Episode Summary
Episode Video
Episode Show Notes & Transcript
Show highlights:
duckbillhq.com
Transcript
1
00:00:00,150 --> 00:00:04,110
For what we're doing and who we wanna talk to, like open source is actually a
2
00:00:04,110 --> 00:00:07,620
value add for them to be able to see what's happening and like what controls
3
00:00:07,620 --> 00:00:11,010
and like what policies they could set before they even get to the conversation.
4
00:00:16,140 --> 00:00:17,730
Welcome to Screaming in the Cloud.
5
00:00:17,910 --> 00:00:19,020
I'm Corey Quinn.
6
00:00:19,200 --> 00:00:23,550
I am joined today by CEO of Paper Compute Brian
7
00:00:23,550 --> 00:00:26,400
Douglas or B Dougie as the kids today call him.
8
00:00:27,255 --> 00:00:28,125
Thank you for joining me.
9
00:00:28,125 --> 00:00:28,875
I appreciate it.
10
00:00:28,935 --> 00:00:29,294
Yeah.
11
00:00:29,294 --> 00:00:30,314
Uh, glad to be here.
12
00:00:31,034 --> 00:00:34,155
This episode is sponsored in part by my day job Duck.
13
00:00:34,155 --> 00:00:37,335
Bill, do you have a horrifying AWS bill?
14
00:00:37,605 --> 00:00:39,495
That can mean a lot of things.
15
00:00:39,705 --> 00:00:42,795
Predicting what it's going to be, determining what it
16
00:00:42,795 --> 00:00:46,629
should be, negotiating your next long-term contract with
17
00:00:46,724 --> 00:00:51,015
AWS, or just figuring out why it increasingly resembles of.
18
00:00:51,130 --> 00:00:54,760
Phone number, but nobody seems to quite know why that is.
19
00:00:55,030 --> 00:00:58,630
To learn more, visit duck bill hq.com.
20
00:00:58,930 --> 00:01:01,780
Remember, you can't duck the duck bill.
21
00:01:01,840 --> 00:01:07,210
Bill, which my CEO reliably informs me is absolutely not our slogan.
22
00:01:07,665 --> 00:01:10,365
So I have to say that the, my discovery of what
23
00:01:10,365 --> 00:01:13,305
you do was a, is a tail rife with disappointment.
24
00:01:13,515 --> 00:01:18,345
I was so excited for a brief instant that when I heard paper compute, oh wow.
25
00:01:18,345 --> 00:01:20,775
It's an e ink company and I finally will have
26
00:01:20,775 --> 00:01:23,505
something other than a Kindle that uses the technology.
27
00:01:23,715 --> 00:01:25,935
But no, it's an AI thing instead.
28
00:01:26,295 --> 00:01:26,715
Yeah.
29
00:01:26,720 --> 00:01:28,065
Uh, sorry to rug pull you on that one.
30
00:01:28,245 --> 00:01:29,265
It was a quick one though.
31
00:01:29,355 --> 00:01:29,955
Uh, yeah.
32
00:01:29,955 --> 00:01:30,705
Paper compute.
33
00:01:30,735 --> 00:01:30,914
Yeah.
34
00:01:30,914 --> 00:01:32,565
We were doing, uh, distributed systems for
35
00:01:32,565 --> 00:01:34,335
agents, well, infrastructure for agents.
36
00:01:34,335 --> 00:01:37,875
So we've been at it for about two months at this point and been quite the ride.
37
00:01:38,175 --> 00:01:39,675
Obviously, AI is moving pretty quick.
38
00:01:40,305 --> 00:01:44,055
I am hesitant to say too much about the current state of AI because between
39
00:01:44,055 --> 00:01:47,805
the time we record this and the time we publish this, which is not that long,
40
00:01:48,045 --> 00:01:50,865
there will doubtless be a bunch of things that come out that made me look.
41
00:01:50,975 --> 00:01:55,115
Foolish, but when you say infrastructure for AI agents, what does that mean?
42
00:01:55,294 --> 00:01:57,964
Yeah, so it's, we started with two different open source projects.
43
00:01:57,964 --> 00:02:01,414
One is telemetry, so traces, if you think about O Hotel, we started
44
00:02:01,414 --> 00:02:05,615
this back in January and at the time O Hotel did not work with LLMs.
45
00:02:05,794 --> 00:02:07,539
I, it works, but it was a, it was kind of hamfisted.
46
00:02:08,164 --> 00:02:08,975
No, that's just hotel.
47
00:02:08,975 --> 00:02:10,535
That's basically its entire tagline.
48
00:02:10,535 --> 00:02:12,095
It works, but it's kind of hamfisted.
49
00:02:12,095 --> 00:02:12,545
It's right on their
50
00:02:12,545 --> 00:02:12,815
website.
51
00:02:12,820 --> 00:02:13,385
Oh, oh, fair enough.
52
00:02:13,415 --> 00:02:13,595
Yeah.
53
00:02:13,595 --> 00:02:13,834
Yeah.
54
00:02:13,894 --> 00:02:17,434
Uh, not from my mouth, but they have been sitting around and like doing their
55
00:02:17,434 --> 00:02:20,975
sort of weekly meetings and they're gonna like, have better structure for this.
56
00:02:21,065 --> 00:02:24,094
But what we wanna do is like telemetry for agents in
57
00:02:24,094 --> 00:02:28,535
particular, and that is the first project, open source, HEPL.
58
00:02:28,780 --> 00:02:30,549
The other one is stereos, which is the play
59
00:02:30,549 --> 00:02:32,350
you the, the place that you play tapes.
60
00:02:32,440 --> 00:02:34,239
And the idea is having a agent runtime.
61
00:02:34,239 --> 00:02:38,049
So it's a VM built on Nix Os, and that's kind of the TLDR.
62
00:02:38,079 --> 00:02:40,150
I could go into like way more detail on the why and
63
00:02:40,209 --> 00:02:42,579
how, but I, I realize we also only only have 30 minutes.
64
00:02:42,820 --> 00:02:45,459
Well, I do wanna dive into this a little bit because when
65
00:02:45,459 --> 00:02:49,240
you say telemetry for ai, that usually means a few things.
66
00:02:49,240 --> 00:02:51,010
There's the metadata around it.
67
00:02:51,040 --> 00:02:52,840
What was the duration of the call?
68
00:02:52,895 --> 00:02:55,745
An application was this tied into what was the provider?
69
00:02:55,745 --> 00:02:58,145
And notably the one I like, what were the tokens?
70
00:02:58,415 --> 00:03:00,305
But then you can also get into it what was the actual
71
00:03:00,305 --> 00:03:03,245
conversation, and that's where I start to feel a little bit hunted.
72
00:03:03,485 --> 00:03:05,255
Which side of that divide do you come down on?
73
00:03:05,285 --> 00:03:07,415
Yeah, so we're doing a bit more of the telemetry where
74
00:03:07,415 --> 00:03:10,415
you have like the tokens and the, the time duration.
75
00:03:10,565 --> 00:03:13,475
We do have the prompt stored from the agent and from the human.
76
00:03:13,475 --> 00:03:15,455
So like the agent does have a bit of the,
77
00:03:15,485 --> 00:03:17,380
with the thinking we'll have like, hey.
78
00:03:17,785 --> 00:03:20,185
Brian asked this thing, and I'm looking at this thing and I
79
00:03:20,185 --> 00:03:22,765
remember this folder here, uh, which is always fun to look at.
80
00:03:22,795 --> 00:03:25,975
I think a lot of that gets squashed in Quad Code and Codex now, but back in,
81
00:03:25,975 --> 00:03:30,055
back in my day last summer, uh, agents would spit that out when it was thinking
82
00:03:30,445 --> 00:03:32,305
it, it's also dangerous on some level
83
00:03:32,305 --> 00:03:34,315
because it it because this has been the case.
84
00:03:34,415 --> 00:03:37,445
20 some odd years where if I wanna make one of the smartest
85
00:03:37,445 --> 00:03:39,545
people I know look like one of the dumbest people on the
86
00:03:39,545 --> 00:03:42,605
planet, all you have to do is just do a read back of things.
