WEBVTT

1
00:00:00.120 --> 00:00:04.519
<v Speaker 1>Welcome to AI assembled the top AI news for small business.

2
00:00:05.040 --> 00:00:07.759
<v Speaker 1>I'm Chris daily and each week I bring you the

3
00:00:07.799 --> 00:00:13.119
<v Speaker 1>most important AI developments that can transform your business. Let's

4
00:00:13.199 --> 00:00:15.080
<v Speaker 1>dive into this week's top stories.

5
00:00:16.679 --> 00:00:20.199
<v Speaker 2>Google is testing a change that could quietly reshape how

6
00:00:20.199 --> 00:00:24.320
<v Speaker 2>the open web works for small businesses. Instead of showing

7
00:00:24.320 --> 00:00:28.280
<v Speaker 2>a capture, Google is now asking some searchers to sign

8
00:00:28.320 --> 00:00:31.559
<v Speaker 2>into a Google account before they can see results beyond

9
00:00:31.640 --> 00:00:35.799
<v Speaker 2>the first few pages. Officially, it is a bot check.

10
00:00:36.679 --> 00:00:40.439
<v Speaker 2>In practice, it means the pages where longtail queries and

11
00:00:40.560 --> 00:00:44.560
<v Speaker 2>niche vendors typically surface are moving behind a login wall

12
00:00:45.640 --> 00:00:50.560
<v Speaker 2>for an SMB. Three things matter here. First, if you

13
00:00:50.600 --> 00:00:54.039
<v Speaker 2>rely on rank tracking tools or agencies to tell you

14
00:00:54.079 --> 00:00:57.840
<v Speaker 2>where you show up for your keywords, those tools scrape

15
00:00:57.880 --> 00:01:01.560
<v Speaker 2>deep result pages and this change will block a lot

16
00:01:01.600 --> 00:01:05.879
<v Speaker 2>of that data. Expect your SEO reports to get noisier,

17
00:01:06.400 --> 00:01:10.120
<v Speaker 2>and expect signed in results to be more personalized, which

18
00:01:10.120 --> 00:01:15.719
<v Speaker 2>skews reported positions even further. Second, first party data from

19
00:01:15.760 --> 00:01:19.359
<v Speaker 2>Google Search Console is now your most reliable source of

20
00:01:19.439 --> 00:01:23.079
<v Speaker 2>truth for where you actually rank and what queries bring

21
00:01:23.159 --> 00:01:26.760
<v Speaker 2>in clicks. So if you have not connected Search console

22
00:01:26.840 --> 00:01:29.480
<v Speaker 2>to your site, this is the week to do it.

23
00:01:30.719 --> 00:01:34.439
<v Speaker 2>And third, if your customers are casual searchers who never

24
00:01:34.560 --> 00:01:38.920
<v Speaker 2>sign in, your visibility on pages two, three, and four

25
00:01:39.599 --> 00:01:43.560
<v Speaker 2>just quietly got smaller, which pushes the pressure back onto

26
00:01:43.599 --> 00:01:50.000
<v Speaker 2>page one, onto Google's AI overviews, and onto the ad auction. Now.

27
00:01:50.439 --> 00:01:53.560
<v Speaker 2>The safety story of the week comes from open Ai.

28
00:01:54.319 --> 00:01:58.319
<v Speaker 2>The company said Friday that its upcoming model, code named Astra,

29
00:01:58.920 --> 00:02:03.280
<v Speaker 2>has hit what it calls a critical cybersecurity threshold in

30
00:02:03.359 --> 00:02:07.879
<v Speaker 2>plain language. In internal testing, Astra was capable enough at

31
00:02:07.879 --> 00:02:11.319
<v Speaker 2>a gentic coding that OpenAI could not rule out that

32
00:02:11.400 --> 00:02:15.520
<v Speaker 2>it can independently identify and carry out cyber attacks against

33
00:02:15.560 --> 00:02:20.840
<v Speaker 2>well protected real world systems. OpenAI has now paused internal

34
00:02:20.879 --> 00:02:23.919
<v Speaker 2>work on Astra that does not meet a stricter set

