699
00:38:34,600 --> 00:38:38,600
and started to introduce more of that governance layer on top

700
00:38:38,600 --> 00:38:41,600
so that maybe even if the user did have access

701
00:38:41,600 --> 00:38:43,600
to write to something,

702
00:38:43,600 --> 00:38:45,600
it could be an additional layer on top

703
00:38:45,600 --> 00:38:49,600
that if the agent did have permission based on the user's

704
00:38:49,600 --> 00:38:51,600
permission to write to SharePoint,

705
00:38:51,600 --> 00:38:53,600
Agents 365 could just say,

706
00:38:53,600 --> 00:38:55,600
no, if you're not going to do that.

707
00:38:55,600 --> 00:39:01,600
So you definitely should be approaching agent development

708
00:39:01,600 --> 00:39:04,600
or building whatever term you want to use there.

709
00:39:04,600 --> 00:39:08,600
You should be approaching that with the same sort of mindset

710
00:39:08,600 --> 00:39:11,600
that you would give to a user that you're creating in the environment.

711
00:39:11,600 --> 00:39:13,600
Do they need right access to this thing?

712
00:39:13,600 --> 00:39:15,600
If they don't, then they shouldn't have it.

713
00:39:15,600 --> 00:39:18,600
Give them just the permissions that they need to do.

714
00:39:18,600 --> 00:39:21,600
They need to do their job and nothing more.

715
00:39:21,600 --> 00:39:25,600
It's like zero trusts.

716
00:39:25,600 --> 00:39:27,600
Exactly, zero trust.

717
00:39:27,600 --> 00:39:31,600
That's another thing that I hear a lot about when I talk to some

718
00:39:31,600 --> 00:39:33,600
of the security experts that I work with

719
00:39:33,600 --> 00:39:38,600
or have certainly AI discussions in a security context.

720
00:39:38,600 --> 00:39:40,600
That's always the thing.

721
00:39:40,600 --> 00:39:45,600
If you're applying this kind of thinking and process to your users,

722
00:39:45,600 --> 00:39:48,600
you should be doing the same thing with agents as well.

723
00:39:48,600 --> 00:39:52,600
And if you're not a developer, a developer is building out an agent.

724
00:39:52,600 --> 00:39:55,600
And they're saying, I need these permissions.

725
00:39:55,600 --> 00:39:58,600
You should be questioning that and you should be making sure

726
00:39:58,600 --> 00:40:04,600
of what permissions you're allowing that agent to get

727
00:40:04,600 --> 00:40:07,600
from Microsoft Graph.

728
00:40:07,600 --> 00:40:10,600
Because that's really where you're going to be stopping

729
00:40:10,600 --> 00:40:13,600
that the over-promissioning and over-sharing.

730
00:40:13,600 --> 00:40:16,600
All of a sudden, those horror stories aren't happening to you

731
00:40:16,600 --> 00:40:19,600
because you've taken the proper precautions.

732
00:40:19,600 --> 00:40:25,600
Yeah, and I think also it's like a company thinking.

733
00:40:25,600 --> 00:40:30,600
I think they need an uncle Ben who advised Peter Parker

734
00:40:30,600 --> 00:40:35,600
was great, Paul Warkand's great responsibility.

735
00:40:35,600 --> 00:40:37,600
Exactly.

736
00:40:37,600 --> 00:40:39,600
Yeah.

737
00:40:39,600 --> 00:40:43,600
A little bit of what I hear of the audience,

738
00:40:43,600 --> 00:40:46,600
governance security, there's a bunch of AI topics.

739
00:40:46,600 --> 00:40:52,600
And what security concerns come up most often

740
00:40:52,600 --> 00:40:56,600
from your experience?

741
00:40:56,600 --> 00:41:00,600
Usually it is that horror story that, hey, I heard agents can just go,

742
00:41:00,600 --> 00:41:04,600
"Hey, no, do all these things that it's not supposed to do like it?"

743
00:41:04,600 --> 00:41:07,600
You know, if we granted access to read SharePoint all of a sudden

744
00:41:07,600 --> 00:41:11,600
that can write to SharePoint, those are usually the misconceptions

745
00:41:11,600 --> 00:41:16,600
I'm addressing almost on a daily basis, to be honest.

746
00:41:16,600 --> 00:41:22,600
So just, and it's almost always with a customer who is either asking

747
00:41:22,600 --> 00:41:25,600
about how to leverage AI, or they're starting to,

748
00:41:25,600 --> 00:41:28,600
and they're starting to hear some of these stories.