87
00:03:42,605 --> 00:03:46,295
They have typed into search engines, like things that, even things that I
88
00:03:46,295 --> 00:03:50,135
know off the top of my head, but I wanna validate like what does S3 cost?
89
00:03:50,135 --> 00:03:52,445
Yeah, I do know that off the top of my head, but every once
90
00:03:52,445 --> 00:03:54,995
in a while I like to make sure that it hasn't changed on me.
91
00:03:55,355 --> 00:03:59,495
And also I have brain farts, which tends to happen as we age.
92
00:03:59,495 --> 00:04:01,055
And yeah, let's just make that work.
93
00:04:01,325 --> 00:04:04,114
The idea now of, oh, something's watching me and actually.
94
00:04:04,290 --> 00:04:06,600
Seeing the dumb question I'm asking the robot.
95
00:04:06,975 --> 00:04:07,695
That's scary.
96
00:04:07,815 --> 00:04:09,915
I'm gonna have to make sure that, uh, I grease some palms.
97
00:04:09,960 --> 00:04:10,250
Yeah.
98
00:04:10,255 --> 00:04:10,425
Yeah.
99
00:04:10,425 --> 00:04:12,825
So like the way it works today, so we're, we're still open
100
00:04:12,825 --> 00:04:15,975
source first and only right now, and it works SQL light.
101
00:04:16,065 --> 00:04:18,255
It basically writes to your, your home directory.
102
00:04:18,735 --> 00:04:18,885
Hmm.
103
00:04:18,975 --> 00:04:19,935
Oh, so it's local telemetry.
104
00:04:20,144 --> 00:04:23,805
Yeah, so it is local, local first, and that's what we have today.
105
00:04:23,865 --> 00:04:25,995
By the time this comes out, we will have a cloud product, we'll
106
00:04:25,995 --> 00:04:29,205
will be multiplayer, uh, where if you choose, you can send your tape
107
00:04:29,205 --> 00:04:33,435
sessions to the cloud for your, your coworkers to also comment on.
108
00:04:33,660 --> 00:04:38,280
You are funded by Heavybit, as am I, and so is Atin, uh, Ellie Huxtable.
109
00:04:38,340 --> 00:04:40,590
And that, that's always been an interesting problem too.
110
00:04:40,590 --> 00:04:42,330
It's, oh, you can take your Shell history
111
00:04:42,480 --> 00:04:44,730
and share that to your colleagues as well.
112
00:04:44,730 --> 00:04:47,790
It's like, oh God, can I pay you extra to not do that?
113
00:04:48,090 --> 00:04:51,390
Like, even now if I've run at two Unwrapped, which is the,
114
00:04:51,510 --> 00:04:55,290
their whole summary thing on this, where it tells me that.
115
00:04:55,545 --> 00:04:57,105
Or the number of unique commands.
116
00:04:57,195 --> 00:04:57,405
Yeah.
117
00:04:57,435 --> 00:05:02,985
My commands failed 23.1% of the time, which as someone who spent 25
118
00:05:02,985 --> 00:05:06,015
years in the Linux admin space, that that's kind of a high number.
119
00:05:06,525 --> 00:05:06,765
Yeah.
120
00:05:06,770 --> 00:05:08,265
It's, it's high number, but I imagine it's
121
00:05:08,265 --> 00:05:11,295
more like golf where, nah, I have no idea.
122
00:05:11,505 --> 00:05:14,985
But I feel like with like, 'cause I'm, I'm a pretty heavy Vim user, so
123
00:05:14,985 --> 00:05:18,165
like I'm definitely hitting like leader key and like hammering like a
124
00:05:18,165 --> 00:05:21,135
bunch of stuff that doesn't exist because it's just all muscle memory.
125
00:05:21,525 --> 00:05:22,785
I'm a big fan of.
126
00:05:22,845 --> 00:05:25,335
Like seeing the data and like you can interpret
127
00:05:25,335 --> 00:05:27,645
the data in whatever good and bad it is.
128
00:05:27,885 --> 00:05:29,925
Uh, so like I'm not a big fan of like, oh, well
129
00:05:29,955 --> 00:05:33,735
90% of other companies are, are failing only 40%.
130
00:05:33,915 --> 00:05:35,505
Like, your company's failing and now you don't,
131
00:05:35,535 --> 00:05:38,175
you're never gonna make it to unicorn status.
132
00:05:38,175 --> 00:05:40,725
Like, I, I think that's a, yeah, it's a bit of a myth, but I,
133
00:05:40,725 --> 00:05:43,155
I think I could see what a competitive nature and leaderboards
134
00:05:43,155 --> 00:05:45,345
would come in there, which a big fan of that too, as well.
135
00:05:45,745 --> 00:05:48,505
Yeah, which I highly recommend not looking at the AT two and wrapped summary.
136
00:05:48,505 --> 00:05:52,465
For example, in 2025, apparently I typed 17 and half thousand commands.
137
00:05:52,615 --> 00:05:55,465
1700 of them were cd, which is my number one.
138
00:05:55,615 --> 00:06:00,414
Number two is NPM Run, and the third is ls, which is just, hmm.
139
00:06:00,935 --> 00:06:01,535
Alright.
140
00:06:01,895 --> 00:06:02,405
Alright.
141
00:06:02,405 --> 00:06:02,615
Yeah.
142
00:06:02,615 --> 00:06:04,985
Where am I and what else is here with me is
143
00:06:04,985 --> 00:06:06,455
sort of the thing that I'm always wondering.
144
00:06:06,785 --> 00:06:07,025
Yeah.
145
00:06:07,025 --> 00:06:09,755
LS is de, that's definitely top three for me for sure.
146
00:06:09,755 --> 00:06:11,165
I don't think I actually, I have no idea.
147
00:06:11,195 --> 00:06:14,345
Uh, I don't run Atune actively right now, so I just wiped the
148
00:06:14,345 --> 00:06:17,255
Mach Miss machine, so I'm still catching up with all the stuff.
149
00:06:17,975 --> 00:06:18,545
It's pretty great.
150
00:06:18,935 --> 00:06:23,345
So one thing I find interesting is folks who do a painful thing
151
00:06:23,345 --> 00:06:26,255
and get hurt and decide, you know what, let's do that again.
152
00:06:26,615 --> 00:06:28,685
This is of course a terrible segue into the reality.
153
00:06:28,685 --> 00:06:30,005
You are a second time founder.
154
00:06:30,344 --> 00:06:31,005
Tell me about that.
155
00:06:31,005 --> 00:06:32,025
Why would you do this again?
156
00:06:32,354 --> 00:06:35,895
Yeah, so I'll mention my first time founder project, which is
157
00:06:36,015 --> 00:06:39,104
Opens sauce pizza, which now pizza is actually not working anymore.
158
00:06:39,135 --> 00:06:41,534
Dot com is the only way you can see it, unfortunately.
159
00:06:41,745 --> 00:06:43,905
But we built insights in the GitHub repos, so
160
00:06:43,905 --> 00:06:45,885
I'm, I'm just a big fan of looking at data.
161
00:06:46,065 --> 00:06:47,655
I gotta see in statistics in college.
162
00:06:47,655 --> 00:06:50,385
So for whatever reason, it's been a chip on my shoulder
163
00:06:50,715 --> 00:06:52,935
to build more stats and the developer products.
164
00:06:53,145 --> 00:06:55,784
But yeah, so built that, sold it to Lennox Foundation.
165
00:06:56,125 --> 00:06:57,685
It worked out pretty well for us.
166
00:06:57,925 --> 00:07:00,594
I thought I was never gonna do it again because I felt like
167
00:07:00,594 --> 00:07:03,145
I got like a, a sort of boulder lifted off my shoulder.
168
00:07:03,445 --> 00:07:07,104
I went and worked at another company and then immediately saw in the AI space,
169
00:07:07,104 --> 00:07:09,865
infrastructure was just like not even a thing we were thinking about in 25.
170
00:07:10,015 --> 00:07:12,175
So now in 2026, we're like, oh.
