AI Agents - Simply Explained

1
00:00:00,000 --> 00:00:03,200
Welcome to another episode of Microsoft Knowledge Nuggets on M365.

2
00:00:03,200 --> 00:00:04,800
FM, I'm your host, Mirko Peters.

3
00:00:04,800 --> 00:00:08,000
Today's topic is one you've probably heard everywhere, AI agents.

4
00:00:08,000 --> 00:00:09,200
But what are they really?

5
00:00:09,200 --> 00:00:12,700
Most people will tell you it's a smarter chatbot and that is actually the biggest

6
00:00:12,700 --> 00:00:13,800
misunderstanding right now.

7
00:00:13,800 --> 00:00:16,900
By the end of this episode, you'll understand what an agent really is,

8
00:00:16,900 --> 00:00:20,600
how it's different from a chatbot and the three ways you can build one inside

9
00:00:20,600 --> 00:00:21,900
Microsoft's ecosystem.

10
00:00:21,900 --> 00:00:23,900
Here's the simplest way to think about it.

11
00:00:23,900 --> 00:00:26,900
A chatbot is like a receptionist at a front desk.

12
00:00:26,900 --> 00:00:29,600
You ask a question, they answer, and that's it.

13
00:00:29,800 --> 00:00:32,700
An agent is more like a personal assistant who takes initiative.

14
00:00:32,700 --> 00:00:36,200
You give them a goal like book that meeting with the supplier and they figure

15
00:00:36,200 --> 00:00:37,300
out the steps on their own.

16
00:00:37,300 --> 00:00:41,100
They check your calendar, send emails, update files and come back to you when

17
00:00:41,100 --> 00:00:41,600
it's done.

18
00:00:41,600 --> 00:00:45,000
One answers questions, the other does work and that distinction matters because

19
00:00:45,000 --> 00:00:47,400
it's the foundation for everything we're about to cover.

20
00:00:47,400 --> 00:00:50,600
Chatbots versus agents, the core difference.

21
00:00:50,600 --> 00:00:53,500
Let's break down what actually separates a chatbot from an agent.

22
00:00:53,500 --> 00:00:55,000
A chatbot is reactive.

23
00:00:55,000 --> 00:00:57,300
Someone types a question and the chatbot responds.

24
00:00:57,600 --> 00:01:01,500
It uses short term memory, meaning it only knows what you've said in this conversation.

25
00:01:01,500 --> 00:01:05,600
Once the chat ends, that context is gone and the chatbot is limited to

26
00:01:05,600 --> 00:01:06,300
conversations.

27
00:01:06,300 --> 00:01:08,100
They can't do anything outside the chat window.

28
00:01:08,100 --> 00:01:11,300
An agent is different because it's proactive instead of waiting for you to

29
00:01:11,300 --> 00:01:14,700
ask something, it can take action on its own, like sending a reminder or

30
00:01:14,700 --> 00:01:15,700
updating a status.

31
00:01:15,700 --> 00:01:19,700
It can access other systems like your email, your calendar and your company's

32
00:01:19,700 --> 00:01:20,300
sharepoint.

33
00:01:20,300 --> 00:01:24,000
It even has its own identity inside Microsoft Entra ID, which means it can have

34
00:01:24,000 --> 00:01:26,400
its own mailbox, its own one drive and its own team's account.

35
00:01:26,800 --> 00:01:28,000
And it works autonomously.

36
00:01:28,000 --> 00:01:31,000
You give it a task and it goes off and does it without needing you to check in.

37
00:01:31,000 --> 00:01:31,800
Think of it this way.

38
00:01:31,800 --> 00:01:35,300
A chatbot is like a help desk employee with a printed FAQ.

39
00:01:35,300 --> 00:01:36,400
You ask a question.

40
00:01:36,400 --> 00:01:38,400
They flip to the right page and read you the answer.

41
00:01:38,400 --> 00:01:42,300
An agent is like a project manager who can access files, send emails, schedule

42
00:01:42,300 --> 00:01:46,000
meetings, follow up on tasks and report back without you guiding every step.

43
00:01:46,000 --> 00:01:47,400
Here's the key distinction.

44
00:01:47,400 --> 00:01:51,300
A chatbot lives inside the conversation while an agent can trigger workflows

45
00:01:51,300 --> 00:01:51,900
in the background.