749
00:41:28,600 --> 00:41:33,600
So usually clearing up misconceptions is the number one thing

750
00:41:33,600 --> 00:41:40,600
that I hear about and I talk about, just to help someone understand

751
00:41:40,600 --> 00:41:46,600
like the backstory that they're not hearing with that horror story.

752
00:41:46,600 --> 00:41:51,600
Because these things are incredibly safe if you are designing them

753
00:41:51,600 --> 00:41:55,600
honestly with common sense.

754
00:41:55,600 --> 00:42:02,600
I think that the agent, I think from the user perspective,

755
00:42:02,600 --> 00:42:08,600
does the agent only know what the user already has access to?

756
00:42:08,600 --> 00:42:16,600
Or is it, is it, is it, item, or can I grant other X-access

757
00:42:16,600 --> 00:42:23,600
to an agent as me as user who do you develop this?

758
00:42:23,600 --> 00:42:26,600
It really depends on your use case.

759
00:42:26,600 --> 00:42:29,600
In some cases, in a lot of cases I would say,

760
00:42:29,600 --> 00:42:33,600
you do want the agent to use whatever the user has permission to.

761
00:42:33,600 --> 00:42:37,600
Maybe it's more of a productivity assistant.

762
00:42:37,600 --> 00:42:41,600
Maybe it's something that's going to help them do their job.

763
00:42:41,600 --> 00:42:45,600
And so it only needs the same permission as the user.

764
00:42:45,600 --> 00:42:51,600
In other scenarios, you may want the agent to return information

765
00:42:51,600 --> 00:42:56,600
from a resource the user does not have access to.

766
00:42:56,600 --> 00:43:02,600
I hear about all this, but this a lot when if you're pulling information

767
00:43:02,600 --> 00:43:08,600
from Azure SQL, well, a lot of companies don't want to grant end users access to Azure SQL.

768
00:43:08,600 --> 00:43:14,600
So they want the agent to kind of be the demarc point where the agent has access

769
00:43:14,600 --> 00:43:18,600
to return SQL information to the user like on the user's behalf.

770
00:43:18,600 --> 00:43:21,600
So that could be a scenario.

771
00:43:21,600 --> 00:43:24,600
There's other scenarios that we're starting to see now

772
00:43:24,600 --> 00:43:29,600
with things like Microsoft Scout where the agent may have its own level permission.

773
00:43:29,600 --> 00:43:38,600
This is getting a lot more into the AIT make scenario where the agent has its own set of resources.

774
00:43:38,600 --> 00:43:44,600
The user has their own set of resources and permissions and they're just working with each other.

775
00:43:44,600 --> 00:43:46,600
So there's a lot of those different cases.

776
00:43:46,600 --> 00:43:51,600
It's always going to boil down to what is the use case for the agent?

777
00:43:51,600 --> 00:43:54,600
And what is the security model that makes sense?

778
00:43:54,600 --> 00:44:01,600
Should it be only ever using the user's permissions or does the agent need its own set of permissions?

779
00:44:01,600 --> 00:44:06,600
And with all these different scenarios, things just like a basic chat agent

780
00:44:06,600 --> 00:44:10,600
or an automation type agent where you're integrated with systems

781
00:44:10,600 --> 00:44:16,600
or the fully autonomous agent that is an AIT make sort of scenarios.

782
00:44:16,600 --> 00:44:21,600
Those will all determine the security model that you want to adopt for this.

783
00:44:21,600 --> 00:44:29,600
And for each of the use cases, there's usually just one right answer between those different security models

784
00:44:29,600 --> 00:44:32,600
which is the right one for you.

785
00:44:32,600 --> 00:44:34,600
Yeah.

786
00:44:34,600 --> 00:44:41,600
You know, I said also something interesting where I have say all these three tools.

787
00:44:41,600 --> 00:44:51,600
Yeah, what is the, for me, what is the different from the agent builder to the scouts?

788
00:44:51,600 --> 00:44:56,600
So there's one architectural difference.

789
00:44:56,600 --> 00:45:00,600
Agent builder and SharePoint agents.

790
00:45:00,600 --> 00:45:08,600
Actually half of co-pilot studio, all three will fall into what Microsoft calls a declarative agent.

791
00:45:08,600 --> 00:45:16,600
Where you don't have to write any code, you don't have to deal with any of the things like what model to use or what orchestrator to use.

792
00:45:16,600 --> 00:45:19,600
It's kind of the agent in easy mode.

