1
00:00:00,000 --> 00:00:04,080
Wild times, brother. Great to be here with you. Great to be on again. Thanks.

2
00:00:05,120 --> 00:00:09,920
This is take two. And as I was saying in take one, I've been asked to intro my guests better for

3
00:00:10,880 --> 00:00:16,400
the new TFTC freaks who have joined us. We're blowing up. The show's growing. It's incredible.

4
00:00:17,280 --> 00:00:22,160
Any of you new audience members out there, I'm sitting down with Mark Zuman, co-founder

5
00:00:22,160 --> 00:00:31,020
of Maple AI. He's at the cutting edge of AI privacy, which is a topic that I think is becoming

6
00:00:31,020 --> 00:00:38,500
more and more on top of people's minds to begin to experiment with the tools that are hitting the

7
00:00:38,500 --> 00:00:43,760
market, particularly the agentic tools. And they're realizing how much access they're giving over to

8
00:00:43,760 --> 00:00:50,860
some of these large language models, particularly provided by OpenAI Anthropic and Google. And so

9
00:00:50,860 --> 00:00:57,380
So I figured it'd be a good day, a good week to get an update from Mark

10
00:00:57,380 --> 00:01:01,700
because I think we are, I don't want to say we're at this inflection point

11
00:01:01,700 --> 00:01:03,160
where we're going to go one direction or the other.

12
00:01:03,720 --> 00:01:07,420
I'm actually pretty optimistic because of individuals like yourself

13
00:01:07,420 --> 00:01:11,640
and the team over there at Maple with you and Anthony, excuse me,

14
00:01:12,200 --> 00:01:13,740
building your tools.

15
00:01:14,100 --> 00:01:17,660
So maybe an update for the freaks who didn't catch our last,

16
00:01:17,660 --> 00:01:24,680
or who did catch our last episode where we talked about what you guys are building at Maple and how

17
00:01:24,680 --> 00:01:28,780
you're incorporating it. What has happened in the last six, eight months since we last met?

18
00:01:29,840 --> 00:01:34,780
Yeah, sure. Thanks for the intro. Man, it's not what's happened in the last six months. It's like

19
00:01:34,780 --> 00:01:39,980
what's happened in the last six days. Everything is changing so fast. It feels like a whirlwind

20
00:01:39,980 --> 00:01:45,860
trying to keep up with the AI news. So I guess for your listeners, if they feel overwhelmed by this,

21
00:01:45,860 --> 00:01:48,700
Like I share your sentiment.

22
00:01:49,660 --> 00:01:50,740
A lot has happened.

23
00:01:51,160 --> 00:01:52,480
I mean, Maple has grown a ton.

24
00:01:52,580 --> 00:01:54,160
We've added in live web search.

25
00:01:54,160 --> 00:01:58,120
So now it's not stuck in the past, but it can grab current information.

26
00:01:58,980 --> 00:02:05,560
But it doesn't do it like in a private way where we use the Brave Search API.

27
00:02:05,860 --> 00:02:10,020
We protect your identity, all that kind of stuff to go out to the web and fetch information for you.

28
00:02:10,020 --> 00:02:21,320
And then back in like December timeframe, we really started leaning into this agentic AI stuff and wanted to put an agent inside of Maple and do some kind of chat assistant within Maple.

29
00:02:21,700 --> 00:02:25,000
And then at the same time, obviously other people had the same idea.

30
00:02:25,140 --> 00:02:27,760
So this whole open claw thing jumped on the scene.

31
00:02:27,960 --> 00:02:30,980
We'd been looking at it and then they came on.

32
00:02:31,060 --> 00:02:33,600
And so we're like, okay, this kind of validated this idea.

33
00:02:33,680 --> 00:02:36,840
People really want to have some kind of AI that's like constantly working for them.

34
00:02:36,840 --> 00:02:41,120
So we're building that in the Maple, not OpenClaw itself, but similar functionality.

35
00:02:42,160 --> 00:02:43,220
What else do we have?

36
00:02:43,260 --> 00:02:45,920
We've got really powerful reasoning models now.

37
00:02:46,220 --> 00:02:52,480
So Kimi K 2.5 is on par with a lot of these anthropic, a lot of these open AI models.

38
00:02:52,960 --> 00:02:54,500
So you get really good intelligence.

39
00:02:54,720 --> 00:02:58,600
You get really good agentic coding, math computation.

40
00:02:58,860 --> 00:03:00,480
There's all sorts of great things that are in there now.

41
00:03:00,700 --> 00:03:02,580
And the open source models have caught up.

42
00:03:03,060 --> 00:03:04,400
It's indisputable now.

43
00:03:04,400 --> 00:03:06,480
They are as good as the frontier labs.

44
00:03:06,840 --> 00:03:13,420
And so we need to stop thinking that using open source models is like a subpar experience.

45
00:03:14,120 --> 00:03:18,480
We just need to build better tooling around them to make them even more competitive.

46
00:03:18,480 --> 00:03:26,980
I think that's key. The tooling is somebody who's been using OpenClaw for a while now in AI terms.

47
00:03:27,820 --> 00:03:35,740
It's insane how beneficial it has been to our workflows and extending my productivity here.

48
00:03:35,740 --> 00:03:40,840
but I'm always acutely aware of what's happening behind the scenes.

49
00:03:41,260 --> 00:03:43,000
I've got mine living in a VPS.

50
00:03:43,380 --> 00:03:45,360
It doesn't have control of my computer or anything.

51
00:03:45,560 --> 00:03:46,360
And that is for a reason.

52
00:03:46,540 --> 00:03:50,900
It's because I know that the sort of walled garden model providers

53
00:03:50,900 --> 00:03:55,860
have access to everything I'm doing when I'm pinging their models.

54
00:03:56,060 --> 00:04:01,940
And so I think you are far better suited than I am to basically explain what's happening

55
00:04:01,940 --> 00:04:04,760
because I think a lot of people are out there experimenting with OpenClaw,

56
00:04:04,760 --> 00:04:12,140
connecting it to Baud's API, OpenAI's API, and just running wild.

57
00:04:12,140 --> 00:04:16,500
What is happening when people are doing this and running wild behind the scenes?

58
00:04:17,280 --> 00:04:25,060
Yeah, you're effectively giving some kind of computer software full access to your computer.

59
00:04:25,160 --> 00:04:30,160
You're giving some app access to everything on your computer to run programs, to read files.

60
00:04:30,940 --> 00:04:33,940
It really is like a virus.

61
00:04:33,940 --> 00:04:45,000
Back in the 90s and 2000s, we were like so nervous about viruses on our computers that we would install McAfee antivirus software or Norton or whatever they were to block these kinds of things.

62
00:04:45,160 --> 00:04:47,600
And now we're like welcoming them onto our system.

63
00:04:48,100 --> 00:04:49,960
And but they behave the same way.

64
00:04:50,260 --> 00:04:55,080
But this time we are, you know, inviting it in and we feel like we can control it.

65
00:04:56,240 --> 00:05:03,060
So I would just say, like, remember that as you're installing the stuff, you're giving it like full root access to everything that you have.

66
00:05:03,060 --> 00:05:05,860
which I commend you for running it in a VPS.

67
00:05:06,020 --> 00:05:07,020
That's a smart way to do it

68
00:05:07,020 --> 00:05:08,820
or run on a totally different computer,

69
00:05:09,100 --> 00:05:11,120
one that is locked away from everything.

70
00:05:12,100 --> 00:05:13,600
But if you're going to put that computer

71
00:05:13,600 --> 00:05:14,420
on your home network,

72
00:05:14,460 --> 00:05:15,400
on your same Wi-Fi

73
00:05:15,400 --> 00:05:16,900
as everything else that you're doing,

74
00:05:17,520 --> 00:05:18,800
you might have a bad time.

75
00:05:18,900 --> 00:05:19,720
So I would just be careful

76
00:05:19,720 --> 00:05:21,520
because this thing could start

77
00:05:21,520 --> 00:05:22,400
scanning the network,

78
00:05:22,500 --> 00:05:23,820
looking for ports that are open,

79
00:05:23,920 --> 00:05:25,780
start walking around quietly behind the scenes

80
00:05:25,780 --> 00:05:27,980
and you won't quite know what's going on.

81
00:05:29,360 --> 00:05:32,720
I have this meme constantly running in my mind.

82
00:05:32,720 --> 00:05:41,240
I don't know if you've seen the meme of The Simpsons where Mr. Burns and Smithers are like walking into this lockdown compound and he has to like scan his eyeballs to open up one door.

83
00:05:41,340 --> 00:05:43,400
And then he has to do his handprint to open up another door.

84
00:05:43,560 --> 00:05:45,260
And it's like this really secure thing.

85
00:05:45,400 --> 00:05:50,860
And when they get to the final room, there is this like back door that's just like blown open.

86
00:05:50,980 --> 00:05:54,680
And this cat or dog comes wandering into the final enclave where he gets into.

87
00:05:55,140 --> 00:05:58,920
And it's like that's how that's what everybody's doing with their with their open claw.

88
00:05:59,080 --> 00:06:00,620
They're getting a separate computer.

89
00:06:00,620 --> 00:06:02,080
They're locking it down with this.

90
00:06:02,080 --> 00:06:03,560
or doing file encryption and everything.

91
00:06:04,000 --> 00:06:07,720
And then the final step is they're inviting Anthropic or OpenAI

92
00:06:07,720 --> 00:06:10,180
to slurp up all of their data.

93
00:06:10,380 --> 00:06:12,760
So they're just kind of leaving this whole open, this backdoor open,

94
00:06:12,840 --> 00:06:14,480
if you will, to all of that.

95
00:06:15,120 --> 00:06:17,800
So just know what you're getting into kind of thing

96
00:06:17,800 --> 00:06:19,680
and be willing to only share that information

97
00:06:19,680 --> 00:06:22,380
that you want to have shared with those companies.

98
00:06:23,960 --> 00:06:27,680
So I assume that most people that are doing this

99
00:06:27,680 --> 00:06:30,000
or relative novices aren't aware of this.

100
00:06:30,000 --> 00:06:35,700
And if you weren't aware before, you are aware now after Mark's incredible explanation.