171
00:07:12,655 --> 00:07:16,164
Maybe computers should like be more agentic friendly and like,
172
00:07:16,164 --> 00:07:18,205
we should probably lock these things down and like, they'll
173
00:07:18,205 --> 00:07:21,025
give it access to your Gmail or like, time out the, the runners.
174
00:07:21,145 --> 00:07:23,395
So all those questions that I had around like building stuff
175
00:07:23,395 --> 00:07:26,484
in the AI space, it had to be answered with some software.
176
00:07:26,484 --> 00:07:29,575
And like in December when I had the ideas, none of it was built.
177
00:07:29,635 --> 00:07:31,344
Now it feels like everyone's building it right now.
178
00:07:31,405 --> 00:07:33,534
It, it's almost like if you go back in time.
179
00:07:33,730 --> 00:07:39,400
And there's folks of us in a certain generational era where we struggled
180
00:07:39,400 --> 00:07:43,300
mightily with the advent of the internet and social media in some
181
00:07:43,300 --> 00:07:46,480
ways that kids today don't because they grew up knowing that they're
182
00:07:46,480 --> 00:07:49,870
at a fishbowl and everything they do is going to be discoverable.
183
00:07:49,960 --> 00:07:51,250
It feels like.
184
00:07:51,735 --> 00:07:55,335
I have been using my computer in such a way that I assumed
185
00:07:55,335 --> 00:07:58,844
I would be the only person using it, and that AI would not
186
00:07:58,844 --> 00:08:01,905
come along and start trying to pick up the pieces behind me.
187
00:08:02,025 --> 00:08:06,344
And if you give me a wiped machine, and now this is my new computer, and from
188
00:08:06,344 --> 00:08:10,095
day one, the expectation is is that the AI is gonna be looking over my shoulder
189
00:08:10,095 --> 00:08:14,025
and using any of the email that I interact with and the browser and the rest.
190
00:08:14,385 --> 00:08:17,235
I can use it more effectively in ways that aren't
191
00:08:17,235 --> 00:08:20,385
potentially disastrous when a robot gets confused.
192
00:08:20,715 --> 00:08:24,315
Yeah, we're in like a weird world, like about a year and a half ago.
193
00:08:24,315 --> 00:08:26,534
Everyone had like these pendants that you'd like record
194
00:08:26,534 --> 00:08:30,495
conversations with, or you have like the MetaFrame glasses that
195
00:08:30,495 --> 00:08:33,585
you just like recording TikTok videos all nonstop, all the time.
196
00:08:34,335 --> 00:08:36,615
Now we're in a space where like now our shell history is
197
00:08:36,615 --> 00:08:38,805
being recorded, or now our prompts are being recorded.
198
00:08:38,805 --> 00:08:41,475
Now we're basically being judged on how many tokens we're leveraging,
199
00:08:41,715 --> 00:08:44,355
or if we're not spending enough or we're spending too much or
200
00:08:44,355 --> 00:08:46,845
we're doing side work on our main machine and all this other stuff.
201
00:08:46,845 --> 00:08:49,725
Like it is very, very interesting in this
202
00:08:49,725 --> 00:08:51,405
current state of the world we are in today.
203
00:08:51,405 --> 00:08:53,535
And I know like the world's gonna change in like three weeks.
204
00:08:53,535 --> 00:08:55,515
So like today, like.
205
00:08:55,910 --> 00:08:58,939
I, I think we don't know what we don't know yet.
206
00:08:59,030 --> 00:09:01,850
And I think what's interesting in this space is like I, I want to
207
00:09:01,850 --> 00:09:04,640
build something that we can at least understand like what we're
208
00:09:04,640 --> 00:09:06,800
sort of passing through and like what tokens we're leveraging.
209
00:09:06,950 --> 00:09:09,080
I think I've been actually super hyper fixated on
210
00:09:09,439 --> 00:09:12,950
is Quad Code has a Max Max subscription for $200.
211
00:09:13,760 --> 00:09:17,510
I easily will spend $4,000 in tokens every week.
212
00:09:17,840 --> 00:09:20,090
So, Mike, I, I get that in Q4.
213
00:09:20,090 --> 00:09:20,840
They're probably not IPO.
214
00:09:21,475 --> 00:09:23,275
So at least I should start tracking to see how much
215
00:09:23,275 --> 00:09:25,285
I'm spending so I can sort of brace for impact.
216
00:09:25,765 --> 00:09:28,165
Uh, but what I found out also in the last couple weeks is
217
00:09:28,165 --> 00:09:31,705
prompt caching and quad code is saving 93% of those tokens.
218
00:09:32,095 --> 00:09:35,125
So, uh, this is basically all the stuff we're building.
219
00:09:35,125 --> 00:09:37,645
We'll have, we'll have this, um, actually you can see this in the tape stack.
220
00:09:37,645 --> 00:09:40,585
If you record your tape sessions, tape system product, the, the open
221
00:09:40,585 --> 00:09:44,095
source projects tape sessions, tape deck is the command line interface.
222
00:09:44,095 --> 00:09:45,955
You could see a dashboard to see your tokens.
223
00:09:46,285 --> 00:09:48,055
Uh, and what's cool about that is you could see the
224
00:09:48,055 --> 00:09:50,455
prompt caching and you could kind of brace for that.
225
00:09:50,455 --> 00:09:50,995
But then.
226
00:09:51,510 --> 00:09:53,849
Just last weekend they shipped this thing called Fast Mode,
227
00:09:54,300 --> 00:09:58,050
where you can get faster l um, inference, uh, but you'll
228
00:09:58,050 --> 00:10:01,020
just pay extra money on top of your Max Max subscription.
229
00:10:01,229 --> 00:10:02,010
It's pretty wild.
230
00:10:02,099 --> 00:10:02,339
Like
231
00:10:02,339 --> 00:10:04,709
yeah, people did some experimentation with it and said the results
232
00:10:04,709 --> 00:10:08,099
were not materially different and it was expensive at the top of it.
233
00:10:08,130 --> 00:10:09,180
So I, I get it.
234
00:10:09,180 --> 00:10:12,300
If you're production is down, we really need to figure this out.
235
00:10:12,329 --> 00:10:12,750
Okay.
236
00:10:12,750 --> 00:10:16,050
Turn on the, hit the turbo button on the old tower, but.
237
00:10:16,325 --> 00:10:17,825
It's a strange thing.
238
00:10:18,065 --> 00:10:21,305
Uh, there's also a psychology part of this where we
239
00:10:21,305 --> 00:10:23,825
saw back in the day when cell phone companies would
240
00:10:23,825 --> 00:10:27,335
offer fixed amounts of data instead of unlimited plans.
241
00:10:27,875 --> 00:10:30,635
A number of folks, I'm one of them, would wind up going for the
242
00:10:30,635 --> 00:10:34,235
unlimited plan every time just because I don't want the mental overhead.
243
00:10:34,445 --> 00:10:37,930
I still find myself thinking like that when things that are charged
244
00:10:37,990 --> 00:10:41,170
on a usage basis, even though the actual amount of money is.
245
00:10:41,714 --> 00:10:42,464
Irrelevant.
246
00:10:42,555 --> 00:10:44,475
It, we've, we've raised millions of dollars.
247
00:10:44,475 --> 00:10:45,045
Terrific.
248
00:10:45,225 --> 00:10:47,564
I don't necessarily need to hyper optimize
249
00:10:47,564 --> 00:10:51,015
around $4 and 68 cents worth of tokens every day.
250
00:10:51,285 --> 00:10:54,015
Maybe that's not the biggest bang for buck of what I should be
251
00:10:54,015 --> 00:10:56,444
focusing on right now, but I don't know how to turn that off.
252
00:10:56,655 --> 00:10:59,805
I, I guess for, for us, I, I could speak for myself, for us who don't
253
00:10:59,805 --> 00:11:03,704
come from money, like sometimes you've gotta hold onto money longer.
254
00:11:03,824 --> 00:11:06,915
I was also an at t like unlimited plan for years
255
00:11:06,915 --> 00:11:09,824
with the iPhone until eventually it scro like.
256
00:11:10,195 --> 00:11:14,275
Up until, actually I was on T-Mobile up until last summer and I had
257
00:11:14,275 --> 00:11:17,215
like this grandfathered like $80 a month plan for the entire family.
