Why building for few beats building for masses | Vaughn Tan (Executive Director of Tangent)

WEBVTT
00:00:00.000 --> 00:00:05.402
I'm a qualitative sociologist who has never built software in the past, ever, and it is possible
00:00:05.713 --> 00:00:09.132
why do you find it so important to build boring, tiny tools?
00:00:09.215 --> 00:00:14.916
in almost any ecosystem, the vast majority of the biomass is all the boring small stuff that you can barely even see.
00:00:14.916 --> 00:00:16.806
The fungi, the bacteria, the plants.
00:00:17.196 --> 00:00:19.025
That makes up almost everything.
00:00:19.596 --> 00:00:20.826
So boring, tiny tools are important.
00:00:20.826 --> 00:00:24.006
'cause they are probably going to be the vast majority of the tools that we use.
00:00:24.486 --> 00:00:26.946
I built this in four and a half hours.
00:00:27.126 --> 00:00:46.146
it is very weird that this boring, tiny tool has changed my life, but it has changed my life Why are we not encouraging more people to do the unsexy, but actually potentially quite enjoyable and certainly much more likely to be lucrative task of building boring, tiny tools for the enormous number of SMEs that exist in this country.
00:00:46.445 --> 00:00:47.136
Makes no sense to me.
00:00:47.435 --> 00:00:50.735
So we have to also train humans to be better at meaning making.
00:00:50.735 --> 00:00:51.396
We don't do that.
00:00:53.503 --> 00:00:54.162
Hi everyone.
00:00:54.192 --> 00:00:56.485
I'm Eric, host of the Code Riff Podcast.
00:00:56.485 --> 00:00:57.162
Hello Yaohong?
00:00:57.162 --> 00:00:57.362
Yeah,
00:00:57.453 --> 00:00:59.747
and we have a very special guest, professor Vaughn Tan.
00:01:00.232 --> 00:01:01.173
Just normal Vaughn Tan.
00:01:02.313 --> 00:01:03.307
I, I should be looking there.
00:01:03.493 --> 00:01:04.688
Uh, yeah, you can look there.
00:01:04.688 --> 00:01:04.927
Yeah.
00:01:05.188 --> 00:01:05.727
Vaughn Tan.
00:01:05.798 --> 00:01:06.283
Um.
00:01:06.302 --> 00:01:12.322
Vaughn you are the author of a book, uh, the Uncertainty Mindset, you're also co-founder of, of a few things.
00:01:12.412 --> 00:01:13.132
Poetic.
00:01:13.183 --> 00:01:15.612
you're executive director at Tangent.
00:01:15.972 --> 00:01:19.153
You are now a fellow at the Singapore Center of Strategic Futures.
00:01:19.543 --> 00:01:28.543
Uh, before that you were a founding faculty of the UCL School of Management, and you also did your PhD in organization Behavior and sociology at Harvard.
00:01:28.873 --> 00:01:29.802
Uh, so very warm.
00:01:29.802 --> 00:01:30.237
Welcome to the show.
00:01:30.882 --> 00:01:31.513
Thanks for having me.
00:01:31.873 --> 00:01:32.112
Yeah.
00:01:32.412 --> 00:01:35.022
So today we're gonna take the conversation in three parts.
00:01:35.082 --> 00:01:38.293
Uh, we'll talk about the boring tiny tools you've been writing about.
00:01:38.653 --> 00:01:47.718
Uh, secondly on tangent, the incubator, And also the last part will be on, you know, meaning making, where's the role of humans in, in the ai age Yeah.
00:01:48.108 --> 00:01:53.992
So just to kick off, uh, the uncertainty mindset that you wrote couple of years back, um.
00:01:54.673 --> 00:01:55.543
it was that time during COVID.
00:01:55.992 --> 00:01:56.832
Oh, came out during COVID.
00:01:56.953 --> 00:01:57.162
Yeah.
00:01:57.283 --> 00:01:57.643
Yeah.
00:01:57.912 --> 00:02:01.722
So, so you wrote that book, uh, maybe you can, uh, just touch very lightly on it.
00:02:01.813 --> 00:02:04.632
'cause I think you spoke about it in quite a few other places as well.
