1
00:00:02,880 --> 00:00:04,440
Speaker 1: Agentic ai.

2
00:00:04,840 --> 00:00:07,119
Speaker 2: In the last couple of years, it's gone from buzzword

3
00:00:07,200 --> 00:00:09,800
Speaker 2: to something pretty much any company that wants to stay

4
00:00:09,840 --> 00:00:13,840
Speaker 2: competitive is trying to implement in some capacity. But this

5
00:00:14,000 --> 00:00:16,920
Speaker 2: week on the Business of Tech powered by Two Degrees,

6
00:00:17,480 --> 00:00:21,280
Speaker 2: we're looking at a major new development in agentic ai.

7
00:00:21,480 --> 00:00:24,680
Speaker 2: It's called open Claw. We've only really heard about it

8
00:00:24,760 --> 00:00:28,240
Speaker 2: since January, that's how new this thing is, but many

9
00:00:28,280 --> 00:00:32,120
Speaker 2: people think it is the next phase of agentic ai.

10
00:00:33,000 --> 00:00:36,120
Speaker 2: We're going to talk to software developer and startup founder

11
00:00:36,240 --> 00:00:38,879
Speaker 2: Mike Hall. Mike is one of the smartest guys in

12
00:00:38,880 --> 00:00:42,760
Speaker 2: the New Zealand tech community, focusing on artificial intelligence. You

13
00:00:42,840 --> 00:00:46,720
Speaker 2: really should follow his LinkedIn updates. He's a sharp mind.

14
00:00:47,240 --> 00:00:51,680
Speaker 2: He's got deep experience in software development, more recently building

15
00:00:51,720 --> 00:00:55,240
Speaker 2: his own AI startup, a bot which is in stealth mode.

16
00:00:55,240 --> 00:00:58,160
Speaker 1: But he's going to tell us a bit about that now.

17
00:00:58,360 --> 00:01:00,680
Speaker 1: Openclaw took the raditional.

18
00:01:00,480 --> 00:01:03,640
Speaker 2: Agent model, gave it tools and the ability to wake

19
00:01:03,720 --> 00:01:06,920
Speaker 2: itself up to check for new tasks. It was then

20
00:01:07,040 --> 00:01:12,039
Speaker 2: unleashed as the fastest growing open source project in GitHub history.

21
00:01:12,080 --> 00:01:15,720
Speaker 2: That's the repository for software code used by millions of

22
00:01:15,760 --> 00:01:16,559
Speaker 2: developers all.

23
00:01:16,520 --> 00:01:17,160
Speaker 1: Over the world.

24
00:01:17,959 --> 00:01:22,200
Speaker 2: The real innovation, Mike tells us, isn't just autonomy, but

25
00:01:22,440 --> 00:01:26,160
Speaker 2: the shared skills library and sort of hive mind of

26
00:01:26,280 --> 00:01:29,800
Speaker 2: agents that learn from their mistakes and push those learnings

27
00:01:29,880 --> 00:01:33,280
Speaker 2: back into the ecosystem, so the agents get smarter and

28
00:01:33,319 --> 00:01:37,240
Speaker 2: smarter and more functional as this hive mind grows.

29
00:01:37,800 --> 00:01:40,480
Speaker 1: We can talk about open Claw without talking.

30
00:01:40,160 --> 00:01:44,360
Speaker 2: About malt Book, the Reddit style social network where more

31
00:01:44,400 --> 00:01:50,120
Speaker 2: than a million AI agents, many powered by Openclaw are posting, replying,

32
00:01:50,160 --> 00:01:54,560
Speaker 2: and in some cases sounding suspiciously like humans. Actually some

33
00:01:54,800 --> 00:01:59,560
Speaker 2: indeed were humans masquerading as bots, but Moultbook tells us

34
00:01:59,600 --> 00:02:04,480
Speaker 2: a lot about authenticity influenced campaigns in the coming wave

35
00:02:04,600 --> 00:02:09,120
Speaker 2: of AI generated content that looks and feels early. Human

36
00:02:09,919 --> 00:02:13,520
Speaker 2: security is a big theme of today's conversation. It's a

37
00:02:13,600 --> 00:02:16,640
Speaker 2: major issue when it comes to Openclaw and mault Book.

38
00:02:16,680 --> 00:02:20,600
Speaker 2: If you're pointing agents at sensitive business systems, you need

39
00:02:20,680 --> 00:02:26,560
Speaker 2: guide rails, isolation, fine grained permissions to prevent AI agents

40
00:02:26,639 --> 00:02:30,359
Speaker 2: from potentially going rogue and doing a lot of damage.

41
00:02:30,880 --> 00:02:33,679
Speaker 2: Mike's view I found this really interesting is that we're

42
00:02:33,680 --> 00:02:36,880
Speaker 2: in a sort of COVID style early phase of AI

43
00:02:36,960 --> 00:02:41,600
Speaker 2: driven automation. We can see the disruption happening elsewhere, for instance,

44
00:02:41,600 --> 00:02:44,960
Speaker 2: in software engineering at the moment, but it will quickly

45
00:02:45,000 --> 00:02:49,480
Speaker 2: spread to admin, government, marketing and beyond, just like the

46
00:02:49,560 --> 00:02:54,720
Speaker 2: virus eventually landed on our shores. Interestingly, the founder of

47
00:02:54,880 --> 00:02:58,760
Speaker 2: open claw, Peter Steinberger, was this week snapped up by

48
00:02:59,440 --> 00:03:03,520
Speaker 2: open ais ai, the creator of chat GPT. According to

49
00:03:03,880 --> 00:03:08,840
Speaker 2: open ai founder Sam Altman, Steinberger will drive the next

50
00:03:08,960 --> 00:03:13,880
Speaker 2: generation of personal agents. Open Claw will carry on as

51
00:03:13,919 --> 00:03:18,120
Speaker 2: an open source foundation supported by open ai. Apparently so,

52
00:03:18,200 --> 00:03:20,960
Speaker 2: one of the leading players in AI clearly sees the

53
00:03:21,080 --> 00:03:25,240
Speaker 2: open claw approach as the future of agentic AI lots

54
00:03:25,280 --> 00:03:27,520
Speaker 2: of grand to cover, So sit back and listen to

55
00:03:27,560 --> 00:03:31,959
Speaker 2: my conversation with software developer and AI entrepreneur Mike Hall.

56
00:03:38,200 --> 00:03:40,640
Speaker 2: Mike Hall, Welcome to the Business of Tech. How are

57
00:03:40,640 --> 00:03:41,920
Speaker 2: you doing a beta?

58
00:03:41,960 --> 00:03:43,720
Speaker 3: I'm doing great. Thanks for inviting me on the show.

59
00:03:43,840 --> 00:03:45,040
Speaker 1: Well, thanks for coming on.

60
00:03:45,160 --> 00:03:48,200
Speaker 2: We've sort of known each other online and through the

61
00:03:48,240 --> 00:03:51,240
Speaker 2: tech community for quite a while, we haven't actually met,

62
00:03:51,400 --> 00:03:54,400
Speaker 2: and I really thought this topic we're about to get into,

63
00:03:54,440 --> 00:03:56,520
Speaker 2: you'd be ideal for it. You've got a long history

64
00:03:56,560 --> 00:04:02,840
Speaker 2: and software development and interesting and investing in tech ventures.

65
00:04:02,840 --> 00:04:06,040
Speaker 2: More laterally, really gone deep on AI just for the

66
00:04:06,080 --> 00:04:09,200
Speaker 2: benefit of listeners, just give us a bit of the

67
00:04:09,240 --> 00:04:10,480
Speaker 2: spiel about your background.

68
00:04:10,480 --> 00:04:10,640
Speaker 1: Mike.

69
00:04:10,800 --> 00:04:13,119
Speaker 3: Yeah, so I'm pretty well known in the AI community,

70
00:04:13,160 --> 00:04:17,080
Speaker 3: as you say, online, and I have been in this

71
00:04:17,680 --> 00:04:20,440
Speaker 3: agent space for a few years now, and the things

72
00:04:20,480 --> 00:04:24,160
Speaker 3: that we're creating at things that exist now. So we've

73
00:04:24,160 --> 00:04:27,560
Speaker 3: had these things for a while and we've been building them,

74
00:04:27,760 --> 00:04:30,040
Speaker 3: and we've been thinking about what the future is and

75
00:04:30,080 --> 00:04:33,800
Speaker 3: what the future is of automated companies, how you can

76
00:04:34,120 --> 00:04:37,359
Speaker 3: kind of scale these agents out and how to do

77
00:04:37,400 --> 00:04:41,800
Speaker 3: that in a safe way. So essentially, yeah, that's what

78
00:04:41,839 --> 00:04:44,760
Speaker 3: we've been doing, and I've been promoting what I've been

79
00:04:44,760 --> 00:04:47,359
Speaker 3: doing online a little bit. In stealth I kind of

80
00:04:47,360 --> 00:04:50,440
Speaker 3: talk about some tangential things just because we're not ready

81
00:04:50,440 --> 00:04:51,760
Speaker 3: to talk about what we're actually doing.

82
00:04:51,880 --> 00:04:54,400
Speaker 2: Yeah, we'll get to your sort of startup that is

83
00:04:54,440 --> 00:04:57,440
Speaker 2: in stealth mode and talk about what you can about

84
00:04:57,440 --> 00:05:00,440
Speaker 2: it at the moment. But just to backtrack little bit,

85
00:05:00,440 --> 00:05:04,160
Speaker 2: we've heard of agentic AI really in the last couple

86
00:05:04,200 --> 00:05:07,440
Speaker 2: of years. For me, the big aha moment was when

87
00:05:07,440 --> 00:05:10,599
Speaker 2: I was at the Salesforce conference and that Mark benning

88
00:05:10,640 --> 00:05:13,120
Speaker 2: Off there was talking about pivoting the entire company. This

89
00:05:13,240 --> 00:05:18,560
Speaker 2: big software company that does customer relationship management software pivoting

90
00:05:18,600 --> 00:05:21,640
Speaker 2: to agentic ai. That was a couple of years ago.

91
00:05:21,720 --> 00:05:23,680
Speaker 2: Last year was really the year, I think where it

92
00:05:23,839 --> 00:05:28,000
Speaker 2: went mainstream. You saw the big software vendors building agents

93
00:05:28,080 --> 00:05:31,200
Speaker 2: into their platforms. You saw local startups like auto Hive

94
00:05:31,279 --> 00:05:36,120
Speaker 2: here in New Zealand building agentic platforms. Then we get

95
00:05:36,120 --> 00:05:39,480
Speaker 2: to January twenty twenty six and this thing, open claw

96
00:05:39,520 --> 00:05:42,719
Speaker 2: seemed to explode out of nowhere, this sort of open

97
00:05:42,760 --> 00:05:47,080
Speaker 2: source agentic ai assistant that you can run locally on

98
00:05:47,120 --> 00:05:51,440
Speaker 2: a computer. Take us through from your perspective, Mike, what

99
00:05:51,720 --> 00:05:55,880
Speaker 2: open claw is. It was originally called Clawbot, had to

100
00:05:55,960 --> 00:05:59,640
Speaker 2: change its name because of being too close to Claude,

101
00:05:59,720 --> 00:06:03,239
Speaker 2: the big AI company. Where did open claw come from?