258
00:11:17,245 --> 00:11:17,635
So
259
00:11:17,665 --> 00:11:19,165
if they were to keep first trying to beg you into
260
00:11:19,165 --> 00:11:21,595
changing and then trick you into changing over the years,
261
00:11:21,595 --> 00:11:23,125
that was the thing is like when you go through the BART
262
00:11:23,125 --> 00:11:25,885
tunnel, you get no fast internet, you get nothing basically.
263
00:11:26,035 --> 00:11:27,985
And it's because you were on the lower tier plan.
264
00:11:28,574 --> 00:11:30,435
That they can't give you like high throughput.
265
00:11:30,495 --> 00:11:32,385
It's a crowded network in the bar tu tunnel.
266
00:11:32,505 --> 00:11:34,214
And that's the point you really feel.
267
00:11:34,214 --> 00:11:36,255
It is like at the moment you're sitting at GitHub issue
268
00:11:36,255 --> 00:11:38,925
and then GitHub like this freezes on you on your phone.
269
00:11:39,314 --> 00:11:42,165
So I'm like, oh, I guess I'll pay the $225 a month to
270
00:11:42,165 --> 00:11:45,435
like get up my modern plan so that way I can like actually
271
00:11:45,435 --> 00:11:47,150
send a GitHub issue on my, uh, in the BART tunnel.
272
00:11:47,445 --> 00:11:49,515
For the five, 10 minutes that I'm in there.
273
00:11:50,025 --> 00:11:53,475
Yeah, I, I wound up adding Google Fi as a secondary,
274
00:11:53,475 --> 00:11:55,305
uh, sim for a while and then switched over to using it
275
00:11:55,305 --> 00:11:57,585
full time, just 'cause it was half the price of Verizon.
276
00:11:57,585 --> 00:12:00,135
It seems to work reasonably well, but it's still the, it's
277
00:12:00,135 --> 00:12:02,865
unlimited, but they'll slow you down after a hundred gigs
278
00:12:02,865 --> 00:12:05,765
and I think on my busiest month I wound up using 20 of it.
279
00:12:06,105 --> 00:12:07,545
So it's, it's fine.
280
00:12:07,605 --> 00:12:11,715
I feel that way about the Claude Max plans as well, where I have
281
00:12:11,715 --> 00:12:14,715
hit session limits before when I've been working on something.
282
00:12:14,895 --> 00:12:17,745
But I'm not one of those folks that winds up waking up in the
283
00:12:17,745 --> 00:12:20,985
middle of the night to prompt the fleet of agents to do things.
284
00:12:20,985 --> 00:12:23,685
It's when I'm sitting at my desk and working on something.
285
00:12:23,685 --> 00:12:25,305
I'll make the robot go build something.
286
00:12:25,425 --> 00:12:26,325
Okay, fine.
287
00:12:26,750 --> 00:12:29,089
It's not, I'm not the heavy user and in fact, I
288
00:12:29,089 --> 00:12:31,040
could probably cut my plan further than I have.
289
00:12:31,099 --> 00:12:32,209
I'd like not having to think about it.
290
00:12:32,209 --> 00:12:33,380
And I'll be sad when that goes away.
291
00:12:33,469 --> 00:12:35,300
Yeah, and And it likely will go away.
292
00:12:35,300 --> 00:12:37,939
I think they're kind of like preparing us for this moment.
293
00:12:38,000 --> 00:12:39,980
I imagine we still get this unlimited plan, but it's
294
00:12:39,980 --> 00:12:42,110
not gonna be 200 bucks, it might be 2000 a month.
295
00:12:42,530 --> 00:12:45,170
And I think that's the point where we're all like the haves and have nots.
296
00:12:45,170 --> 00:12:45,949
We're probably.
297
00:12:47,370 --> 00:12:48,390
Showing themselves up.
298
00:12:48,450 --> 00:12:51,120
Yeah, I've been playing a lot with a lot of local models as well to see how far
299
00:12:51,120 --> 00:12:54,540
I could take those, which, yeah, there's a lot of stuff you could do to like,
300
00:12:54,600 --> 00:12:57,510
and almost a prompt router where great, go ahead and, uh, do
301
00:12:57,510 --> 00:13:00,930
this basically complicated find, replace in this entire code base
302
00:13:01,170 --> 00:13:04,680
that can be done with a slower local model, whereas architect the
303
00:13:04,680 --> 00:13:07,620
whole system, maybe I want one of the frontier models to do that.
304
00:13:08,040 --> 00:13:10,890
Yeah, and I've seen a lot of tools that, uh, outside of like the, the agent
305
00:13:10,890 --> 00:13:14,040
harnesses that have approached that problem but haven't done it very well.
306
00:13:14,130 --> 00:13:16,439
Uh, I think Open Router has a, a bit of a feature around
307
00:13:16,439 --> 00:13:19,620
this as well if you just go straight, open router proxy.
308
00:13:19,979 --> 00:13:21,900
But yeah, something I'll be experimenting with a ton.
309
00:13:22,199 --> 00:13:22,469
Yeah.
310
00:13:22,560 --> 00:13:23,130
Exciting times.
311
00:13:23,520 --> 00:13:25,050
And I think you're addressing the right part of the problem, which
312
00:13:25,050 --> 00:13:27,720
is first I, I don't know what cloud code is doing under the hood.
313
00:13:27,720 --> 00:13:29,640
I have no idea what the usage looks like.
314
00:13:29,790 --> 00:13:31,350
I know sometimes it's genius.
315
00:13:31,350 --> 00:13:35,190
Other times it feels like it's not that, and I wonder
316
00:13:35,190 --> 00:13:38,100
how that's playing out under the hood or what's going on.
317
00:13:38,100 --> 00:13:42,420
I just know that I need to be careful enough around this when
318
00:13:42,420 --> 00:13:45,810
I'm having it do things that if this is gonna make a mark,
319
00:13:45,900 --> 00:13:48,840
maybe make sure that that test coverage is where I think it is.
320
00:13:49,545 --> 00:13:53,085
This episode is sponsored by my own company, duck Bill.
321
00:13:53,385 --> 00:13:56,295
Having trouble with your AWS bill, perhaps
322
00:13:56,295 --> 00:13:59,025
it's time to renegotiate a contract with them.
323
00:13:59,355 --> 00:14:01,515
Maybe you're just wondering how to predict
324
00:14:01,515 --> 00:14:04,725
what's going on in the wide world of AWS.
325
00:14:04,814 --> 00:14:07,425
Well, that's where Duck Bill comes in to help.
326
00:14:07,635 --> 00:14:10,365
Remember, you can't duck the duck bill.
327
00:14:10,365 --> 00:14:13,035
Bill, which I am reliably informed by my
328
00:14:13,035 --> 00:14:16,575
business partner is absolutely not our motto.
329
00:14:16,650 --> 00:14:19,829
To learn more, visit doc bill hq.com.
330
00:14:20,939 --> 00:14:22,770
Yeah, and the one thing I'll mention as well, so I mentioned
331
00:14:22,770 --> 00:14:25,140
tapes in like the telemetry part, but stereos is the
332
00:14:25,140 --> 00:14:28,890
vm, so the actual computer, like essentially it's N os.
333
00:14:28,920 --> 00:14:30,660
You could run it on your Mac Os, you could
334
00:14:30,660 --> 00:14:32,790
run analytics currently not Windows yet.
335
00:14:32,790 --> 00:14:36,480
So we just need more people to care about Windows and we'll focus on it.
336
00:14:36,660 --> 00:14:37,860
So open the issue please.
337
00:14:37,860 --> 00:14:38,760
That's what I keep telling people.
338
00:14:38,760 --> 00:14:39,990
Open the issue and then we'll track it.
339
00:14:40,260 --> 00:14:43,050
But the idea there is like the same way Quad Code does sub subagents,
340
00:14:43,230 --> 00:14:46,770
it's great because it takes that token usage onto its own channel.
341
00:14:46,770 --> 00:14:47,850
So you don't get that sort of.
