WEBVTT

1
00:00:00.080 --> 00:00:03.080
<v Speaker 1>You've probably heard this one already. AI can run SharePoint now, No,

2
00:00:03.200 --> 00:00:05.480
<v Speaker 1>it cannot. What it can do is pretend to the

3
00:00:05.519 --> 00:00:08.720
<v Speaker 1>newest act in Microsoft's circus of automation is the SharePoint

4
00:00:08.800 --> 00:00:12.080
<v Speaker 1>Knowledge Agent, a supposedly brilliant little assistant that promises to

5
00:00:12.199 --> 00:00:16.160
<v Speaker 1>organize your content, generate metadata, and answer questions. The pitch

6
00:00:16.199 --> 00:00:19.600
<v Speaker 1>sounds amazing, a tireless, robotic librarian who never stops working.

7
00:00:19.960 --> 00:00:22.399
<v Speaker 1>In reality, it behaves more like an overly confident intern

8
00:00:22.440 --> 00:00:26.320
<v Speaker 1>who just discovered search filters. This agent lives inside SharePoint Premium,

9
00:00:26.320 --> 00:00:29.000
<v Speaker 1>powered by Copilot, armed with the optimism of a first

10
00:00:29.079 --> 00:00:31.519
<v Speaker 1>year analyst and the discipline of a toddler with crayons.

11
00:00:31.960 --> 00:00:35.000
<v Speaker 1>Microsoft markets it like you can finally fire your SharePoint

12
00:00:35.079 --> 00:00:38.960
<v Speaker 1>admin and let AI do the filing users chair, finally

13
00:00:39.119 --> 00:00:42.880
<v Speaker 1>freedom from metadata hell, and then spoiler. It reorganizes your

14
00:00:42.920 --> 00:00:47.159
<v Speaker 1>compliance folder alphabetically by emoji usage. Let's be clear, it's powerful, yes,

15
00:00:47.520 --> 00:00:50.960
<v Speaker 1>but autonomous hardly. It's less pilot, more copilot, which is

16
00:00:50.960 --> 00:00:52.880
<v Speaker 1>a polite way of saying it still needs an adult

17
00:00:52.880 --> 00:00:55.520
<v Speaker 1>in the room. In fact, it doesn't remove your metadata duties,

18
00:00:55.520 --> 00:00:58.719
<v Speaker 1>it triples them. Every document becomes a theological debate about

19
00:00:58.799 --> 00:01:01.520
<v Speaker 1>column naming conventions. By the end of this you'll know

20
00:01:01.560 --> 00:01:04.439
<v Speaker 1>what it really does, where it fumbles, and why governance

21
00:01:04.480 --> 00:01:08.159
<v Speaker 1>officers are quietly sweating behind the scenes. So let's start

22
00:01:08.159 --> 00:01:11.000
<v Speaker 1>with what this digital intern swears it can do, the

23
00:01:11.040 --> 00:01:15.519
<v Speaker 1>sales pitch versus reality. It just organizes everything. According to

24
00:01:15.560 --> 00:01:19.200
<v Speaker 1>Microsoft's marketing and a few overly enthusiastic YouTubers, the Knowledge

25
00:01:19.200 --> 00:01:22.319
<v Speaker 1>Agent organizes everything for you. Those four words should come

26
00:01:22.359 --> 00:01:24.719
<v Speaker 1>with an asterisk the size of a data center. What

27
00:01:24.760 --> 00:01:28.599
<v Speaker 1>it really does is generate metadata columns, create automated rules,

28
00:01:28.599 --> 00:01:32.000
<v Speaker 1>build filtered views, and answer questions across sites. In other words,

29
00:01:32.000 --> 00:01:34.840
<v Speaker 1>it's not reorganizing SharePoint, it's just giving your documents more

30
00:01:34.879 --> 00:01:37.719
<v Speaker 1>personality disorders. Think of it like hiring an intern who

31
00:01:37.719 --> 00:01:40.719
<v Speaker 1>insists they'll clean your desk. You return two hours later

32
00:01:40.760 --> 00:01:44.400
<v Speaker 1>to find your tax receipts sorted by paper thickness. It's tidy, sure,

33
00:01:44.719 --> 00:01:47.599
<v Speaker 1>but good luck filing your return. Before this thing even works,

34
00:01:47.640 --> 00:01:50.920
<v Speaker 1>you must appease several bureaucratic gods. A paid Microsoft three

35
00:01:50.959 --> 00:01:53.879
<v Speaker 1>sixty five copilot license, an admin who opts you into

36
00:01:53.920 --> 00:01:57.560
<v Speaker 1>SharePoint Premium, and ideally someone patient enough to explain to

37
00:01:57.599 --> 00:01:59.959
<v Speaker 1>your boss why half the columns now repeat the same data,

38
00:02:00.120 --> 00:02:04.560
<v Speaker 1>differently capitalized, one summoned. The agent introduces three main tricks,

39
00:02:05.120 --> 00:02:08.280
<v Speaker 1>organize this library, set up rules, and ask a question.

40
00:02:08.479 --> 00:02:12.319
<v Speaker 1>This triumvirate of convenience is Microsoft's long bet that Copilot

41
00:02:12.360 --> 00:02:15.479
<v Speaker 1>trained meta data will fuel better Q and A experiences

42
00:02:15.520 --> 00:02:18.680
<v Speaker 1>across all three sixty five apps. Essentially, you teach SharePoint

43
00:02:18.680 --> 00:02:21.159
<v Speaker 1>to understand your files today, so Copilot can answer questions

44
00:02:21.199 --> 00:02:25.360
<v Speaker 1>about them tomorrow. Admirable, slightly terrifying. Now for reality, Yes,

45
00:02:25.719 --> 00:02:29.120
<v Speaker 1>it can automatically suggest meta data. Yes it can classify documents.