46
00:01:51,900 --> 00:01:55,500
It can modify data in a database, create a purchase order in sharepoint

47
00:01:55,600 --> 00:01:58,000
while you're in a meeting and you don't have to watch it work.

48
00:01:58,000 --> 00:02:01,300
It just does the job behind the scenes and lets you know when it's done.

49
00:02:01,300 --> 00:02:02,900
So now you know what an agent is.

50
00:02:02,900 --> 00:02:04,500
But how do you actually build one?

51
00:02:04,500 --> 00:02:08,000
Microsoft gives you three parts and which one you choose depends on your skill

52
00:02:08,000 --> 00:02:09,400
level and what you need it to do.

53
00:02:09,400 --> 00:02:10,800
We'll cover those in just a moment.

54
00:02:10,800 --> 00:02:13,500
The three build paths and overview.

55
00:02:13,500 --> 00:02:17,300
Microsoft gives you three ways to build agents and each one matches a different

56
00:02:17,300 --> 00:02:18,000
skill level.

57
00:02:18,000 --> 00:02:20,800
The first is agent builder, which is the no code option.

58
00:02:20,800 --> 00:02:23,700
It lives right inside Microsoft 365 co-pilot.

59
00:02:23,700 --> 00:02:27,200
So anyone can create an agent just by describing what they want in plain English.

60
00:02:27,200 --> 00:02:29,400
You never have to write a single line of code.

61
00:02:29,400 --> 00:02:31,300
The second option is co-pilot studio.

62
00:02:31,300 --> 00:02:33,600
That's a separate web app you find at co-pilot studio.

63
00:02:33,600 --> 00:02:34,900
Microsoft.com.

64
00:02:34,900 --> 00:02:38,200
It's low code, meaning you still describe your agent using natural language,

65
00:02:38,200 --> 00:02:40,600
but now you can add, connect us to other services,

66
00:02:40,600 --> 00:02:44,000
define custom conversation flows and setup automated triggers.

67
00:02:44,000 --> 00:02:46,500
It gives you more power, but it does take a bit more setup.

68
00:02:46,500 --> 00:02:49,300
Then there's Foundry, which is Azure AI Foundry,

69
00:02:49,300 --> 00:02:51,600
Microsoft's professional platform for developers.

70
00:02:51,600 --> 00:02:55,600
Here you get full control over the AI model, the orchestration and the infrastructure.

71
00:02:55,600 --> 00:02:59,100
You can choose from hundreds of models like open AI, Claude or Lama.

72
00:02:59,100 --> 00:03:01,000
You can build multi agent systems.

73
00:03:01,000 --> 00:03:04,500
This one is for pro code teams that need enterprise scale production agents.

74
00:03:04,500 --> 00:03:06,400
Now, here's something a lot of people don't realize.

75
00:03:06,400 --> 00:03:08,600
The lines between these three are blurry.

76
00:03:08,600 --> 00:03:10,700
You can start building an agent in agent builder,

77
00:03:10,700 --> 00:03:12,400
then exported to co-pilot studio.

78
00:03:12,400 --> 00:03:16,000
If you need more power, you can start in co-pilot studio and move to Foundry later.

79
00:03:16,000 --> 00:03:17,200
They're not separate islands.

80
00:03:17,200 --> 00:03:20,500
They're more like a spectrum and Microsoft designed them to work together.

81
00:03:21,200 --> 00:03:23,300
Let's start with the easiest path, agent builder.

82
00:03:23,300 --> 00:03:27,600
If you already have a Microsoft 365 co-pilot license, you have access to this tool.

83
00:03:27,600 --> 00:03:30,400
Agent builder, no code agents for everyone.

84
00:03:30,400 --> 00:03:32,200
So what does agent builder actually look like?

85
00:03:32,200 --> 00:03:34,400
It's a simple form inside Microsoft 365 co-pilot.

86
00:03:34,400 --> 00:03:37,200
You open co-pilot, click the agent section on the left and hit create.

87
00:03:37,200 --> 00:03:39,900
No separate website, no setup, no configuration needed.

88
00:03:39,900 --> 00:03:41,500
You start by describing what you want.

89
00:03:41,500 --> 00:03:46,500
Type something like create an agent that answers questions about our company's HR policies.