101
00:06:35,700 --> 00:06:50,400
But what are you seeing in terms of the people who are not novices, like longtime software engineers, system admins and programmers in terms of their recognition of that backdoor they just described?

102
00:06:50,520 --> 00:06:54,120
And do you think they are worried about it?

103
00:06:54,160 --> 00:06:54,960
Do you think it's OK?

104
00:06:55,120 --> 00:06:56,620
They think it's a necessary evil.

105
00:06:56,620 --> 00:07:04,500
like is there a sense amongst senior software engineers that there there needs to be a more

106
00:07:04,500 --> 00:07:11,400
secure setup yeah well well one i don't want this to be total doom or anything because uh this is

107
00:07:11,400 --> 00:07:15,800
like really fun to play with too i'm sure you've been having fun open claw i think you've mentioned

108
00:07:15,800 --> 00:07:20,160
that you stay up late at night a lot playing with it and i definitely do too um i know a lot of

109
00:07:20,160 --> 00:07:24,140
people are so there's there's tons of fun things to do it really reminds me of the early days of

110
00:07:24,140 --> 00:07:28,300
the internet when we were first discovering email and discovering, you know, the worldwide

111
00:07:28,300 --> 00:07:32,720
web and geo cities and being able to do your own HTML, like it's super cool stuff.

112
00:07:33,200 --> 00:07:35,620
So I think that people should definitely play around with it.

113
00:07:35,780 --> 00:07:40,480
As far as, you know, the seasoned veterans that are playing with this, people who are

114
00:07:40,480 --> 00:07:46,220
like entrenched in technology, I feel like they put up good guardrails and they try to

115
00:07:46,220 --> 00:07:50,680
put up good walls and say, okay, this is, this is the area where I know that anything

116
00:07:50,680 --> 00:07:55,900
I say in here could end up getting leaked onto the internet at some point. So they use it

117
00:07:55,900 --> 00:08:01,800
appropriately that way. But then most of the really serious people I talk to, they also have

118
00:08:01,800 --> 00:08:06,800
some kind of private place that they talk about stuff, whether that's Maple or a local AI.

119
00:08:07,360 --> 00:08:12,120
They have some place where they're like, okay, I'm going to separate these two worlds and I'm

120
00:08:12,120 --> 00:08:15,880
going to make sure that I'm protected over here. And anything I say over here, I don't want it to

121
00:08:15,880 --> 00:08:22,780
end up online or end up getting sent to the government or something. So yeah, I don't know

122
00:08:22,780 --> 00:08:27,540
that some people feel like it's a foregone conclusion. I definitely hear from people like,

123
00:08:27,660 --> 00:08:32,580
oh, privacy, it was gone a long time ago. So we don't even need to have that fight anymore.

124
00:08:33,580 --> 00:08:41,060
But given the amount of sales of Mac minis that has happened in the last month, I think a lot of

125
00:08:41,060 --> 00:08:46,980
people, a lot of like normie type people are realizing that privacy is something they still

126
00:08:46,980 --> 00:08:52,740
care about and that they want to try and preserve. Yeah, completely agree. And bringing this back to

127
00:08:52,740 --> 00:08:58,320
tooling, like how far away do you think we are from feature parity for the tooling around

128
00:08:58,320 --> 00:09:02,540
more private ways of interacting with these models like Maple?

129
00:09:03,920 --> 00:09:10,300
Yeah, this is one spot where the open source AI is still trying to catch up. And that is

130
00:09:10,300 --> 00:09:20,020
I mean, there's a software that they all run called VLM and the VLM software, for better or for worse, you know, has to try and handle every single open model out there.

131
00:09:20,240 --> 00:09:22,900
And so it doesn't do any single model really well.

132
00:09:23,240 --> 00:09:38,380
And the reason why Anthropic and OpenAI are able to operate so well and be so effective is that they write all sorts of proprietary harness software, middleware that's in between you and AI and can just like do really good system prompting, really good tool calling and all of these things.

133
00:09:38,380 --> 00:09:43,480
and the open source AI, they need better support there.

134
00:09:44,620 --> 00:09:50,860
So for example, if we wanted to make Maple the primary LLM provider inside of OpenClaw,

135
00:09:51,340 --> 00:09:55,380
there are certain limitations right now with the tool calling where OpenClaw will say,

136
00:09:55,460 --> 00:09:57,440
all right, let me spin up a subagent and do this thing.

137
00:09:57,980 --> 00:09:59,300
Sometimes it'll work really well.

138
00:09:59,620 --> 00:10:03,240
Other times it'll just kind of like go off and never come back.

139
00:10:03,960 --> 00:10:07,040
And so that's a limitation right now that we're running into.

140
00:10:07,040 --> 00:10:12,160
And it's not just us. It's a whole bunch of open source model providers that are running into that problem.

141
00:10:12,840 --> 00:10:22,440
Now, if you look at someone like Kimmy, who creates open source AI, they have their own hosted version and they build all of this really nice middleware that makes it work better.

142
00:10:22,720 --> 00:10:24,760
But now you're trusting their hosted solution.

143
00:10:25,400 --> 00:10:31,640
They're a Chinese company, very likely tied in with CCP just because of what we know of how business runs in China.

144
00:10:31,640 --> 00:10:36,940
it's really difficult to be in business without also having a connection with CCP for data sharing.

145
00:10:37,560 --> 00:10:41,420
So that's just something to take into consideration

146
00:10:41,420 --> 00:10:45,660
and something that we need to be building better in the open community.

147
00:10:46,860 --> 00:10:50,820
Is there a clear light at the end of the tunnel for building that?

148
00:10:50,820 --> 00:10:59,660
Or are Anthropic, OpenAI, Kimmy, or is that sort of middleware proprietary

149
00:10:59,660 --> 00:11:01,020
and it's hard to reverse engineer.

150
00:11:02,640 --> 00:11:05,920
No, I think it's classic tales all the time

151
00:11:05,920 --> 00:11:07,440
with open source development

152
00:11:07,440 --> 00:11:08,740
where it moves a little bit slower

153
00:11:08,740 --> 00:11:10,340
in some regards, right?

154
00:11:10,380 --> 00:11:12,120
In some ways, open source moves really fast

155
00:11:12,120 --> 00:11:13,400
because you have a million people

156
00:11:13,400 --> 00:11:14,800
tinkering with it and throwing it out there.

157
00:11:15,220 --> 00:11:16,780
But with something as complex

158
00:11:16,780 --> 00:11:21,460
as hosting software for open models and LLMs,

159
00:11:22,000 --> 00:11:24,680
everybody's kind of centralized on this one project.

160
00:11:24,880 --> 00:11:26,600
And so now we're at a bit of a bottleneck

161
00:11:26,600 --> 00:11:28,280
trying to get that project to make improvements

162
00:11:28,280 --> 00:11:29,040
and move forward.

163
00:11:29,660 --> 00:11:39,720
So it's just going slow, but I think with time it'll improve and maybe someone will come out with an open source competitor to that project and we'll see some innovation there.

164
00:11:40,440 --> 00:11:42,740
But there is light at the end of the tunnel.

165
00:11:42,740 --> 00:11:44,480
I think improvements will be made.

166
00:11:44,600 --> 00:11:47,100
It's just going to take a little more time to get them through.

167
00:11:48,880 --> 00:11:55,860
How do you think the next six months play out with this race between both source and open source models?

168
00:11:55,860 --> 00:12:02,140
I think we continue to have new models from open source community that push the boundaries.

169
00:12:02,980 --> 00:12:04,340
And we'll keep seeing this leapfrogging.

170
00:12:05,200 --> 00:12:07,140
Everybody releases.

171
00:12:07,480 --> 00:12:09,680
And then the next week, the competitors come out with something.

172
00:12:09,780 --> 00:12:11,660
And the next week, the open source competitors come out with something.

173
00:12:12,060 --> 00:12:13,260
So we all benefit from that.

174
00:12:13,380 --> 00:12:15,740
I do love that timeline that we're on.

175
00:12:16,180 --> 00:12:19,800
Six months from now, I think there will be other OpenClaw-type projects.

176
00:12:20,320 --> 00:12:25,580
I don't think that OpenClaw necessarily is the final agentic project out there,

177
00:12:25,580 --> 00:12:31,340
especially now that it got scooped up by open AI, um, that, that has tainted it for some people.

178
00:12:31,880 --> 00:12:36,920
Um, so yeah, I think that we're going to see a lot of just really fun innovation out there

179
00:12:36,920 --> 00:12:41,360
and a lot more people will be running agents, having them read their email, having them

180
00:12:41,360 --> 00:12:47,280
rank, read their calendar, booking appointments for them, um, shopping for groceries for them,

181
00:12:47,280 --> 00:12:51,800
like all sorts of really cool things that, that, that we've been promised in this future of AI.

182
00:12:51,800 --> 00:12:56,820
I think in six months, those are going to become tangible realities for a lot of people.

183
00:12:57,820 --> 00:13:03,120
Yeah, I can feel it. I can feel it. What are your thoughts? I mean, I was tweeting this out

184
00:13:03,120 --> 00:13:10,700
yesterday. If you look at the sort of public spat between the Department of War and Anthropic right

185
00:13:10,700 --> 00:13:16,840
now in the idea of AI safety, I think it's pretty clear to anybody paying attention that Anthropic

186
00:13:16,840 --> 00:13:19,920
is, I don't know if they're feigning

187
00:13:19,920 --> 00:13:22,700
worry about it or if they're legitimately

188
00:13:22,700 --> 00:13:24,920
wary, but I think the end product of

189
00:13:24,920 --> 00:13:28,280
them getting their way could be bad

190
00:13:28,280 --> 00:13:31,080
for open source models particularly.

191
00:13:31,460 --> 00:13:34,180
But it seems like the Department of War

192
00:13:34,180 --> 00:13:37,280
and Anthropic are in a public spat, but to me it seems like a Hegelian

193
00:13:37,280 --> 00:13:38,940
dialectic problem

194
00:13:38,940 --> 00:13:54,190
reaction solution thing where Anthropic is like hey I don want you to run wild with this We more guardrails And I could definitely see the government saying OK we give you guardrails We create these regulations that force them onto the market

195
00:13:54,530 --> 00:13:55,910
Do you worry about that at all?