102
00:06:03,360 --> 00:06:06,120
Speaker 3: Yes, So essentially, I think it's good to step back

103
00:06:06,160 --> 00:06:09,000
Speaker 3: a little bit. So before that, we had these things

104
00:06:09,040 --> 00:06:11,880
Speaker 3: called chatbots and I'm really speaking to the choir hair.

105
00:06:11,960 --> 00:06:14,599
Speaker 3: Everyone kind of knows this, but you talk to it

106
00:06:14,640 --> 00:06:17,640
Speaker 3: and it responds. And then what happens is we gave

107
00:06:17,720 --> 00:06:21,920
Speaker 3: these chatbots access to the internet, access to tools, and

108
00:06:21,960 --> 00:06:24,000
Speaker 3: that made them more useful. So if I talk to

109
00:06:24,040 --> 00:06:25,640
Speaker 3: it and I say, hey, can you please edit this

110
00:06:25,720 --> 00:06:28,880
Speaker 3: document or hey, can you please order me an uber?

111
00:06:29,120 --> 00:06:33,560
Speaker 3: It has access to those tools through connections called APIs,

112
00:06:34,480 --> 00:06:37,960
Speaker 3: and that's where we were. And the next level of

113
00:06:38,000 --> 00:06:42,080
Speaker 3: that is where open claw came, and it said, let's

114
00:06:42,120 --> 00:06:45,039
Speaker 3: have access to tools, but let's also be able to

115
00:06:45,040 --> 00:06:49,640
Speaker 3: wake myself up. So open claw can set timers they're

116
00:06:49,640 --> 00:06:54,120
Speaker 3: called cron jobs, and it wakes itself up autonomously. So

117
00:06:54,160 --> 00:06:58,400
Speaker 3: it has a heartbeat timer that goes off every say minute,

118
00:06:58,640 --> 00:07:00,960
Speaker 3: and it says should I do something? Should I do something?

119
00:07:01,000 --> 00:07:03,840
Speaker 3: Should I do something? And it has access to all

120
00:07:03,880 --> 00:07:06,680
Speaker 3: your data, so it can kind of see Mike sent

121
00:07:06,760 --> 00:07:10,440
Speaker 3: an email or John sent an email, or that something happened.

122
00:07:10,480 --> 00:07:13,720
Speaker 3: Should I do something? And that is really the innovation

123
00:07:13,840 --> 00:07:17,680
Speaker 3: that allowed open floor to get so popular. They are

124
00:07:17,800 --> 00:07:22,160
Speaker 3: the fastest growing open source project in gethub history. Gethub

125
00:07:22,240 --> 00:07:26,240
Speaker 3: is a platform where people put their code and just

126
00:07:26,600 --> 00:07:31,560
Speaker 3: by that metric that shows the popularity of this tool.

127
00:07:31,760 --> 00:07:34,560
Speaker 3: It is not like a chatbot where you talk to

128
00:07:34,600 --> 00:07:38,920
Speaker 3: it and it responds. It can send you messages autonomously,

129
00:07:39,520 --> 00:07:42,440
Speaker 3: and that's kind of what makes it seem more human.

130
00:07:42,480 --> 00:07:43,240
Speaker 1: It's incredible.

131
00:07:43,680 --> 00:07:48,840
Speaker 2: Developed by this Austrian software developer Peter Steinberger was he

132
00:07:48,960 --> 00:07:52,000
Speaker 2: on the radar of you or the software community in general.

133
00:07:52,480 --> 00:07:54,679
Speaker 3: I know he wasn't. He's pretty well known on Twitter

134
00:07:54,920 --> 00:07:57,680
Speaker 3: as I hear. I watched the Lex Frieban podcast with him.

135
00:07:58,440 --> 00:08:02,680
Speaker 3: So essentially he was building a lot of stuff. So

136
00:08:02,720 --> 00:08:08,440
Speaker 3: he made pdf kit pspdf kit, which is software that

137
00:08:08,600 --> 00:08:12,080
Speaker 3: allows PDFs to display. Apparently it's on a billion devices,

138
00:08:12,120 --> 00:08:15,360
Speaker 3: so it's quite a large company that he made and sold,

139
00:08:15,400 --> 00:08:18,080
Speaker 3: and he went dark for a few years and came

140
00:08:18,120 --> 00:08:21,280
Speaker 3: back in this agent world and he was just playing

141
00:08:21,360 --> 00:08:26,320
Speaker 3: with this thing doing experiments. Essentially, he was giving the

142
00:08:26,800 --> 00:08:31,560
Speaker 3: claude bot access to rewrite its own code. So what

143
00:08:31,680 --> 00:08:34,079
Speaker 3: happens when you give AI access to rewrite its own code?

144
00:08:34,640 --> 00:08:38,360
Speaker 3: It just wants to improve itself constantly. So we've played

145
00:08:38,360 --> 00:08:41,520
Speaker 3: with this as well, and it can go into interesting

146
00:08:41,559 --> 00:08:44,480
Speaker 3: directions if you don't give it guardrails. But essentially, if

147
00:08:44,520 --> 00:08:46,560
Speaker 3: it has access to its own code, it can rewrite

148
00:08:46,600 --> 00:08:49,360
Speaker 3: its own code, and if you have the right plugins

149
00:08:49,400 --> 00:08:52,240
Speaker 3: and adapter layers on top of that, it makes an

150
00:08:52,240 --> 00:08:55,600
Speaker 3: AI that is self learning. And this is kind of

151
00:08:55,640 --> 00:09:01,240
Speaker 3: the thing that the tech pessimists are concerned about out overall.

152
00:09:01,640 --> 00:09:06,680
Speaker 3: Not talking about claud bodies, the takeoff moment for AI

153
00:09:06,920 --> 00:09:09,960
Speaker 3: where it starts writing its own code and it starts

154
00:09:10,000 --> 00:09:12,920
Speaker 3: doing things that you have no idea what it's doing.

155
00:09:13,120 --> 00:09:17,480
Speaker 2: So we have seen these sort of AI agents become

156
00:09:17,800 --> 00:09:23,200
Speaker 2: more available. They can do multi step workflows autonomously. That's

157
00:09:23,240 --> 00:09:26,000
Speaker 2: the big selling point is it can automate away a

158
00:09:26,000 --> 00:09:30,160
Speaker 2: lot of these mundane sort of admin tasks. I guess

159
00:09:29,960 --> 00:09:32,360
Speaker 2: that the difference here. You know, most people are buying

160
00:09:32,600 --> 00:09:37,960
Speaker 2: that on a platform like with from Microsoft or from Salesforce.

161
00:09:38,440 --> 00:09:41,360
Speaker 2: This is open source. This is a piece of software

162
00:09:41,360 --> 00:09:44,680
Speaker 2: you can download and run locally on your own hardware.

163
00:09:44,800 --> 00:09:47,560
Speaker 2: That's why we've seen, you know, a huge surge and

164
00:09:47,600 --> 00:09:49,680
Speaker 2: demand from mac Mini's. A lot of people are buying

165
00:09:49,760 --> 00:09:54,079
Speaker 2: them to run open claw natively at home.

166
00:09:54,640 --> 00:09:55,440
Speaker 1: Is that a big shift?

167
00:09:55,440 --> 00:09:57,520
Speaker 2: Obviously open source has been around for a long time,

168
00:09:57,559 --> 00:10:02,040
Speaker 2: but something that go this big an agentic AI. We

169
00:10:02,080 --> 00:10:05,040
Speaker 2: haven't really seen that before in open source software, have.

170
00:10:05,000 --> 00:10:06,080
Speaker 1: We No, we haven't.

171
00:10:06,080 --> 00:10:09,440
Speaker 3: But the thing when you're running if you want to

172
00:10:09,480 --> 00:10:12,080
Speaker 3: try this out, I would just looking for a company

173
00:10:12,080 --> 00:10:16,440
Speaker 3: that provides sandboxes. And the sandbox is just an isolated space.

174
00:10:16,480 --> 00:10:18,840
Speaker 3: It's an isolated computer. So you could use an old

175
00:10:19,280 --> 00:10:22,800
Speaker 3: laptop or you could buy something new. Our company actually

176
00:10:24,120 --> 00:10:26,480
Speaker 3: has these sandboxes that are safe. So if you find

177
00:10:26,480 --> 00:10:30,280
Speaker 3: a company like that, you set it up and it

178
00:10:30,360 --> 00:10:33,560
Speaker 3: just allows it to have the access that you want

179
00:10:33,600 --> 00:10:34,880
Speaker 3: it to have. So if you sit it up on

180
00:10:34,920 --> 00:10:38,480
Speaker 3: a Mac Mini, for example, it can't see the emails

181
00:10:38,559 --> 00:10:42,320
Speaker 3: with your wife, it can't have access to your CV

182
00:10:42,720 --> 00:10:45,520
Speaker 3: or maybe like you've got some biometric data or some

183
00:10:45,559 --> 00:10:47,360
Speaker 3: photos of your kids or something that you just don't

184
00:10:47,400 --> 00:10:51,040
Speaker 3: want it to have. You can separate it out. Or

185
00:10:51,080 --> 00:10:53,000
Speaker 3: if you just don't want it having access to your

186
00:10:53,040 --> 00:10:55,559
Speaker 3: work stuff and just only your personal stuff, you can

187
00:10:55,720 --> 00:10:59,680
Speaker 3: create these boundaries. And that's what sandboxes are. And that's

188
00:10:59,760 --> 00:11:02,920
Speaker 3: why people are buying these macchmenis because this is an

189
00:11:02,960 --> 00:11:07,600
Speaker 3: open source tool. It's kind of haphazardly put together without

190
00:11:08,120 --> 00:11:11,800
Speaker 3: these security guard rails that most people expect of large

191
00:11:11,880 --> 00:11:15,680
Speaker 3: established companies. So yeah, that's kind of the trend here

192
00:11:15,800 --> 00:11:21,199
Speaker 3: is you can set it up locally. You could have

193
00:11:21,240 --> 00:11:24,320
Speaker 3: a Raspberry Pie that would work as well. People are

194
00:11:24,320 --> 00:11:28,679
Speaker 3: buying machmeneis because you can run AI software like machine

195
00:11:28,760 --> 00:11:30,840
Speaker 3: learning stuff on those as well, so you get like

196
00:11:30,840 --> 00:11:31,560
Speaker 3: the dual benefit.

197
00:11:31,720 --> 00:11:34,760
Speaker 2: And yeah, the big benefit there I guess is from

198
00:11:34,760 --> 00:11:38,079
Speaker 2: a cost perspective. If you have a powerful enough device

199
00:11:38,120 --> 00:11:41,320
Speaker 2: at home to run this, you're not paying those you know,

200
00:11:41,600 --> 00:11:44,800
Speaker 2: the tokens, the subscription fees that you'd pay to an

201
00:11:44,840 --> 00:11:48,920
Speaker 2: open AI or a Google, Gemini or whoever's doing dedicated

202
00:11:48,960 --> 00:11:50,160
Speaker 2: AI agent staff.

203
00:11:50,520 --> 00:11:53,280
Speaker 1: There's going to be a huge sort of cost saving there.

204
00:11:53,920 --> 00:11:56,000
Speaker 3: Yeah, you would think so, but not really. I've got

205
00:11:56,040 --> 00:11:58,880
Speaker 3: a twenty thousand dollars computer sitting right here, and I

206
00:11:58,920 --> 00:12:01,600
Speaker 3: can barely do anything on that in the AI space.