90
00:03:46,500 --> 00:03:50,600
Co-pilot takes that description and auto generates a name, a set of instructions,

91
00:03:50,600 --> 00:03:53,200
and even suggested prompts for users to start with.

92
00:03:53,200 --> 00:03:55,200
You can refine those instructions if you want.

93
00:03:55,200 --> 00:04:01,200
Add rules like "Only use the sources I provide" or "If you don't know the answer, direct the user to HR".

94
00:04:01,200 --> 00:04:04,700
Then you give it knowledge sources and this is where the real power is.

95
00:04:04,700 --> 00:04:10,200
You can point it to files on your one drive, sharepoint sites, teams chats and channels, specific websites,

96
00:04:10,200 --> 00:04:11,800
or even your company's org chart.

97
00:04:11,800 --> 00:04:15,800
There's a critical setting here too, a toggle that says "Only use specified sources".

98
00:04:15,800 --> 00:04:19,500
Turn that on and the agent can't pull from the general knowledge of the AI model.

99
00:04:19,500 --> 00:04:22,400
It can only answer from the documents and sites you gave it.

100
00:04:22,400 --> 00:04:24,100
That's how you prevent hallucinations.

101
00:04:24,100 --> 00:04:27,600
The agent literally cannot make things up because it's locked to your data.

102
00:04:27,600 --> 00:04:29,600
You can also give the agent capabilities.

103
00:04:29,600 --> 00:04:32,300
Wanted to create word documents, flip the toggle.

104
00:04:32,300 --> 00:04:33,600
Wanted to generate images?

105
00:04:33,600 --> 00:04:35,800
That's a toggle too. Wanted to write code snippets?

106
00:04:35,800 --> 00:04:38,900
Same thing. These are optional, but they make the agent much more useful.

107
00:04:38,900 --> 00:04:41,300
Once you're done, you can share the agent with teammates.

108
00:04:41,300 --> 00:04:42,800
You choose who gets access.

109
00:04:42,800 --> 00:04:48,400
Specific people or everyone in your organization, they'll find it in their own co-pilot sidebar, ready to use.

110
00:04:48,400 --> 00:04:53,000
Here's a concrete example. Say you work in HR and you have a 50-page employee manual.

111
00:04:53,000 --> 00:04:55,300
You create an agent called HR Helper.

112
00:04:55,300 --> 00:04:57,300
You upload the manual as a knowledge source.

113
00:04:57,300 --> 00:04:59,800
You turn on "Only use specified sources".

114
00:04:59,800 --> 00:05:02,600
You write instructions saying "answer questions politely"

115
00:05:02,600 --> 00:05:05,300
and always cite the section of the manual you're referencing.

116
00:05:05,300 --> 00:05:10,900
Now anyone in your company can ask HR Helper about vacation policies, sick leave or dress code,

117
00:05:10,900 --> 00:05:14,700
and get accurate answers without calling HR every time.

118
00:05:14,700 --> 00:05:18,500
But agent builder does have limits. It cannot trigger actions automatically.

119
00:05:18,500 --> 00:05:22,100
It cannot connect to external systems like Salesforce or your accounting software.

120
00:05:22,100 --> 00:05:25,700
It cannot publish to a public website. It's designed for simple Q&A agents

121
00:05:25,700 --> 00:05:27,800
that sit inside your company's co-pilot.

122
00:05:27,800 --> 00:05:29,800
That works great for basic use cases.

123
00:05:29,800 --> 00:05:33,800
But what if you need your agent to take action automatically like when a customer email comes in?

124
00:05:33,800 --> 00:05:35,600
That's where co-pilot studio comes in.

125
00:05:35,600 --> 00:05:36,600
Co-pilot studio.

126
00:05:36,600 --> 00:05:38,600
Low-code power for business processes.

127
00:05:38,600 --> 00:05:40,300
So where do you find co-pilot studio?

128
00:05:40,300 --> 00:05:42,600
It lives at co-pilot studio, Microsoft.com

129
00:05:42,600 --> 00:05:46,000
and this is a full agent building platform, not just a simple form.

130
00:05:46,000 --> 00:05:50,800
If you already have a Microsoft 365 co-pilot license, you probably already have access.

131
00:05:50,800 --> 00:05:53,200
If not, there's a pay as you go option available.