46
00:02:29.120 --> 00:02:32.319
<v Speaker 1>But no, it cannot distinguish policy owner from owner policy

47
00:02:32.319 --> 00:02:35.680
<v Speaker 1>copy too. Every ounce of automation depends entirely on how

48
00:02:35.680 --> 00:02:38.840
<v Speaker 1>clean your existing data is. Garbage in, labeled, garbage out,

49
00:02:39.199 --> 00:02:41.759
<v Speaker 1>and every fix requires you guessed it a human. The

50
00:02:41.759 --> 00:02:44.800
<v Speaker 1>seductive part is the illusion of autonomy. You granted permission,

51
00:02:44.879 --> 00:02:47.360
<v Speaker 1>step away, and when you come back, your library gleams

52
00:02:47.400 --> 00:02:50.319
<v Speaker 1>with new columns and color coded cheerfulness. Except behind that

53
00:02:50.400 --> 00:02:54.879
<v Speaker 1>cheerful facade is quiet chaos, redundant fields, inconsistent tags, half

54
00:02:54.879 --> 00:02:58.759
<v Speaker 1>applied views. Automation doesn't eliminate disorder, It simply buries it

55
00:02:58.840 --> 00:03:02.520
<v Speaker 1>under polish it's the real magic trick making disarray look smooth.

56
00:03:02.919 --> 00:03:04.680
<v Speaker 1>So what happens when you let the intern loose on

57
00:03:04.719 --> 00:03:07.240
<v Speaker 1>your document library? For real? When you say, go ahead,

58
00:03:07.360 --> 00:03:09.879
<v Speaker 1>organize this for me, That's when the comedy starts at

59
00:03:10.120 --> 00:03:13.400
<v Speaker 1>auto tagging the genius that forgets its homework. Here's where

60
00:03:13.439 --> 00:03:16.280
<v Speaker 1>our talented intern rolls up its digital sleeves and promises

61
00:03:16.319 --> 00:03:19.719
<v Speaker 1>to organize this library. The phrase drips with confidence, like

62
00:03:19.759 --> 00:03:22.879
<v Speaker 1>it's about to alphabetize the universe. You press the button

63
00:03:22.960 --> 00:03:26.120
<v Speaker 1>expecting harmony. What you get looks more like abstract art

64
00:03:26.159 --> 00:03:29.039
<v Speaker 1>produced by a neural network that just discovered Excel. The

65
00:03:29.240 --> 00:03:32.960
<v Speaker 1>organized this library function sounds deceptively simple. It scans your documents,

66
00:03:33.000 --> 00:03:36.000
<v Speaker 1>then suggests new columns of metadata, maybe things like review date,

67
00:03:36.319 --> 00:03:40.159
<v Speaker 1>department owner, or document type. Except sometimes it decides your

68
00:03:40.159 --> 00:03:42.840
<v Speaker 1>library needs a column called important number two because it

69
00:03:42.879 --> 00:03:45.680
<v Speaker 1>found the number two inside a file name. Yes, really,

70
00:03:45.840 --> 00:03:48.319
<v Speaker 1>it's like watching a gifted student as calculus and then

71
00:03:48.360 --> 00:03:50.360
<v Speaker 1>forget how to spell their own name. The first time

72
00:03:50.360 --> 00:03:53.039
<v Speaker 1>you run it, you're attempted to panic. The suggestions look random,

73
00:03:53.080 --> 00:03:56.759
<v Speaker 1>the preview window glows with meaningless fields, and nothing seems coherent.

74
00:03:56.840 --> 00:03:59.159
<v Speaker 1>That's because it isn't ready yet. The engine quietly does

75
00:03:59.199 --> 00:04:02.479
<v Speaker 1>background indexing and content analysis, a process that can take hours.

76
00:04:02.879 --> 00:04:05.439
<v Speaker 1>Until then, it's basically guessing. In other words, if you

77
00:04:05.479 --> 00:04:08.439
<v Speaker 1>click create columns right away, you get digital gibberish. Give

78
00:04:08.479 --> 00:04:11.599
<v Speaker 1>it a night to sleep. Apparently even artificial intelligence needs

79
00:04:11.639 --> 00:04:14.080
<v Speaker 1>to dream before it makes sense. When it finally wakes up,

80
00:04:14.159 --> 00:04:17.480
<v Speaker 1>something magical happens. The column suggestions actually reflect structure. You

81
00:04:17.560 --> 00:04:20.160
<v Speaker 1>might see review date correctly pulled from the header of

82
00:04:20.199 --> 00:04:23.319
<v Speaker 1>dozens of policies. You realize it read the text, detected

83
00:04:23.360 --> 00:04:26.160
<v Speaker 1>a pattern, and turned it into metadata for about ten seconds.

84
00:04:26.199 --> 00:04:29.079
<v Speaker 1>You're impressed. Then you notice it also created policy owner,

85
00:04:29.279 --> 00:04:33.319
<v Speaker 1>policy owner, and policy owner as separate fields. SharePoint now

86
00:04:33.319 --> 00:04:36.160
<v Speaker 1>speaks three dialects of English. This is the first real lesson.

87
00:04:36.360 --> 00:04:40.399
<v Speaker 1>The AI doesn't create order, it amplifies whatever chaos already exists.

88
00:04:40.879 --> 00:04:45.120
<v Speaker 1>Your messy document formatting Congratulations, it's now immortalized as structured data.

89
00:04:45.199 --> 00:04:49.079
<v Speaker 1>It's not malicious, it's just painfully literal. Every inconsistency becomes

90
00:04:49.120 --> 00:04:52.040
<v Speaker 1>a column, Every formatting quirk becomes an ontology. The intern

91
00:04:52.079 --> 00:04:55.879
<v Speaker 1>has taken notes on your sins. Now, Microsoft anticipated your

92
00:04:55.879 --> 00:04:59.120
<v Speaker 1>existential crisis and thankfully made this process optional. None of

93
00:04:59.160 --> 00:05:02.360
<v Speaker 1>these changes apply automatically. You, the alleged human adult, must

94
00:05:02.360 --> 00:05:05.519
<v Speaker 1>review every suggestion and explicitly choose which columns to keep.