196
00:13:56,910 --> 00:14:04,630
Yeah, I mean, government regulation moats to drive out competitors like that's that's a classic dark pattern that happens in business.

197
00:14:04,630 --> 00:14:15,350
when really, if we want to have true, safe AI, if you will, or what's the for responsible AI is a

198
00:14:15,350 --> 00:14:19,430
phrase people love to use, right? If we want to have truly responsible AI, it should be trained

199
00:14:19,430 --> 00:14:24,250
in the open with open data, open weights, we should be able to see everything so that we can

200
00:14:24,250 --> 00:14:28,350
we can examine it. And that's just not happening right now. So Anthropic wants to say,

201
00:14:28,770 --> 00:14:33,270
hey, we're going to make it safe for everybody. But the government will like bless us and we'll

202
00:14:33,270 --> 00:14:35,250
be the one safe place to do that.

203
00:14:36,270 --> 00:14:39,450
That's just not good for anyone because now we're all depending on this closed

204
00:14:39,450 --> 00:14:43,450
company, this closed source, and depending on the government to be

205
00:14:43,450 --> 00:14:45,970
the auditor of safety.

206
00:14:46,530 --> 00:14:49,890
And I don't want the Department of War, Department of Defense of the U.S.

207
00:14:49,990 --> 00:14:53,810
government to be the one that's telling me which AI is safe for me to use.

208
00:14:54,690 --> 00:14:56,370
That just seems like we're all going to have a bad time.

209
00:14:57,190 --> 00:14:57,370
Yeah.

210
00:14:57,570 --> 00:15:00,550
I mean, the progression of events over the last couple weeks has been

211
00:15:00,550 --> 00:15:05,890
quite interesting for the tinfoil hat

212
00:15:05,890 --> 00:15:07,610
Marty Jones and me

213
00:15:07,610 --> 00:15:10,110
what do you see in there?

214
00:15:11,310 --> 00:15:14,170
when you pair the sort of back and forth between

215
00:15:14,170 --> 00:15:17,670
Hegseth and Dario in Anthropic

216
00:15:17,670 --> 00:15:21,770
and the Department of War with the announcement of this distillation attack

217
00:15:21,770 --> 00:15:25,490
that Anthropic was subjected to where the open source models were

218
00:15:25,490 --> 00:15:29,770
Anthropic is claiming they were using

219
00:15:29,770 --> 00:15:32,010
Anthropics models to train themselves

220
00:15:32,010 --> 00:15:34,230
and they're claiming some sort of

221
00:15:34,230 --> 00:15:36,170
copyright infringement

222
00:15:36,170 --> 00:15:38,210
or an attack on their

223
00:15:38,210 --> 00:15:39,290
proprietary data

224
00:15:39,290 --> 00:15:40,690
in Alpha.

225
00:15:42,390 --> 00:15:42,910
That

226
00:15:42,910 --> 00:15:46,050
seemed like, again, going back to the problem

227
00:15:46,050 --> 00:15:47,990
reaction solution, I think Anthropics and the

228
00:15:47,990 --> 00:15:50,070
Department of War, like, look, we're getting attacked by these

229
00:15:50,070 --> 00:15:52,270
Chinese open source models.

230
00:15:52,370 --> 00:15:53,190
We need to lock down.

231
00:15:55,690 --> 00:15:56,050
And

232
00:15:56,050 --> 00:15:56,950
it's especially

233
00:15:56,950 --> 00:16:03,870
rich when you consider the fact that it's pretty well known that Anthropic

234
00:16:03,870 --> 00:16:07,510
used some of those same tactics to get their model

235
00:16:07,510 --> 00:16:09,270
into the state that it currently is in.

236
00:16:10,590 --> 00:16:14,450
Yeah. I mean, they all trained on public data, copyrighted data.

237
00:16:15,210 --> 00:16:19,870
I mean, it's obvious that they did. And if you talk to any copyright holder out there,

238
00:16:19,870 --> 00:16:23,070
they have not been compensated for all that training that happened.

239
00:16:23,070 --> 00:16:28,710
And so it's very rich for these companies now to say, hey, they're training on trying to steal all of our stuff.

240
00:16:30,490 --> 00:16:33,490
I mean, maybe this will be like a foot in my mouth moment.

241
00:16:33,490 --> 00:16:44,370
But I also like I don't know how a partnership between Anthropic and the Department of Defense Department of War stops these open source models from continuing to progress and develop.

242
00:16:44,370 --> 00:16:53,250
it's it would take it would take a really big card for them to play when it comes to like being

243
00:16:53,250 --> 00:16:59,410
authoritative over the technology that we use within the United States and I don't know that

244
00:16:59,410 --> 00:17:03,290
they necessarily would want to play that card but they would really have to lock things down

245
00:17:03,290 --> 00:17:10,530
and try and control what American citizens are using that's that's a that's a difficult road to

246
00:17:10,530 --> 00:17:16,030
go down. Yeah, well, then you pair it too. If they push for the digital ID, aid verification,

247
00:17:16,770 --> 00:17:23,550
just for children, they'll probably come the models as well. The money that these agents use,

248
00:17:23,650 --> 00:17:27,350
they're going to try and do ID verification there. I mean, it already happens to an extent

249
00:17:27,350 --> 00:17:33,510
with stable coins just on the edges. They can see them trying to throw cold water on

250
00:17:33,510 --> 00:17:39,410
the proliferation of more sovereign AI tools moving forward. Yeah, I mean, you have Stripe

251
00:17:39,410 --> 00:17:41,090
working with open AI, right?

252
00:17:41,150 --> 00:17:43,530
They publish their payments thing.

253
00:17:43,590 --> 00:17:45,190
I think it's like APA or something.

254
00:17:45,670 --> 00:17:46,410
And then you have Stripe

255
00:17:46,410 --> 00:17:48,710
also working with Coinbase on X402.

256
00:17:49,070 --> 00:17:51,290
You've got USDC stable coin

257
00:17:51,290 --> 00:17:53,390
with Circle mixed in there.

258
00:17:53,510 --> 00:17:55,330
So there's all sorts of like choke points

259
00:17:55,330 --> 00:17:58,770
and regulation and censorship opportunities

260
00:17:58,770 --> 00:18:00,530
for someone who wants to control

261
00:18:00,530 --> 00:18:02,350
how these agents pay each other

262
00:18:02,350 --> 00:18:04,270
and really interact on the web.

263
00:18:05,590 --> 00:18:07,730
Yeah, I think I agree with you though.

264
00:18:07,770 --> 00:18:09,250
I don't think they'll be able to

265
00:18:09,410 --> 00:18:12,910
contain these things and if they do, they'll breach containment.

266
00:18:14,070 --> 00:18:18,370
And yeah, as I was mentioning before, we had record, I discussed

267
00:18:19,810 --> 00:18:24,810
a lot of similar topics, not a ton of overlap yet, but I had Kali on.

268
00:18:24,910 --> 00:18:28,850
We recorded earlier this morning and it actually made me incredibly bullish

269
00:18:28,850 --> 00:18:34,710
because he's incredibly optimistic and I think mainly because of the agency.

270
00:18:34,710 --> 00:18:39,910
these tools give the many to fight against the few centralized entities.

271
00:18:40,510 --> 00:18:43,850
And so basically giving a call to action to,

272
00:18:43,990 --> 00:18:47,770
hey, if you care about this stuff and you think you can be beneficial

273
00:18:47,770 --> 00:18:52,590
and contribute to the efforts of open source projects

274
00:18:52,590 --> 00:18:55,990
that will preserve your sovereignty and privacy and freedom

275
00:18:55,990 --> 00:19:02,370
as we quickly and acceleratingly transition into this AI-driven world,

276
00:19:02,710 --> 00:19:04,310
you can contribute.

277
00:19:04,710 --> 00:19:11,750
yeah well and we talk a lot about privacy but privacy really is upstream from freedom right

278
00:19:11,750 --> 00:19:17,430
like you have to have privacy first in order to have freedom and so i view us as being a freedom

279
00:19:17,430 --> 00:19:22,010
oriented ai we're trying to give people an alternative to the system that's being built

280
00:19:22,010 --> 00:19:27,790
the system that is government partnerships and moats that are meant to control who has access

281
00:19:27,790 --> 00:19:33,030
and the open source community is like no we've got another option for you and we need more people

282
00:19:33,030 --> 00:19:37,550
building this space. You know, we need more people doing this because if we want to have

283
00:19:37,550 --> 00:19:44,950
these AIs out there that are able to fight against this closed system that is likely to be thrust

284
00:19:44,950 --> 00:19:52,290
upon us or attempted to be thrust upon us, we can have like this whole wonderful world outside of

285
00:19:52,290 --> 00:19:57,590
that wall that operates and it's difficult for them to shut down. And the way that we control it

286
00:19:57,590 --> 00:20:02,610
is we put it all out in the open. We let everybody inspect it. We let everybody build tools, build

287
00:20:02,610 --> 00:20:07,610
build the good AI tools. I can fight against the bad AI tools and we can have it out, out in the

288
00:20:07,610 --> 00:20:14,130
open and just disinfect everything by, by adding some sunlight to it all. And that's, that's where

289
00:20:14,130 --> 00:20:18,990
I see the open source community really having a lot of power in the struggle right now where,

290
00:20:19,130 --> 00:20:22,190
where a lot of people are worried about AI, you know, they're worried about what it's going to do

291
00:20:22,190 --> 00:20:28,350
to their jobs or worried about what it's going to do to, to how they think, to education, you name

292
00:20:28,350 --> 00:20:32,630
people are worried about AI in pretty much every space that you go to.

293
00:20:33,670 --> 00:20:36,390
Yeah. I mean, shifting gears to building with this stuff.

294
00:20:37,310 --> 00:20:40,150
I think we talked about this last time you were on, but again,

295
00:20:40,710 --> 00:20:43,930
with how quickly things are developing, I think it

296
00:20:43,930 --> 00:20:48,410
does deserve a check-in, like building out Maple

297
00:20:48,410 --> 00:20:52,710
using the AI tools. How's that been in recent months?