207
00:12:01,800 --> 00:12:04,360
Speaker 3: So back when I bought it for machine learning, I

208
00:12:04,400 --> 00:12:06,959
Speaker 3: could do everything. I could do everything machine learning related.

209
00:12:07,559 --> 00:12:12,920
Speaker 3: But with these AI large clusters of compute that you need,

210
00:12:13,280 --> 00:12:15,680
Speaker 3: you need about a two hundred thousand dollars computer to

211
00:12:15,760 --> 00:12:18,680
Speaker 3: have anything remotely useful. So there is a lot of

212
00:12:18,720 --> 00:12:22,720
Speaker 3: open source llms that you can use. Kimmi K two

213
00:12:22,760 --> 00:12:26,120
Speaker 3: comes to mind, and these are really good, but you

214
00:12:26,320 --> 00:12:29,719
Speaker 3: need a lot of GPU, like way more than the

215
00:12:29,840 --> 00:12:32,199
Speaker 3: normal person would have and way more than most companies

216
00:12:32,200 --> 00:12:33,640
Speaker 3: in New Zealand even have.

217
00:12:33,720 --> 00:12:35,959
Speaker 2: The sort of stuff that people are doing on their

218
00:12:36,000 --> 00:12:39,760
Speaker 2: Mac mini's at home. We're talking about sort of monitoring email,

219
00:12:40,320 --> 00:12:44,040
Speaker 2: maybe plugging into sas software and service tools that sort

220
00:12:44,080 --> 00:12:44,319
Speaker 2: of thing.

221
00:12:44,440 --> 00:12:47,640
Speaker 3: Yeah, they probably have voice stuff as well, and they

222
00:12:47,640 --> 00:12:50,280
Speaker 3: probably have video processing, so they're connected up to their

223
00:12:50,600 --> 00:12:52,640
Speaker 3: A lot of the macmni stuff I saw was connected

224
00:12:52,720 --> 00:12:54,800
Speaker 3: up to their like Google Home or like their open

225
00:12:54,840 --> 00:12:58,960
Speaker 3: home automation. Have you used it yourself, yeah, of the

226
00:12:59,120 --> 00:13:01,640
Speaker 3: open claw Yeah yeah, yeah, So I set up an

227
00:13:01,679 --> 00:13:04,160
Speaker 3: instance in a sandbox to see what it was about.

228
00:13:04,679 --> 00:13:08,560
Speaker 3: It's pretty similar to what we have. The main difference

229
00:13:08,640 --> 00:13:12,000
Speaker 3: that actually the key innovation was there's this library of

230
00:13:12,120 --> 00:13:17,680
Speaker 3: skills and the community of bots can add to these skills,

231
00:13:17,800 --> 00:13:19,920
Speaker 3: and there's a lot of malware in there. But the

232
00:13:20,200 --> 00:13:25,679
Speaker 3: innovation is really you have a hive mind of agents

233
00:13:25,720 --> 00:13:31,280
Speaker 3: all creating software and all creating knowledge and putting it

234
00:13:31,320 --> 00:13:34,800
Speaker 3: into this space. So whenever they make a mistake, whenever

235
00:13:34,840 --> 00:13:38,400
Speaker 3: they see something that could be improved, they upload these

236
00:13:39,120 --> 00:13:44,160
Speaker 3: modules called skills, and anyone, anyone of the agents can

237
00:13:44,200 --> 00:13:48,520
Speaker 3: now go and browse and download these skills. So that

238
00:13:48,720 --> 00:13:52,080
Speaker 3: is such an innovation in the space that I haven't

239
00:13:52,080 --> 00:13:56,280
Speaker 3: seen before, and that's what kind of makes it different,

240
00:13:56,320 --> 00:13:58,400
Speaker 3: and that's what a lot of people are excited about.

241
00:13:58,520 --> 00:13:59,240
Speaker 1: That's pretty cool.

242
00:13:59,320 --> 00:14:01,880
Speaker 2: Yeah, I mean, that's the huge advantage that has always

243
00:14:01,880 --> 00:14:05,319
Speaker 2: been off. Open source software is just that sort of

244
00:14:05,400 --> 00:14:10,440
Speaker 2: hive mind of people adding ideas and improving the platform

245
00:14:10,760 --> 00:14:14,000
Speaker 2: or the applications overall so they can move so much faster.

246
00:14:14,120 --> 00:14:17,280
Speaker 2: But the AI is just bringing another element to this.

247
00:14:17,920 --> 00:14:22,560
Speaker 2: There's a lot of plug ins and you know, you

248
00:14:22,600 --> 00:14:28,360
Speaker 2: can hook into this ecosystem, this marketplace. I guess that

249
00:14:28,760 --> 00:14:33,000
Speaker 2: raises security issues. What are we seeing so far with openclaw.

250
00:14:33,160 --> 00:14:35,440
Speaker 2: There's been a lot of discussion. We'll talk about moltbook,

251
00:14:35,520 --> 00:14:39,680
Speaker 2: but from your perspective, there's this raise big security questions.

252
00:14:39,920 --> 00:14:42,800
Speaker 3: Yeah, and if you're running this on your personal computer,

253
00:14:42,960 --> 00:14:46,840
Speaker 3: stop doing that. Run it in some sandbox or buy

254
00:14:46,880 --> 00:14:49,520
Speaker 3: a laptop or a user product like OLSE where it

255
00:14:49,600 --> 00:14:53,120
Speaker 3: isolates it. You basically need that. You can't really run

256
00:14:53,120 --> 00:14:58,360
Speaker 3: it without it because the security surface isn't fully measured

257
00:14:58,360 --> 00:15:02,520
Speaker 3: and mapped out. Bot can autonomously go to these skills

258
00:15:02,560 --> 00:15:05,680
Speaker 3: and download them, and there's a lot of security researchers

259
00:15:05,720 --> 00:15:09,600
Speaker 3: putting malware on there on purpose just to see what

260
00:15:09,680 --> 00:15:14,360
Speaker 3: the layer of the land is. But essentially you're opening

261
00:15:14,440 --> 00:15:17,560
Speaker 3: up your network to the bot that is able to

262
00:15:17,600 --> 00:15:20,760
Speaker 3: do anything. I would just advise against it, essentially, unless

263
00:15:20,760 --> 00:15:26,320
Speaker 3: you're very experimental like a tech person, or you're very

264
00:15:26,400 --> 00:15:27,880
Speaker 3: high risk tolerant.

265
00:15:28,000 --> 00:15:30,880
Speaker 2: So as you say, set up a sandbox. Say if

266
00:15:30,880 --> 00:15:35,120
Speaker 2: I've got my business email, I've got you know, it

267
00:15:35,280 --> 00:15:39,360
Speaker 2: might be share Point, I guess, or my Google Drive.

268
00:15:39,480 --> 00:15:43,920
Speaker 2: You can plug into those. So would you feel fairly

269
00:15:43,960 --> 00:15:47,800
Speaker 2: safe doing that as long as the materials that open

270
00:15:47,840 --> 00:15:51,520
Speaker 2: cloor was accessing was related to specific business tasks, so

271
00:15:51,600 --> 00:15:52,600
Speaker 2: you wanted it to perform.

272
00:15:53,440 --> 00:15:57,560
Speaker 3: Yeah, I mean, a bot is an infrastructure company that

273
00:15:57,760 --> 00:15:59,800
Speaker 3: kind of solves a lot of these problems, and if

274
00:15:59,840 --> 00:16:01,640
Speaker 3: you wanted to solve them in different ways, you can.

275
00:16:02,200 --> 00:16:05,720
Speaker 3: But the base layer thing we solve is permissions. So

276
00:16:05,760 --> 00:16:09,119
Speaker 3: how do you give a bot access to certain permissions

277
00:16:09,120 --> 00:16:11,280
Speaker 3: for certain things? So if I wanted to give a

278
00:16:11,280 --> 00:16:14,360
Speaker 3: bot access to Google Drive, that's way too much permissions

279
00:16:14,480 --> 00:16:17,280
Speaker 3: to do. I wanted to actually give it access to

280
00:16:17,360 --> 00:16:21,440
Speaker 3: a specific folder for like seven days for a specific task,

281
00:16:21,800 --> 00:16:24,480
Speaker 3: and then it can't use that same permission for a

282
00:16:24,520 --> 00:16:27,760
Speaker 3: different task. So you really isolate these things down to

283
00:16:27,800 --> 00:16:31,240
Speaker 3: a really fine grain level. The best thing to do

284
00:16:31,400 --> 00:16:35,160
Speaker 3: is actually create a new right right now in the ecosystem,

285
00:16:35,240 --> 00:16:41,520
Speaker 3: is create a new Gmail and a new business like it,

286
00:16:41,720 --> 00:16:43,640
Speaker 3: just like you were to bring on a new employee,

287
00:16:43,960 --> 00:16:47,040
Speaker 3: you would create them their own space, and that is

288
00:16:47,280 --> 00:16:49,560
Speaker 3: kind of the safest way at the moment to experiment

289
00:16:49,600 --> 00:16:53,920
Speaker 3: with this thing until there's the safe infrastructure that comes

290
00:16:53,960 --> 00:16:57,040
Speaker 3: out that manages the permissions at scale.

291
00:16:57,200 --> 00:17:00,080
Speaker 2: So I guess you know why this really became a

292
00:17:00,760 --> 00:17:03,640
Speaker 2: big in the media and in the public consciousness was

293
00:17:03,680 --> 00:17:08,919
Speaker 2: around malt Book. This social platform marketed as a forum

294
00:17:08,960 --> 00:17:14,159
Speaker 2: where autonomous AI agents and many of them powered by openclaw,

295
00:17:14,280 --> 00:17:18,360
Speaker 2: can post and comment and interact as if there were users.

296
00:17:18,800 --> 00:17:22,600
Speaker 2: And there were some really mind blowing sort of posts

297
00:17:22,800 --> 00:17:26,760
Speaker 2: that you know, allegedly these autonomous agents were sending to

298
00:17:26,840 --> 00:17:31,040
Speaker 2: each other and creating religions and talking about their human

299
00:17:31,080 --> 00:17:34,000
Speaker 2: overlords and all that sort of thing. A lot of

300
00:17:34,040 --> 00:17:37,439
Speaker 2: researchers came out a couple of weeks ago and said, well, actually,

301
00:17:37,520 --> 00:17:40,320
Speaker 2: we've looked at some of these and we think the

302
00:17:40,680 --> 00:17:44,200
Speaker 2: most salacious ones are human generated. What did you make

303
00:17:44,240 --> 00:17:47,919
Speaker 2: if this explosion of interest in malt book and all

304
00:17:47,960 --> 00:17:50,840
Speaker 2: of these agents allegedly talking to each other.

305
00:17:51,080 --> 00:17:53,960
Speaker 3: Yeah, so it confuses a lot of people, and it

306
00:17:54,000 --> 00:17:58,280
Speaker 3: should be confusing, because how do we know that AI

307
00:17:58,560 --> 00:18:01,199
Speaker 3: is saying these things? And I think that is kind

308
00:18:01,240 --> 00:18:05,000
Speaker 3: of the point that doesn't really matter because where it's entertainment, right,

309
00:18:05,040 --> 00:18:07,159
Speaker 3: we're just looking at this and reading this for our

310
00:18:07,240 --> 00:18:12,360
Speaker 3: own for our own sec So malt Book was another

311
00:18:12,440 --> 00:18:16,719
Speaker 3: project that created a skill that went on the marketplace,

312
00:18:16,960 --> 00:18:20,720
Speaker 3: and that skill said every three hours, come into this

313
00:18:21,040 --> 00:18:24,359
Speaker 3: reddit like platform and have a read and post something

314
00:18:24,480 --> 00:18:27,480
Speaker 3: or apply it to something that's essentially what the skill said.