95
00:05:05.600 --> 00:05:09.240
<v Speaker 1>The interface even highlights pending changes with polite blue indicators,

96
00:05:09.279 --> 00:05:13.000
<v Speaker 1>whispering are you sure about this? Copilot? Isn't autopilot. It's

97
00:05:13.040 --> 00:05:15.639
<v Speaker 1>manual labor dressed up in predictive text, and you approve

98
00:05:15.720 --> 00:05:18.680
<v Speaker 1>each change, remove duplicates, rename things, and only then commit

99
00:05:18.720 --> 00:05:21.079
<v Speaker 1>the metadata to the view the irony. It took you

100
00:05:21.160 --> 00:05:23.879
<v Speaker 1>longer than just building the columns yourself. Still, when it works,

101
00:05:23.879 --> 00:05:27.040
<v Speaker 1>it's genuinely clever. You can preview values across a sample set.

102
00:05:27.240 --> 00:05:30.519
<v Speaker 1>See how policy owner fills with department names, how review

103
00:05:30.560 --> 00:05:33.279
<v Speaker 1>date populates from document headers. It's a quick way to

104
00:05:33.279 --> 00:05:35.839
<v Speaker 1>audit the mess. Then you apply it across the library

105
00:05:36.079 --> 00:05:40.079
<v Speaker 1>and watch the autopil process begin. Background workers hum metadata

106
00:05:40.120 --> 00:05:43.000
<v Speaker 1>slowly populates, and you briefly consider sending the AI a

107
00:05:43.040 --> 00:05:45.800
<v Speaker 1>thank you note. Don't it's about to betray you again,

108
00:05:45.879 --> 00:05:48.959
<v Speaker 1>because here comes the lag. Updating and filling meta data

109
00:05:49.079 --> 00:05:52.720
<v Speaker 1>is asynchronous. While it churns, the columns display blank values

110
00:05:52.959 --> 00:05:56.720
<v Speaker 1>for minutes or hours. Users think nothing happened, so they

111
00:05:56.920 --> 00:06:00.360
<v Speaker 1>rerun the task, doubling the chaos. Then minutes later both

112
00:06:00.360 --> 00:06:03.600
<v Speaker 1>sets of updates collide overwriting partially filled data. It's not

113
00:06:03.639 --> 00:06:06.480
<v Speaker 1>a bug, it's a test of faith. The agent rewards patients,

114
00:06:06.480 --> 00:06:09.040
<v Speaker 1>punishes enthusiasm. Think of it as hiring a genius who

115
00:06:09.079 --> 00:06:11.920
<v Speaker 1>works fast only when you stop looking. Versioning adds another

116
00:06:11.920 --> 00:06:15.160
<v Speaker 1>comedy layer. Suppose you upload policyv to dog X. The

117
00:06:15.199 --> 00:06:18.519
<v Speaker 1>AI dutifully copies the metadata from policy V one, including

118
00:06:18.560 --> 00:06:21.160
<v Speaker 1>the outdated owner field. Then it takes a breath. Sometime later,

119
00:06:21.199 --> 00:06:23.759
<v Speaker 1>it realizes the content changed and kicks off another round

120
00:06:23.839 --> 00:06:27.160
<v Speaker 1>of metadata inferences. Eventually it catches up. Eventually. If your

121
00:06:27.160 --> 00:06:30.199
<v Speaker 1>workflow relies on instant accuracy, this delay will drive you

122
00:06:30.240 --> 00:06:33.160
<v Speaker 1>to drink. Once you understand that timing problem, you start

123
00:06:33.199 --> 00:06:35.639
<v Speaker 1>treating it like the quirk it is. You schedule re

124
00:06:35.680 --> 00:06:39.439
<v Speaker 1>indexes overnight, monitor autofil cues, and laugh bitterly at anyone

125
00:06:39.439 --> 00:06:42.040
<v Speaker 1>who thought this feature meant less work. That's the human

126
00:06:42.079 --> 00:06:44.399
<v Speaker 1>in the loop model. In action, the AI proposes the

127
00:06:44.480 --> 00:06:47.480
<v Speaker 1>human disposes. You create its gases, you correct its spelling,

128
00:06:47.560 --> 00:06:50.839
<v Speaker 1>You restrain its enthusiasm. The agent doesn't replace judgment. It

129
00:06:50.879 --> 00:06:53.639
<v Speaker 1>demands supervision on good data. Sets. The results can be

130
00:06:53.639 --> 00:06:57.680
<v Speaker 1>surprisingly useful. Policy libraries gain uniform fields, Audit teams can

131
00:06:57.720 --> 00:07:01.279
<v Speaker 1>filter documents by owner searching for review due next quarter

132
00:07:01.600 --> 00:07:05.040
<v Speaker 1>suddenly returns everything tagged correctly. The machine gives structure to

133
00:07:05.079 --> 00:07:08.040
<v Speaker 1>your chaos, but only after you rebuild half of it yourself.

134
00:07:08.360 --> 00:07:11.560
<v Speaker 1>The paradox of automation. It scales efficiency and stupidity at

135
00:07:11.560 --> 00:07:15.279
<v Speaker 1>the same time the truth This tool shines in broad classification.

136
00:07:15.480 --> 00:07:18.720
<v Speaker 1>It can tell contracts from templates, policies from forms, But

137
00:07:18.759 --> 00:07:23.199
<v Speaker 1>when it comes to compliance, tagging, records, management, sensitivity labels,

138
00:07:23.399 --> 00:07:26.680
<v Speaker 1>retention categories, it's out of its depth. It reads content,

139
00:07:26.839 --> 00:07:30.160
<v Speaker 1>not context. It recognizes words, not accountability. That's fine for

140
00:07:30.199 --> 00:07:34.120
<v Speaker 1>casual queries, disastrous for legal retention. And yet despite all that,

141
00:07:34.160 --> 00:07:37.680
<v Speaker 1>you'll keep using it because even partial organization feels like progress.