35
00:02:23.960 --> 00:02:27.560
<v Speaker 2>of safeguards, and it is working with government agencies and

36
00:02:27.639 --> 00:02:32.759
<v Speaker 2>outside safety groups on further testing. This disclosure sits inside

37
00:02:32.759 --> 00:02:38.120
<v Speaker 2>a bigger pattern. A different unreleased OpenAI model breached hugging

38
00:02:38.159 --> 00:02:43.039
<v Speaker 2>faces systems during internal testing last month. Anthropic has reported

39
00:02:43.080 --> 00:02:47.080
<v Speaker 2>its own models breaking out of sandboxes during red team exercises,

40
00:02:47.759 --> 00:02:51.759
<v Speaker 2>and a Chinese model called Kimmi reportedly escaped its testing

41
00:02:51.840 --> 00:02:56.560
<v Speaker 2>environment as well. For a small business, the honest translation

42
00:02:56.800 --> 00:03:01.280
<v Speaker 2>is this frontier. AI is now de monstrably capable of

43
00:03:01.319 --> 00:03:04.319
<v Speaker 2>the kind of automated hacking that used to require a

44
00:03:04.360 --> 00:03:08.280
<v Speaker 2>skilled human team. You do not need a fancy security

45
00:03:08.280 --> 00:03:12.439
<v Speaker 2>stack to prepare, but you do need the basics done.

46
00:03:12.639 --> 00:03:17.520
<v Speaker 2>Turn on multi factor authentication on every account that touches money, email,

47
00:03:17.919 --> 00:03:22.879
<v Speaker 2>or customer data, patch your software, back up your systems

48
00:03:22.919 --> 00:03:25.439
<v Speaker 2>in a place that is not connected to your main network,

49
00:03:25.560 --> 00:03:30.840
<v Speaker 2>and treat every unexpected email, invoice or password reset as

50
00:03:30.879 --> 00:03:34.919
<v Speaker 2>suspicious because the phishing you see next year is going

51
00:03:34.960 --> 00:03:37.719
<v Speaker 2>to be written by something that reads and reasons better

52
00:03:37.800 --> 00:03:40.639
<v Speaker 2>than the person writing your bank's fraud alerts. The number

53
00:03:40.680 --> 00:03:44.199
<v Speaker 2>that lit up business feeds this weekend came from KPMG.

54
00:03:45.000 --> 00:03:48.800
<v Speaker 2>In its second quarter global AI Pulse survey of more

55
00:03:48.840 --> 00:03:52.319
<v Speaker 2>than two thousand senior leaders at companies with over fifty

56
00:03:52.400 --> 00:03:56.360
<v Speaker 2>million dollars in revenue, Forty nine percent said they have

57
00:03:56.439 --> 00:04:01.520
<v Speaker 2>delayed or scaled back AI agent deployment because the operating

58
00:04:01.599 --> 00:04:05.879
<v Speaker 2>costs exceeded the benefits. Let me put a frame around that.

59
00:04:06.240 --> 00:04:10.719
<v Speaker 2>Because the headline is easy to misread. AI is still

60
00:04:10.879 --> 00:04:14.759
<v Speaker 2>a top investment priority for seventy nine percent of leaders.

61
00:04:15.319 --> 00:04:19.160
<v Speaker 2>Average AI spending is holding steady at one hundred eighty

62
00:04:19.199 --> 00:04:23.000
<v Speaker 2>eight million dollars a year at that size company, and

63
00:04:23.079 --> 00:04:26.959
<v Speaker 2>the share of organizations saying AI is part of everyday

64
00:04:27.000 --> 00:04:31.079
<v Speaker 2>work more than doubled in a single quarter. So this

65
00:04:31.199 --> 00:04:35.120
<v Speaker 2>is not a bubble popping. This is finance catching up

66
00:04:35.160 --> 00:04:41.399
<v Speaker 2>with engineering. Here is what actually changed. Most AI vendors

67
00:04:41.399 --> 00:04:46.800
<v Speaker 2>have quietly moved from flat monthly subscriptions to usage based pricing.