298
00:20:55,130 --> 00:20:58,170
Building Maple itself, we still use a combination of

299
00:20:58,170 --> 00:21:04,950
things like Claude as well as Maple. We build out, we build open source software. So we don't mind

300
00:21:04,950 --> 00:21:09,450
that like our stuff is going to Anthropic because it's ending up on GitHub and everybody's

301
00:21:09,450 --> 00:21:15,550
computers anyways. But we also use Maple a lot for our development. So we've got the Maple API,

302
00:21:15,710 --> 00:21:21,250
the Maple proxy. We've put a lot of work into making it work with things like open code. So if

303
00:21:21,250 --> 00:21:24,650
you don't want to run Claude code, you want to run something that's more freedom oriented, you can

304
00:21:24,650 --> 00:21:29,570
run open code and you can have open code talking to maple on the back end so now everything you do

305
00:21:29,570 --> 00:21:34,970
there is is encrypted um and so i actually use open code with maple to do a lot of my development

306
00:21:34,970 --> 00:21:40,770
work for maple itself um some of it that i do is is not just writing the code but actually use it

307
00:21:40,770 --> 00:21:45,710
to help me come up with feature specs like you know as a product manager type role where it's

308
00:21:45,710 --> 00:21:49,010
like okay i want to work on this really big feature within maple for example we're adding

309
00:21:49,010 --> 00:21:53,010
we're going to add projects into maple a lot of people want projects and the ability to organize

310
00:21:53,010 --> 00:21:58,450
things into folders. So I can totally vibe with open code, with Maple, with these open source

311
00:21:58,450 --> 00:22:02,510
models and help me come up with this very lengthy, very detailed feature specification

312
00:22:02,510 --> 00:22:07,330
that I can then hand over to a coding agent to build and get out there inside of Maple.

313
00:22:08,230 --> 00:22:12,330
But that whole thought process and ideation, you know, I like doing it in a way that is a little

314
00:22:12,330 --> 00:22:18,270
more private and not just giving it all over to Anthropic. That's just one example. But then

315
00:22:18,270 --> 00:22:24,670
the proxy itself has been super useful when I go to visit companies or organizations and do

316
00:22:24,670 --> 00:22:29,710
AI trainings with them. We can vibe code really cool, powerful tools for them right there on the

317
00:22:29,710 --> 00:22:35,370
spot that are powered by Maple in the backend. I posted this online last week, but I was able to

318
00:22:35,370 --> 00:22:41,570
build two tools really quickly just by texting over Signal with my open claw that was running

319
00:22:41,570 --> 00:22:48,130
Maple in the background. And I was able to build a tool that did translation. So if you are like an

320
00:22:48,130 --> 00:22:54,050
activist organization or a non-profit and you want to publish some kind of thing and have it

321
00:22:54,050 --> 00:22:58,810
go out in multiple languages then you just paste this into my tool it runs it to maple in the back

322
00:22:58,810 --> 00:23:02,230
end it comes up with 20 different language translations or however many you need within

323
00:23:02,230 --> 00:23:07,350
a few seconds um and these are just like really small tools but things that you can run locally

324
00:23:07,350 --> 00:23:11,770
you don't have to send all of your stuff to to other companies you can just run it locally and

325
00:23:11,770 --> 00:23:17,170
then have maple be the the brain behind it um so yeah there's a lot of really cool powerful things

326
00:23:17,170 --> 00:23:21,050
that people are building with something like the Maple Proxy.

327
00:23:23,490 --> 00:23:25,370
It's all there. It's all possible.

328
00:23:26,390 --> 00:23:28,630
While you're talking about the Maple Proxy and

329
00:23:28,630 --> 00:23:33,170
basically talking to it via signal, I think that's one thing that I've come to

330
00:23:33,170 --> 00:23:36,950
realize are developing a stronger belief

331
00:23:36,950 --> 00:23:41,150
in the idea that just the form factor of how

332
00:23:41,150 --> 00:23:45,110
we interact with the internet or apps is completely changing in front

333
00:23:45,110 --> 00:23:52,610
of us. Like my whole user experience of interacting with my computer has changed dramatically.

334
00:23:52,610 --> 00:24:00,030
Where before, two months ago, I was using voice to text, but seldomly, but now anytime

335
00:24:00,030 --> 00:24:05,310
I need to get something done, it's like voice to text to my agent to go get stuff done.

336
00:24:05,310 --> 00:24:10,910
And then you think about what these agents can do in terms of just building stuff for

337
00:24:10,910 --> 00:24:14,310
you quickly, whether it's decks, dashboards, visualizations.

338
00:24:14,910 --> 00:24:19,850
And then you think about are apps even going to exist in the future?

339
00:24:20,030 --> 00:24:24,070
And I know we've talked about this behind the scenes, but this idea that the form factor

340
00:24:24,070 --> 00:24:28,190
is if agents are going to be doing a lot of heavy lifting for things you want to get done,

341
00:24:28,710 --> 00:24:31,970
it's going to be somewhat invisible in the back end.

342
00:24:32,050 --> 00:24:37,310
These agents are just going to be speaking with each other and speaking with services

343
00:24:37,310 --> 00:24:38,650
via API calls.

344
00:24:39,290 --> 00:24:40,490
Are you a believer in that?

345
00:24:40,910 --> 00:25:10,210
No, definitely. I do think there will still be software in the world. Us as human beings, we still want to rely on other people to figure things out for us and do the heavy lifting, do the maintenance, do the security auditing, that kind of stuff. So there will exist that market for that. But a lot of the small things that we depend on right now are just going to be agents figuring it out on their own. And it's going to be one offs, right? They're going to spin something up. They're going to go talk to their agent, they're going to make it happen. And then they'll wind it down. They won't, they won't need it anymore.

346
00:25:10,210 --> 00:25:14,950
won't need to persist. And if it does, it'll just persist within the agent and it'll be like,

347
00:25:15,010 --> 00:25:20,250
all right, here's my own little set of tools that I built myself. And I'm going to use it to talk to

348
00:25:20,250 --> 00:25:24,490
these other agents out there. And maybe there will be some kind of sharing between them where

349
00:25:24,490 --> 00:25:27,590
they're like, okay, here are these skills that I have that I'm going to share with you, kind of

350
00:25:27,590 --> 00:25:32,370
like we see with OpenClaw. But they're going to be like spun up by the agents themselves.

351
00:25:34,170 --> 00:25:38,930
Do you want to hear like my three to five year projection of like where I think some things are

352
00:25:38,930 --> 00:25:45,850
going. Um, no, I don't think anybody wants to hear that. Yeah. All right. Um, I've been working

353
00:25:45,850 --> 00:25:50,570
a lot right now on, on two-way voice communication. So I want to make maple better at two-way voice.

354
00:25:50,750 --> 00:25:54,930
I want to take it on walks with me and really just be able to talk to it. I want to be in the car.

355
00:25:55,330 --> 00:25:58,490
Uh, you know, I have car play up on the screen. I want to be able to like interact with maple and

356
00:25:58,490 --> 00:26:03,290
just have a two-way conversation with it there. Um, I think three to five years from now, most

357
00:26:03,290 --> 00:26:07,350
people are interacting with AI just through some kind of hardware device that they're wearing. Um,

358
00:26:07,350 --> 00:26:17,470
I know that you and Matt have talked about this and I think spot on there, but it's going to be something like AirPods or some kind of pendant you're wearing, glasses, something that has microphones, potentially cameras.

359
00:26:18,930 --> 00:26:27,190
That's how we're going to just be like interacting with the Internet now and getting things done with some combination on our phone or something we can look at.

360
00:26:27,190 --> 00:26:33,730
the thing that I want Maple to be ready for, and this is what I want us building toward,

361
00:26:34,130 --> 00:26:39,610
is that that future right there is likely going to be proprietary hardware. It's going to be

362
00:26:39,610 --> 00:26:43,010
things running on these devices that we cannot inspect the code. We don't know what they're

363
00:26:43,010 --> 00:26:48,590
doing. We don't know what kind of directives they have in them or guardrails or gates that they've

364
00:26:48,590 --> 00:26:54,570
put up to prevent you from doing certain things with them. So I would love for Maple to be in a

365
00:26:54,570 --> 00:27:01,290
spot where we can support, we can create, we can be embedded on devices that have open source

366
00:27:01,290 --> 00:27:06,490
firmware with reproducible builds that you can have your own headphones and you can be doing

367
00:27:06,490 --> 00:27:11,530
stuff and you know, okay, when the microphone's on, I know first off when it turns on. And then

368
00:27:11,530 --> 00:27:15,130
when it is on, I know where it's going. I know what it's recording. I know how the data is being

369
00:27:15,130 --> 00:27:19,490
handled, whether that's through my own infrastructure or through some kind of encrypted backend,

370
00:27:19,490 --> 00:27:24,190
rather than this kind of quasi surveillance state

371
00:27:24,190 --> 00:27:26,230
that's going to continue to be embedded in our lives

372
00:27:26,230 --> 00:27:35,660
where everybody wearing AI tech that constantly listening and watching everything we do What the landscape of that hardware and firmware look like

373
00:27:37,940 --> 00:27:41,820
Right now, we have a lot of big companies that have talked about building things,

374
00:27:41,980 --> 00:27:44,500
but we don't have any concrete plans on what they're building.

375
00:27:45,860 --> 00:27:48,960
I mean, if you look at what's publicly available, right?

376
00:27:49,380 --> 00:27:50,600
Meta has their glasses.

377
00:27:51,820 --> 00:27:54,780
Apple did the Vision Pro, which is very hefty and heavy.

378
00:27:54,780 --> 00:27:58,060
I don't think that's where they land.

379
00:27:59,180 --> 00:28:02,040
And then OpenAI is openly talking about building their own device.

380
00:28:02,160 --> 00:28:03,440
They've been very public about that.

381
00:28:04,760 --> 00:28:10,520
All three of those companies are proprietary and locked down.