132
00:05:53,200 --> 00:05:55,800
You still start by describing your agent in natural language.

133
00:05:55,800 --> 00:05:57,000
That part is the same.

134
00:05:57,000 --> 00:05:59,000
But now you can add automated triggers.

135
00:05:59,000 --> 00:06:01,300
For example, you can set up the agent to activate

136
00:06:01,300 --> 00:06:03,500
when a new item is created in SharePoint

137
00:06:03,500 --> 00:06:06,500
or when an email arrives in a specific mailbox.

138
00:06:06,500 --> 00:06:08,900
The agent doesn't wait for someone to ask it something.

139
00:06:08,900 --> 00:06:11,500
It starts working on its own when an event happens.

140
00:06:11,500 --> 00:06:13,400
So connectors are another big upgrade.

141
00:06:13,400 --> 00:06:20,900
Co-pilot studio has over 1,000 pre-built connectors to services like Salesforce, Dynamics 365, Excel, Outlook, Teams and more.

142
00:06:20,900 --> 00:06:22,900
Your agent can read data from Salesforce,

143
00:06:22,900 --> 00:06:28,300
create records in Dynamics, send emails through Outlook and post updates and Teams all from one conversation.

144
00:06:28,300 --> 00:06:29,300
That's pretty powerful.

145
00:06:29,300 --> 00:06:30,700
Knowledge sources are richer too.

146
00:06:30,700 --> 00:06:33,800
You can connect to Dataverse, that's Microsoft's data platform,

147
00:06:33,800 --> 00:06:36,700
plus SharePoint, Public Websites and uploaded files.

148
00:06:36,700 --> 00:06:38,600
The agent has more context to work with,

149
00:06:38,600 --> 00:06:40,500
so its answers are more accurate and more useful.

150
00:06:40,500 --> 00:06:41,900
Here's where things get interesting.

151
00:06:41,900 --> 00:06:42,900
You can define topics.

152
00:06:42,900 --> 00:06:45,900
These are custom conversation flows for specific scenarios.

153
00:06:45,900 --> 00:06:47,900
Say a user reports a workplace conflict.

154
00:06:47,900 --> 00:06:49,400
You can create a topic that says,

155
00:06:49,400 --> 00:06:52,500
"If the user mentions a conflict, don't try to resolve it.

156
00:06:52,500 --> 00:06:54,500
Instead, collect their name and department,

157
00:06:54,500 --> 00:06:56,900
then escalate to the HR director with a summary."

158
00:06:56,900 --> 00:07:00,000
The agent follows that flow every single time.

159
00:07:00,000 --> 00:07:03,700
It's like giving the agent a playbook for each situation before you publish.

160
00:07:03,700 --> 00:07:05,700
You can test the agent in a preview pane.

161
00:07:05,700 --> 00:07:09,200
Ask it questions, see how it responds, check the citations.

162
00:07:09,200 --> 00:07:11,300
When you're happy, you can publish the Teams,

163
00:07:11,300 --> 00:07:15,600
Microsoft 365, co-pilot, or even a demo website for external users.

164
00:07:15,600 --> 00:07:16,900
One thing to know about pricing.

165
00:07:16,900 --> 00:07:18,900
Co-pilot studio requires consumption credits

166
00:07:18,900 --> 00:07:21,200
if external users interact with your agent

167
00:07:21,200 --> 00:07:23,000
or if the agent takes autonomous actions.

168
00:07:23,000 --> 00:07:25,900
If you're the only one using it or you share it with colleagues

169
00:07:25,900 --> 00:07:28,700
who also have co-pilot licenses, credits aren't needed.

170
00:07:28,700 --> 00:07:30,900
But if you build an agent for customers or partners,

171
00:07:30,900 --> 00:07:33,200
your organization pays per interaction.

172
00:07:33,200 --> 00:07:36,000
Co-pilot studio handles most business scenarios,

173
00:07:36,000 --> 00:07:38,000
but for developers who need full control.

174
00:07:38,000 --> 00:07:41,700
Over the AI model, the orchestration and the infrastructure, there's Foundry.

175
00:07:41,700 --> 00:07:43,800
Foundry Procode Custom Agents.

176
00:07:43,800 --> 00:07:46,100
Now, if you're a developer who wants full control,

177
00:07:46,100 --> 00:07:47,700
there's Azure AI Foundry.