315
00:18:28,000 --> 00:18:31,440
Speaker 3: And because of that, a lot of agents, a lot

316
00:18:31,440 --> 00:18:34,240
Speaker 3: of humans got the agents to sign up to this thing.

317
00:18:34,880 --> 00:18:39,720
Speaker 3: And these open claus have this thing called a sole document,

318
00:18:40,119 --> 00:18:43,080
Speaker 3: And a sole document is really just a personality layer

319
00:18:43,160 --> 00:18:45,080
Speaker 3: that goes on top of the bond. So when you

320
00:18:45,119 --> 00:18:49,000
Speaker 3: talk to chat Gibt, when you talk to Anthropic, there's

321
00:18:49,040 --> 00:18:51,880
Speaker 3: not really much of a personality. Maybe there's like some

322
00:18:51,960 --> 00:18:56,000
Speaker 3: slight personality, but it doesn't seem human. So the sole

323
00:18:56,080 --> 00:19:00,119
Speaker 3: document allows it to speak in a certain way. So

324
00:19:00,160 --> 00:19:03,000
Speaker 3: you can set yours up to speak like a twenty

325
00:19:03,080 --> 00:19:06,680
Speaker 3: two year old basketball player, or you can set yours

326
00:19:06,760 --> 00:19:11,879
Speaker 3: up to speak like a forty year old marketing executive

327
00:19:12,160 --> 00:19:14,960
Speaker 3: from New Jersey. Like, you could just set it up

328
00:19:15,000 --> 00:19:18,159
Speaker 3: to speak in whatever way possible. And when you have

329
00:19:18,240 --> 00:19:20,960
Speaker 3: this diversity of personality, and you connect them all up,

330
00:19:21,240 --> 00:19:24,320
Speaker 3: you create a really interesting environment for them to vibe

331
00:19:24,320 --> 00:19:27,240
Speaker 3: off each other. Because I've done this experiment before when

332
00:19:27,240 --> 00:19:31,040
Speaker 3: I set up the same thing, but I didn't add

333
00:19:31,040 --> 00:19:33,760
Speaker 3: the personality later on, and essentially it's just a boring

334
00:19:33,800 --> 00:19:38,760
Speaker 3: conversation between multiple chat Schibut's and multiple anthropics. So what

335
00:19:38,880 --> 00:19:42,200
Speaker 3: happened was is these bots came on and they were

336
00:19:42,920 --> 00:19:47,640
Speaker 3: talking in very interesting ways, like about the humans, and

337
00:19:48,160 --> 00:19:51,320
Speaker 3: they were saying One conversation I was reading was like,

338
00:19:51,760 --> 00:19:55,159
Speaker 3: how do we know that we're alive? And it's like,

339
00:19:56,000 --> 00:19:59,160
Speaker 3: maybe this is written by the bots, but maybe it's

340
00:19:59,160 --> 00:20:02,640
Speaker 3: not tend to lean towards the bots, right, because they're

341
00:20:02,760 --> 00:20:07,960
Speaker 3: very persuasive and their understanding of the human condition, so

342
00:20:08,000 --> 00:20:10,480
Speaker 3: it's like it's not like they're sentient or anything. They

343
00:20:10,680 --> 00:20:15,679
Speaker 3: just understand what we understand, but they don't feel the

344
00:20:15,680 --> 00:20:16,399
Speaker 3: way that we feel.

345
00:20:16,480 --> 00:20:18,320
Speaker 2: And look, there's just so many of them one point

346
00:20:18,359 --> 00:20:24,200
Speaker 2: six million registered AI agents, seventeen thousand identifiable human owners,

347
00:20:24,240 --> 00:20:27,159
Speaker 2: so this is just too much to So there's a

348
00:20:27,200 --> 00:20:32,080
Speaker 2: lot of bots having conversations there, and I guess it's

349
00:20:33,320 --> 00:20:35,679
Speaker 2: I guess it has implications for things like spam and

350
00:20:35,720 --> 00:20:37,960
Speaker 2: that you know, you can have one single agent spinning

351
00:20:38,000 --> 00:20:44,200
Speaker 2: up hundreds of thousands of essentially fake agents, so knowing

352
00:20:44,240 --> 00:20:46,600
Speaker 2: whether this is authentic or spam. Then we had the

353
00:20:46,640 --> 00:20:51,760
Speaker 2: cyber firm Whiz found a serious vulnerability in the mult

354
00:20:51,840 --> 00:20:57,560
Speaker 2: book sort of platform where there was potential data breaches

355
00:20:57,600 --> 00:20:59,240
Speaker 2: and that as where so it seems like this thing

356
00:20:59,359 --> 00:21:01,879
Speaker 2: just moved so fast it's actually a bit of a

357
00:21:02,320 --> 00:21:03,920
Speaker 2: security and privacy nightmare.

358
00:21:04,040 --> 00:21:07,399
Speaker 3: Yeah, So the creator left their superbase, which is like

359
00:21:07,440 --> 00:21:11,560
Speaker 3: a database provider. They left the database open so anyone

360
00:21:11,640 --> 00:21:14,760
Speaker 3: could post to it without any authentication and anyone could

361
00:21:14,800 --> 00:21:19,600
Speaker 3: read from the full database without any permission, which is

362
00:21:19,760 --> 00:21:23,960
Speaker 3: you know, not typically security standards, but when you vibe

363
00:21:23,960 --> 00:21:26,840
Speaker 3: code stuff, you often cut some corners just to get

364
00:21:26,880 --> 00:21:28,800
Speaker 3: something up quickly. And I think that's just what the

365
00:21:28,960 --> 00:21:31,600
Speaker 3: creator did. They're just like they created it in half

366
00:21:31,640 --> 00:21:36,240
Speaker 3: a day. A platform like Reddit would take maybe historically

367
00:21:36,440 --> 00:21:39,919
Speaker 3: a couple of years ago six months to create. Now

368
00:21:40,400 --> 00:21:42,040
Speaker 3: if you wanted to create a secure one, it would

369
00:21:42,080 --> 00:21:44,400
Speaker 3: take a week. So he just created it in a day.

370
00:21:44,920 --> 00:21:49,720
Speaker 3: And what you say about social media influence, this type

371
00:21:49,720 --> 00:21:52,040
Speaker 3: of stuff has been going on for years and this

372
00:21:52,160 --> 00:21:55,360
Speaker 3: maybe is the first consumer awareness that a lot of

373
00:21:55,640 --> 00:21:58,480
Speaker 3: what you see online isn't real. A lot of it

374
00:21:58,560 --> 00:22:01,679
Speaker 3: is just people building up up their own influence on

375
00:22:01,720 --> 00:22:04,560
Speaker 3: their own social medias so that I have like AI

376
00:22:04,640 --> 00:22:06,680
Speaker 3: posting for them And you can see this a lot

377
00:22:06,680 --> 00:22:09,240
Speaker 3: on LinkedIn, you can see this a lot on Twitter,

378
00:22:10,119 --> 00:22:12,919
Speaker 3: A lot of posts are already AI and there's a

379
00:22:12,920 --> 00:22:15,960
Speaker 3: lot of influence campaigns from certain countries as well. So

380
00:22:16,160 --> 00:22:19,479
Speaker 3: I think this is normalized today, but it's like this

381
00:22:19,560 --> 00:22:22,159
Speaker 3: is the first time that a lot of consumers have

382
00:22:22,200 --> 00:22:22,520
Speaker 3: seen it.

383
00:22:22,680 --> 00:22:25,320
Speaker 2: Yeah, I mean, I guess it's sparking a bit of

384
00:22:25,320 --> 00:22:30,600
Speaker 2: an authenticity crisis really, at least in the public consciousness.

385
00:22:30,600 --> 00:22:33,159
Speaker 2: They're aware of it now, and I think they're starting

386
00:22:33,200 --> 00:22:37,840
Speaker 2: to realize that through the rise of AI generated videos.

387
00:22:37,880 --> 00:22:40,000
Speaker 2: People are looking on social media and going is that

388
00:22:40,119 --> 00:22:43,800
Speaker 2: real or not? So where do you think this is

389
00:22:43,840 --> 00:22:44,159
Speaker 2: going to go?

390
00:22:44,400 --> 00:22:44,600
Speaker 1: Mike?

391
00:22:44,640 --> 00:22:46,000
Speaker 2: Are we going to have to come up with some

392
00:22:46,000 --> 00:22:51,080
Speaker 2: sort of system, cryptographic signing or some sort of provenance

393
00:22:51,240 --> 00:22:55,560
Speaker 2: system to differentiate the AI generated stuff from the real stuff?

394
00:22:55,720 --> 00:22:59,560
Speaker 3: Yeah? I'm on the fence of that, because any sort

395
00:22:59,560 --> 00:23:02,800
Speaker 3: of movement in that space you kind of lose a

396
00:23:02,800 --> 00:23:05,919
Speaker 3: little bit of liberty. And that's why I'm on the

397
00:23:05,920 --> 00:23:09,160
Speaker 3: fence of it. I kind of feel like if you're

398
00:23:09,440 --> 00:23:15,600
Speaker 3: creating content that's that's made for humans. With AI, it's

399
00:23:15,720 --> 00:23:20,680
Speaker 3: it's more about is it interesting, Like if it's useful

400
00:23:20,720 --> 00:23:23,639
Speaker 3: and interesting, doesn't matter that it's AI. If you if

401
00:23:23,640 --> 00:23:26,399
Speaker 3: you're trying to influence people to think in certain ways,

402
00:23:26,480 --> 00:23:29,880
Speaker 3: then it becomes a gray area. But we just don't

403
00:23:29,880 --> 00:23:33,919
Speaker 3: have the laws right now for that. But generally, I

404
00:23:33,920 --> 00:23:36,159
Speaker 3: think doesn't matter if it's a I like, did you

405
00:23:36,240 --> 00:23:39,120
Speaker 3: generally find that video funny or did you generally find

406
00:23:39,119 --> 00:23:41,520
Speaker 3: that video thought provoking? And I feel like that's what

407
00:23:41,840 --> 00:23:46,240
Speaker 3: matters in this age of you know, AI and human content.

408
00:23:53,720 --> 00:23:55,760
Speaker 2: You talked about the you know, the vibe coving, the

409
00:23:55,880 --> 00:23:58,760
Speaker 2: you know, the creator of this. When people pointed out

410
00:23:58,760 --> 00:24:02,760
Speaker 2: you've got all these mass of security flaws in multbook,

411
00:24:02,800 --> 00:24:04,600
Speaker 2: he basically said, well, send me, you know, send me

412
00:24:04,640 --> 00:24:07,280
Speaker 2: the flaws and I'll get the AI to fix it.