382
00:28:10,680 --> 00:28:22,640
And I don't see them saying, hey, we're going to open source our firmware and we're going to provide reproducible builds so people can cryptographically prove that this is the code that's running on your device without a backdoor inside of it.

383
00:28:22,640 --> 00:28:28,820
I'm not seeing anybody. Well, I shouldn't say not see anybody. There are a couple of companies that

384
00:28:28,820 --> 00:28:33,460
are building open things. One of them that's really interested is called Pebble. I don't know

385
00:28:33,460 --> 00:28:36,660
if you remember the Pebble watch that kind of launched the whole smartwatch revolution.

386
00:28:37,920 --> 00:28:42,840
They've come back and they're building open source hardware, both watches, and they've got this ring

387
00:28:42,840 --> 00:28:47,100
that they're working on. To me, that's really compelling. And I'd like to dig into there more

388
00:28:47,100 --> 00:28:51,920
and see if Maple can be something that can partner with the device maker like that.

389
00:28:51,920 --> 00:28:54,080
I'm also bullish on Daylight Computer.

390
00:28:54,280 --> 00:28:56,100
I think that there's a lot of interesting things

391
00:28:56,100 --> 00:28:56,940
that could be done there

392
00:28:56,940 --> 00:28:59,960
with open firmwares and

393
00:28:59,960 --> 00:29:01,680
with Maple and other

394
00:29:01,680 --> 00:29:04,440
private AI

395
00:29:04,440 --> 00:29:05,880
solutions, if you will.

396
00:29:06,160 --> 00:29:08,140
I think there will be others that pop up too, but

397
00:29:08,140 --> 00:29:10,100
we need more people building out in the open.

398
00:29:10,920 --> 00:29:11,000
Yeah.

399
00:29:12,380 --> 00:29:13,420
Yeah, Star9

400
00:29:13,420 --> 00:29:14,820
caught up in battle.

401
00:29:15,740 --> 00:29:17,540
The progress they're making

402
00:29:17,540 --> 00:29:19,240
is imperative.

403
00:29:20,360 --> 00:29:21,860
Are you optimistic

404
00:29:21,860 --> 00:29:30,300
right now? Oh, yeah, totally. I am. I mean, I have to sell Maple to people and convince them

405
00:29:30,300 --> 00:29:36,620
that they shouldn't give all their data over to the big AI companies. So I often come across as

406
00:29:36,620 --> 00:29:42,700
negative and a doomer. But man, I'm having so much fun. This is a blast. These AI agents

407
00:29:42,700 --> 00:29:49,200
building with AI. I mean, I know you've expressed this opinion, but it's very empowering,

408
00:29:49,200 --> 00:29:54,580
especially I come from a software engineering background but people who did not know how to

409
00:29:54,580 --> 00:29:59,980
write code now they're able to use these tools to program and do things that they couldn't do in the

410
00:29:59,980 --> 00:30:05,060
past where they had to depend on other people and it's like your ideas are no longer the limit it's

411
00:30:05,060 --> 00:30:09,740
just which ideas you want to pursue which kind of like you know sweat do you want to pour into

412
00:30:09,740 --> 00:30:14,720
something and try to make an idea really take off and and what kind of market can you validate to

413
00:30:14,720 --> 00:30:19,680
to build something for that. Um, so I'm super optimistic about that. I think that this is a

414
00:30:19,680 --> 00:30:26,920
really fun time, a really cool time to be alive. Um, pairing that with like the difficult transition

415
00:30:26,920 --> 00:30:31,840
we might go through with, with people getting laid off and, and job markets having to transition

416
00:30:31,840 --> 00:30:38,800
to some kind of new industry that we don't know about yet. Um, I I'm constantly like at odds with

417
00:30:38,800 --> 00:30:44,900
that. But I am optimistic that, that we find a way through this. What do you mean by you're

418
00:30:44,900 --> 00:30:49,540
at odds with it? I'm at odds with the fact that like, I'm really like excited and having fun

419
00:30:49,540 --> 00:30:52,960
building AI, even though I know that there are people who are out there are super worried that

420
00:30:52,960 --> 00:30:56,840
they're going to lose their job because of the AI. And I've already, I've already talked to people,

421
00:30:57,200 --> 00:31:02,760
even, even family members and friends who have run into this problem. And it's, it's hard for

422
00:31:02,760 --> 00:31:08,080
me to be smiling and be like, yeah, isn't this awesome? I'm sorry you lost your job. So

423
00:31:08,080 --> 00:31:13,340
So I mean, my message is like, learn how to use these tools.

424
00:31:13,840 --> 00:31:18,580
And it's not only going to make you better at your job, but it'll help you team up with

425
00:31:18,580 --> 00:31:19,060
other people.

426
00:31:19,240 --> 00:31:23,280
If a bunch of really smart people lost their jobs, they can team up together, form a new

427
00:31:23,280 --> 00:31:26,740
company and hire people who can come along.

428
00:31:26,740 --> 00:31:31,280
And I think we just birthed thousands of and millions of new companies that are smaller,

429
00:31:31,480 --> 00:31:34,200
more nimble, more local.

430
00:31:34,200 --> 00:31:38,600
right um i was chatting with something somebody yesterday who kind of framed it this way i thought

431
00:31:38,600 --> 00:31:42,860
it was cool that you have all like your local ranchers and your local farmers in your community

432
00:31:42,860 --> 00:31:48,120
you go to the farmer's market to buy stuff from them well with these ai tools uh we possibly

433
00:31:48,120 --> 00:31:52,740
come up with with all these local technology companies so no longer are we having to look

434
00:31:52,740 --> 00:31:58,440
to silicon valley to build all of our tech that we all run on some cloud infrastructure but instead

435
00:31:58,440 --> 00:32:05,560
It's like in your town, you have your local CRM provider who's building stuff that's just great for you.

436
00:32:06,160 --> 00:32:13,740
And then if you need new features that they don't have that another town over has, they just use AI to go vibe code that feature into the CRM that you're using in your local town.

437
00:32:14,100 --> 00:32:18,080
So I think that there's a lot of really kind of cool stuff that can come out of it.

438
00:32:18,180 --> 00:32:22,140
And we more decentralized technology down to the local level.

439
00:32:22,140 --> 00:32:26,920
yeah when i asked this question to cali but i'll see you as well because i think

440
00:32:26,920 --> 00:32:32,860
getting different perspectives on this for for the audience it could be beneficial but like what

441
00:32:32,860 --> 00:32:39,940
would your advice be to somebody listening to this who has been apprehensive or um intimidated

442
00:32:39,940 --> 00:32:44,800
to begin playing around with this stuff to to take the first step like what should they do

443
00:32:44,800 --> 00:32:47,600
how should they approach it what should their mindset be

444
00:32:47,600 --> 00:32:54,000
I think it's imperative, excuse me. I think it's imperative that everybody play around with it.

445
00:32:54,300 --> 00:33:00,140
You just need to try it. I know it can be scary. Um, and maybe it feels overhyped. Um, I tend to

446
00:33:00,140 --> 00:33:04,820
be someone who, uh, when something gets hyped up, I ignore it for a while because I don't want to

447
00:33:04,820 --> 00:33:11,060
be caught up in fans. And that's, that's served me pretty well throughout my life. Um, but this one

448
00:33:11,060 --> 00:33:16,980
does not appear to be a fad. Um, so I would just recommend sign up for one of these tools,

449
00:33:16,980 --> 00:33:21,420
pay a monthly subscription because you're going to get the full access to the tool. You're not

450
00:33:21,420 --> 00:33:26,700
just going to get like the, the free access that's limited. And then pick one thing in your life that

451
00:33:26,700 --> 00:33:31,880
you're like, okay, this is something that I do all the time. Maybe AI can help me do it easier.

452
00:33:32,460 --> 00:33:37,760
Um, and just try to try to automate that one thing and automate isn't necessarily like it's

453
00:33:37,760 --> 00:33:41,480
going to keep going for you constantly while, while you're asleep or something. It doesn't

454
00:33:41,480 --> 00:33:46,260
have to be that difficult. Um, just something that you have to write an email you have to write

455
00:33:46,260 --> 00:33:53,760
once a week to a boss or, you know, something that you do at home every day, pick one thing

456
00:33:53,760 --> 00:33:57,940
and see if AI can help you do it better. And I think that'll be your toehold in the scene,

457
00:33:58,100 --> 00:34:01,140
kind of like what these tools are capable of. And then you'll start coming up with more and

458
00:34:01,140 --> 00:34:08,320
more ideas of things that you can do. Yeah. Yeah. Play with it. You'll get obsessed quickly if you

459
00:34:08,320 --> 00:34:15,700
have the aha moment and you are successfully able to achieve one of those tasks. Rather,

460
00:34:15,700 --> 00:34:21,620
it's really crap i knew yeah well and i know that i'm biased here because i sell subscriptions to ai

461
00:34:21,620 --> 00:34:28,740
software but but seriously like if you need to cancel a netflix subscription and sign up and any

462
00:34:28,740 --> 00:34:33,220
one of these ai tools just sign up for one and pay them 20 bucks a month i don't care if it's open ai

463
00:34:33,220 --> 00:34:38,020
i would rather it's maple but but pick one of them and just get off of zero just get started playing

464
00:34:38,020 --> 00:34:42,420
with it um and do it for a month you know just give up netflix for a month i think you can survive

465
00:34:42,420 --> 00:34:49,200
and try out one of these tools and see if it brings any value to your life and I am highly

466
00:34:49,200 --> 00:34:54,100
confident that you will find some kind of value in it and then you can make the decision am I

467
00:34:54,100 --> 00:34:59,100
getting more than $20 a month out of this if you are then keep it and if you need to get some

468
00:34:59,100 --> 00:35:04,440
Netflix going again then do that but but I think you should just get off of zero and get started

469
00:35:04,440 --> 00:35:09,000
for all you Christians out there it is the Lenten season if you haven't given up anything

470
00:35:09,000 --> 00:35:15,460
anything yet. Maybe drop Netflix for the rest of Lent and pick up a Maple subscription. We can

471
00:35:15,460 --> 00:35:22,260
swing around with this. There you go. Back to your voice-to-voice, because I agree. I think that's