00:02:04.632 --> 00:02:04.813
Yes.
00:02:05.233 --> 00:02:05.563
Um.
00:02:06.132 --> 00:02:10.543
Actually that's all the more needed now with, you know, AI reshaping knowledge work.
00:02:10.543 --> 00:02:24.372
And, and I'm not sure if you agree with that, but, uh, you've gone from innovating at Google to researching the world's most innovative kitchens, to building your own tools, uh, and, and now tangent, the incubator.
00:02:25.242 --> 00:02:32.832
And it's, it's quite interesting how you, you went from, you know, management and, and now I'm very, very deep into tech as well.
00:02:33.138 --> 00:02:34.818
Um, and very fluent in that.
00:02:34.818 --> 00:02:39.407
How did that journey happen and why, how, yeah, why was, you know, building things?
00:02:39.647 --> 00:02:39.858
Okay.
00:02:39.858 --> 00:02:40.997
How did that come into the picture?
00:02:41.177 --> 00:02:43.397
Uh, okay, so all very good questions.
00:02:43.698 --> 00:02:48.737
I think the, the answer is like, you know, it, it sounds like it's very like planned or weird or something.
00:02:49.157 --> 00:02:57.527
There are all these particular points, like first there's tech at Google, then after that, go and do a PhD and then teach somewhere and then come back here and do this incubator.
00:02:58.217 --> 00:03:00.737
Uh, but actually it was all extremely accidental.
00:03:01.087 --> 00:03:11.108
Uh, always just sort of thinking, uh, what would be fun and interesting to do at that particular time without having any kind of long-term plan about where things are gonna end up.
00:03:11.587 --> 00:03:17.288
And I think your question about, uh, building, I think along the way, I've always been making stuff.
00:03:17.907 --> 00:03:21.418
Uh, sometimes the stuff takes the form of some theory somewhere.
00:03:22.168 --> 00:03:29.277
Sometimes that stuff takes the form of a book or some writing about that theory or some practice, uh, that is built out that theory.
00:03:30.027 --> 00:03:31.617
Sometimes the stuff is physical stuff.
00:03:31.617 --> 00:03:43.978
So I, I also, in between working at Google and, uh, doing all this sociological research, I also worked at an art school in Colorado and I was in a furniture shop making actual furniture, and I think.
00:03:44.367 --> 00:03:47.728
And then now obviously one of the things I hope we talk about is these boring, tiny tools, right?
00:03:47.728 --> 00:03:47.818
Mm-hmm.
00:03:48.057 --> 00:03:54.568
Which are pieces of software that are very, very small and really kind of boring in the sense that they do something which is not exciting.
00:03:54.568 --> 00:03:56.487
They're just doing things that are necessary to do.
00:03:56.818 --> 00:03:59.157
And they're very small pieces of software that just do that.
00:03:59.848 --> 00:04:08.937
And I think the thing about all of them is they all seem, in some cases, like they might be particularly interesting, but furniture making is not exciting.
00:04:09.177 --> 00:04:09.298
Hmm.
00:04:09.328 --> 00:04:10.948
Even working at a big tech company.
00:04:11.353 --> 00:04:16.872
For the most part, the idea of what the tech company is trying to do may be exciting, but the working inside is not that exciting.
00:04:16.872 --> 00:04:18.343
It's just you gotta do some stuff.
00:04:18.882 --> 00:04:25.153
And I think early on I realized all the really boring stuff ends up being where all the interesting stuff happens, right?
00:04:25.153 --> 00:04:27.583
So that's why I eventually became an organizational sociologist.
00:04:27.822 --> 00:04:30.432
I'm interested in the boring, everyday work of people just doing.
00:04:31.103 --> 00:04:32.692
And that's actually what the book is about.
00:04:32.692 --> 00:04:38.362
It's about people doing boring everyday work to do all the new stuff that you see in restaurants.
00:04:38.362 --> 00:04:46.642
Like when you see a new dish come out or a new ingredient at a restaurant or uh, a new way of cooking that's super exciting.
00:04:46.942 --> 00:04:54.562
Most of the time behind it is a really long series of boring things that people are doing that fail a lot and eventually turn into something that you find Exactly.