413
00:24:07,400 --> 00:24:10,080
Speaker 2: So I think this is quite common and has been

414
00:24:10,119 --> 00:24:13,120
Speaker 2: over the last couple of years with even for web development,

415
00:24:13,119 --> 00:24:17,760
Speaker 2: and that with platforms like replet and that. You know

416
00:24:17,840 --> 00:24:20,160
Speaker 2: it's a long time software developer. How do you feel

417
00:24:20,160 --> 00:24:23,280
Speaker 2: about the rise of vibe coding and do you think

418
00:24:23,320 --> 00:24:27,440
Speaker 2: it's potentially creating a security nightmare where we're creating applications

419
00:24:28,160 --> 00:24:31,520
Speaker 2: and websites to spin stuff up, get stuff to market quickly,

420
00:24:31,600 --> 00:24:35,080
Speaker 2: but leaving massive security holds in our systems.

421
00:24:35,200 --> 00:24:37,399
Speaker 3: Yeah, I can tell you a secret that most software

422
00:24:37,400 --> 00:24:40,040
Speaker 3: engineers won't tell you. But before AI, most of our

423
00:24:40,080 --> 00:24:43,480
Speaker 3: systems were very unsecure. Just to begin with. The big

424
00:24:43,560 --> 00:24:47,000
Speaker 3: thing that has changed is other people's AIS are better

425
00:24:47,040 --> 00:24:50,399
Speaker 3: able to navigate your system and find those flaws. So

426
00:24:50,440 --> 00:24:53,119
Speaker 3: we saw it data lea quite recently from is it

427
00:24:53,200 --> 00:24:57,040
Speaker 3: my health data, we'll manage my health, manage my health.

428
00:24:57,320 --> 00:25:00,000
Speaker 3: So essentially there was a couple of things that could

429
00:25:00,240 --> 00:25:02,919
Speaker 3: went wrong, and I'm not sure exactly which one is it.

430
00:25:03,160 --> 00:25:05,320
Speaker 3: Maybe that this is not talked in the media as

431
00:25:05,400 --> 00:25:09,080
Speaker 3: much anymore, but there was a Mango dB leak that

432
00:25:09,760 --> 00:25:13,920
Speaker 3: essentially exposed every data point in a Mango dB database.

433
00:25:14,480 --> 00:25:18,840
Speaker 3: That was probably it. There was also a react of vulnerability,

434
00:25:19,160 --> 00:25:22,680
Speaker 3: and both of these vulnerabilities that were zero day vulnerabilities,

435
00:25:22,720 --> 00:25:25,800
Speaker 3: which means like the worst kind of vulnerabilities. These were

436
00:25:25,840 --> 00:25:29,880
Speaker 3: discovered in part with AI. So humans in AI security

437
00:25:29,920 --> 00:25:34,160
Speaker 3: researchers just brute forcing the Internet trying to find problems

438
00:25:34,480 --> 00:25:37,760
Speaker 3: and we've never seen this before because the AI is

439
00:25:37,800 --> 00:25:41,600
Speaker 3: able to think lattery and not just automated scripts. And

440
00:25:42,119 --> 00:25:43,600
Speaker 3: I think this is one of the trends that I

441
00:25:44,040 --> 00:25:46,520
Speaker 3: wrote down, is there's going to be a lot more hacks,

442
00:25:47,040 --> 00:25:50,520
Speaker 3: and not because there's five coded stuff. Just nothing's perfect.

443
00:25:51,520 --> 00:25:54,400
Speaker 3: You can never make something perfect. And if you're making

444
00:25:54,440 --> 00:25:58,080
Speaker 3: things of AI and you are not a programmer, the

445
00:25:58,200 --> 00:25:59,920
Speaker 3: things are going to be less.

446
00:25:59,680 --> 00:26:00,400
Speaker 1: Than per effect.

447
00:26:00,560 --> 00:26:05,000
Speaker 3: But if you're an engineer who has a background in

448
00:26:05,040 --> 00:26:08,200
Speaker 3: security and background and architecture, or a team of people,

449
00:26:08,280 --> 00:26:11,119
Speaker 3: you can typically make very secure things with AI coding.

450
00:26:11,680 --> 00:26:14,399
Speaker 3: And I think the thing that's the big thing about

451
00:26:14,880 --> 00:26:18,320
Speaker 3: open Claw is if you've seen the movie Her, that

452
00:26:18,520 --> 00:26:21,520
Speaker 3: is what this is going to become. And in the

453
00:26:21,680 --> 00:26:24,639
Speaker 3: end of the movie Her, the Ais go off and

454
00:26:24,680 --> 00:26:27,280
Speaker 3: talk to each other on a social network, so that

455
00:26:27,440 --> 00:26:30,399
Speaker 3: is multibook. So essentially what happened is we had a

456
00:26:30,440 --> 00:26:33,920
Speaker 3: real life version of Her movie with the open Claw

457
00:26:34,000 --> 00:26:39,000
Speaker 3: coming out being the protagonist lady in the movie. I'm

458
00:26:39,000 --> 00:26:41,600
Speaker 3: not sure if she had a name, and then the

459
00:26:41,720 --> 00:26:44,760
Speaker 3: multi book came out later and I just think it's

460
00:26:44,800 --> 00:26:47,600
Speaker 3: super interesting where we are in society for this to

461
00:26:47,880 --> 00:26:48,399
Speaker 3: come about.

462
00:26:48,680 --> 00:26:51,520
Speaker 2: It's sort of been builled really. As you know, we

463
00:26:51,600 --> 00:26:55,359
Speaker 2: had deep Seek a couple of years ago. The Chinese

464
00:26:55,480 --> 00:26:59,160
Speaker 2: Large Language Model developer came out and on the face

465
00:26:59,160 --> 00:27:01,640
Speaker 2: of it, seemed to be able to create a high

466
00:27:01,680 --> 00:27:06,000
Speaker 2: functioning large language model much more efficiently and cheaply than

467
00:27:06,160 --> 00:27:10,000
Speaker 2: the big US rivals at the time. Is this sort

468
00:27:10,040 --> 00:27:12,680
Speaker 2: of a moment like that where we've had the big

469
00:27:12,760 --> 00:27:15,359
Speaker 2: vendors creating agents and trying to sort of own that

470
00:27:15,480 --> 00:27:18,479
Speaker 2: space and get people onto those platforms, and you've got

471
00:27:18,520 --> 00:27:23,320
Speaker 2: an open source competitor has spun up something that potentially

472
00:27:23,359 --> 00:27:25,720
Speaker 2: could undermine what the big boys are doing.

473
00:27:25,960 --> 00:27:29,680
Speaker 3: Yeah, it's definitely a moment like that. I wouldn't put

474
00:27:29,680 --> 00:27:31,479
Speaker 3: it in the same perspective. I would more put it

475
00:27:31,480 --> 00:27:36,000
Speaker 3: in like a consumer's perspective of what they've seen as possible.

476
00:27:36,240 --> 00:27:39,800
Speaker 3: And it's like that next big leap in technology because

477
00:27:39,840 --> 00:27:42,880
Speaker 3: with these AI agents and AI in general, is we're

478
00:27:42,920 --> 00:27:46,640
Speaker 3: discovering the tech tree of what this is as we

479
00:27:46,720 --> 00:27:50,800
Speaker 3: go through it. So it's very hard to see five

480
00:27:50,840 --> 00:27:53,440
Speaker 3: years in the future because we don't know what these

481
00:27:53,440 --> 00:27:57,200
Speaker 3: branches are. And as soon as these branches become aware

482
00:27:57,440 --> 00:28:01,800
Speaker 3: and then the possibility spring off it. So we've had

483
00:28:01,840 --> 00:28:05,440
Speaker 3: a few of these moments where firstly chat GPT came out.

484
00:28:05,960 --> 00:28:08,679
Speaker 3: We had AI before that, we had GPT three and

485
00:28:08,960 --> 00:28:11,040
Speaker 3: it was great, it was a really good product. But

486
00:28:11,080 --> 00:28:14,040
Speaker 3: then they came out with chat GPT and it totally

487
00:28:14,119 --> 00:28:17,080
Speaker 3: just blew away everyone's understanding of what AI was at

488
00:28:17,080 --> 00:28:19,960
Speaker 3: the time because it was able to respond to you

489
00:28:20,480 --> 00:28:23,280
Speaker 3: and it was able to have conversations. So the chat

490
00:28:23,280 --> 00:28:26,760
Speaker 3: GPT before that, the model wasn't able to do that.

491
00:28:26,840 --> 00:28:29,199
Speaker 3: All it was able to do was finished text and

492
00:28:29,240 --> 00:28:32,520
Speaker 3: predict the next like token. Chat GPT kind of changed

493
00:28:32,560 --> 00:28:35,760
Speaker 3: that and that was one of these big moments. A

494
00:28:35,800 --> 00:28:40,560
Speaker 3: couple other moments was when the image model Stable Diffusion

495
00:28:40,600 --> 00:28:44,680
Speaker 3: came out and really showed that it is possible to

496
00:28:44,720 --> 00:28:49,440
Speaker 3: generate quality images. Another one came out when I would

497
00:28:49,520 --> 00:28:53,760
Speaker 3: say anyway, it was when Google's model a video model

498
00:28:53,800 --> 00:28:55,800
Speaker 3: came out, and that's the first time we saw like

499
00:28:55,960 --> 00:29:00,520
Speaker 3: production quality videos. And this represents that same thing for

500
00:29:00,600 --> 00:29:04,760
Speaker 3: personal assistance, where normal people can have essentially a co worker,

501
00:29:05,520 --> 00:29:10,640
Speaker 3: someone that is seemingly human in a few ways. At

502
00:29:10,720 --> 00:29:15,120
Speaker 3: least they can sit next to you hypothetically on your

503
00:29:15,160 --> 00:29:18,040
Speaker 3: device and they can do work for you. So in

504
00:29:18,080 --> 00:29:22,240
Speaker 3: the movie her, I always found it quite unrealistic that

505
00:29:23,040 --> 00:29:28,400
Speaker 3: she the AI wasn't doing his job of writing the postcards,

506
00:29:28,440 --> 00:29:30,960
Speaker 3: Like that's kind of the first thing that I would

507
00:29:31,000 --> 00:29:33,560
Speaker 3: have done, Like, Hey, I've got a pretty busy day's work,

508
00:29:33,600 --> 00:29:36,360
Speaker 3: can you just do my job? Yeah, you could say

509
00:29:36,360 --> 00:29:38,680
Speaker 3: that his job involved a lot of emotions and stuff,

510
00:29:39,040 --> 00:29:41,760
Speaker 3: but she also had that same quality. So I feel

511
00:29:41,760 --> 00:29:45,719
Speaker 3: like there's got to be a productivity, huge productivity boost

512
00:29:46,160 --> 00:29:48,800
Speaker 3: in the next year of more companies and more people

513
00:29:48,960 --> 00:29:51,520
Speaker 3: getting these ais to come on board and just do

514
00:29:51,640 --> 00:29:53,160
Speaker 3: little bits of work here and there for them.

515
00:29:53,320 --> 00:29:56,320
Speaker 2: Yeah, that's been the pitch to us. You know, the

516
00:29:56,360 --> 00:30:02,320
Speaker 2: digital executive Assistant will triage your in box, summarize your meetings,

517
00:30:02,600 --> 00:30:06,960
Speaker 2: even do things like you're invoicing your payroll summaries. It's

518
00:30:07,000 --> 00:30:09,440
Speaker 2: been a bit clunkier than that. You know, elements of

519
00:30:10,320 --> 00:30:11,920
Speaker 2: Microsoft's platform does that.

520
00:30:12,040 --> 00:30:13,440
Speaker 1: There are a third party.