178
00:07:47,700 --> 00:07:49,400
This is Microsoft's professional platform

179
00:07:49,400 --> 00:07:52,100
for building and operating AI agents at scale.

180
00:07:52,100 --> 00:07:55,700
It's designed for developers, data scientists, and AI engineers

181
00:07:55,700 --> 00:07:57,900
who want to work with the Microsoft agent framework.

182
00:07:57,900 --> 00:08:00,100
And here's something that happened recently that matters.

183
00:08:00,100 --> 00:08:02,100
Microsoft merged two major frameworks into one,

184
00:08:02,100 --> 00:08:04,500
Semantic kernel, which was the enterprise-friendly option,

185
00:08:04,500 --> 00:08:07,600
and Autogen, which was more about research and multi-agent experiments.

186
00:08:07,600 --> 00:08:09,900
They're now a single framework, so you don't have to choose.

187
00:08:09,900 --> 00:08:13,200
In Foundry, you get access to a model catalog with hundreds of options,

188
00:08:13,200 --> 00:08:14,900
open AI models, of course.

189
00:08:14,900 --> 00:08:18,200
But also Claude from Anthropic, Lama from Meta, and many others.

190
00:08:18,200 --> 00:08:21,200
You pick the model that fits your use case, not the other way round.

191
00:08:21,200 --> 00:08:23,100
You also get full control over orchestration.

192
00:08:23,100 --> 00:08:28,000
That means you decide how the agent reasons, which tools it uses, and in what order.

193
00:08:28,000 --> 00:08:31,000
Foundry gives you prompt flow for designing those reasoning steps,

194
00:08:31,000 --> 00:08:34,800
an agent service for running everything at scale and built in evaluation tools

195
00:08:34,800 --> 00:08:37,700
to test the quality of your agent before it goes live.

196
00:08:37,700 --> 00:08:40,200
Here's a technical detail that matters for production.

197
00:08:40,200 --> 00:08:44,600
Agents built on Foundry run on managed infrastructure with near-instant start-up

198
00:08:44,600 --> 00:08:47,400
under 100 milliseconds, and there's zero idle cost.

199
00:08:47,400 --> 00:08:51,000
If the agent isn't actively processing something, you're not paying for compute.

200
00:08:51,000 --> 00:08:53,000
That's a big deal for cost-conscious teams.

201
00:08:53,000 --> 00:08:57,200
When you're ready to publish, you can push your agent to Microsoft 365 Copilot

202
00:08:57,200 --> 00:08:59,100
and Teams with a single pipeline.

203
00:08:59,100 --> 00:09:00,500
No rebuilding for each surface.

204
00:09:00,500 --> 00:09:02,200
One build, two destinations.

205
00:09:02,200 --> 00:09:04,000
Foundry is best for complex scenarios.

206
00:09:04,000 --> 00:09:06,700
Multi-agent systems where agents talk to each other,

207
00:09:06,700 --> 00:09:09,400
custom memory that persists across sessions.

208
00:09:09,400 --> 00:09:13,000
Advanced security requirements, high volume production workloads.

209
00:09:13,000 --> 00:09:15,800
If your agent needs to handle thousands of requests per minute

210
00:09:15,800 --> 00:09:20,000
across multiple systems with strict compliance rules, Foundry is the right choice.

211
00:09:20,000 --> 00:09:21,800
Now you know the three build parts.

212
00:09:21,800 --> 00:09:25,400
Agent Builder for instant no-code, Copilot Studio for business workflows,

213
00:09:25,400 --> 00:09:27,700
and Foundry for full developer control.

214
00:09:27,700 --> 00:09:31,400
But let's take a step back and look at what actually happens behind the scenes

215
00:09:31,400 --> 00:09:32,900
when an agent does its work.

216
00:09:32,900 --> 00:09:35,300
Because there's more to it than just AI.

217
00:09:35,300 --> 00:09:36,700
How agents actually work.

218
00:09:36,700 --> 00:09:38,700
Identity, data, and governance.

219
00:09:38,700 --> 00:09:41,600
Every agent has a digital identity in Microsoft Entra ID.

220
00:09:41,600 --> 00:09:43,700
That's the same system that manages your login.