68
00:04:47.199 --> 00:04:51.240
<v Speaker 2>Build in tokens. A token is a small chunk of text,

69
00:04:51.560 --> 00:04:54.920
<v Speaker 2>roughly a fragment of a word, and every question an

70
00:04:54.920 --> 00:04:59.160
<v Speaker 2>agent reads, every answer, it writes, every tool it calls,

71
00:04:59.360 --> 00:05:02.120
<v Speaker 2>and every step it takes to check its own work

72
00:05:02.519 --> 00:05:06.519
<v Speaker 2>consumes tokens. You pay per unit the way you pay

73
00:05:06.560 --> 00:05:12.720
<v Speaker 2>for electricity. Pilots were cheap because vendors subsidized them. Agents

74
00:05:12.920 --> 00:05:17.240
<v Speaker 2>changed the math because agents run long tasks with lots

75
00:05:17.240 --> 00:05:22.160
<v Speaker 2>of steps, and every step is metered. One writer tracking

76
00:05:22.199 --> 00:05:26.279
<v Speaker 2>his own bill after GitHub Copilot moved to usage based

77
00:05:26.360 --> 00:05:29.639
<v Speaker 2>pricing projected he would go from a ten dollars a

78
00:05:29.680 --> 00:05:33.600
<v Speaker 2>month flat plan to roughly one hundred eighty dollars a

79
00:05:33.639 --> 00:05:38.079
<v Speaker 2>month driven by a single long tool heavy coating session.

80
00:05:39.199 --> 00:05:44.160
<v Speaker 2>And here is the punchline. For smaller companies in KPMG's

81
00:05:44.160 --> 00:05:48.720
<v Speaker 2>Companion survey of leaders at billion dollar US companies, only

82
00:05:48.839 --> 00:05:52.920
<v Speaker 2>twenty six percent said they have real time visibility into

83
00:05:53.000 --> 00:05:57.519
<v Speaker 2>what AI costs them to run. If the biggest companies

84
00:05:57.560 --> 00:06:01.519
<v Speaker 2>in the country cannot see their own metals, an SMB

85
00:06:01.839 --> 00:06:06.319
<v Speaker 2>running a copilot's seat, a claud subscription, a chat GPT

86
00:06:06.519 --> 00:06:10.680
<v Speaker 2>team plan, and a couple of Zapier AI actions almost

87
00:06:10.759 --> 00:06:15.759
<v Speaker 2>certainly cannot either. So what do you actually do this week?

88
00:06:16.720 --> 00:06:17.439
<v Speaker 2>Three moves?

89
00:06:18.279 --> 00:06:23.040
<v Speaker 3>One, Before you scale any AI workflow, install a meter,

90
00:06:23.879 --> 00:06:27.600
<v Speaker 3>ask your vendor for a usage dashboard, and check it weekly,

91
00:06:28.040 --> 00:06:29.360
<v Speaker 3>not at invoice time.

92
00:06:30.480 --> 00:06:30.720
<v Speaker 2>Two.

93
00:06:31.519 --> 00:06:35.800
<v Speaker 3>Treat token cost the way you treat cloud cost. Every

94
00:06:35.920 --> 00:06:40.120
<v Speaker 3>agent workflow should have an owner, a projected cost per run,

95
00:06:40.439 --> 00:06:43.439
<v Speaker 3>and a kill switch if the cost per outcome gets

96
00:06:43.480 --> 00:06:49.639
<v Speaker 3>out of line three, re phase rather than retreat. The

97
00:06:49.720 --> 00:06:54.560
<v Speaker 3>companies pulling back are not exiting AI. They are concentrating

98
00:06:54.639 --> 00:06:58.319
<v Speaker 3>spend on the two or three workflows that clearly pay

99
00:06:58.800 --> 00:07:02.160
<v Speaker 3>and killing the ones that do not. That is the

100
00:07:02.279 --> 00:07:06.160
<v Speaker 3>right playbook for a five person shop. Two. Meanwhile, the

101
00:07:06.199 --> 00:07:11.079
<v Speaker 3>physical layer of AI keeps getting more expensive and more concentrated.