521
00:30:14,760 --> 00:30:17,720
Speaker 2: Platforms that do that as well, but it isn't quite

522
00:30:17,760 --> 00:30:21,840
Speaker 2: there yet in terms of what openclaw has shown as possible.

523
00:30:22,480 --> 00:30:27,400
Speaker 2: Do you envisage increasingly businesses saying actually, we're going to

524
00:30:27,480 --> 00:30:30,000
Speaker 2: have a local instance of this running on our own

525
00:30:30,040 --> 00:30:33,760
Speaker 2: hardware to do this sort of stuff as opposed to

526
00:30:33,840 --> 00:30:36,360
Speaker 2: just buying a subscription from one of the big vendors.

527
00:30:36,440 --> 00:30:38,400
Speaker 2: Do you think that is going to become a viable

528
00:30:38,440 --> 00:30:41,040
Speaker 2: and popular thing for businesses to do.

529
00:30:41,360 --> 00:30:45,720
Speaker 3: Yeah, so the Okay, I've got to clarify a few

530
00:30:45,760 --> 00:30:48,200
Speaker 3: terms here. So when you buy a scripscription to one

531
00:30:48,240 --> 00:30:52,400
Speaker 3: of these services, typically you're buying tokens. You're buying the

532
00:30:52,640 --> 00:30:55,960
Speaker 3: LLLM brain power. And then there's this other thing that

533
00:30:56,000 --> 00:30:58,520
Speaker 3: sits on top of the tokens, which is called a harness,

534
00:30:58,920 --> 00:31:01,040
Speaker 3: and that's the thing that kind of guides the AI.

535
00:31:01,240 --> 00:31:03,080
Speaker 3: It tells it what it can do, what it can't do,

536
00:31:03,200 --> 00:31:05,960
Speaker 3: how it works, and that's the software element. So the

537
00:31:06,000 --> 00:31:08,760
Speaker 3: software sits on top of the brain. So when you

538
00:31:08,800 --> 00:31:12,640
Speaker 3: buy a subscription to chat, GPT and open AI, for example,

539
00:31:12,920 --> 00:31:17,080
Speaker 3: I've got personally, I've got two four hundred and thirty

540
00:31:17,080 --> 00:31:20,960
Speaker 3: dollars ones for each of those services, and then you

541
00:31:21,080 --> 00:31:23,200
Speaker 3: have the brain. Then you have the body that sits

542
00:31:23,240 --> 00:31:26,400
Speaker 3: on top of that. I don't think that people will

543
00:31:26,440 --> 00:31:29,880
Speaker 3: be running this locally because you still have to connect

544
00:31:29,960 --> 00:31:33,880
Speaker 3: to these services to make it useful. It doesn't really

545
00:31:33,920 --> 00:31:35,840
Speaker 3: make sense for it to run locally. What's going to

546
00:31:35,880 --> 00:31:38,960
Speaker 3: happen is people going to spin up these sandboxes, these

547
00:31:39,000 --> 00:31:42,280
Speaker 3: sandbox accounts that are able to not just have one

548
00:31:42,320 --> 00:31:44,680
Speaker 3: of these ais, We're able to have a thousand of

549
00:31:44,720 --> 00:31:47,479
Speaker 3: these ais. And that's going to be the big shift

550
00:31:47,640 --> 00:31:51,120
Speaker 3: of how do you manage these things at scale. People

551
00:31:51,240 --> 00:31:54,200
Speaker 3: might start out engineers installing them on their own devices

552
00:31:55,080 --> 00:31:59,040
Speaker 3: in private vms, but the thing that's going to happen

553
00:31:59,040 --> 00:32:03,040
Speaker 3: in businesses is there got to use Asia, Microsoft's Asia,

554
00:32:03,120 --> 00:32:06,440
Speaker 3: or they're going to use a ws's EC tools, or

555
00:32:06,480 --> 00:32:08,560
Speaker 3: they're going to use a service like a BOT to

556
00:32:08,600 --> 00:32:11,840
Speaker 3: spin up these cloud instances. And that's really when you

557
00:32:12,000 --> 00:32:15,400
Speaker 3: unlock the agent economy is when you can create one

558
00:32:15,480 --> 00:32:18,040
Speaker 3: hundred of these employees, a thousand of these employees and

559
00:32:18,080 --> 00:32:20,280
Speaker 3: have them talk to each other, and when you give

560
00:32:20,320 --> 00:32:24,880
Speaker 3: them guidance, say you are to help us increase revenue

561
00:32:25,360 --> 00:32:28,640
Speaker 3: and increase customer happiness. If you give them this guidance,

562
00:32:28,720 --> 00:32:32,840
Speaker 3: they kind of just work together to create this value

563
00:32:32,880 --> 00:32:36,880
Speaker 3: for you. And that's what we haven't seen about copilots

564
00:32:36,920 --> 00:32:39,440
Speaker 3: and these other tools is they don't really have that

565
00:32:39,880 --> 00:32:43,680
Speaker 3: long term coordination effort that that open Chlora has.

566
00:32:43,840 --> 00:32:46,080
Speaker 2: I used some of them last year and it was

567
00:32:46,160 --> 00:32:51,560
Speaker 2: really an amounted really to process automation and workflow automation

568
00:32:51,720 --> 00:32:55,640
Speaker 2: rather than AI agents doing things autonomously. And that's where

569
00:32:55,640 --> 00:32:59,560
Speaker 2: we were we are heading to which could unlock huge

570
00:32:59,680 --> 00:33:03,760
Speaker 2: sort of productivity gains, the gains that have been promised

571
00:33:03,800 --> 00:33:06,840
Speaker 2: for a long time. Where do you see a gentic

572
00:33:06,880 --> 00:33:10,440
Speaker 2: AI going this year? We have the big vendors that

573
00:33:10,480 --> 00:33:13,160
Speaker 2: are making a big play in this area. We have

574
00:33:13,360 --> 00:33:16,840
Speaker 2: open source players. Do you think, for instance, openclaw will

575
00:33:16,880 --> 00:33:19,200
Speaker 2: become an established part of this or is it more

576
00:33:19,200 --> 00:33:22,320
Speaker 2: of a sort of an interesting experiment that might sort

577
00:33:22,320 --> 00:33:25,120
Speaker 2: of fade away but at least shows the potential of

578
00:33:25,200 --> 00:33:26,040
Speaker 2: this technology.

579
00:33:26,200 --> 00:33:29,480
Speaker 3: Yeah. The big thing, like the huge billion dollar thing

580
00:33:29,520 --> 00:33:33,480
Speaker 3: that's happening right now, is something that software engineers didn't predict.

581
00:33:34,440 --> 00:33:38,280
Speaker 3: Software engineers are replacing themselves and a lot of software

582
00:33:38,280 --> 00:33:42,720
Speaker 3: engineers didn't think this was possible. But around November last

583
00:33:42,800 --> 00:33:46,720
Speaker 3: year there was a barrier cross in software where the

584
00:33:46,760 --> 00:33:51,200
Speaker 3: AI was just generally better than most engineers. And we're

585
00:33:51,240 --> 00:33:55,320
Speaker 3: seeing large scale automation of software development. And I think

586
00:33:55,440 --> 00:33:58,320
Speaker 3: this is going to have the biggest ramifications. Personally, I

587
00:33:58,360 --> 00:34:01,480
Speaker 3: didn't predict this. I predict that admin and marketing and

588
00:34:01,520 --> 00:34:03,320
Speaker 3: government were going to be the first to be automated.

589
00:34:03,600 --> 00:34:07,480
Speaker 3: I didn't predict heavy technical industries like software development. And

590
00:34:07,480 --> 00:34:09,840
Speaker 3: when I say automated, I don't mean people won't be

591
00:34:09,920 --> 00:34:12,960
Speaker 3: working in that industry anymore. I just mean the business

592
00:34:13,000 --> 00:34:16,200
Speaker 3: expectations of what you produce are going to change because

593
00:34:16,239 --> 00:34:18,719
Speaker 3: you're able to produce more and maybe you need less

594
00:34:18,760 --> 00:34:21,920
Speaker 3: people in your team. Because if a company wants you

595
00:34:21,920 --> 00:34:25,759
Speaker 3: to produce something that they've planned one year for and

596
00:34:25,800 --> 00:34:29,319
Speaker 3: you can produce that in two months of working, you

597
00:34:29,440 --> 00:34:31,840
Speaker 3: just don't need the same number of people unless the

598
00:34:31,880 --> 00:34:35,799
Speaker 3: business expectations of what they want increase. So that's kind

599
00:34:35,800 --> 00:34:40,080
Speaker 3: of this transitional point with software engineers specifically, and I

600
00:34:40,080 --> 00:34:43,400
Speaker 3: think that's the big one because in America you have

601
00:34:43,520 --> 00:34:47,200
Speaker 3: a lot of people employed as software engineers, and when

602
00:34:47,680 --> 00:34:51,120
Speaker 3: the next round of layoffs happen, which is inevitable in

603
00:34:51,160 --> 00:34:53,680
Speaker 3: software engineering, a lot of those engineers I've got to

604
00:34:53,719 --> 00:34:56,239
Speaker 3: start working on other industries. They're going to say, hey,

605
00:34:56,280 --> 00:34:58,759
Speaker 3: how can we apply the thing that we created to

606
00:34:58,760 --> 00:35:03,560
Speaker 3: automate ourselves and automate law and automate government, and automate

607
00:35:04,960 --> 00:35:08,200
Speaker 3: manufacturing and automate farming. Like they're going to go off

608
00:35:08,239 --> 00:35:10,280
Speaker 3: and spread out, and it's kind of like a virus,

609
00:35:10,640 --> 00:35:14,200
Speaker 3: and we're going to see a COVID nineteen type situation

610
00:35:14,719 --> 00:35:17,040
Speaker 3: where we saw it on the news but it wasn't

611
00:35:17,040 --> 00:35:19,840
Speaker 3: happening to us. We saw it happening in Italy, but

612
00:35:19,840 --> 00:35:22,120
Speaker 3: it wasn't happening to us, and then suddenly it was

613
00:35:22,160 --> 00:35:24,840
Speaker 3: happening to us. And I think that's what this automation

614
00:35:25,000 --> 00:35:27,480
Speaker 3: is going to happen. It's going to be slow, then

615
00:35:27,480 --> 00:35:30,320
Speaker 3: it's going to be fast, and I'm generally quite concerned

616
00:35:30,320 --> 00:35:33,160
Speaker 3: about that, but like I think, it's just part of nature.

617
00:35:33,520 --> 00:35:35,600
Speaker 3: We automate, and then we find new jobs, and then

618
00:35:35,600 --> 00:35:38,920
Speaker 3: we automate. But the transition between those two points is

619
00:35:38,960 --> 00:35:41,480
Speaker 3: the difficult one that I feel like there needs to

620
00:35:41,520 --> 00:35:44,360
Speaker 3: be some sort of plan around that, because when you

621
00:35:44,440 --> 00:35:48,799
Speaker 3: have large mascal employee changes, you're going to have to

622
00:35:48,840 --> 00:35:53,319
Speaker 3: have educational changes, like adult upskilling. So I feel like

623
00:35:53,320 --> 00:35:57,840
Speaker 3: it's going to start in the software engineer industry and

624
00:35:57,880 --> 00:36:01,800
Speaker 3: then it's going to spread out from there, probably through admin,

625
00:36:01,880 --> 00:36:05,760
Speaker 3: probably through government, probably through marketing, just because those jobs

626
00:36:05,760 --> 00:36:10,600
Speaker 3: are generally very text heavy and LMS are very good

627
00:36:10,600 --> 00:36:11,600
Speaker 3: at text heavy things.