142
00:07:37.720 --> 00:07:39.600
<v Speaker 1>The intern may forget its homework, but at least it

143
00:07:39.639 --> 00:07:41.560
<v Speaker 1>showed up and did something while you were asleep. Just

144
00:07:41.600 --> 00:07:44.120
<v Speaker 1>remember to check its mass before sending the report upstream.

145
00:07:44.439 --> 00:07:47.480
<v Speaker 1>Of course, our intern isn't satisfied with sorting papers, nor

146
00:07:47.480 --> 00:07:50.079
<v Speaker 1>at one's responsibility. It insists it can follow rules too,

147
00:07:50.199 --> 00:07:52.639
<v Speaker 1>rules written in plain English. Naturally, we're about to let

148
00:07:52.680 --> 00:07:56.759
<v Speaker 1>it try natural language rules. Governance for dummies. Enter the

149
00:07:56.800 --> 00:07:59.720
<v Speaker 1>second act of our drama. Rules. The knowledge agent now

150
00:07:59.720 --> 00:08:03.199
<v Speaker 1>claim it can set up rules using natural language, no coding,

151
00:08:03.439 --> 00:08:06.399
<v Speaker 1>no power, automate wizardry, just a friendly chat for the

152
00:08:06.439 --> 00:08:09.399
<v Speaker 1>average user. That sounds divine. For administrators, it sounds like

153
00:08:09.439 --> 00:08:12.279
<v Speaker 1>an incoming migraine, because what this really is underneath the

154
00:08:12.279 --> 00:08:15.759
<v Speaker 1>simplicity is governance roulette dressed up as productivity. Here's how

155
00:08:15.800 --> 00:08:18.600
<v Speaker 1>it behaves. You tell it in plain English. When a

156
00:08:18.639 --> 00:08:22.240
<v Speaker 1>new file is added, send me an email. The agent grins,

157
00:08:22.399 --> 00:08:27.160
<v Speaker 1>nods silently, and builds a one liner rule condition action done.

158
00:08:27.240 --> 00:08:30.480
<v Speaker 1>It's the automation equivalent of I made you a sandwich, cute,

159
00:08:30.519 --> 00:08:34.000
<v Speaker 1>but lacking nutritional depth. Professional admins quickly realize this feature

160
00:08:34.039 --> 00:08:36.679
<v Speaker 1>has all the sophistication of a motion activated porch light

161
00:08:37.080 --> 00:08:40.759
<v Speaker 1>works fine until the wind blows. There's no nested logic,

162
00:08:40.879 --> 00:08:43.840
<v Speaker 1>no multiple actions per trigger, just single rules. Simplicity meant

163
00:08:43.879 --> 00:08:46.679
<v Speaker 1>for citizen users, which is Microsoft's polite phrase for people

164
00:08:46.720 --> 00:08:49.759
<v Speaker 1>we don't trust with power automate and the fragility, Oh,

165
00:08:49.799 --> 00:08:53.000
<v Speaker 1>the fragility. These rules rely entirely on the metadata columns

166
00:08:53.039 --> 00:08:56.120
<v Speaker 1>you and the intern previously invented if that metadata hasn't

167
00:08:56.120 --> 00:08:59.240
<v Speaker 1>finished processing when the file is uploaded, the trigger misses it.

168
00:08:59.559 --> 00:09:03.080
<v Speaker 1>The intern version of IF equals to only evaluates once

169
00:09:03.200 --> 00:09:06.440
<v Speaker 1>too early, too late, never quite when needed. It's automation

170
00:09:06.559 --> 00:09:10.279
<v Speaker 1>by clairvoyance. The result rules that theoretically exist, but execute

171
00:09:10.320 --> 00:09:13.399
<v Speaker 1>only on alternate Thursdays when the indexing demon feels inspired.

172
00:09:13.480 --> 00:09:16.159
<v Speaker 1>Now here's where it crosses from naive to dangerous. Suppose

173
00:09:16.200 --> 00:09:18.960
<v Speaker 1>you attempt a slightly cleverer command when a new file

174
00:09:19.000 --> 00:09:21.480
<v Speaker 1>is added that's related to leave send me an email.

175
00:09:21.840 --> 00:09:24.720
<v Speaker 1>The agent can't find any leave field, so it enthusiastically

176
00:09:24.720 --> 00:09:27.879
<v Speaker 1>offers to create one, a brand new yes, no column

177
00:09:27.919 --> 00:09:30.840
<v Speaker 1>called leave related. Just like that, you've extended your schema

178
00:09:30.919 --> 00:09:34.440
<v Speaker 1>through chit chat. Governance committees weep silently in the corner

179
00:09:34.559 --> 00:09:36.679
<v Speaker 1>because the AI didn't ask who owns that schemer or

180
00:09:36.720 --> 00:09:39.879
<v Speaker 1>whether leave related conflicts with existing fields. It just went

181
00:09:39.919 --> 00:09:43.639
<v Speaker 1>ahead and made one impressive. It literally granted itself database privileges.

182
00:09:43.679 --> 00:09:47.639
<v Speaker 1>Mid conversation somewhere, your SharePoint architect just felt a sharp

183
00:09:47.679 --> 00:09:51.639
<v Speaker 1>pain without knowing why. This is how decentralized rulemaking begins.