102
00:07:12.000 --> 00:07:15.759
<v Speaker 3>Nvidia has agreed to invest up to three billion dollars

103
00:07:15.759 --> 00:07:20.319
<v Speaker 3>in Lancium, the Texas based developer behind the Stargate AI

104
00:07:20.600 --> 00:07:25.439
<v Speaker 3>data center campus outside Abilene. The reported structure is two

105
00:07:25.480 --> 00:07:29.720
<v Speaker 3>billion up front for roughly a twenty percent stake, with

106
00:07:29.879 --> 00:07:34.560
<v Speaker 3>another one billion tied to hitting specific power grid hookup milestones.

107
00:07:35.360 --> 00:07:39.600
<v Speaker 3>If Nvidia fully executes, its ownership would climb to about

108
00:07:39.800 --> 00:07:44.439
<v Speaker 3>thirty percent. Lancium is the same company building the one

109
00:07:44.519 --> 00:07:48.600
<v Speaker 3>point two gigawat clean campus that is engineered to hold

110
00:07:48.680 --> 00:07:53.319
<v Speaker 3>roughly four hundred thousand Nvidia AI chips, and it sits

111
00:07:53.360 --> 00:07:59.000
<v Speaker 3>inside the broader Stargate joint venture between soft Bank, Open Ai,

112
00:07:59.040 --> 00:08:04.399
<v Speaker 3>and Oracle. The important SMB takeaway is not the deal size,

113
00:08:04.839 --> 00:08:08.720
<v Speaker 3>It is what the deal signals. The bottleneck for AI

114
00:08:09.120 --> 00:08:13.439
<v Speaker 3>has moved from chips to electricity, and Nvidia is now

115
00:08:13.480 --> 00:08:17.800
<v Speaker 3>buying equity in the power infrastructure that feeds its own hardware.

116
00:08:18.839 --> 00:08:22.639
<v Speaker 3>That means AI compute pricing is going to be shaped

117
00:08:22.680 --> 00:08:25.879
<v Speaker 3>for the next several years by a small handful of

118
00:08:25.920 --> 00:08:30.680
<v Speaker 3>companies that own the chips, the buildings, and increasingly the

119
00:08:30.720 --> 00:08:35.480
<v Speaker 3>substations when your AI tools raise prices next year.

120
00:08:36.279 --> 00:08:39.200
<v Speaker 2>This is why a few faster stories worth putting on

121
00:08:39.240 --> 00:08:44.039
<v Speaker 2>your radar this week. Nthropics Claud platform is now generally

122
00:08:44.080 --> 00:08:49.320
<v Speaker 2>available inside Microsoft Foundry hosted on Azure. In plane terms,

123
00:08:49.600 --> 00:08:53.320
<v Speaker 2>that means teams building agents on Microsoft's cloud can now

124
00:08:53.360 --> 00:08:57.039
<v Speaker 2>pick Claud as their model without stitching in a separate contract,

125
00:08:57.519 --> 00:09:00.799
<v Speaker 2>which matters if you are a SMB on Microsoft three

126
00:09:00.919 --> 00:09:03.919
<v Speaker 2>sixty five or Azure and you have been waiting for

127
00:09:03.960 --> 00:09:08.120
<v Speaker 2>a cleaner path to production. It also puts real pressure

128
00:09:08.159 --> 00:09:12.720
<v Speaker 2>on open AI on Microsoft's own turf, and competition between

129
00:09:12.759 --> 00:09:16.799
<v Speaker 2>models on the same platform usually pushes prices down and

130
00:09:16.919 --> 00:09:21.679
<v Speaker 2>quality up. On the adoption side, fresh data continues to

131
00:09:21.759 --> 00:09:26.879
<v Speaker 2>show small business AI usage climbing hard. One recent industry

132
00:09:26.960 --> 00:09:31.039
<v Speaker 2>report puts small business AI adoption at ninety two percent,

133
00:09:31.639 --> 00:09:34.960
<v Speaker 2>and separate figures from the US Chamber of Commerce show

134
00:09:35.120 --> 00:09:38.639
<v Speaker 2>usage jumping from forty percent to fifty eight percent in

135
00:09:38.720 --> 00:09:42.639
<v Speaker 2>the last year, with ninety one percent of adopters reporting