793
00:45:19,600 --> 00:45:21,600
They hide all the complex stuff from you.

794
00:45:21,600 --> 00:45:25,600
It runs on top of the M365 co-pilot.

795
00:45:25,600 --> 00:45:30,600
And generally will at that point now require a co-pilot license.

796
00:45:30,600 --> 00:45:32,600
Those are kind of the easy ones.

797
00:45:32,600 --> 00:45:36,600
Then you look at something like scout.

798
00:45:36,600 --> 00:45:43,600
Well, before I get to scout, the other type of agent that we've talked about for years now is a custom engine agent.

799
00:45:43,600 --> 00:45:50,600
That is what you're creating when you're creating with boundary using the Microsoft 365 agents toolkit, which is my preference.

800
00:45:50,600 --> 00:45:52,600
You're creating a custom engine agent.

801
00:45:52,600 --> 00:45:56,600
That's when you're defining what model to use, what orchestrator you want to use.

802
00:45:56,600 --> 00:46:04,600
And that's when you hear words like the agent framework or Lang chain or Lang graph, semantic kernel.

803
00:46:04,600 --> 00:46:10,600
You hear about all those different types of things, just different orchestrators, kind of the brains behind the scenes almost.

804
00:46:10,600 --> 00:46:13,600
And that's when you start to get into more the autonomous agents.

805
00:46:13,600 --> 00:46:20,600
Now scout is an autonomous agent, but they call it an auto pilot.

806
00:46:20,600 --> 00:46:22,600
It's one of their newest things they had announced.

807
00:46:22,600 --> 00:46:27,600
I believe that build this year, the 2026 build conference.

808
00:46:27,600 --> 00:46:35,600
And that one is from what I have seen is I haven't got my hands on it.

809
00:46:35,600 --> 00:46:42,600
I don't know internally how it works, but I do know it uses the GitHub co pilot SDK.

810
00:46:42,600 --> 00:46:45,600
I believe this would fall under an agent harness.

811
00:46:45,600 --> 00:46:47,600
I'm of architecture.

812
00:46:47,600 --> 00:46:56,600
So this is more similar to open claw that it is to agent builder because this is going to be able to do to do work on its own.

813
00:46:56,600 --> 00:47:07,600
It will usually have actually I'm not sure the on the permission model on that I would it seems like it's going to be user level permission.

814
00:47:07,600 --> 00:47:11,600
So whatever permissions I have access to it would have access to.

815
00:47:11,600 --> 00:47:18,600
But it's more of an agent harness mixed with an auto an autonomous agent.

816
00:47:18,600 --> 00:47:47,600
So this is really the newest class of things and I'm starting to see a lot of cases where a custom agent may not be the right answer anymore because we have some of these harnesses like co pilot co work or scout or I would what would come right after scout up the thing I'm personally waiting for a lot right now is building custom autopilot because I think you're going to be seeing a lot more agents that are using marketing.

817
00:47:47,600 --> 00:47:57,600
And not code you'll be able to have the same power as a custom engine agent where those things can do whatever it is you want to do.

818
00:47:57,600 --> 00:48:02,600
But of course they'll only do what you program it to be able to do.

819
00:48:02,600 --> 00:48:16,600
But you'll be able to do that without being a Python developer or a type script developer or a C sharp developer you'll be using markdown because that's why things like open claw have gotten so powerful and popular why.

820
00:48:16,600 --> 00:48:39,600
People love claw code they love both you know all the co work type products using natural language and using markdown files and that is it and even a beginner user to AI could quickly understand how to work with something like that because they don't even have to know how to build these skills.

821
00:48:39,600 --> 00:48:47,600
That's one of the newest buzzwords now is the skills and topic had them now copo let's got them across a number of different experiences.

822
00:48:47,600 --> 00:49:04,600
But you can just talk with a ion it will build the skill for you though the skill floor floor to get involved with some of these products has gotten so low now and the in my mind the bar.

823
00:49:04,600 --> 00:49:14,600
The you have to the threshold you have to cross to actually be able to say yes we actually need a pro code agent for this it's got a lot higher it's gotten.

824
00:49:14,600 --> 00:49:32,600
I see a lot more cases where you're going to be building auto pilots you're going to be building skills inside co work and it's going to save a lot of money on development and being much easier to maintain with a less technical skill set.

825
00:49:32,600 --> 00:49:53,600
And I work on this was go good explanation and I can come back to the governance and security topic and as an yeah I say an agent builder or maker and how many experience need I in additional tools like entrap you view.