342
00:14:48,204 --> 00:14:50,485
Now it's a million tokens context for like a session.
343
00:14:50,935 --> 00:14:53,905
Previously it was like 250, but what I'm getting at is like, it will take
344
00:14:53,905 --> 00:14:58,045
that session off to its own session, uh, and then come back with a hash.
345
00:14:58,045 --> 00:15:01,225
So like with tapes, we're using a Merkel dag, just like get, you could
346
00:15:01,225 --> 00:15:03,715
take a session, you could hash it, you could search it based on the hash.
347
00:15:03,780 --> 00:15:06,720
So you can do lookups, you could generate skills, but then you can also have
348
00:15:06,720 --> 00:15:10,380
separate sessions within subagents, and that's what we're doing with stereos.
349
00:15:10,380 --> 00:15:13,770
So stereos is more likely something you use on your own metal.
350
00:15:13,829 --> 00:15:15,930
Um, there's a lot of these like sandboxes and
351
00:15:15,930 --> 00:15:18,209
computers that people are leveraging in the cloud.
352
00:15:18,540 --> 00:15:21,480
Our bet is that you probably wanna build your own cloud or your own enterprise
353
00:15:21,480 --> 00:15:24,390
that wants to do this, and that's what stereos comes in in the question.
354
00:15:25,020 --> 00:15:25,230
Yeah.
355
00:15:25,230 --> 00:15:29,760
What I've been doing is there are risks to it and I'm aware of that, but I
356
00:15:29,760 --> 00:15:34,440
have a dedicated AWS account called Superfund because it is expensive and toxic
357
00:15:34,500 --> 00:15:39,390
and it has no access to any data in it or so I thought, and then I just run
358
00:15:39,390 --> 00:15:42,690
an EC2 instance there that is, and run Claude and dangerous permissions mode.
359
00:15:42,690 --> 00:15:44,430
It has full root access to the box.
360
00:15:44,580 --> 00:15:44,970
Go ahead.
361
00:15:44,970 --> 00:15:46,455
All all you're gonna blow up is my bill.
362
00:15:47,115 --> 00:15:49,395
Then one day I realized it had access to read
363
00:15:49,395 --> 00:15:52,185
things off of my calendar, which was not expected.
364
00:15:52,185 --> 00:15:54,944
It turns out that at some point in the last few months, Claude's
365
00:15:54,944 --> 00:15:59,355
skills and connections were now tied to your Anthropic account, not
366
00:15:59,355 --> 00:16:03,135
to your actual, uh, not, not to just the machine you installed them
367
00:16:03,135 --> 00:16:06,824
on, which is great most of the time and terrible in this one use case.
368
00:16:06,944 --> 00:16:09,584
So first thing I did is slap the control into the settings
369
00:16:09,615 --> 00:16:12,535
json to make sure it doesn't do that anymore, just for.
370
00:16:13,320 --> 00:16:16,440
Data isolation reasons, but yeah, that is
371
00:16:16,440 --> 00:16:17,790
not something most people are going to do.
372
00:16:17,850 --> 00:16:19,800
They're not going to build a box that does this.
373
00:16:19,980 --> 00:16:22,500
Claude Code gets very greedy with the Rams.
374
00:16:22,500 --> 00:16:24,720
So this is a couple hundred month bucks a month for the EC2
375
00:16:24,720 --> 00:16:28,110
instance to sit there running, uh, four or five of these things and.
376
00:16:28,454 --> 00:16:30,015
There needs to be a better solution.
377
00:16:30,285 --> 00:16:32,505
I'd love to have something like that on the laptop, since
378
00:16:32,505 --> 00:16:35,535
I'm mostly around when I want this stuff doing things.
379
00:16:35,864 --> 00:16:36,165
Yeah.
380
00:16:36,170 --> 00:16:38,925
I, I wrote it on a, on a, I wrote it actually on this machine I'm talking to
381
00:16:38,925 --> 00:16:42,915
you on, just as of last weekend I did open claw on A box is a, uh, a repo.
382
00:16:42,915 --> 00:16:45,375
I put together just to like, if you wanna spin up this
383
00:16:45,375 --> 00:16:47,415
and you wanna have like constraints and you wanna have
384
00:16:47,415 --> 00:16:49,814
like network shut down on certain interactions, so.
385
00:16:50,580 --> 00:16:52,110
The use case I use is on Sundays.
386
00:16:52,110 --> 00:16:54,060
I go through my email, it labels a bunch of stuff
387
00:16:54,060 --> 00:16:56,520
for me, and it only, it times out at two hours.
388
00:16:56,610 --> 00:16:58,740
So if it can't get it done in two hours, then it's
389
00:16:58,740 --> 00:17:01,470
probably just, it's, it's too long or I need less email.
390
00:17:01,500 --> 00:17:02,640
So like, I can figure that out.
391
00:17:02,640 --> 00:17:04,650
I'll go call the government and tell 'em to sit, stop
392
00:17:04,650 --> 00:17:07,710
setting up email or whatever the, uh, I'm thinking of the
393
00:17:07,710 --> 00:17:10,380
Seinfeld, uh, when, when, uh, Kramer said, no more mail.
394
00:17:10,680 --> 00:17:11,819
Not sure if that's possible yet.
395
00:17:12,410 --> 00:17:15,650
Yeah, I have a, uh, in the next room I have a Kubernete
396
00:17:15,680 --> 00:17:19,040
singular running on a cluster of 10 raspberry pies.
397
00:17:19,280 --> 00:17:21,980
And one of the things that runs on there is Billy the Platypus,
398
00:17:21,980 --> 00:17:25,520
who's now my virtual ea, where he'll wind up receiving the
399
00:17:25,520 --> 00:17:28,940
email writing a barely professional sarcastic response.
400
00:17:29,210 --> 00:17:32,810
And now with an additional fun feature, he'll go out to a data enrichment,
401
00:17:32,810 --> 00:17:37,160
API figure out what you look like, and then integrate a cartoon of him
402
00:17:37,160 --> 00:17:41,420
dunking on you in some way into the insulting response, which is just.
403
00:17:41,844 --> 00:17:42,445
Amazing.
404
00:17:42,445 --> 00:17:42,985
It is terrific.
405
00:17:42,985 --> 00:17:47,455
People are trying to sell me things or be obnoxious in email in various ways.
406
00:17:47,574 --> 00:17:50,905
I feel a little sorry for some of the, uh, entry level,
407
00:17:50,935 --> 00:17:53,365
uh, business development types and SDRs who are reaching
408
00:17:53,365 --> 00:17:55,584
out, trying to sell me something that doesn't make sense.
409
00:17:55,705 --> 00:17:58,254
But other times it's, this stuff is so scattershot and
410
00:17:58,375 --> 00:18:01,135
send to the entire world where I like to buy land in Napa.
411
00:18:01,195 --> 00:18:02,125
No, no.
412
00:18:02,125 --> 00:18:02,995
Thank you for asking.
413
00:18:02,995 --> 00:18:05,064
I would not, so yeah.
414
00:18:05,064 --> 00:18:06,834
Have the insulting POTUS smack back.
415
00:18:07,195 --> 00:18:09,325
I love that, but there is risk to this if
416
00:18:09,325 --> 00:18:11,275
it starts tying into all the other stuff.
417
00:18:11,275 --> 00:18:14,125
I mean, the scary thing that I need to defend against and why there's a
418
00:18:14,125 --> 00:18:18,235
human in the loop is it can now schedule things for me, which is great,
419
00:18:18,565 --> 00:18:21,235
but it'll casually look at my calendar and then say, oh, he can't.
420
00:18:21,235 --> 00:18:23,425
Then he's meeting with insert company here.
421
00:18:23,575 --> 00:18:26,575
It's ah, how about no.
422
00:18:27,250 --> 00:18:29,170
That maybe that's, maybe it's fine.
423
00:18:29,230 --> 00:18:31,510
Always having another meeting to yell at AWS.
424
00:18:31,510 --> 00:18:32,590
Yeah, that's a real rarity.
425
00:18:32,770 --> 00:18:34,630
But maybe I'm meeting with a confidential client.