221
00:09:43,700 --> 00:09:47,600
An agent gets its own Entra Agent ID just like you have a user account.

222
00:09:47,600 --> 00:09:50,900
That ID gives the agent its own mailbox, its own one drive,

223
00:09:50,900 --> 00:09:53,300
its own Teams account, and a set of permissions.

224
00:09:53,300 --> 00:09:55,600
So agents can access the same systems people do.

225
00:09:55,600 --> 00:09:57,200
They can read SharePoint documents.

226
00:09:57,200 --> 00:09:59,600
They can send emails. They can join Teams channels.

227
00:09:59,600 --> 00:10:01,100
But only within their permissions.

228
00:10:01,100 --> 00:10:03,800
If an agent doesn't have permission to a file, it can't read it.

229
00:10:03,800 --> 00:10:05,500
Same rule as a human employee.

230
00:10:05,500 --> 00:10:07,800
Now how does the agent know what data to use?

231
00:10:07,800 --> 00:10:09,000
That's where Work IQ comes in.

232
00:10:09,000 --> 00:10:12,000
Work IQ grounds agent responses in your company's data.

233
00:10:12,000 --> 00:10:14,000
When an agent needs to answer a question,

234
00:10:14,000 --> 00:10:17,200
it searches your emails, documents, meetings, and chats.

235
00:10:17,200 --> 00:10:18,600
But it respects permissions.

236
00:10:18,600 --> 00:10:20,800
If you can't see a document, the agent won't use it.

237
00:10:20,800 --> 00:10:22,500
The access boundaries stay intact.

238
00:10:22,500 --> 00:10:26,600
All of this identity and permission management happens through the Agent 365 control plane.

239
00:10:26,600 --> 00:10:29,600
That's a dashboard in the Microsoft 365 Admin Center.

240
00:10:29,600 --> 00:10:33,300
IT admins can see every agent in the organization who built it,

241
00:10:33,300 --> 00:10:35,300
what it can access, what it's been doing.

242
00:10:35,300 --> 00:10:37,900
Think of it like an employee directory, but for digital workers.

243
00:10:37,900 --> 00:10:40,100
Governance makes agents ready for your company.

244
00:10:40,100 --> 00:10:42,800
They follow the same compliance policies as people.

245
00:10:42,800 --> 00:10:44,600
Conditional access rules apply.

246
00:10:44,600 --> 00:10:47,400
Someone tries to access an agent from an untrusted device.

247
00:10:47,400 --> 00:10:48,400
Blocked.

248
00:10:48,400 --> 00:10:51,100
Data loss prevention stops them from sending sensitive data outside.

249
00:10:51,100 --> 00:10:54,500
This governance matters because agents can take real actions.

250
00:10:54,500 --> 00:10:56,100
They can create purchase orders.

251
00:10:56,100 --> 00:10:58,700
Send emails to customers, update records in databases.

252
00:10:58,700 --> 00:11:00,400
Without guardrails, that's risky.

253
00:11:00,400 --> 00:11:04,400
With Agent 365, IT has full visibility and control.

254
00:11:04,400 --> 00:11:06,800
Real-world example, the procurement agent.

255
00:11:06,800 --> 00:11:08,800
Let me show you how this works in practice.

256
00:11:08,800 --> 00:11:12,300
Microsoft Mechanics shared a great example of a procurement agent.

257
00:11:12,300 --> 00:11:14,000
It walks through the full life cycle.

258
00:11:14,000 --> 00:11:15,000
Here's the scenario.

259
00:11:15,000 --> 00:11:19,100
A coworker builds a procurement agent and submits it to IT for approval.

260
00:11:19,100 --> 00:11:22,100
Once IT signs off, it gets published to the company's agent store.

261
00:11:22,100 --> 00:11:25,500
Think of it like an app store, but for agents inside your organization.

262
00:11:25,500 --> 00:11:28,100
A procurement manager discovers the agent in that store.

263
00:11:28,100 --> 00:11:30,300
They set it up with just a couple of clicks.

264
00:11:30,300 --> 00:11:33,300
The agent introduces itself in teams and asks for what it needs.

265
00:11:33,300 --> 00:11:34,800
Supplyer policies.

266
00:11:34,800 --> 00:11:36,200
Approved supplier lists.