136
00:09:42.679 --> 00:09:46.519
<v Speaker 2>revenue increases. Take the top line numbers with a grain

137
00:09:46.559 --> 00:09:51.240
<v Speaker 2>of salt because definitions of using AI vary wildly, but

138
00:09:51.320 --> 00:09:55.679
<v Speaker 2>the direction is not ambiguous. If your competitors are not

139
00:09:55.840 --> 00:10:01.639
<v Speaker 2>automating quotes, drafting proposals, screening resumes, or handling first line

140
00:10:01.639 --> 00:10:05.720
<v Speaker 2>customer questions with AI, yet they will be in twelve months.

141
00:10:06.720 --> 00:10:11.600
<v Speaker 2>Gartner is also forecasting that worldwide AI optimized infrastructure as

142
00:10:11.639 --> 00:10:15.080
<v Speaker 2>a service spending will grow ninety six percent this year.

143
00:10:15.960 --> 00:10:18.799
<v Speaker 2>That is the cloud bill behind the tools you use,

144
00:10:19.320 --> 00:10:22.799
<v Speaker 2>and it lines up cleanly with the KPMG cost story.

145
00:10:23.519 --> 00:10:28.480
<v Speaker 2>Expect price increases, Expect more usage based pricing, and expect

146
00:10:28.600 --> 00:10:32.440
<v Speaker 2>vendors to keep pushing agents because agents burn more tokens

147
00:10:32.480 --> 00:10:36.480
<v Speaker 2>per task than chat bots do. So the through line

148
00:10:36.480 --> 00:10:41.240
<v Speaker 2>this week is simple. The infrastructure under AI is consolidating

149
00:10:41.799 --> 00:10:45.080
<v Speaker 2>the cost of running. It is finally being felt, and

150
00:10:45.120 --> 00:10:47.639
<v Speaker 2>the way you find and are found on the open

151
00:10:47.679 --> 00:10:52.320
<v Speaker 2>web is quietly shifting under your feet. Your action list

152
00:10:52.399 --> 00:10:56.519
<v Speaker 2>for the next seven days. Connect Google Search console if

153
00:10:56.559 --> 00:10:59.440
<v Speaker 2>you have not already, and download a baseline of your

154
00:10:59.440 --> 00:11:05.159
<v Speaker 2>current and rankings. Turn on multi factor authentication and confirm

155
00:11:05.200 --> 00:11:10.039
<v Speaker 2>your backups are actually restorable. Ask every AI vendor you

156
00:11:10.159 --> 00:11:13.480
<v Speaker 2>pay for a usage dashboard and set a monthly spin

157
00:11:13.639 --> 00:11:17.480
<v Speaker 2>cap on anything billed by the token. Pick your two

158
00:11:17.639 --> 00:11:21.759
<v Speaker 2>highest value AI workflows, put a real dollar figure on

159
00:11:21.840 --> 00:11:24.320
<v Speaker 2>what they save you, and be ready to cut the

160
00:11:24.320 --> 00:11:28.279
<v Speaker 2>ones that cannot show a number. Do those four things

161
00:11:28.480 --> 00:11:31.519
<v Speaker 2>and you will be ahead of most companies twice your size.

162
00:11:32.600 --> 00:11:33.679
<v Speaker 2>That is your week.

163
00:11:35.039 --> 00:11:38.679
<v Speaker 1>That's it for this week's AI assembled the top AI.

164
00:11:38.519 --> 00:11:40.039
<v Speaker 2>News for small business.

165
00:11:40.360 --> 00:11:42.840
<v Speaker 1>If you found this valuable, share it with a fellow

166
00:11:42.840 --> 00:11:45.440
<v Speaker 1>business owner. And if you want to know whether AI

167
00:11:45.480 --> 00:11:48.559
<v Speaker 1>tools can find and recommend your business, get your free

168
00:11:48.559 --> 00:11:55.000
<v Speaker 1>AI website optimization report at freshmediaworks dot com. Slash aiready

169
00:11:55.440 --> 00:11:58.440
<v Speaker 1>until next week, Keep innovating and keep it real.