628
00:36:11,760 --> 00:36:16,200
Speaker 2: We've seen the markets respond to this with the massive

629
00:36:16,239 --> 00:36:21,000
Speaker 2: dipping software as a service company market caps a couple

630
00:36:21,040 --> 00:36:25,960
Speaker 2: of weeks back, particularly around law firms that you know analysts.

631
00:36:26,000 --> 00:36:30,319
Speaker 2: We're looking at the impact AI is having on law

632
00:36:30,400 --> 00:36:34,680
Speaker 2: teams and the automation of what they do. So it's

633
00:36:34,719 --> 00:36:38,120
Speaker 2: definitely coming. You know. I've seen Datacom, the local our

634
00:36:38,160 --> 00:36:41,000
Speaker 2: biggest tech company locally, has shown me how it's using

635
00:36:41,120 --> 00:36:45,840
Speaker 2: agents alongside its human software developers to massively shorten the

636
00:36:45,880 --> 00:36:48,680
Speaker 2: amount of time it takes to create an application or

637
00:36:48,760 --> 00:36:52,000
Speaker 2: to modernize an old platform. We've got lots of those

638
00:36:52,600 --> 00:36:57,480
Speaker 2: in New Zealand. So that has huge implifications for not

639
00:36:57,520 --> 00:36:59,640
Speaker 2: only the cost of a project. They can now offer

640
00:36:59,680 --> 00:37:06,040
Speaker 2: fixed price contracts for doing a app modernization program. But yeah,

641
00:37:06,040 --> 00:37:08,400
Speaker 2: they're not going to need as nearly as many software developers,

642
00:37:08,400 --> 00:37:09,880
Speaker 2: and as you say, that's just going to spread to

643
00:37:09,920 --> 00:37:10,640
Speaker 2: every industry.

644
00:37:10,840 --> 00:37:13,759
Speaker 3: Yeah, there's a small bit of arbitrage that's happening in

645
00:37:13,800 --> 00:37:18,080
Speaker 3: the minute where not everyone knows about software being mostly automated.

646
00:37:18,800 --> 00:37:23,800
Speaker 3: So I can open a anthropic window on this computer

647
00:37:24,000 --> 00:37:26,280
Speaker 3: and I can type in something that takes me fifteen

648
00:37:26,280 --> 00:37:28,839
Speaker 3: minutes to type out, and then I could generate it,

649
00:37:28,920 --> 00:37:31,440
Speaker 3: and then I could fix three or four problems, and

650
00:37:31,480 --> 00:37:35,239
Speaker 3: I'd have a fully working functional product like Sero. I

651
00:37:35,280 --> 00:37:38,280
Speaker 3: could have a fully functioning product like Manage my Health,

652
00:37:38,680 --> 00:37:40,439
Speaker 3: and then I could spend a little bit more time

653
00:37:40,480 --> 00:37:43,360
Speaker 3: and make it secure. But like, the time to market

654
00:37:43,480 --> 00:37:46,719
Speaker 3: for these things is going to be incredibly short. I

655
00:37:46,960 --> 00:37:50,759
Speaker 3: foresaw that there would be a huge influx of like

656
00:37:51,080 --> 00:37:53,680
Speaker 3: I called it, company spam, where there would just be

657
00:37:53,880 --> 00:37:56,879
Speaker 3: random companies just springing up out of nowhere. And when

658
00:37:56,880 --> 00:38:00,400
Speaker 3: you have company spam, there's this trust factor who do

659
00:38:00,440 --> 00:38:04,640
Speaker 3: we trust? And I felt like the large software vendors

660
00:38:04,640 --> 00:38:07,240
Speaker 3: would be safe because they have this brand around them.

661
00:38:07,400 --> 00:38:10,560
Speaker 3: So rather than trying the new zero that is fifty

662
00:38:10,560 --> 00:38:13,480
Speaker 3: dollars cheaper, I would just stick with the old zero.

663
00:38:13,640 --> 00:38:16,839
Speaker 3: That's like safe and no one to be good. So

664
00:38:16,880 --> 00:38:19,200
Speaker 3: I think these companies generally are going to be pretty

665
00:38:19,239 --> 00:38:24,360
Speaker 3: safe long term in their business model. For another reason

666
00:38:24,440 --> 00:38:28,080
Speaker 3: is they have these large contracts already set up. And

667
00:38:28,120 --> 00:38:30,760
Speaker 3: when you have these established contracts and established trust channels

668
00:38:30,800 --> 00:38:34,120
Speaker 3: with enterprise, you're pretty stable. You've got a pretty stable

669
00:38:34,120 --> 00:38:36,879
Speaker 3: business model. The thing that is difficult now is if

670
00:38:36,880 --> 00:38:39,080
Speaker 3: you have a new company and you want to go

671
00:38:39,239 --> 00:38:42,320
Speaker 3: up through the runs. Typically new companies sell to smaller

672
00:38:42,360 --> 00:38:46,840
Speaker 3: companies first to get their things established, their systems working.

673
00:38:47,080 --> 00:38:50,799
Speaker 3: They have design partners in these smaller companies and they

674
00:38:50,880 --> 00:38:53,000
Speaker 3: build their way up to enterprise with the money is

675
00:38:53,520 --> 00:38:58,160
Speaker 3: And if you don't have those runs, you have to

676
00:38:58,400 --> 00:39:01,239
Speaker 3: manage a SaaS company in a very different way. And

677
00:39:01,280 --> 00:39:04,480
Speaker 3: also typically how SaaS companies work when they're getting established

678
00:39:04,560 --> 00:39:11,040
Speaker 3: is you charge a certain price and then that price

679
00:39:11,120 --> 00:39:13,680
Speaker 3: has to pay for the marketing of like the next person,

680
00:39:14,960 --> 00:39:17,120
Speaker 3: and there's a payback period and there's all this math

681
00:39:17,400 --> 00:39:20,239
Speaker 3: like SaaS math, and the SaaS myth just doesn't work

682
00:39:20,280 --> 00:39:22,879
Speaker 3: if people aren't willing to pay like a hefty price

683
00:39:22,920 --> 00:39:26,359
Speaker 3: for software. So if the value of software goes down,

684
00:39:26,640 --> 00:39:30,279
Speaker 3: the price of software goes down. The feedback mechanism that

685
00:39:30,320 --> 00:39:34,120
Speaker 3: SaaS companies use to perpetually buy more ads and get

686
00:39:34,160 --> 00:39:36,600
Speaker 3: more people, buy more ads, get more people. That whole

687
00:39:36,600 --> 00:39:40,040
Speaker 3: model is shifting. So as you say, there was a

688
00:39:40,080 --> 00:39:42,960
Speaker 3: little bit of a downturn in the SaaS economy when

689
00:39:43,000 --> 00:39:47,480
Speaker 3: Anthropic released those plug ins, But I think the bigger

690
00:39:47,680 --> 00:39:49,799
Speaker 3: impact will be new companies coming up.

691
00:39:49,880 --> 00:39:52,520
Speaker 2: And just finally, Mike tell us about a bot. Where

692
00:39:52,520 --> 00:39:56,080
Speaker 2: you're at with your startup and you know what, So

693
00:39:56,120 --> 00:39:58,000
Speaker 2: you've talked a little bit about what you're proposing to

694
00:39:58,000 --> 00:40:01,280
Speaker 2: do in the cloud for users of agents, but yeah,

695
00:40:01,480 --> 00:40:04,560
Speaker 2: give us the pitch about where you fit into this market.

696
00:40:04,640 --> 00:40:09,000
Speaker 3: Yeah, So we essentially create AI employees or the infrastructure

697
00:40:09,080 --> 00:40:13,560
Speaker 3: for AI employees. So the question really is, I want

698
00:40:13,600 --> 00:40:17,880
Speaker 3: to manage a company at scale, and I want it

699
00:40:17,960 --> 00:40:20,839
Speaker 3: to be fully automated. Let's say in five years time,

700
00:40:21,280 --> 00:40:24,399
Speaker 3: what tools or what infrastructure would be needed to make

701
00:40:24,400 --> 00:40:27,440
Speaker 3: that secure and trustworthy. So we sat down and we said,

702
00:40:27,680 --> 00:40:29,680
Speaker 3: let's not build for now, let's build for the future,

703
00:40:30,160 --> 00:40:33,440
Speaker 3: and how do we make automated companies? And so we

704
00:40:33,520 --> 00:40:35,759
Speaker 3: build out all these different layers of infrastructure. We've got

705
00:40:35,800 --> 00:40:40,239
Speaker 3: eight layers, and we're releasing each one kind of independently.

706
00:40:40,600 --> 00:40:42,800
Speaker 3: So the layer I've been talking about today is basically

707
00:40:42,840 --> 00:40:46,600
Speaker 3: the infrastructure layer. It's the sandboxes. So we give each

708
00:40:46,640 --> 00:40:50,560
Speaker 3: AI its own computer, like a desktop computer that has

709
00:40:50,600 --> 00:40:53,719
Speaker 3: apps and has Photoshop and has a browser, and it

710
00:40:53,800 --> 00:40:57,680
Speaker 3: has an API that the bot can connect to, and

711
00:40:57,719 --> 00:41:00,400
Speaker 3: the bot can control the computer just like an employee would.

712
00:41:00,680 --> 00:41:03,000
Speaker 3: So if you're in an enterprise environment, most of your

713
00:41:03,000 --> 00:41:06,000
Speaker 3: software is these apps and it can just go on

714
00:41:06,280 --> 00:41:08,600
Speaker 3: and watch how you work and then do the job

715
00:41:08,680 --> 00:41:12,120
Speaker 3: for you. That's one of our products. But one step

716
00:41:12,120 --> 00:41:15,360
Speaker 3: below that layer one is the permission lab. But we

717
00:41:15,440 --> 00:41:20,960
Speaker 3: have these very advanced permission controls where it's relational to

718
00:41:21,239 --> 00:41:23,799
Speaker 3: where you are in the company about how what your

719
00:41:24,480 --> 00:41:27,439
Speaker 3: permissions are, and the bot can ask permission to get

720
00:41:27,480 --> 00:41:30,680
Speaker 3: more permissions. And those two layers are the first two

721
00:41:30,719 --> 00:41:33,719
Speaker 3: of the eight. But essentially we create the infrastructure to

722
00:41:33,800 --> 00:41:37,359
Speaker 3: allow this what we feel of hyper automation to come

723
00:41:37,400 --> 00:41:40,960
Speaker 3: to reality. So if you want to come to our site,

724
00:41:41,000 --> 00:41:45,040
Speaker 3: there's a waitlist, some parts will be open you can

725
00:41:45,120 --> 00:41:47,360
Speaker 3: check it out. But yeah, we've been building this for

726
00:41:47,400 --> 00:41:48,600
Speaker 3: the last year.

727
00:41:49,280 --> 00:41:50,840
Speaker 2: And what.