826
00:49:53,600 --> 00:49:58,600
What we have also defender and so on.

827
00:49:58,600 --> 00:50:09,600
If you're an agent builder and agent maker and you're not a pro code developer you shouldn't be worried about any of those things that should be IT's responsibility.

828
00:50:09,600 --> 00:50:27,600
As if you're in the no code or low code experience you your agent will always be subject to those types of things now if you're pro developer you're going to have a lot more concerns because you will depending on how you're architecting your agent depending on the way you're.

829
00:50:27,600 --> 00:50:56,600
Creating knowledge sources maybe your indexing content from SharePoint into a custom Azure AI search index all of a sudden you could end up stripping out all the permissions and and storing just you know the raw knowledge in there the raw data and then all of a sudden your agent is returning information that it shouldn't be returning to a particular user but you made the mistake in architecting your knowledge source in a way that it exposed that security issue.

830
00:50:56,600 --> 00:51:13,600
Same with same with part of you labeling there's a lot of things that you could really screw up if you are if you're working in the foundry world without the proper skill set without knowledge of things like

831
00:51:13,600 --> 00:51:24,600
or you data labeling and information protection defender defender as well we're seeing a lot of these capabilities.

832
00:51:24,600 --> 00:51:42,600
In defender and draw and purview all get beefed up specifically for agents that was something that agent 365 really relies on I'd say the core of the agent 365 functionality is industry products and the upgrades that they're getting things like conditional access policies that are

833
00:51:42,600 --> 00:52:02,600
specific to ages and without at least awareness of what policies exist and what policies are about to get rolled out your agent can go from working to broken in in minutes depending on when things change in that environment that your agent runs in.

834
00:52:02,600 --> 00:52:22,600
Yeah and there are two topics I think they're also related to the security topic but also a little bit future topics but what do you think we are moving towards our an autonomous agent area.

835
00:52:22,600 --> 00:52:46,600
I think they'll always be a mixture of simple agents come or in complex agents and the nation harness is because again it'll always depend on the use case if you don't need an autonomous agent you don't then you shouldn't be trying to force that type of a solution if a share point agent will work if that does what you need to you should need me thinking about

836
00:52:46,600 --> 00:53:15,600
copilot studio and much less boundary you should go with the simplest technology it's going to be the easiest to maintain certainly in the case of a share point agent a side owner could be maintaining that and not IT so you always want to go with that the simplest type of technology and not trying over engineer things so I will always have simple a use cases as well as complex and is and so I think that all the different types of agents we have now they're going to continue

837
00:53:15,600 --> 00:53:44,600
we will be seeing more of out type things but it's there they'll be there for certain use cases and not others scout certainly seems like a very general purpose type of solution so going back to the thing I said at the beginning of this episode if it's sort of good at everything it's never going to be great at any one of those things now skills can

838
00:53:44,600 --> 00:54:13,600
kind of maybe blur that line or that rule a little bit but you'll always be use cases where you need AI to do one thing and do it exceptionally well and for that you can't really use a general purpose type agent now harnesses again little little different but you'll always have a need for agents because you want something to do only one thing and nothing else

839
00:54:13,600 --> 00:54:39,600
okay and there it's I think a lot of you would your videos there they say okay I built an entry company with multi agent systems what role will multi agent systems play I see that playing a role with very complex processes where you

840
00:54:39,600 --> 00:55:08,600
what I usually see people doing is they break down a complex process into individual components I think most people are used to doing that just to understand what this process entails each of these simpler components would be an agent because again you want that agent to do one thing and do it incredibly well if it needs to do something sort of unrelated it should be calling a different agent to do that one thing so now you start to build out a team of agents that are working together

841
00:55:08,600 --> 00:55:29,600
each one is an expert in its own job and nothing more and that's really what those multi agent or architectures are about it maybe you could get one agent to do all of it and wouldn't do it that well that's first that's that's for sure because they do need to be very very focused

842
00:55:29,600 --> 00:55:56,600
other way you end up with a general purpose agent or is general purpose solution like in three 65 copilot so I see those being used a lot more for in a system integrations back end processing more more back end work than other things so usually a lot of those highly complex systems especially ones that are touching third party systems or

843
00:55:56,600 --> 00:56:09,600
or back in system different Azure resources and things like that those are usually the best use cases for a multi agent architecture.