426
00:18:34,660 --> 00:18:37,510
Maybe that's not something that needs to be disclosed in an email.
427
00:18:37,690 --> 00:18:39,730
My challenge is like, I just started paying $20
428
00:18:39,730 --> 00:18:41,860
a month for chat GBT, which gives it memory.
429
00:18:41,920 --> 00:18:42,790
I think you have to enable it.
430
00:18:42,790 --> 00:18:46,960
I did enable it 'cause I was building a, a, a computer, uh, with A GPU in it.
431
00:18:46,960 --> 00:18:49,900
I was like, oh, I'm gonna just like prep the chat GBT to
432
00:18:49,900 --> 00:18:52,150
like, know everything I know about my computer and then like.
433
00:18:52,520 --> 00:18:55,639
Recommend, like, well not Ram at this point, but like, uh,
434
00:18:55,639 --> 00:18:57,560
this was actually last year before RAM was, yeah, good luck.
435
00:18:57,560 --> 00:18:57,739
Money
436
00:18:57,739 --> 00:18:58,070
bags.
437
00:18:58,129 --> 00:18:58,520
Yeah.
438
00:18:58,729 --> 00:19:00,949
But, uh, I was just like looking for different parts and like
439
00:19:00,949 --> 00:19:03,469
seeing if what would work or what, but then like randomly I'll
440
00:19:03,469 --> 00:19:05,959
ask a question about like something like actual work related
441
00:19:05,959 --> 00:19:08,060
and be like, oh yeah, why don't you run on your GPU machine?
442
00:19:08,060 --> 00:19:08,689
They're like, what do you mean?
443
00:19:09,110 --> 00:19:10,250
Oh, the machine you built last year.
444
00:19:10,250 --> 00:19:12,679
It's like, I, well, completely outta context.
445
00:19:12,800 --> 00:19:14,750
Not as bad as your, your sort of confidential.
446
00:19:14,815 --> 00:19:17,245
Call, but like it feels like, it's like, oh, remember
447
00:19:17,245 --> 00:19:18,805
that one like thing you built for your daughter?
448
00:19:18,805 --> 00:19:20,005
It's like, dude, stop.
449
00:19:20,005 --> 00:19:20,815
Not the time of the place.
450
00:19:20,815 --> 00:19:21,415
Read the room.
451
00:19:22,345 --> 00:19:22,435
Yeah.
452
00:19:22,435 --> 00:19:24,505
That's part of the problem that all these agent things have.
453
00:19:24,505 --> 00:19:28,375
Like there are different personas that I take on when I talk to folks when
454
00:19:28,375 --> 00:19:32,545
I'm in a board meeting versus when I am pitching to a customer, when versus
455
00:19:32,545 --> 00:19:35,875
when I'm writing a sarcastic newsletter versus when I am being a parent,
456
00:19:35,875 --> 00:19:38,365
versus when I'm shit posting on the thing known, formerly known as Twitter.
457
00:19:38,725 --> 00:19:41,575
I act very differently, and if you cross
458
00:19:41,575 --> 00:19:43,675
those streams, it doesn't go super well.
459
00:19:43,925 --> 00:19:45,784
Like there's a time and a place for all of this.
460
00:19:45,784 --> 00:19:49,804
Like I, I do a mean conference talk and I do some great standup jokes.
461
00:19:50,375 --> 00:19:53,375
Maybe don't do that when I'm putting together
462
00:19:53,405 --> 00:19:55,655
a wedding ceremony or something like that.
463
00:19:55,685 --> 00:19:58,504
There's a time and a place, and that's something that gets lost sometimes.
464
00:19:59,020 --> 00:20:00,700
Yeah, but it's great for a LinkedIn post.
465
00:20:00,700 --> 00:20:03,610
I think it, it, it could be at the LinkedIn, whatever.
466
00:20:03,610 --> 00:20:06,760
I dunno what that, that vibe LinkedIn has on posting, but, uh, I know
467
00:20:06,760 --> 00:20:11,080
Claude has a good, does a good job of, uh, self a referent and posting.
468
00:20:11,260 --> 00:20:12,100
Oh, absolutely.
469
00:20:12,100 --> 00:20:13,810
And the problem is, is you can't even say that.
470
00:20:13,810 --> 00:20:15,340
That's not what LinkedIn is designed for.
471
00:20:15,340 --> 00:20:15,490
You.
472
00:20:15,490 --> 00:20:18,399
You go to LinkedIn now and it interrupts you when you're writing stuff.
473
00:20:18,399 --> 00:20:21,429
Hey, you wanna rewrite that with ai, it's, what are you?
474
00:20:21,645 --> 00:20:24,525
I think this network is for Exactly.
475
00:20:24,525 --> 00:20:26,535
And then I read some of the insipid stuff on there
476
00:20:26,535 --> 00:20:28,995
and you wanna hope it's ai, but you're kind of scared.
477
00:20:28,995 --> 00:20:29,415
It's not,
478
00:20:30,014 --> 00:20:31,095
well, I guess we'll, we'll figure it out.
479
00:20:31,095 --> 00:20:32,565
Someone's gonna have to ship a this.
480
00:20:32,565 --> 00:20:34,485
Is this ai not true or false.
481
00:20:34,754 --> 00:20:39,075
So I am curious what would possess you, there might be a less confrontational
482
00:20:39,075 --> 00:20:42,675
way to phrase that, but we're going with possess you to start a company
483
00:20:42,675 --> 00:20:45,855
where effectively everything you're putting out so far is open source.
484
00:20:45,920 --> 00:20:48,920
That feels like it has some definite advantages to it,
485
00:20:49,310 --> 00:20:51,470
but it's also the sort of thing that can be fraught.
486
00:20:51,740 --> 00:20:54,140
Like I come from GitHub, I come from open source.
487
00:20:54,140 --> 00:20:56,300
I, I learn how to code because of open source.
488
00:20:56,330 --> 00:21:00,080
So like, it's in, in my ethos, but specifically for what we're building with
489
00:21:00,080 --> 00:21:03,680
tapes and stereos, like I feel like the telemetry parts, it's table stakes.
490
00:21:03,680 --> 00:21:06,080
If you ask any sort of YC company or any sort of
491
00:21:06,110 --> 00:21:08,180
VC of like, Hey, we're doing telemetry for agents.
492
00:21:08,180 --> 00:21:10,850
Like, cool, so we're a hundred other companies, and then
493
00:21:10,850 --> 00:21:12,800
now we thought we were ahead of the game when it came
494
00:21:12,800 --> 00:21:15,290
to like the VM and like building the actual runtime.
495
00:21:15,580 --> 00:21:18,040
But then, uh, like a bazillion other companies
496
00:21:18,040 --> 00:21:19,389
came out and said, we have a runtime too.
497
00:21:19,389 --> 00:21:20,860
Like, Deno has a runtime.
498
00:21:20,860 --> 00:21:22,389
Deno is doing something completely different
499
00:21:22,600 --> 00:21:25,240
for what we're doing and who we wanna talk to.
500
00:21:25,270 --> 00:21:28,270
Like open source is actually a value add for them to be able to see what's
501
00:21:28,270 --> 00:21:31,185
happening and like what controls and like what policies they can set.
502
00:21:31,835 --> 00:21:33,485
Before they even get to the conversation.
503
00:21:33,695 --> 00:21:35,945
I spent a lot of time in my first startup where we spent way more
504
00:21:35,945 --> 00:21:38,345
time trying to convince people and having long conversations.
505
00:21:38,675 --> 00:21:40,925
Today we're just like, Hey, use tape stereos if that
506
00:21:40,925 --> 00:21:43,145
kind of fits within the thing that works for you.
507
00:21:43,595 --> 00:21:45,725
Then we have this cloud, or we have this self-hosted on-prem
508
00:21:45,725 --> 00:21:48,905
thing that works one and the same, like not one and the same,
509
00:21:48,905 --> 00:21:51,875
but obviously one's gonna cost more and be more complicated.
510
00:21:52,264 --> 00:21:54,815
But we feel like open source is a lot easier for people to try
511
00:21:54,815 --> 00:21:58,205
before they buy without setting up a 14 day trial or whatever.