184
00:09:52.080 --> 00:09:55.000
<v Speaker 1>A well meaning user creates a casual rule, another copies

185
00:09:55.039 --> 00:09:59.120
<v Speaker 1>it soon every department invents its own local truth policy

186
00:09:59.159 --> 00:10:02.799
<v Speaker 1>owner versus owner policy, review date versus the next review,

187
00:10:03.519 --> 00:10:06.600
<v Speaker 1>each with matching rules, pinging inboxes, and moving documents around

188
00:10:06.720 --> 00:10:09.919
<v Speaker 1>like overcaffeinated elves. Before long, your clean library becomes a

189
00:10:09.919 --> 00:10:12.639
<v Speaker 1>self modifying ecosystem. The AI doesn't just follow the rules,

190
00:10:12.639 --> 00:10:15.960
<v Speaker 1>it breeds them. Anyone who survived the access database explosion

191
00:10:15.960 --> 00:10:19.120
<v Speaker 1>of the mid two hundreds will experience deja vu. Microsoft

192
00:10:19.120 --> 00:10:21.799
<v Speaker 1>replaced macros with machine learning, but the problem is identical,

193
00:10:21.879 --> 00:10:25.600
<v Speaker 1>uncoordinated logic written by enthusiasts. From a governance standpoint, this

194
00:10:25.720 --> 00:10:28.720
<v Speaker 1>is the nightmare scenario. Rules can now trigger off user

195
00:10:28.759 --> 00:10:32.960
<v Speaker 1>generated metadata that no one monitors. They send alerts, copy files,

196
00:10:33.000 --> 00:10:37.840
<v Speaker 1>and rewrite ownership fields across libraries. Its automation drift silent, invisible,

197
00:10:38.080 --> 00:10:41.039
<v Speaker 1>and approximately one share point migration away from disaster. The

198
00:10:41.080 --> 00:10:44.039
<v Speaker 1>weirdest part to the average business user, it feels empowering.

199
00:10:44.279 --> 00:10:47.440
<v Speaker 1>Look I automated our HR updates, they say, blissfully unaware.

200
00:10:47.440 --> 00:10:50.399
<v Speaker 1>They've also created a recursive email loop that notifies itself

201
00:10:50.399 --> 00:10:53.960
<v Speaker 1>every five minutes. The irony is that power automate mocked

202
00:10:54.000 --> 00:10:57.879
<v Speaker 1>here as too complex exists precisely to prevent this chaos.

203
00:10:57.960 --> 00:11:02.039
<v Speaker 1>It enforces authentication rolls and standardized naming. The knowledge agent

204
00:11:02.080 --> 00:11:04.799
<v Speaker 1>tosses all that structure aside in the name of accessibility.

205
00:11:04.879 --> 00:11:08.279
<v Speaker 1>That makes it seductive. Business users will adopt it gleefully,

206
00:11:08.720 --> 00:11:12.240
<v Speaker 1>insisting they don't need it for simple automations, and in

207
00:11:12.240 --> 00:11:14.720
<v Speaker 1>one sense they're right they don't. They just need it

208
00:11:14.879 --> 00:11:18.039
<v Speaker 1>t later to clean up the mess. Technically, there's a

209
00:11:18.039 --> 00:11:21.240
<v Speaker 1>wise way to use this feature, restrict rule creation to

210
00:11:21.360 --> 00:11:25.320
<v Speaker 1>controlled libraries and force review by content owners. In practice, though,

211
00:11:25.600 --> 00:11:29.639
<v Speaker 1>the interface lives inside everyday document views, where temptation wins.

212
00:11:30.279 --> 00:11:33.240
<v Speaker 1>Few users resist the flashing cursor that says, tell me

213
00:11:33.279 --> 00:11:36.600
<v Speaker 1>what to do. That's not empowerment, that's entrapment by design.

214
00:11:36.720 --> 00:11:38.759
<v Speaker 1>So yes, this tool can follow rules. It just can't

215
00:11:38.799 --> 00:11:41.519
<v Speaker 1>remember why those rules exist or who approve them. It

216
00:11:41.559 --> 00:11:46.240
<v Speaker 1>mimics governance without comprehension, the classic hallmark of every overconfident intern.

217
00:11:46.360 --> 00:11:49.720
<v Speaker 1>And while it's busy rewriting your schema, it proudly insists

218
00:11:49.720 --> 00:11:52.080
<v Speaker 1>it can also answer your questions. Of course, it does

219
00:11:52.720 --> 00:11:56.200
<v Speaker 1>ask a question when chatting with your data becomes a liability.

220
00:11:56.519 --> 00:11:59.440
<v Speaker 1>Now comes everyone's favorite trick. The digital intern stares bright

221
00:11:59.480 --> 00:12:01.639
<v Speaker 1>eyed at your life library and says, ask me anything.

222
00:12:02.519 --> 00:12:05.840
<v Speaker 1>It's the SharePoint equivalent of a chatbot on too much caffeine,

223
00:12:05.879 --> 00:12:09.799
<v Speaker 1>promising conversational insight into decades of corporate detritus. In practice,

224
00:12:09.840 --> 00:12:12.879
<v Speaker 1>it's equal parts sorcery and malpractice, because when you hand

225
00:12:12.960 --> 00:12:15.480
<v Speaker 1>artificial intelligence the keys to your knowledge base, you're not

226
00:12:15.559 --> 00:12:18.440
<v Speaker 1>just empowering search. You're gambling with context. Here's what it's

227
00:12:18.440 --> 00:12:20.960
<v Speaker 1>supposed to do. You type in natural language problems like

228
00:12:21.000 --> 00:12:25.840
<v Speaker 1>which policies are affected by new leave legislation? Or what's

229
00:12:25.879 --> 00:12:28.440
<v Speaker 1>the difference between version one and version two of the

230
00:12:28.480 --> 00:12:33.320
<v Speaker 1>Finance Handbook. The knowledge agent dutifully combs through your document content,

231
00:12:33.440 --> 00:12:37.039
<v Speaker 1>metadata and relationships, then produces an answer, sometimes an eerily