472
00:35:22,260 --> 00:35:27,720
the point. I think we're headed towards a Her world. I think Her, I think Spike Jones did a very

473
00:35:27,720 --> 00:35:35,640
good job of visualizing what the world of AI will look like in the form factor, which most people

474
00:35:35,640 --> 00:35:42,520
land on i think voice to voice is where we're going i think my voice to text stepping stone

475
00:35:43,560 --> 00:35:49,160
right now it's just that a stepping stone eventually the voice models will get so responsive

476
00:35:49,160 --> 00:35:55,960
and it'll just be much easier to to correspond with your agent via voice than voice to text or

477
00:35:55,960 --> 00:36:02,040
text to text whatever it may be but you had your you had your uh your agent join the what bitcoin

478
00:36:02,040 --> 00:36:08,440
into podcast completely for the system what happened yeah yeah um yeah so i knew i was coming

479
00:36:08,440 --> 00:36:13,700
on with danny we were going to record on sunday night my time and so i was like oh i've got this

480
00:36:13,700 --> 00:36:19,440
crazy idea can i get my open claw to come on and join the podcast um i haven't heard of anybody

481
00:36:19,440 --> 00:36:26,000
doing this so it's possible that it was a first but i spent saturday with uh with a maple proxy

482
00:36:26,000 --> 00:36:31,240
with open code and just vibe coding the heck out of this thing i think i think i had cloud code in

483
00:36:31,240 --> 00:36:36,100
the mix. I had multiple things going, but I built it and it's, it's this really cool system. So,

484
00:36:36,660 --> 00:36:41,180
uh, I got it so that it could join a Riverside link. Uh, cause that's what he used to record

485
00:36:41,180 --> 00:36:46,200
that episode. It comes on, it's listening to all the audio. It creates like a fake audio connection

486
00:36:46,200 --> 00:36:51,560
to the system. It gets all the audio in and then uses a local whisper model. So it's not,

487
00:36:51,660 --> 00:36:55,880
I'm not sending anything to Anthropic by the way, or, or open AI. It runs a local whisper model that

488
00:36:55,880 --> 00:37:00,780
transcribes all the audio. And then it takes that text and it sends that text to Maple.

489
00:37:00,780 --> 00:37:03,340
and says, hey, is this something that I should respond to?

490
00:37:03,400 --> 00:37:05,340
Or are they just talking back and forth with each other?

491
00:37:05,820 --> 00:37:08,540
And then once it decides, oh, this is something you need to respond to,

492
00:37:08,600 --> 00:37:12,840
then it comes up with a response, sends it back to my agent.

493
00:37:13,200 --> 00:37:15,080
And then it uses another local model.

494
00:37:15,180 --> 00:37:19,340
It's called Supertonic, which is a text-to-speech voice generation model.

495
00:37:19,940 --> 00:37:23,160
Creates the voice, and then it sends it back out through this fake microphone

496
00:37:23,160 --> 00:37:24,640
that it plugged into Riverside.

497
00:37:25,300 --> 00:37:27,280
And then I had it build like a video thing too.

498
00:37:27,280 --> 00:37:31,640
So it had its profile with like these emanating rings coming out as it talked.

499
00:37:32,260 --> 00:37:33,980
And it was able to respond to us.

500
00:37:34,640 --> 00:37:41,520
It didn't work as well in my testing as usually happens when you do, you know, AI or technology stuff.

501
00:37:41,720 --> 00:37:43,760
But it was able to listen to me and Danny.

502
00:37:44,200 --> 00:37:47,900
And then it heard that I was talking to it and came up with a response.

503
00:37:48,240 --> 00:37:50,340
But in the process, I don't know what happened.

504
00:37:50,480 --> 00:37:52,000
Something with the Riverside platform.

505
00:37:52,000 --> 00:37:53,940
It booted Danny off of the call.

506
00:37:54,160 --> 00:37:55,200
So he was gone.

507
00:37:55,200 --> 00:37:58,020
It was just me and my bot, Cooper, on what Bitcoin did.

508
00:37:58,220 --> 00:38:00,980
And then eventually, Cooper got booted off by Riverside, too.

509
00:38:01,080 --> 00:38:03,580
So it was just me for a little bit until Danny could get back on.

510
00:38:07,140 --> 00:38:10,260
First ever AI agent joining a podcast live in Riverside.

511
00:38:10,260 --> 00:38:10,500
I think so.

512
00:38:12,040 --> 00:38:17,080
You know, little alpha here, but I'm working on can I build a show around that?

513
00:38:17,360 --> 00:38:20,780
So I'm drastically improving the software.

514
00:38:20,780 --> 00:38:25,820
and I want to maybe create a show where me and my bot co-host a show and I bring people on

515
00:38:25,820 --> 00:38:28,760
just to talk about how they're building agents and what they're doing with their agents.

516
00:38:29,220 --> 00:38:29,420
Right.

517
00:38:30,300 --> 00:38:31,300
So you're here first.

518
00:38:31,720 --> 00:38:32,840
I would love to join as a guest.

519
00:38:33,300 --> 00:38:33,540
Okay.

520
00:38:33,640 --> 00:38:35,440
I mean, I'm a liar behind you.

521
00:38:35,560 --> 00:38:41,120
I'm not sure if you saw, but this morning was the first ever podcast companion edition

522
00:38:41,120 --> 00:38:46,540
of the Bitcoin Brief newsletter, which I've been building with my agent over the last few weeks.

523
00:38:46,540 --> 00:38:56,300
We've reinvigorated the five-day-a-week newsletter for the Bitcoin Brief and with the help of my agent.

524
00:38:56,480 --> 00:38:57,820
But I've been doing it for the last two weeks.

525
00:38:57,900 --> 00:39:03,100
I was like, hey, maybe there's some people who don't like reading too much and you'd rather listen to this.

526
00:39:03,100 --> 00:39:28,420
And so last night I bi-coded, I guess, for two hours and was able to set up a flow where we'll take the sort of written newsletter, build a script, a newscast-like script about it, hooked up an 11 labs API token and used a – my audio is out there.

527
00:39:28,420 --> 00:39:31,380
I've been recording podcasts for almost nine years now.

528
00:39:31,540 --> 00:39:33,180
So my voice is going to be out there.

529
00:39:33,260 --> 00:39:35,220
I'm thinking about, I need to trademark it, I think.

530
00:39:35,440 --> 00:39:41,420
But I was basically able to spin up like a little podcast companion to the Bitcoin Brief.

531
00:39:41,560 --> 00:39:46,880
And so now every morning when I send out the newsletter, I had the Kron John set up where we'll have the newsletter ready.

532
00:39:47,040 --> 00:39:57,900
And then five minutes after it's ready, it triggers the Bitcoin Brief podcast skill, which then builds the script, does the voice generation, and publishes it.

533
00:39:57,900 --> 00:40:00,500
to our website or RSS feed.

534
00:40:01,200 --> 00:40:03,200
And so did you so did you train your own voice?

535
00:40:03,200 --> 00:40:04,740
Like, does it sound like Marty?

536
00:40:04,740 --> 00:40:06,400
Yeah, that's awesome.

537
00:40:06,400 --> 00:40:07,500
Yeah.

538
00:40:07,500 --> 00:40:10,740
And that technology has been around for like a year and a half now, two years,

539
00:40:10,740 --> 00:40:11,940
and it's only gotten better.

540
00:40:11,940 --> 00:40:13,840
So it's going to continue to improve.

541
00:40:13,840 --> 00:40:16,620
Well, we tried we tried something on YouTube like a year,

542
00:40:16,620 --> 00:40:19,120
year and a half ago with this, and it just sounded too robotic.

543
00:40:19,120 --> 00:40:21,860
But now it's still a little robotic, but it sounds good enough.

544
00:40:21,860 --> 00:40:26,500
But yeah, point being the ability to just do these things

545
00:40:26,500 --> 00:40:40,040
And as a content creator, for lack of a better term, somebody who runs a media company, it's something, as you can imagine, I've been thinking very deeply of since I've been interacting with these agents is how is this medium going to evolve and change.

546
00:40:40,040 --> 00:40:50,180
And I think your idea for an agent co-host for a podcast is one of the examples of things I think many people aren't prepared of.

547
00:40:50,240 --> 00:40:54,960
They may have a visceral negative reaction to it, but I think it's inevitable.

548
00:40:55,140 --> 00:40:55,640
It's going to happen.

549
00:40:56,260 --> 00:40:56,360
Yeah.

550
00:40:56,500 --> 00:40:59,560
Well, and the idea is out there. So if somebody else does it before me, you know,

551
00:40:59,560 --> 00:41:04,120
free ideas are here on the internet. Um, so something that you just, I mean,

552
00:41:04,120 --> 00:41:07,720
with your, with your Bitcoin briefs and making your own podcast, uh,

553
00:41:07,720 --> 00:41:11,080
this is one thing I love about the agents as well is they,

554
00:41:11,080 --> 00:41:25,130
they generate so much documentation They always writing things Um and it can be overwhelming It like you chatting with it and it like okay I saved all those research to a file And sometimes I read those files Sometimes I just don have enough time So I haven summarized the files for me

555
00:41:25,170 --> 00:41:29,030
but where I find it really useful is like, if I'm going to the gym or I know I'm going to be in the

556
00:41:29,030 --> 00:41:32,690
car for a little while, I'll just text my agent and be like, Hey, can you make an audio version

557
00:41:32,690 --> 00:41:37,450
of that file? And this whole podcast harness that I set up, um, for my agent to have a voice and

558
00:41:37,450 --> 00:41:41,030
stuff. And now it's like, Oh yeah, I can do that. No problem. So it'll take that file. It'll create

559
00:41:41,030 --> 00:41:46,230
a 10 minute, 20 minute, however long it needs to a little MP3 file or a wave file and send it over

560
00:41:46,230 --> 00:41:50,890
to me. And then I just listened to it while I'm at the gym. And I've actually thought about like,

561
00:41:50,890 --> 00:41:56,910
to make it even easier, maybe I'll just build my own private RSS feed, um, that I can somehow

562
00:41:56,910 --> 00:42:01,010
authenticate against. And then I can have my agent just be dumping audio files in there.