512
00:21:58,655 --> 00:22:01,805
What about the current zeitgeist impression that folks are giving off
513
00:22:01,805 --> 00:22:05,705
where this is the, the bad days of SaaS where companies are gonna start
514
00:22:05,705 --> 00:22:09,935
vibe coding SaaS solutions instead of paying, uh, third parties for it?
515
00:22:10,145 --> 00:22:13,504
Yeah, I think that's the same thing of like, I can write a bunch of c plus
516
00:22:13,504 --> 00:22:16,445
plus code, but also I can't because I've never written c plus plus code.
517
00:22:16,445 --> 00:22:18,905
So yes, I can vibe it, but also I don't know
518
00:22:18,905 --> 00:22:22,085
what bad, like unsafe c plus plus code is.
519
00:22:22,355 --> 00:22:24,605
Uh, so when it comes to like SaaS products, like you could probably
520
00:22:24,605 --> 00:22:27,824
build the thing I built like a. Look at my LinkedIn connections
521
00:22:27,824 --> 00:22:30,554
and give me like a list of companies that I can like go reach out
522
00:22:30,554 --> 00:22:32,655
to and hang out with a bunch of friends and show 'em my new thing.
523
00:22:33,014 --> 00:22:35,955
Like I don't wanna pay LinkedIn premium or sales thing
524
00:22:35,955 --> 00:22:38,294
for 12 KA month when I can just like vibe code a thing.
525
00:22:38,595 --> 00:22:40,905
But there will be a point where I bring on a sales person who
526
00:22:40,905 --> 00:22:43,784
I don't want to like use my vibe code a thing and I wanna share
527
00:22:43,784 --> 00:22:47,625
information properly in like a Salesforce manner or HubSpot manner.
528
00:22:47,985 --> 00:22:50,054
Like there's a place where you kind of grow up
529
00:22:50,145 --> 00:22:52,605
into, like, again, I, I don't want a 14 day trial.
530
00:22:52,910 --> 00:22:53,690
I know what I'm doing.
531
00:22:53,720 --> 00:22:56,330
I'm just like running through the, the system and plugging and playing.
532
00:22:56,750 --> 00:22:58,850
And then on the, the vein of like open source, like
533
00:22:59,120 --> 00:23:01,310
I feel like open source is table stakes as well.
534
00:23:01,310 --> 00:23:03,350
Like where you can close source it and like
535
00:23:03,350 --> 00:23:05,150
keep your, keep everything close to chess.
536
00:23:05,480 --> 00:23:08,420
But also I think what we need is like more people building infrastructure.
537
00:23:08,420 --> 00:23:10,760
So like if anybody clones our repo and like says,
538
00:23:10,790 --> 00:23:13,370
I'm gonna kickstart from this and let's get started.
539
00:23:13,670 --> 00:23:15,350
Maybe we're protected with A GPL.
540
00:23:15,470 --> 00:23:16,160
Maybe we're not.
541
00:23:16,470 --> 00:23:18,300
But at the end of the day, it's like what we need is this stuff
542
00:23:18,300 --> 00:23:21,240
to be out in the system and out in the, um, infrastructure.
543
00:23:21,540 --> 00:23:23,430
And hopefully we're the ones that build it, but
544
00:23:23,430 --> 00:23:25,770
like someone else builds it, then obviously we'll,
545
00:23:25,980 --> 00:23:27,510
we at least not going in the right direction.
546
00:23:27,630 --> 00:23:27,900
Yeah.
547
00:23:27,930 --> 00:23:30,630
Like I've built a bunch of internal SaaS tooling that's,
548
00:23:30,630 --> 00:23:33,390
uh, scratches, certain itches I have that I'm not willing
549
00:23:33,390 --> 00:23:36,720
to pay, uh, professional to get up and running for me.
550
00:23:36,900 --> 00:23:39,120
But when we're building Skyway over at Duck Bill
551
00:23:39,420 --> 00:23:42,600
that it's not, it's not hard software in that.
552
00:23:42,910 --> 00:23:47,020
It is effectively a data platform that manipulates data and normalizes it.
553
00:23:47,080 --> 00:23:48,400
Yeah, we, we know how to do that.
554
00:23:48,400 --> 00:23:50,290
We are not pushing the bounds of computer
555
00:23:50,290 --> 00:23:52,600
science here, but the data must be correct.
556
00:23:52,600 --> 00:23:54,880
It's not the sort of thing that you can vibe,
557
00:23:54,880 --> 00:23:57,250
code, and expect good things to fall out of.
558
00:23:57,370 --> 00:23:59,980
Lord knows we've done experiments around that.
559
00:24:00,280 --> 00:24:02,440
Now you can come out with something that sounds very good.
560
00:24:03,220 --> 00:24:05,950
Looks good at first glance until you start digging into it.
561
00:24:05,950 --> 00:24:09,400
But if you're making sincere business decisions based on that data, and
562
00:24:09,400 --> 00:24:13,690
it turns out that data's wrong, uh, no one wants to be in that position.
563
00:24:13,900 --> 00:24:16,120
It's the same reason that Anthropic vibe codes a lot
564
00:24:16,120 --> 00:24:18,880
of stuff, but they pay, I think it's a DP for payroll.
565
00:24:19,060 --> 00:24:21,910
They, they're not just paying for the software that arranges the paychecks.
566
00:24:21,910 --> 00:24:23,980
They're, they're basically paying for expertise.
567
00:24:24,220 --> 00:24:26,140
You don't want your super base table to be
568
00:24:26,140 --> 00:24:28,120
completely open and like read and write.
569
00:24:28,360 --> 00:24:30,940
Well, obviously the, the thing that happened in the last week is delve like.
570
00:24:31,590 --> 00:24:34,679
Everything was accessible and I imagine a lot of the stuff was
571
00:24:34,679 --> 00:24:37,050
just kind of shoot from the hip and a certain point you kind of
572
00:24:37,050 --> 00:24:39,929
wanna bring in the experts to come clean up and polish the stuff.
573
00:24:40,200 --> 00:24:42,540
Naively, I thought compliance was one of those
574
00:24:42,540 --> 00:24:45,480
spaces where you wanted to have expertise in shows.
575
00:24:45,480 --> 00:24:45,629
What?
576
00:24:45,629 --> 00:24:48,270
I know, that's why I've never been on the 30, under 30,
577
00:24:48,419 --> 00:24:49,290
40 under 40.
578
00:24:49,740 --> 00:24:50,189
Perhaps
579
00:24:50,189 --> 00:24:50,939
not even anymore.
580
00:24:50,945 --> 00:24:53,970
I have to be 50 under 50 at some point where I think we just call that folks.
581
00:24:54,645 --> 00:24:55,845
So what's, what's next?
582
00:24:55,845 --> 00:24:57,705
You mentioned you're doing a cloud offering.
583
00:24:57,705 --> 00:24:59,955
Probably by the time that this winds up shipping, I mean
584
00:25:00,015 --> 00:25:02,505
that will presumably address my biggest problem with your
585
00:25:02,505 --> 00:25:05,175
website right now, which is it doesn't have a pricing page.
586
00:25:05,460 --> 00:25:05,750
Yeah.
587
00:25:05,755 --> 00:25:05,895
Yeah.
588
00:25:05,895 --> 00:25:08,775
So we will have a pricing page hopefully in the next two weeks.
589
00:25:08,775 --> 00:25:10,425
Uh, we're fastly approaching that.
590
00:25:10,425 --> 00:25:11,715
We have the infrastructure that works.
591
00:25:11,925 --> 00:25:14,385
Uh, we're sort of pending a. UI on top of it.
592
00:25:14,445 --> 00:25:18,495
So next is basically you have tapes, you have sessions, you have agents.
593
00:25:18,495 --> 00:25:20,385
You wanna play those in a multiplayer fashion.
594
00:25:20,595 --> 00:25:23,835
So like you and your team could all look at your weird prompts and all the
595
00:25:23,835 --> 00:25:27,345
tokens you're using, uh, and then you sort of like dictate what what is next.