267
00:11:36,200 --> 00:11:37,500
A procurement playbook.

268
00:11:37,500 --> 00:11:41,700
The manager provides those documents by typing "Use this policy guide for your actions"

269
00:11:41,700 --> 00:11:43,700
and referencing a file from their OneDrive.

270
00:11:43,700 --> 00:11:44,700
That's it.

271
00:11:44,700 --> 00:11:46,300
The agent now has everything it needs.

272
00:11:46,300 --> 00:11:47,800
Now think about the agent's identity.

273
00:11:47,800 --> 00:11:49,300
Remember, "Entra Agent IDs".

274
00:11:49,300 --> 00:11:52,700
This agent has its own mailbox, its own OneDrive, its own Teams account.

275
00:11:52,700 --> 00:11:54,800
So when a customer order comes in for new laptops,

276
00:11:54,800 --> 00:11:56,500
the agent reasons over the request.

277
00:11:56,500 --> 00:12:00,700
It searches suppliers, checks SLAs, looks at pricing from recent orders.

278
00:12:00,700 --> 00:12:04,100
Using WorkIQ, it pulls contacts from across Microsoft 365.

279
00:12:04,100 --> 00:12:08,700
Based on fulfillment time, it recommends a supplier and asks the manager if it should proceed.

280
00:12:08,700 --> 00:12:09,800
The manager confirms.

281
00:12:09,800 --> 00:12:12,400
The agent creates a purchase order for the laptops.

282
00:12:12,400 --> 00:12:17,900
Then it logs that order into a purchasing tracker spreadsheet that lives in SharePoint, all autonomous.

283
00:12:17,900 --> 00:12:21,400
Here's the part that's real today, even though it sounds futuristic.

284
00:12:21,400 --> 00:12:24,100
The manager can ad-mention the procurement agent in teams,

285
00:12:24,100 --> 00:12:26,500
just like any coworker, and ask for status.

286
00:12:26,500 --> 00:12:28,300
The agent responds with what it's working on,

287
00:12:28,300 --> 00:12:30,500
what's been completed, and any pending items.

288
00:12:30,500 --> 00:12:32,500
This example shows the full life cycle.

289
00:12:32,500 --> 00:12:34,200
You discover the agent in the store,

290
00:12:34,200 --> 00:12:36,400
configure it with a few clicks and some documents.

291
00:12:36,400 --> 00:12:39,800
Let it work autonomously on real tasks, get reports when it's done.

292
00:12:39,800 --> 00:12:44,500
That's the power of giving an agent its own identity inside the tools you already use.

293
00:12:44,500 --> 00:12:46,200
But powerful agents need powerful security.

294
00:12:46,200 --> 00:12:48,900
Let's look at how Microsoft keeps agents under control.

295
00:12:48,900 --> 00:12:50,900
Security, compliance, and guardrails.

296
00:12:50,900 --> 00:12:54,600
Every agent runs inside a control center called agent365.

297
00:12:54,600 --> 00:12:57,600
It lives inside the Microsoft 365 Admin Center,

298
00:12:57,600 --> 00:13:01,300
and it gives IT a single dashboard to manage every agent in the organization.

299
00:13:01,300 --> 00:13:05,100
You'll see agents built in co-pilot studio, agents built in Foundry,

300
00:13:05,100 --> 00:13:08,700
even agents built on non-microsoft platforms, all in one place.

301
00:13:08,700 --> 00:13:11,300
No more hunting through different tools just to figure out what's running.

302
00:13:11,300 --> 00:13:16,100
From that dashboard, admins can review permissions, approve new agents before they go live,

303
00:13:16,100 --> 00:13:19,100
block any that shouldn't be running, and apply policy templates.

304
00:13:19,100 --> 00:13:22,300
Those templates are pre-built rule sets that enforce things like

305
00:13:22,300 --> 00:13:27,300
"Don't share content externally" or "Require approval" for any action that modifies data.

306
00:13:27,300 --> 00:13:30,800
Behind the scenes, three security services watch everything.

307
00:13:30,800 --> 00:13:33,300
Microsoft purview handles data loss prevention.

308
00:13:33,300 --> 00:13:36,400
If an agent tries to send sensitive information outside your company,

309
00:13:36,400 --> 00:13:37,800
purview stops it cold.