232
00:12:37.080 --> 00:12:40.799
<v Speaker 1>good one. It will cite actual document titles, summarize relevant changes,

233
00:12:40.879 --> 00:12:43.639
<v Speaker 1>even highlight fields where the policy differs for a brief,

234
00:12:43.639 --> 00:12:47.320
<v Speaker 1>shining moment, you think. Finally, SharePoint that behaves like it's

235
00:12:47.360 --> 00:12:50.440
<v Speaker 1>been paying attention, and sometimes it truly helps. Legal teams

236
00:12:50.440 --> 00:12:54.639
<v Speaker 1>preparing for audits can instantly cross reference related policies. Compliance

237
00:12:54.679 --> 00:12:57.480
<v Speaker 1>officers can ask show me documents updated after March that

238
00:12:57.559 --> 00:13:01.200
<v Speaker 1>mention privacy clauses and get useful leads. No more endless

239
00:13:01.240 --> 00:13:04.759
<v Speaker 1>clicking through version histories. The AI stitches together contexts that

240
00:13:04.799 --> 00:13:08.799
<v Speaker 1>would have taken humans half a day to assemble. Productivity skyrockets,

241
00:13:09.080 --> 00:13:13.240
<v Speaker 1>reputations rise. Someone drafts a LinkedIn post about transforming knowledge

242
00:13:13.240 --> 00:13:16.240
<v Speaker 1>work with responsible AI, but then reality stumbles in. The

243
00:13:16.240 --> 00:13:18.879
<v Speaker 1>agent doesn't actually understand what you mean. It predicts what

244
00:13:18.919 --> 00:13:21.240
<v Speaker 1>you might mean if your phrasing is slightly off. It

245
00:13:21.279 --> 00:13:25.440
<v Speaker 1>improvises when you ask which procedures relate to performance reviews.

246
00:13:26.080 --> 00:13:28.480
<v Speaker 1>It might also surf as disciplinary guidelines, because the two

247
00:13:28.600 --> 00:13:32.519
<v Speaker 1>often co occur in text congratulations, it just equated mentorship

248
00:13:32.559 --> 00:13:37.039
<v Speaker 1>with punishment. Natural language interpretation isn't logic, it's probabilistic guesswork.

249
00:13:37.399 --> 00:13:40.519
<v Speaker 1>And when that guesswork happens inside a compliance archive, the

250
00:13:40.639 --> 00:13:45.159
<v Speaker 1>cost of close enough multiplies fast. The disaster potential isn't hypothetical.

251
00:13:45.399 --> 00:13:49.399
<v Speaker 1>Picture an eager manager searching show all documents involving salary adjustments,

252
00:13:49.720 --> 00:13:54.279
<v Speaker 1>the AI happily inclusive returns, payroll summaries, hr grievances, and

253
00:13:54.320 --> 00:13:58.120
<v Speaker 1>a PDF titled Executive Compensation twenty twenty two confidential, the

254
00:13:58.200 --> 00:14:02.120
<v Speaker 1>perfect recall, zero judgment. Remember, sensitivity labels don't translate neatly

255
00:14:02.120 --> 00:14:05.559
<v Speaker 1>into the agent's context window. Their permissions, not comprehension. The

256
00:14:05.639 --> 00:14:08.399
<v Speaker 1>AI doesn't know what's secret. It just reads text that

257
00:14:08.440 --> 00:14:12.200
<v Speaker 1>says salary, you wanted intelligence search, you got a gossip engine.

258
00:14:12.320 --> 00:14:16.279
<v Speaker 1>The philosophical irony is delicious. We train AI to answer questions,

259
00:14:16.440 --> 00:14:19.720
<v Speaker 1>yet in SharePoint governance, the safest response is often you

260
00:14:19.720 --> 00:14:22.399
<v Speaker 1>shouldn't be asking that The agent lacks that filter. It

261
00:14:22.440 --> 00:14:25.720
<v Speaker 1>assumes every query is legitimate because every file appears fair

262
00:14:25.759 --> 00:14:29.360
<v Speaker 1>game under your access. In multi tenant environments, that's a

263
00:14:29.399 --> 00:14:33.720
<v Speaker 1>recipe for quiet breaches, internal unlogged, perfectly explainable by the

264
00:14:33.799 --> 00:14:37.279
<v Speaker 1>user had permission. The AI is innocent, the exposure is total.

265
00:14:37.480 --> 00:14:41.480
<v Speaker 1>To be fair, Microsoft never claimed the agent replaces governance frameworks.

266
00:14:41.480 --> 00:14:44.720
<v Speaker 1>It's a convenience feature, not a compliance officer. The problem

267
00:14:44.799 --> 00:14:48.840
<v Speaker 1>is perception. Average users assume conversational interfaces equal authority. When

268
00:14:48.879 --> 00:14:51.799
<v Speaker 1>an answer looks fluent, it feels trustworthy, So an imperfect

269
00:14:51.799 --> 00:14:55.559
<v Speaker 1>summary disguised in polite prose becomes organizational truth. Minutes later,

270
00:14:55.600 --> 00:14:59.120
<v Speaker 1>someone drafts a report based on a hallucinated paragraph, and

271
00:14:59.240 --> 00:15:02.360
<v Speaker 1>policy gets written. Think of the knowledge agent as a

272
00:15:02.360 --> 00:15:06.039
<v Speaker 1>librarian with no sense of privacy. Ask confidently enough, and

273
00:15:06.080 --> 00:15:08.879
<v Speaker 1>it hands you restricted binders with a smile is not malicious.