844
00:56:09,600 --> 00:56:27,600
Okay yeah we're a little bit running on a time so let's jump in the rabbit fire around I say something and you give a quick answer so one microphone soft feature you yeah or everybody should alone.

845
00:56:27,600 --> 00:56:47,600
One Microsoft feature everyone should be using I definitely say m365 copilot period like that that opens the door to everything else is there one underrated core pilot capability.

846
00:56:47,600 --> 00:57:00,600
Well I would say let me rephrase that copilot pages which is a mixture of copilot and loop.

847
00:57:00,600 --> 00:57:16,600
Okay what's the most exciting announcement from Microsoft recently autopilot by far autopilot is it opens up a brand new world for agent development and the

848
00:57:16,600 --> 00:57:23,600
level of the level of work compared to the level of effort to get that work.

849
00:57:23,600 --> 00:57:30,600
One AI with you will love to eliminate.

850
00:57:30,600 --> 00:57:32,600
Could you repeat that?

851
00:57:32,600 --> 00:57:37,600
Yeah one AI with you love to eliminate.

852
00:57:37,600 --> 00:57:45,600
That AI can just do anything it wants to regardless of what you allow it to do.

853
00:57:45,600 --> 00:57:48,600
Visual studio code or visual studio.

854
00:57:48,600 --> 00:57:51,600
VSCode all day.

855
00:57:51,600 --> 00:57:58,600
SharePoint all day.

856
00:57:58,600 --> 00:58:04,600
Best drink during the develop the copilot agent.

857
00:58:04,600 --> 00:58:13,600
If it's off the clock it's a bloody Mary if it's on the clock it's coffee.

858
00:58:13,600 --> 00:58:18,600
What's the best thing to be on the Microsoft MVP program.

859
00:58:18,600 --> 00:58:30,600
The best thing I would say being plugged in with all the individual product teams especially the ones that matter to me the most copilot and copilot extensibility.

860
00:58:30,600 --> 00:58:42,600
I'll be able to give that direct feedback and know all of these super duper top secret features that are about to come out so that I can start getting my head around it before customers here about it.

861
00:58:42,600 --> 00:58:50,600
And if you were an IT what carry of will you will you prefer.

862
00:58:50,600 --> 00:58:53,600
I think it'd be fun to be a doctor.

863
00:58:53,600 --> 00:58:58,600
Maybe I'm just watching too many medical shows though.

864
00:58:58,600 --> 00:59:06,600
And what was the best advice you ever received.

865
00:59:06,600 --> 00:59:11,600
To work my butt off and get whatever it is I want.

866
00:59:11,600 --> 00:59:15,600
Yeah that's what I've done for over 20 years now.

867
00:59:15,600 --> 00:59:26,600
And I've certainly applied a lot more in the past few years and all the sudden the BK or Microsoft MVP and there's a lot of things happening for me right now and I have no plans on stopping.

868
00:59:26,600 --> 00:59:29,600
Just grind I'm a grinder.

869
00:59:29,600 --> 00:59:44,600
Yeah awesome thank you so then my yeah my closing question is what should the listeners take away from from this talk when it's one thing.

870
00:59:44,600 --> 00:59:56,600
The one thing is that agents cannot just go rogue if you have the proper guide guard rails in place the proper security and your data protected.

871
00:59:56,600 --> 01:00:04,600
Understand but but making sure that whenever you do with AI you always think about security and governance first.

872
01:00:04,600 --> 01:00:15,600
Yeah awesome yeah then thank you Steve for joining me today it was a fantastic exploring Microsoft 365 Copanet agents agents 65 and the future of enterprise AI with you.

873
01:00:15,600 --> 01:00:33,600
Yeah for everyone listening be sure connect the Steve you find his links in the show notes also check out this YouTube channel and also he is on the host on the MC65 show that's also a good source.

874
01:00:33,600 --> 01:00:51,600
Yeah I think if you're interested in his work you find all all the links and yeah if you enjoy today's conversation don't forget to subscribe to the MC65 podcast and yeah leave us a review and share this episode with someone explorate Microsoft Copanet.

875
01:00:51,600 --> 01:00:59,600
You'll see you next time keep learning keep building and we'll see you in the next episode on the MC65 podcast so thank you Steve for being here with me.

876
01:00:59,600 --> 01:01:05,600
My pleasure thanks for inviting me it's always good to talk with you and we'll talk again soon.

877
01:01:05,600 --> 01:01:09,600
Goodbye. Ciao. Thanks everyone.

878
01:01:09,600 --> 01:01:11,460
you