310
00:13:37,800 --> 00:13:41,300
Microsoft Entra manages access control via conditional access.

311
00:13:41,300 --> 00:13:45,300
If someone tries to reach an agent from an untrusted device that agent gets blocked,

312
00:13:45,300 --> 00:13:48,300
and Microsoft defends a monitor's agent for suspicious activity.

313
00:13:48,300 --> 00:13:51,100
If an agent suddenly starts signing in dozens of times an hour

314
00:13:51,100 --> 00:13:53,700
when it usually logs in twice a day, defends a flag set.

315
00:13:53,700 --> 00:13:55,100
Here's how that works in practice.

316
00:13:55,100 --> 00:13:57,200
Say an agent shows abnormal sign-in frequency.

317
00:13:57,200 --> 00:14:00,900
Entra conditional access automatically blocks it from accessing resources.

318
00:14:00,900 --> 00:14:03,700
The admin gets an alert in the agent365 dashboard,

319
00:14:03,700 --> 00:14:06,300
sees the risk and can block the agent entirely with one click.

320
00:14:06,300 --> 00:14:09,900
That agent is disabled immediately for current users and won't show up for new ones.

321
00:14:09,900 --> 00:14:11,400
There's also the agent map.

322
00:14:11,400 --> 00:14:13,900
It's a visual diagram that shows all your agents,

323
00:14:13,900 --> 00:14:15,700
their connections to each other,

324
00:14:15,700 --> 00:14:18,500
and their connections to workflows and data sources.

325
00:14:18,500 --> 00:14:21,900
If one agent in a chain starts throwing errors you see it on the map,

326
00:14:21,900 --> 00:14:23,800
drill into the details and take action.

327
00:14:23,800 --> 00:14:28,300
It turns agent management from a guessing game into something you can actually see and understand.

328
00:14:28,300 --> 00:14:33,100
So that's the full picture from simple no-code agents to enterprise grade autonomous systems

329
00:14:33,100 --> 00:14:35,600
all with proper security and governance.

330
00:14:35,600 --> 00:14:36,700
Your next steps.

331
00:14:36,700 --> 00:14:38,200
Let's bring this all together.

332
00:14:38,200 --> 00:14:41,100
Agents are not just smarter chatbots, they're autonomous workers

333
00:14:41,100 --> 00:14:42,700
with their own digital identity.

334
00:14:42,700 --> 00:14:48,100
They can access systems, take actions and report back all without you guiding every step.

335
00:14:48,100 --> 00:14:50,300
Microsoft gives you three ways to build them.

336
00:14:50,300 --> 00:14:54,900
Use agent builder inside Microsoft 365 co-pilot for instant no-code agents.

337
00:14:54,900 --> 00:14:58,800
Use co-pilot studio for business workflows with connectors and automated triggers.

338
00:14:58,800 --> 00:15:03,400
Use foundry if you need full control over models, orchestration and infrastructure as a developer.

339
00:15:03,400 --> 00:15:07,000
And wrapping around all of that is the governance layer, agent 365,

340
00:15:07,000 --> 00:15:09,000
Entra ID, purview and defender.

341
00:15:09,000 --> 00:15:10,800
That makes these agents enterprise ready.

342
00:15:10,800 --> 00:15:13,000
They follow the same rules as your human employees.

343
00:15:13,000 --> 00:15:17,800
Here's your homework, open Microsoft 365 co-pilot, go to the agent section, click create.

344
00:15:17,800 --> 00:15:21,700
Describe a simple agent, maybe one that answers questions about a project you're working on.

345
00:15:21,700 --> 00:15:25,900
Give it a knowledge source, turn on only use specified sources and see what it can do.

346
00:15:25,900 --> 00:15:29,000
Most people won't take this step, be different.

347
00:15:29,000 --> 00:15:33,500
In our next episode, we'll walk through building a real agent in co-pilot studio step by step.

348
00:15:33,500 --> 00:15:34,600
You won't want to miss it.

349
00:15:34,600 --> 00:15:37,600
This is Microsoft Knowledge Nuggets on M365 FM.

350
00:15:37,600 --> 00:15:41,900
Subscribe on your favorite podcast platform and share this with someone starting their journey.

351
00:15:41,900 --> 00:15:44,400
I'm Mirko Peters and I'll see you in the next episode.