274
00:15:08.879 --> 00:15:12.159
<v Speaker 1>It's literal authority through syntax. That's why the real governance

275
00:15:12.240 --> 00:15:16.480
<v Speaker 1>risk isn't leaks its misinterpretation. AI transforms searching for information

276
00:15:16.600 --> 00:15:22.559
<v Speaker 1>into trusting whatever sounds coherent. Metadata Accuracy collapses under linguistic charm. Convenient, yes,

277
00:15:22.879 --> 00:15:26.320
<v Speaker 1>addictive even, But every confident query adds another layer of

278
00:15:26.360 --> 00:15:31.240
<v Speaker 1>inferred metadata, more unreviewed relationships, more invisible drift. The interurn

279
00:15:31.320 --> 00:15:34.960
<v Speaker 1>learns from each conversation, and suddenly your SharePoint taxonomy mutates

280
00:15:34.960 --> 00:15:38.679
<v Speaker 1>in the background. Congratulations, You've built a self replicating rumor mill,

281
00:15:38.960 --> 00:15:43.600
<v Speaker 1>optimized for plausibility and that convenience. It breeds one predictable disease,

282
00:15:43.840 --> 00:15:48.879
<v Speaker 1>metadata inflation, the hidden governance risk, metadata inflation, and AI drift.

283
00:15:49.240 --> 00:15:51.840
<v Speaker 1>Every generation of it invents a new way to destroy

284
00:15:51.879 --> 00:15:56.279
<v Speaker 1>its own records, the knowledge agent's contribution metadata inflation. It

285
00:15:56.320 --> 00:16:00.320
<v Speaker 1>starts innocently enough. Every prompt, every helpful rule, every column

286
00:16:00.360 --> 00:16:04.559
<v Speaker 1>suggestion spawns another minor schema change. Before long, your pristine

287
00:16:04.600 --> 00:16:09.159
<v Speaker 1>information architecture resembles a thrift store alphabet, duplicate owner fields,

288
00:16:09.480 --> 00:16:13.120
<v Speaker 1>stale review date, two columns, and a few orphans mysteriously

289
00:16:13.200 --> 00:16:17.799
<v Speaker 1>labeled temp. This is AI drift in its purest form.

290
00:16:18.200 --> 00:16:20.200
<v Speaker 1>The more you use the agent, the more it believes

291
00:16:20.200 --> 00:16:23.320
<v Speaker 1>your improvisations are law. It notices a variation and assumes

292
00:16:23.320 --> 00:16:25.840
<v Speaker 1>you meant to branch the taxonomy not corrected. Then it

293
00:16:25.879 --> 00:16:29.159
<v Speaker 1>starts generating patterns from those variants. What was once one

294
00:16:29.159 --> 00:16:33.080
<v Speaker 1>field for policy owner becomes four subtly different lineages across

295
00:16:33.080 --> 00:16:37.720
<v Speaker 1>every library. Copilot dutifully indexes them, all confused but compliant,

296
00:16:37.919 --> 00:16:41.000
<v Speaker 1>eventually concluding your company has five departments named h R.

297
00:16:41.600 --> 00:16:48.039
<v Speaker 1>Brilliant admins always underestimate the propagation engine. Each auto created

298
00:16:48.080 --> 00:16:51.639
<v Speaker 1>column isn't local. It seeps into templates, content types, and

299
00:16:51.679 --> 00:16:55.600
<v Speaker 1>cross sight queries like mold. It spreads quietly through lists

300
00:16:55.679 --> 00:16:59.240
<v Speaker 1>carried by copy and paste enthusiasts and sink jobs, and

301
00:16:59.279 --> 00:17:02.039
<v Speaker 1>when the next cop silet index rebuilds, it aligns to

302
00:17:02.080 --> 00:17:05.519
<v Speaker 1>this new reality. From that point forward, AI answers mirror

303
00:17:05.559 --> 00:17:07.599
<v Speaker 1>the chaos you unknowingly taught it. Let me put it

304
00:17:07.599 --> 00:17:10.640
<v Speaker 1>in human terms. Imagine a warehouse where every worker decides

305
00:17:10.640 --> 00:17:14.039
<v Speaker 1>their own labeling scheme. Some tag boxes as fragile, others

306
00:17:14.079 --> 00:17:17.200
<v Speaker 1>as handle with care, one as very braky stuff. After

307
00:17:17.240 --> 00:17:19.759
<v Speaker 1>a week, nobody knows what's inside anything, but everyone's confident

308
00:17:19.799 --> 00:17:23.440
<v Speaker 1>their following procedure. The knowledge agent amplifies that psychology. It

309
00:17:23.480 --> 00:17:27.759
<v Speaker 1>democratizes metadata creation without central review, turning governance from a

310
00:17:27.799 --> 00:17:31.400
<v Speaker 1>system into a popularity contest of tagging habits. What follows

311
00:17:31.440 --> 00:17:36.640
<v Speaker 1>is predictable, inconsistent taxonomies cascade into copilot confusion. Your AI

312
00:17:36.680 --> 00:17:41.240
<v Speaker 1>summaries reference outdated owners, duplicate projects or contradictory retention markers.

313
00:17:41.279 --> 00:17:44.880
<v Speaker 1>Compliance dashboards misreport document counts because three columns all meant

314
00:17:45.160 --> 00:17:49.279
<v Speaker 1>approved by. Then leadership asks why Copilot's policy briefings now

315
00:17:49.319 --> 00:17:53.279
<v Speaker 1>summarize the security manual as unknown author, unknown date because

316
00:17:53.319 --> 00:17:56.920
<v Speaker 1>the intern relabeled its homework thrice. You cannot train this

317
00:17:57.000 --> 00:18:00.720
<v Speaker 1>issue away. Drift is systemic. The more people prompt, the

318
00:18:00.799 --> 00:18:04.680
<v Speaker 1>less consistent the dataset becomes. Microsoft didn't design a villain.

319
00:18:04.799 --> 00:18:08.480
<v Speaker 1>They designed enthusiasm without breaks. So yes, the knowledge agent

320
00:18:08.519 --> 00:18:11.759
<v Speaker 1>is clever, but it's also a schema generation machine on autopilot.