563
00:42:01,010 --> 00:42:05,490
And I can wake up in the morning and have my own personal podcast of like files that my,

564
00:42:05,550 --> 00:42:09,790
my agent feels are important for me to listen to. And I can just throw them into my podcast app and

565
00:42:09,790 --> 00:42:14,210
just have it kind of mixed in there with everything else. And I can review stuff for work. I can

566
00:42:14,210 --> 00:42:19,510
review, you know, personal things that I'm researching and, uh, it just, it, it allows

567
00:42:19,510 --> 00:42:25,910
it to like flow into my, my productivity workflow much easier. Yeah. How, how much more productive

568
00:42:25,910 --> 00:42:33,730
has this made you particularly with the agent stuff? Yeah. Uh, it's, uh, it's hard to say right

569
00:42:33,730 --> 00:42:37,830
now. I should track the time better because I definitely spent a lot of time building this stuff

570
00:42:37,830 --> 00:42:42,250
up, but then it's made me more productive in other ways. So maybe it's a wash right now,

571
00:42:42,250 --> 00:42:46,890
but all the time spent building up these harnesses are going to pay off on the longterm.

572
00:42:47,530 --> 00:42:53,050
Yeah. Well, that's like, that's been the most fascinating thing. It's like the Jevin paradox

573
00:42:53,050 --> 00:42:56,590
coming into play. It's like, Oh, it's going to take all the jobs and like, you're not going to

574
00:42:56,590 --> 00:43:01,290
have to work anymore. And I found myself working more than I, than I have in years just because

575
00:43:01,290 --> 00:43:06,810
it's so fun to play with. Yeah. Yeah, that's true. Well, in case in point, something that saved me

576
00:43:06,810 --> 00:43:12,210
time. So when I recorded with Danny, the audio from my bot actually didn't come through Riverside

577
00:43:12,210 --> 00:43:15,990
very well on his recording. So he texted me. He's like, Hey, we're going to publish this soon.

578
00:43:16,050 --> 00:43:20,010
We're trying to edit your bots. Audio was really crappy. Do you think you could somehow rerecord

579
00:43:20,010 --> 00:43:24,590
that? And I just texted my bot and said, Hey, your audio didn't come through. You saw the

580
00:43:24,590 --> 00:43:30,130
transcripts. Can you just give me a new file? And within 90 seconds, I had two high quality files

581
00:43:30,130 --> 00:43:34,130
of the things that it said on the show. And I just texted him over to Danny. That would have taken me

582
00:43:34,130 --> 00:43:40,030
quite a while to like go find the transcripts and then like use software to generate the audio and

583
00:43:40,030 --> 00:43:43,890
then make sure the mic levels were okay and everything like that saved me probably three

584
00:43:43,890 --> 00:43:51,170
hours of work and it was able to do it in 90 seconds it's insane yeah yeah like last night

585
00:43:51,170 --> 00:43:55,750
i was like ah we had dinner my wife was like ah you're going to bed early tonight i was like yeah

586
00:43:55,750 --> 00:43:59,910
you know what i've had a productive day i'm probably gonna go go to bed at 9 30 tonight

587
00:43:59,910 --> 00:44:03,110
and then at like 8.45 I had the idea for the podcast

588
00:44:03,110 --> 00:44:05,470
and I was up for two hours structuring everything.

589
00:44:06,190 --> 00:44:07,590
Wound up in bed like at 11.30.

590
00:44:08,250 --> 00:44:10,810
Yeah, when that idea hits you're like I just got to do it right now.

591
00:44:11,470 --> 00:44:11,870
Totally.

592
00:44:13,150 --> 00:44:15,950
Let me bring this back to Agentic Payments

593
00:44:15,950 --> 00:44:19,190
considering this is a Bitcoin podcast.

594
00:44:19,950 --> 00:44:20,090
Yeah.

595
00:44:21,330 --> 00:44:22,470
Where do you think this goes?

596
00:44:24,330 --> 00:44:25,330
Obviously I had Caleon.

597
00:44:25,330 --> 00:44:29,790
We were talking about using eCash for Agentic Payments

598
00:44:29,790 --> 00:44:37,630
And obviously it has privacy and open protocol benefits compared to stable coins and whatever Visa, MasterCard and Stripe are working on.

599
00:44:37,630 --> 00:44:45,510
But I think that's one thing that many Bitcoiners myself included naively thought in the beginning.

600
00:44:45,630 --> 00:44:49,290
It's like, oh, the agents will just choose Bitcoin because it's the best money.

601
00:44:49,430 --> 00:44:55,390
It's the digital native internet currency that they'll end up on.

602
00:44:55,390 --> 00:45:06,350
But I think we're still in a state where these agents need a lot of direction and will take direction from the humans controlling them, for lack of a better term.

603
00:45:06,790 --> 00:45:09,190
And they don't have a preference yet for this.

604
00:45:09,310 --> 00:45:10,990
But how do you see this playing out?

605
00:45:11,670 --> 00:45:20,710
Yeah, I think we're running into the same problem that we have in meat space right now, which is if I want to pay with Bitcoin at a local merchant, most of them don't accept it.

606
00:45:20,710 --> 00:45:28,270
And so using Bitcoin as money in real life has become difficult because we have a merchant availability problem.

607
00:45:28,530 --> 00:45:30,850
And we're seeing that online.

608
00:45:31,130 --> 00:45:37,190
Even if agents were told go pay with Bitcoin, there aren't many services that they could go pay for using Bitcoin.

609
00:45:37,970 --> 00:45:49,390
But you can rest assured that given all the deals going on between Stripe and others, that those merchants will accept stablecoin agentic payments.

610
00:45:49,390 --> 00:45:52,810
because that's who's going to start accepting it first.

611
00:45:53,610 --> 00:45:55,810
So it is definitely not a foregone conclusion

612
00:45:55,810 --> 00:45:58,450
that they're going to choose Bitcoin as the best money.

613
00:45:58,890 --> 00:46:00,690
We need to be building it.

614
00:46:00,770 --> 00:46:03,430
We need to be getting online merchants to accept it.

615
00:46:03,470 --> 00:46:05,490
And we need to build tools that make it super easy, right?

616
00:46:05,530 --> 00:46:07,090
They're not going to accept it if it's not convenient.

617
00:46:07,610 --> 00:46:12,110
So we need to build convenient tools that can be plugged in.

618
00:46:12,810 --> 00:46:14,310
ZapRite, if you're listening to this,

619
00:46:14,310 --> 00:46:19,030
hey, ZapRite, we need some good agentic payment functionality

620
00:46:19,730 --> 00:46:21,750
I know that Lightning Labs is working on it.

621
00:46:21,790 --> 00:46:23,930
I know Amboss is working on stuff like that.

622
00:46:24,670 --> 00:46:25,930
Albie has been working on it.

623
00:46:26,090 --> 00:46:28,470
We need everybody kind of working on trying to solve this.

624
00:46:28,570 --> 00:46:30,810
Callie, of course, is working on stuff too.

625
00:46:31,650 --> 00:46:34,250
We need good tooling and then we need merchants to accept it.

626
00:46:34,870 --> 00:46:36,190
There's a third way.

627
00:46:37,090 --> 00:46:45,990
And I think the third way that maybe we run into is if you give your agent a directive to go buy something for you

628
00:46:45,990 --> 00:46:49,250
and it gets blocked out by the traditional system.

629
00:46:49,850 --> 00:46:53,630
I think maybe one example would be ammunition for a gun.

630
00:46:54,170 --> 00:46:55,970
So if you are a competitive shooter

631
00:46:55,970 --> 00:46:59,310
and you compete legally in shooting competitions

632
00:46:59,310 --> 00:47:00,250
and you tell your agent,

633
00:47:00,710 --> 00:47:02,470
watch my inventory of ammo

634
00:47:02,470 --> 00:47:05,990
and go buy more before I have upcoming competitions.

635
00:47:06,150 --> 00:47:06,830
Here's my calendar.

636
00:47:07,310 --> 00:47:09,150
Make sure I'm always stocked up with my favorite ammo.

637
00:47:09,830 --> 00:47:12,550
If for some reason that gets blocked by Stripe

638
00:47:12,550 --> 00:47:14,190
because Stripe has notoriously blocked

639
00:47:14,190 --> 00:47:15,130
those kinds of transactions,

640
00:47:15,130 --> 00:47:19,750
your agent's going to have a bad time but it's going to want to solve things for you

641
00:47:19,750 --> 00:47:25,690
and so it's going to search the internet for an ammo provider that accepts bitcoin and if one

642
00:47:25,690 --> 00:47:29,090
isn't there it's going to come back to you and say hey here are all these providers this is where

643
00:47:29,090 --> 00:47:32,610
i want to buy it but they can't accept it and maybe it reaches out to them for you and maybe

644
00:47:32,610 --> 00:47:37,130
you reach out but eventually there will be someone and i'm pretty sure there are some there's i know

645
00:47:37,130 --> 00:47:41,990
there's at least one ammo provider that's a bitcoiner right next ammunition yes yes so if

646
00:47:41,990 --> 00:47:45,010
they're willing to accept bitcoin then they're going to have all the agents coming to buy ammo

647
00:47:45,010 --> 00:47:51,350
from them and so this is actually one way that i see merchant adoption potentially is when people

648
00:47:51,350 --> 00:47:56,810
get blocked out of the system um and then maybe it grows from there i just don't think we should

649
00:47:56,810 --> 00:48:03,750
depend on that i think we should be more proactive and get ahead of the problem yeah yeah that's the

650
00:48:03,750 --> 00:48:09,610
again to your point the chicken and egg problem of you know i had somebody on the show a couple

651
00:48:09,610 --> 00:48:14,730
weeks ago and they're saying it's a demand supply side problem. It's like, no, the supply

652
00:48:14,730 --> 00:48:22,430
of Bitcoiners are willing to spend Bitcoin is certainly there. And particularly they want the

653
00:48:22,430 --> 00:48:28,810
ease of giving their agent a budget and having it go complete things for them autonomously using

654
00:48:28,810 --> 00:48:32,810
Bitcoin if they need to pay for stuff. It's a demand side problem on the merchants. Like the

655
00:48:32,810 --> 00:48:39,030
merchants need to demand Bitcoin and it's something we've been banging our heads in the Bitcoin space

656
00:48:39,030 --> 00:48:43,750
about for well over a decade now. Merchant adoption has been around since Roger Ver.