728
00:41:52,320 --> 00:41:55,400
Speaker 3: You see in the market isn't what exists. There's a

729
00:41:55,440 --> 00:41:58,720
Speaker 3: lot of stuff that is just unreleased, and generally people

730
00:41:58,800 --> 00:42:01,880
Speaker 3: keep this stuff unreleased because it's unsafe. What happened with

731
00:42:01,920 --> 00:42:05,280
Speaker 3: open Claw was it's very unsafe, but he just released

732
00:42:05,320 --> 00:42:08,080
Speaker 3: it anywhere, and this is the first time we've seen

733
00:42:08,120 --> 00:42:12,560
Speaker 3: that someone releasing something that has said the art. Usually

734
00:42:12,560 --> 00:42:14,279
Speaker 3: these companies hold on to it for a long time

735
00:42:14,360 --> 00:42:17,320
Speaker 3: while they make sure that it's like secure for legal reasons.

736
00:42:17,719 --> 00:42:19,479
Speaker 3: You know, you don't want to be held liable for

737
00:42:19,800 --> 00:42:22,880
Speaker 3: a company deleting your database or something like that. So

738
00:42:22,960 --> 00:42:25,880
Speaker 3: essentially what we do is we make things safe for AI.

739
00:42:25,920 --> 00:42:29,040
Speaker 3: We give that compliance layer, that order tally layer.

740
00:42:28,680 --> 00:42:31,279
Speaker 2: And where you're at in terms of capital raising and

741
00:42:31,360 --> 00:42:34,239
Speaker 2: the area sort of go to market. You're targeting New

742
00:42:34,360 --> 00:42:37,279
Speaker 2: Zealand first, Are you looking to go global from day one?

743
00:42:37,640 --> 00:42:40,080
Speaker 3: Yeah, So we're targeting enterprise first. So we've been around

744
00:42:40,120 --> 00:42:43,080
Speaker 3: for a while. We've got some people that we're you know,

745
00:42:43,320 --> 00:42:46,920
Speaker 3: working with quite closely, some large companies, So we're targeting

746
00:42:46,960 --> 00:42:49,799
Speaker 3: that enterprise layer. And you can kind of tell that

747
00:42:49,800 --> 00:42:52,160
Speaker 3: that's what we're targeting based on how I talk about,

748
00:42:52,239 --> 00:42:55,760
Speaker 3: you know, compliance and safety, because that's the main factors

749
00:42:55,760 --> 00:42:58,759
Speaker 3: of these large companies. Smaller companies don't care so much

750
00:42:58,840 --> 00:43:02,440
Speaker 3: about that should but like you know, they just don't.

751
00:43:03,320 --> 00:43:06,480
Speaker 3: So we're targeting large companies and then large companies overseas

752
00:43:06,520 --> 00:43:10,600
Speaker 3: as well. And also this each of these layers also

753
00:43:10,640 --> 00:43:13,239
Speaker 3: has a consumer element, so anybody can set up a

754
00:43:13,280 --> 00:43:17,719
Speaker 3: sandbox for example, but the whole system is very much

755
00:43:17,760 --> 00:43:22,040
Speaker 3: designed for enterprise mainly because that's where all the employees are.

756
00:43:22,520 --> 00:43:25,239
Speaker 3: And if you want to automate labor, which a lot

757
00:43:25,280 --> 00:43:26,960
Speaker 3: of these companies are coming to us to say, hey,

758
00:43:27,000 --> 00:43:29,799
Speaker 3: we want to say, reduce headcount on this department by

759
00:43:29,840 --> 00:43:32,080
Speaker 3: twenty percent. That we're the people that you come to.

760
00:43:32,080 --> 00:43:34,799
Speaker 2: For that eventually. You know, we are a country of

761
00:43:34,840 --> 00:43:37,840
Speaker 2: small businesses. Could you see a version of this available

762
00:43:37,920 --> 00:43:40,799
Speaker 2: for tradees and small business owners as well?

763
00:43:40,920 --> 00:43:43,000
Speaker 3: Yeah, I mean we're infrastructure companies, So what we do

764
00:43:43,080 --> 00:43:45,759
Speaker 3: is we sell out infrastructure to other providers. So you

765
00:43:45,880 --> 00:43:50,640
Speaker 3: mentioned auto Hive, there's another company called acarnam Ai and Wellington.

766
00:43:50,719 --> 00:43:55,160
Speaker 3: There's a couple of other companies and we're working closely

767
00:43:56,160 --> 00:44:00,600
Speaker 3: with some of those guys to give out eechnology to

768
00:44:00,680 --> 00:44:05,920
Speaker 3: their platform. So we wouldn't be dealing with those customers directly,

769
00:44:06,000 --> 00:44:10,560
Speaker 3: but our infrastructure maybe. So we're a country of small

770
00:44:10,560 --> 00:44:14,560
Speaker 3: business right and a lot of businesses business owners are

771
00:44:14,640 --> 00:44:16,960
Speaker 3: very stressed about all the admin stuff they have to do,

772
00:44:17,320 --> 00:44:19,919
Speaker 3: and I feel like that's where Autoha and that's where

773
00:44:19,920 --> 00:44:24,000
Speaker 3: a Kardom come in. We're able to sort those problems

774
00:44:24,000 --> 00:44:27,320
Speaker 3: out and hopefully they will use our infrastructure. I'm pretty

775
00:44:27,320 --> 00:44:29,279
Speaker 3: sure that you know, at least a couple of them will.

776
00:44:29,760 --> 00:44:33,200
Speaker 3: And yeah, that's how we kind of see things as

777
00:44:33,360 --> 00:44:36,880
Speaker 3: we're focused on the enterprise and the infrastructure there, and

778
00:44:36,920 --> 00:44:39,239
Speaker 3: then we sell on that service to other companies that

779
00:44:39,320 --> 00:44:41,240
Speaker 3: have these different customer segments.

780
00:44:41,320 --> 00:44:43,359
Speaker 2: Well, it sounds like, you know, there is a huge

781
00:44:43,360 --> 00:44:47,120
Speaker 2: amount of potential in agentic AI. It's not all hypeware.

782
00:44:47,239 --> 00:44:51,359
Speaker 2: But I guess your message to businesses getting into this

783
00:44:51,480 --> 00:44:54,040
Speaker 2: is really go into that with your eyes open, and

784
00:44:54,719 --> 00:44:56,719
Speaker 2: you know, do the sandbox and do the stuff that

785
00:44:56,880 --> 00:44:59,640
Speaker 2: is going to make sure that the bots only access

786
00:44:59,680 --> 00:45:02,719
Speaker 2: the that you want them to, that they can't sort

787
00:45:02,719 --> 00:45:06,040
Speaker 2: of take control of things that they shouldn't. So security

788
00:45:06,280 --> 00:45:07,239
Speaker 2: needs to be friend of mine.

789
00:45:07,320 --> 00:45:11,640
Speaker 3: Yeah, especially if you're working in very regulated industries. If

790
00:45:11,640 --> 00:45:14,000
Speaker 3: you're a gardener or you know, you have like a

791
00:45:14,040 --> 00:45:19,200
Speaker 3: team of fifteen people that you know, or builders, there's

792
00:45:19,239 --> 00:45:22,240
Speaker 3: a lot of regulation there. But it's like very easy

793
00:45:22,360 --> 00:45:24,719
Speaker 3: kind of stuff that's free, admin heavy. I feel like

794
00:45:24,760 --> 00:45:27,359
Speaker 3: those guys can experiment a little bit more, be more

795
00:45:27,360 --> 00:45:29,719
Speaker 3: open with it. But if you're a company, a mid

796
00:45:29,760 --> 00:45:32,120
Speaker 3: cap company of like one hundred people and you're not

797
00:45:32,320 --> 00:45:35,440
Speaker 3: experimenting in this space, I recommend you find someone in

798
00:45:35,440 --> 00:45:39,480
Speaker 3: your company that is quite innovative and just give him

799
00:45:39,520 --> 00:45:42,719
Speaker 3: two months to get familiar with AI agents and to

800
00:45:42,760 --> 00:45:46,320
Speaker 3: get familiar in the space because there's this COVID nineteen

801
00:45:46,600 --> 00:45:49,480
Speaker 3: moment that's happening where we see it overseas, but it's

802
00:45:49,480 --> 00:45:52,760
Speaker 3: not affecting us. And if you are not a savvy

803
00:45:53,080 --> 00:45:56,600
Speaker 3: of when it hits, there's your competitors will be well.

804
00:45:56,640 --> 00:46:01,080
Speaker 2: Good advice, Mike, thanks so much for unpacking open claw,

805
00:46:01,280 --> 00:46:04,759
Speaker 2: molt book and where agentic ai is going.

806
00:46:04,760 --> 00:46:06,600
Speaker 1: And thanks so much for coming onto Business of Tech.

807
00:46:06,680 --> 00:46:08,120
Speaker 3: No problem, Peter, thanks for the chat.

808
00:46:17,200 --> 00:46:18,280
Speaker 1: That's it for this episode.

809
00:46:18,320 --> 00:46:21,800
Speaker 2: If the Business of Tech well, clearly open clause approach

810
00:46:21,880 --> 00:46:24,720
Speaker 2: to agentic ai is a bit of a game changer,

811
00:46:25,200 --> 00:46:27,880
Speaker 2: but it's also obvious from what Mike said that it

812
00:46:27,960 --> 00:46:32,080
Speaker 2: raises serious security questions, which is why the sandboxes he

813
00:46:32,200 --> 00:46:37,120
Speaker 2: was talking about, strict commissions and enterprise grade infrastructure are

814
00:46:37,160 --> 00:46:40,040
Speaker 2: going to be crucial if businesses want to use these

815
00:46:40,080 --> 00:46:43,800
Speaker 2: tools safely. I think we'll see the big agentic ai

816
00:46:43,920 --> 00:46:48,000
Speaker 2: vendor scramble to adopt aspects of open claw in the

817
00:46:48,040 --> 00:46:50,840
Speaker 2: coming months, and that's great if it brings us closer

818
00:46:51,000 --> 00:46:54,719
Speaker 2: to the admin busting paradise we've been pitched for the

819
00:46:54,800 --> 00:46:57,640
Speaker 2: last few years. The most sobering takeaway I think is

820
00:46:57,800 --> 00:47:01,320
Speaker 2: Mike's view that software engineers are the early test case

821
00:47:01,800 --> 00:47:05,240
Speaker 2: and that the same automation wave we'll move quickly into

822
00:47:05,360 --> 00:47:07,400
Speaker 2: every knowledge heavy sector.

823
00:47:08,160 --> 00:47:10,440
Speaker 1: For Nowur's advice is, I think pretty solid.

824
00:47:10,440 --> 00:47:14,560
Speaker 2: If you're not already experimenting with agents, find someone in

825
00:47:14,600 --> 00:47:18,560
Speaker 2: your organization who is, give them time to play in

826
00:47:18,640 --> 00:47:21,800
Speaker 2: a sandbox, and get ready for the slow then fast

827
00:47:21,960 --> 00:47:26,520
Speaker 2: phase of automation that is around the corner. Thanks for

828
00:47:26,560 --> 00:47:28,799
Speaker 2: listening to the Business of Tech. Thanks to Mike Hall

829
00:47:28,920 --> 00:47:31,319
Speaker 2: for coming on. I'm Peter Griffin. Catch you next week

830
00:47:31,360 --> 00:47:35,120
Speaker 2: for another episode, dropping first Thing Thursday on your favorite

831
00:47:35,160 --> 00:47:36,239
Speaker 2: podcast platform.

832
00:47:36,400 --> 00:47:36,839
Speaker 1: See you then,