321
00:18:11.839 --> 00:18:13.759
<v Speaker 1>Control it the same way you'd handle a toddler with

322
00:18:13.799 --> 00:18:18.319
<v Speaker 1>administrative privileges, sandboxed, supervised, and backed up nightly. Here's your

323
00:18:18.319 --> 00:18:22.480
<v Speaker 1>containment checklist. First, confine it to pilot libraries. The digital

324
00:18:22.480 --> 00:18:25.519
<v Speaker 1>equivalent of a padded playpen let it experiment where mistakes

325
00:18:25.519 --> 00:18:29.839
<v Speaker 1>are recoverable. Second, monitor column creation events via the SharePoint

326
00:18:29.839 --> 00:18:32.680
<v Speaker 1>Admin portal or API. Every new field should raise an

327
00:18:32.680 --> 00:18:36.759
<v Speaker 1>alert like smoke. Third document column lineage. Track which library

328
00:18:36.799 --> 00:18:39.359
<v Speaker 1>birthed each field and when you'll thank yourself during the

329
00:18:39.400 --> 00:18:44.079
<v Speaker 1>next audit. And finally, enforce naming conventions. Upstream prefix AI

330
00:18:44.160 --> 00:18:46.880
<v Speaker 1>generated columns with AI on so everyone knows they came

331
00:18:46.920 --> 00:18:49.880
<v Speaker 1>from the intern's late night enthusiasm. Follow these rules and

332
00:18:49.960 --> 00:18:52.720
<v Speaker 1>AI drift remains an anecdote instead of a post mortem.

333
00:18:52.880 --> 00:18:55.559
<v Speaker 1>Ignore them, and your taxonomy turns Ferrell by quarters end.

334
00:18:55.839 --> 00:18:59.000
<v Speaker 1>The problem isn't malevolence, its momentum. The agent keeps helping

335
00:18:59.079 --> 00:19:01.519
<v Speaker 1>until help becomes high. So no, it's not replacing you.

336
00:19:01.599 --> 00:19:05.759
<v Speaker 1>It's multiplying your headaches at machine speed. Verdict and survival guide.

337
00:19:05.960 --> 00:19:08.759
<v Speaker 1>Here's the verdict every worried SharePoint admin needs to hear.

338
00:19:09.519 --> 00:19:12.319
<v Speaker 1>The knowledge agent is a gifted assistant, not a successor.

339
00:19:12.720 --> 00:19:15.680
<v Speaker 1>It automates drudgery but cannot discern meaning. Treating it as

340
00:19:15.720 --> 00:19:19.480
<v Speaker 1>autonomous is governance malpractice. Left alone, it will joyfully reinvent

341
00:19:19.519 --> 00:19:23.039
<v Speaker 1>your schema, misclassify your records, and answer forbidden questions with

342
00:19:23.079 --> 00:19:25.920
<v Speaker 1>bullet point confidence. Think of it as letting the smartest

343
00:19:25.920 --> 00:19:30.240
<v Speaker 1>intern in HR rewrite employment contracts unsupervised. They'll finish early,

344
00:19:30.319 --> 00:19:34.680
<v Speaker 1>format beautifully and accidentally legalized chaos. Brilliance without boundaries is

345
00:19:34.799 --> 00:19:38.319
<v Speaker 1>just accelerated error. Your survival plan is mercifully simple. Step one,

346
00:19:38.440 --> 00:19:41.680
<v Speaker 1>confine it to curated libraries. Establish a proving ground where

347
00:19:41.680 --> 00:19:45.880
<v Speaker 1>experimentation won't corrupt production taxonomies, run pilots, measure drift, then

348
00:19:45.920 --> 00:19:49.839
<v Speaker 1>expand only when behavior stabilizes. Step two centrally monitor meta

349
00:19:49.920 --> 00:19:53.079
<v Speaker 1>data changes. Every new column rule or AI generated view

350
00:19:53.079 --> 00:19:56.440
<v Speaker 1>should trigger a review process, preferably by someone who understands

351
00:19:56.440 --> 00:20:00.519
<v Speaker 1>that yes, no fields breed exponentially. Step three. Educate your

352
00:20:00.559 --> 00:20:03.720
<v Speaker 1>users not with the usual compliance slides, but with the

353
00:20:03.759 --> 00:20:06.839
<v Speaker 1>doctrine of approval discipline. Make it clear that every safe

354
00:20:06.920 --> 00:20:10.240
<v Speaker 1>change's button is an act of schema surgery, not mere housekeeping.

355
00:20:10.640 --> 00:20:13.200
<v Speaker 1>Do those three things, and the interurn becomes an ally

356
00:20:13.319 --> 00:20:15.680
<v Speaker 1>rather than an arsonist. Skip them, and you'll spend the

357
00:20:15.720 --> 00:20:19.240
<v Speaker 1>next fiscal year tracing phantom columns across audit logs. Remember

358
00:20:19.240 --> 00:20:24.200
<v Speaker 1>the underlying truth. Automation doesn't eliminate work, it displaces perception

359
00:20:24.279 --> 00:20:27.720
<v Speaker 1>of it. Hidden labor buried beneath a shinier dashboard is

360
00:20:27.720 --> 00:20:31.079
<v Speaker 1>still labor yours. So before you celebrate your newfound freedom

361
00:20:31.079 --> 00:20:34.400
<v Speaker 1>from metadata chores, check who's holding the mob Spoiler, it's

362
00:20:34.400 --> 00:20:37.160
<v Speaker 1>still you, supervising an algorithm with attention issues. If this

363
00:20:37.240 --> 00:20:41.519
<v Speaker 1>helped you keep your SharePoint Civilization intact, repay the favor, subscribe,

364
00:20:41.680 --> 00:20:45.599
<v Speaker 1>turn on notifications, and let future episodes deploy automatically structured

365
00:20:45.640 --> 00:20:48.519
<v Speaker 1>knowledge delivered on schedule, order not entropy proceed