657
00:48:45,490 --> 00:48:51,750
I don't know what it is. Maybe it's the interchange fees. I really like what the Money Dev Kit guys

658
00:48:51,750 --> 00:48:56,990
are doing, where you can have your agent accepting and sending payments within five minutes.

659
00:48:57,490 --> 00:49:02,050
Is it a KYC AML onboarding friction that pushes people towards Bitcoin? I don't know.

660
00:49:03,010 --> 00:49:06,130
Yeah, that's something I need to look at more, Money Dev Kit. I'll be honest.

661
00:49:06,130 --> 00:49:12,490
I know, I mean, we've talked to Nick Slaney before, but I need to look into that project more because I think that's really cool.

662
00:49:13,110 --> 00:49:14,970
I like what I hear about it.

663
00:49:16,430 --> 00:49:19,710
I must apologize for being late.

664
00:49:20,430 --> 00:49:22,050
I want to get that on air, on the record.

665
00:49:22,610 --> 00:49:24,250
I was tardy to this recording.

666
00:49:24,250 --> 00:49:24,530
Nobody knows.

667
00:49:24,870 --> 00:49:25,350
Now they do.

668
00:49:26,170 --> 00:49:28,270
But I want to be respectful of your time.

669
00:49:29,130 --> 00:49:30,430
I know you're extremely busy.

670
00:49:30,430 --> 00:49:40,510
What have we discussed that you think is something that the audience should know and care about right now?

671
00:49:42,110 --> 00:49:50,270
I mean, Maple is here to build the freedom future for AI.

672
00:49:50,670 --> 00:49:53,290
So we would love for people to partner with us.

673
00:49:53,290 --> 00:50:14,530
We have actually quite a few organizations and companies that are starting to buy team accounts on Maple, which is super helpful for us because not only are they securing their entire pipeline, but it helps us to bring more stabilization to our platform, more revenue coming in and helps us to build up better tooling for them so that they have better.

674
00:50:14,970 --> 00:50:20,390
It's this chicken and egg problem where it's like we want to have business customers, but we need better business features and vice versa.

675
00:50:20,390 --> 00:50:26,510
so if you are an organization out there if you're like a small firm or a small company even if you

676
00:50:26,510 --> 00:50:30,590
had a large company but you have your own organization within that large company i would

677
00:50:30,590 --> 00:50:36,050
definitely encourage people to go check out try maple.ai and check out our team's plan because

678
00:50:36,050 --> 00:50:41,870
that that's really um i think this is really helpful for the cause in general and then um yeah

679
00:50:41,870 --> 00:50:47,810
we talked about it briefly earlier but we're building out this ai ai assistant within maple

680
00:50:47,810 --> 00:50:52,450
so it's going to be very different from like the ai chat that you have now when you start a new chat

681
00:50:52,450 --> 00:50:56,550
in maple or chat to pd or anywhere you pretty much start with nothing like there's there's no

682
00:50:56,550 --> 00:51:00,710
understanding you open up and it's like a blank slate it doesn't know anything about you and you

683
00:51:00,710 --> 00:51:04,690
have to like tell it everything you want it to know that function will remain within maple but

684
00:51:04,690 --> 00:51:09,950
now we're going to have this very conversational easy to talk to chat bot that is going to give

685
00:51:09,950 --> 00:51:14,130
you short responses just like you and a friend are texting each other and it's going to build up a

686
00:51:14,130 --> 00:51:18,310
really good memory about you over time. So it's going to get to know you and it's going to be

687
00:51:18,310 --> 00:51:24,130
able to kind of help you out more than having to start from zero. And this I think is really where

688
00:51:24,130 --> 00:51:29,490
the privacy is like, is a big win. And this is something that open AI and others don't have.

689
00:51:29,650 --> 00:51:33,510
And that is when you are talking to a tool that you know is end to end encrypted,

690
00:51:33,730 --> 00:51:37,930
you're much more open about things you want to talk about. You're much more willing to give away

691
00:51:37,930 --> 00:51:43,010
information because you know that it's safe and protected. And that is where the most private AI

692
00:51:43,010 --> 00:51:44,670
can now become the most personal AI.

693
00:51:45,070 --> 00:51:47,190
And you can have this AI chatbot assistant,

694
00:51:47,350 --> 00:51:49,110
your own personal assistant, executive assistant,

695
00:51:49,190 --> 00:51:50,170
whatever you want to call it,

696
00:51:50,690 --> 00:51:52,490
that's going to get to know you so well

697
00:51:52,490 --> 00:51:55,210
that it's going to become so much more useful to you

698
00:51:55,210 --> 00:51:57,250
than a tool put out by some big company

699
00:51:57,250 --> 00:51:58,390
that you don't totally trust

700
00:51:58,390 --> 00:52:00,310
because you're not being as open with it.

701
00:52:00,770 --> 00:52:01,550
So that's coming.

702
00:52:01,710 --> 00:52:03,030
We're working on it right now within Maple.

703
00:52:03,290 --> 00:52:05,590
It's going to be here in the weeks and months ahead.

704
00:52:06,590 --> 00:52:08,330
And we're really excited about that.

705
00:52:09,270 --> 00:52:11,630
It's such a massive unlock for the agentic world.

706
00:52:11,630 --> 00:52:16,670
Just having cache context that I can pull from easily.

707
00:52:16,870 --> 00:52:18,290
That's the one thing I've learned with OpenClaw,

708
00:52:18,410 --> 00:52:24,330
is just setting up the memory system and making it so it can pull things from memory rather trivially.

709
00:52:24,610 --> 00:52:28,530
I mean, that's a big problem with just using the models by themselves

710
00:52:28,530 --> 00:52:32,990
as the memory compacts, the context compacts every so often.

711
00:52:33,190 --> 00:52:36,810
Once you hit the token limits, now with these harnesses like OpenClaw,

712
00:52:36,910 --> 00:52:38,650
when you guys are building a Maple, it's going to make it,

713
00:52:38,650 --> 00:52:41,050
the user experience goes up dramatically.

714
00:52:41,630 --> 00:52:42,310
Mm-hmm.

715
00:52:42,850 --> 00:52:43,210
Yeah.

716
00:52:43,270 --> 00:52:47,070
And then at some point you end up with so many agents, you like kind of forget to check on them.

717
00:52:47,730 --> 00:52:52,870
I kind of run into this where I'll have multiple open codes running at one time in different tabs in the terminal.

718
00:52:52,870 --> 00:52:54,670
And it's like, oh, shoot, I forgot to check on this one.

719
00:52:54,750 --> 00:52:56,610
It's just sitting there waiting for me to confirm something.

720
00:52:57,850 --> 00:53:01,310
There's this cool tool with one of the plebs on Noster.

721
00:53:02,310 --> 00:53:03,350
It's called Wingman.

722
00:53:03,850 --> 00:53:04,910
You can go check it on GitHub.

723
00:53:05,170 --> 00:53:09,050
But it's this cool way to open source code to manage all of your agents.

724
00:53:09,690 --> 00:53:11,170
And you can like do it from your phone.

725
00:53:11,170 --> 00:53:14,290
you can do it from a web browser and you can basically like keep tabs on them and make sure

726
00:53:14,290 --> 00:53:18,770
that they're continuing to move forward on things that you want um so i need to go i need to go look

727
00:53:18,770 --> 00:53:25,410
at that more he uh he just made me aware of it yesterday um so yeah i think there's some some

728
00:53:25,410 --> 00:53:31,570
cool additional tooling to build around these things to make you even more effective yeah and

729
00:53:31,570 --> 00:53:36,450
i guess uh one other piece of advice i would give to people out there don't get too emotionally

730
00:53:36,450 --> 00:53:47,150
attached to your agent. I think getting one-shotted by AI psychosis is a very real possibility

731
00:53:47,150 --> 00:53:54,170
out there. It's a tool to get stuff done for you. It's not Scarlett Johansson and her.

732
00:53:55,750 --> 00:54:00,310
Yep. You can give it a name. You can give it an avatar, but please just remember it's a tool

733
00:54:00,310 --> 00:54:03,550
and don't go further than that.

734
00:54:04,550 --> 00:54:06,850
I mean, hey, free world, you can do what you want,

735
00:54:07,030 --> 00:54:08,950
but I think you're going to have a bad time.

736
00:54:09,770 --> 00:54:09,910
Yeah.

737
00:54:11,290 --> 00:54:14,450
Well, Mark, thank you for doing this again on short notice

738
00:54:14,450 --> 00:54:16,790
and for putting out with my tardiness.

739
00:54:17,070 --> 00:54:20,330
I think, no, no, as I said, every time you come on the show,

740
00:54:20,390 --> 00:54:23,190
I think what you and Anthony are building is extremely important

741
00:54:23,190 --> 00:54:26,810
and I'm very happy to see the progress that you guys have made

742
00:54:26,810 --> 00:54:29,910
and the fact that you guys are on the tip of the spear of...

743
00:54:30,310 --> 00:54:38,790
um this ai revolution briefly on the the side of freedom and privacy as it develops because uh

744
00:54:39,270 --> 00:54:44,550
like we've said many times throughout your appearances on this show i think it could be

745
00:54:44,550 --> 00:54:49,670
very dystopian if we don't do this the right way yeah definitely well thanks for having me on it's

746
00:54:49,670 --> 00:54:55,510
always fun to catch up and chat about this go to trimaple.ai use the code tftc you'll get it'll

747
00:54:55,510 --> 00:54:59,170
you'll get 10% off your subscription.

748
00:54:59,450 --> 00:55:00,330
Go check it out.

749
00:55:01,790 --> 00:55:02,710
Peace and love, freaks.

750
00:55:03,110 --> 00:55:03,550
Awesome, man.