596
00:25:27,345 --> 00:25:29,415
So something actually very cool I've been working on is like
597
00:25:29,415 --> 00:25:31,515
taking your prompts and then training a model with that.
598
00:25:31,515 --> 00:25:32,220
So if you wanna like.
599
00:25:32,595 --> 00:25:35,835
Do you like some fine tuning off down the road for your local models?
600
00:25:36,345 --> 00:25:38,325
How heavy a lift is it to do something like that?
601
00:25:38,325 --> 00:25:41,145
Because I've toyed with the idea, but every experiment I come up with is like,
602
00:25:41,145 --> 00:25:45,435
step one, you're gonna need about $10,000 worth of GPU time, which, okay,
603
00:25:45,555 --> 00:25:47,625
if you know anybody, Nvidia, some friends that
604
00:25:47,625 --> 00:25:49,995
will, well give you some, some rental space.
605
00:25:50,235 --> 00:25:53,085
But yeah, if you have data, data's one of the hardest parts to get for training.
606
00:25:53,145 --> 00:25:55,065
I, uh, particularly have not done it myself.
607
00:25:55,390 --> 00:25:58,480
I've been in the room when this is happening at my last role at continue.
608
00:25:58,510 --> 00:26:00,400
So I have a good idea of how this works.
609
00:26:00,400 --> 00:26:03,370
But then folks like Unsought, they can help you do the fine tuning part.
610
00:26:03,640 --> 00:26:05,065
Uh, so like there's, there's a bunch of like.
611
00:26:05,879 --> 00:26:10,949
Prosumer, like ification, ways to go about it, but also quad's
612
00:26:10,949 --> 00:26:13,439
my friend as well, so I might just like chat with, with them
613
00:26:13,500 --> 00:26:17,129
and, uh, build out a six week pipeline to, to make this work.
614
00:26:17,250 --> 00:26:19,889
I'm still trying to figure out why, when I prompt for
615
00:26:19,889 --> 00:26:22,800
certain things versus my teammates prompt for certain things.
616
00:26:23,155 --> 00:26:27,835
I get such vastly different results in terms of tone, sarcasm,
617
00:26:27,835 --> 00:26:31,105
the way it builds things, and I, I have no answer for that.
618
00:26:31,105 --> 00:26:33,235
It seems like a telemetry play might be the right answer here.
619
00:26:33,325 --> 00:26:35,814
Yeah, I think it's, it's got your taste down to, uh,
620
00:26:36,085 --> 00:26:38,754
let's, let's get the first thing out to Corey and then.
621
00:26:39,065 --> 00:26:41,255
Co-founder gets the better, the better taste.
622
00:26:41,585 --> 00:26:42,005
Exactly.
623
00:26:42,005 --> 00:26:42,155
Yeah.
624
00:26:42,155 --> 00:26:45,515
Mike is the one that winds up doing all the serious button down stuff.
625
00:26:45,515 --> 00:26:49,115
Like we can make this talk to Microsoft Excel and I'm sitting over
626
00:26:49,115 --> 00:26:52,625
here going, we can make it rank the US Presidents by absorbency.
627
00:26:52,865 --> 00:26:57,215
And it's, it's one of those things where it, some point between the two of
628
00:26:57,215 --> 00:27:00,635
us, we're gonna reach a happy medium where it's both whimsical and useful.
629
00:27:01,755 --> 00:27:02,475
We're not there today.
630
00:27:02,625 --> 00:27:06,225
The litmus stuff I have right now is WW JD, which is what would John do?
631
00:27:06,615 --> 00:27:09,675
And I ask that question every time I go write in one of his
632
00:27:09,675 --> 00:27:12,015
projects, which is like, I don't write a lot of Go code.
633
00:27:12,015 --> 00:27:13,125
I'd like professionally.
634
00:27:13,125 --> 00:27:16,215
I've done some go code, but really not to the level that John has.
635
00:27:16,455 --> 00:27:19,005
So what I wanna do is take John's sessions to help inform.
636
00:27:19,085 --> 00:27:20,375
My prompt sessions.
637
00:27:20,615 --> 00:27:23,765
Uh, so I've been generating skills from his sessions and so far going well
638
00:27:23,765 --> 00:27:25,295
shared team skill of Corey Quinn voice.
639
00:27:25,295 --> 00:27:28,325
I need to go in and sabotage soon if some of it gets uncomfortably close.
640
00:27:28,325 --> 00:27:29,705
And that's my space in the universe.
641
00:27:29,705 --> 00:27:30,575
Thank you very much.
642
00:27:31,535 --> 00:27:31,955
Amazing.
643
00:27:32,705 --> 00:27:34,835
So I wanna thank you for taking the time to speak with me.
644
00:27:34,835 --> 00:27:37,505
If, if people wanna learn more, where's the best place for them to go?
645
00:27:37,865 --> 00:27:41,465
Uh, best place to go is uh, I have a site called be dougie.dev.
646
00:27:41,465 --> 00:27:44,615
That's my, this catchall, well I guess they would call it win trees
647
00:27:44,615 --> 00:27:47,645
back in the day, but LinkedIn, I'm happy to like take a DM there.
648
00:27:47,910 --> 00:27:52,680
No promises on response and then X at BW and then blue sky at pizza.
649
00:27:52,830 --> 00:27:55,410
Pizza because I wasn't taking that platform
650
00:27:55,410 --> 00:27:57,030
seriously and I just picked the random BRL.
651
00:27:57,660 --> 00:28:02,040
Yeah, I am having some challenges myself with the answer of where do I direct
652
00:28:02,040 --> 00:28:06,030
people to for all of these things that I, that I have going on, and every time
653
00:28:06,030 --> 00:28:09,630
it feels like I wind up narrowing it down by collapsing two projects into one.
654
00:28:09,960 --> 00:28:10,980
Three more spring off.
655
00:28:11,160 --> 00:28:12,930
I don't know that there's ever a way to get away from it.
656
00:28:13,405 --> 00:28:13,695
Yeah.
657
00:28:13,700 --> 00:28:13,980
Yeah.
658
00:28:14,160 --> 00:28:15,060
Five man Minecraft.
659
00:28:16,020 --> 00:28:16,530
There we go.
660
00:28:16,710 --> 00:28:21,330
Uh, it's similar to, someone showed me a SSO page screenshot once,
661
00:28:21,480 --> 00:28:24,060
and it showed all the different services you could use to log into it.
662
00:28:24,300 --> 00:28:26,880
And like they had GitHub, they had LinkedIn,
663
00:28:26,880 --> 00:28:29,970
they had Google, and they had the Lego website.
664
00:28:30,354 --> 00:28:33,955
Apparently, which acts as an identity provider, which is just absurd to me.
665
00:28:34,165 --> 00:28:39,685
I want to have that as a enterprise class SSO login, just to see the reaction.
666
00:28:39,774 --> 00:28:42,024
The, the Lego Envoy proxy would be amazing.
667
00:28:42,024 --> 00:28:43,524
Like, let's, let's get that to work.
668
00:28:44,665 --> 00:28:46,074
Thanks again for taking the time to speak with me.
669
00:28:46,074 --> 00:28:46,945
I appreciate it.
670
00:28:47,034 --> 00:28:47,274
Yeah.
671
00:28:47,274 --> 00:28:47,814
Have a good one.
672
00:28:48,084 --> 00:28:48,865
Brian Douglas.
673
00:28:49,135 --> 00:28:51,355
CEO at Paper Compute.
674
00:28:51,655 --> 00:28:55,495
I'm Cloud economist Corey Quinn, and this is Screaming In the Cloud.
675
00:28:55,705 --> 00:28:57,655
If you've enjoyed this podcast, please leave a five
676
00:28:57,655 --> 00:28:59,935
star review on your podcast platform of choice.
677
00:29:00,175 --> 00:29:03,595
Whereas if you hated this podcast, please leave a five star review on
678
00:29:03,595 --> 00:29:07,165
your podcast platform of choice along with an angry, insulting comment.
679
00:29:07,285 --> 00:29:09,700
Be sure to include the tapes replay because there's no
680
00:29:09,700 --> 00:29:12,235
way you were able to be that clever all on your own.