00:04:54.562 --> 00:04:55.913
'cause you coming to it for the first time.
00:04:56.002 --> 00:04:58.192
Mm. I think that's the common thread throughout all of this stuff.
00:04:58.552 --> 00:04:58.822
Mm. Yeah.
00:04:58.822 --> 00:05:00.142
I didn't realize how.
00:05:00.208 --> 00:05:20.333
How difficult it was to run an f and b outlet or even, uh, one of the top restaurants in the world until I read your book and, you know, everything needs to be so precise and yeah, it's, it's hard to see how being an expert innovation in restaurants, you know, is something that's very relevant in, in today's time as well.
00:05:20.548 --> 00:05:21.333
Yeah, I think so.
00:05:21.392 --> 00:05:28.257
Uh, so I, I think when, when I was deciding where to go look for.
00:05:28.812 --> 00:05:30.882
Insights into how you build innovation teams.
00:05:30.913 --> 00:05:31.002
Mm-hmm.
00:05:31.512 --> 00:05:38.923
And I, I, I know it seems like it's a strange thing to say, look at our r and d team at a restaurant to understand how you build an innovation team somewhere else.
00:05:38.923 --> 00:05:39.012
Mm-hmm.
00:05:39.252 --> 00:05:48.913
But actually, if you think about r and d teams at restaurants, all they're doing is trying to figure out a new thing to do where we don't know what success looks like.
00:05:49.572 --> 00:05:55.242
And we don't know how to achieve it, which is the same thing if you're in pharmaceutical research, you're trying to build a new AI model.
00:05:55.512 --> 00:06:00.523
You're trying to develop like a new public policy for a part of Singapore that you know doesn't exist yet.
00:06:00.552 --> 00:06:02.923
Or you're trying to build bongo, for example, like 20 years ago.
00:06:03.163 --> 00:06:12.223
Mm. So all of those activities that we think of as innovation, but we think of them in particular verticals like semiconductors or pharmaceuticals or ai.
00:06:12.747 --> 00:06:16.468
They're all innovation work, but we think about them in these narrow silos.
00:06:16.497 --> 00:06:21.538
Even though the work is kind of the same, like everyone who does innovation work has to be okay with failing all the time.
00:06:22.168 --> 00:06:24.958
They have to be okay with not knowing what success looks like at the beginning.
00:06:25.468 --> 00:06:29.127
They have to be okay with not knowing how to achieve the success that they can't even describe.
00:06:29.757 --> 00:06:40.288
And if you're good at that, it turns out you can be pretty good at doing it in almost any space as long as you then bring in the domain expertise of, you know, if you're trying to do pharmaceutical research, you have to have biologists.
00:06:40.588 --> 00:06:50.697
You have to have chemists, but those biologists and chemists have to have the same mindset, right, of knowing when something is not risky, but instead it's uncertain in a way that you don't know how to do it.
00:06:50.937 --> 00:07:01.858
That is scary, but also creates this opportunity for doing new things, and that's the one thing that's consistent across all of them that you understand by looking at restaurant RD chefs who are just very good at doing this because they never were trained.
00:07:02.262 --> 00:07:04.752
Buy business schools to do it, the risk management way.
00:07:04.903 --> 00:07:05.533
Mm-hmm.
00:07:05.713 --> 00:07:10.483
Yeah, that's a very good summary of the book, uh, at least for those who are interested in reading it.
00:07:10.723 --> 00:07:13.692
But today we'll focus a bit more on the boring, tiny tools.
00:07:13.692 --> 00:07:13.783
Cool.
00:07:14.023 --> 00:07:18.012
Just wanted to ask before diving into your screen sharing.
00:07:18.103 --> 00:07:18.252
Yeah.
00:07:18.283 --> 00:07:23.653
Um, how do you get into building, because, you know, you've built quite technical stuff.
00:07:23.713 --> 00:07:27.612
Um, you've built a agent platform on GitHub or something.
00:07:27.793 --> 00:07:28.093
Yes.
00:07:28.122 --> 00:07:29.773
Um, you've, you've been building.
00:07:30.002 --> 00:07:30.333
A lot.
00:07:30.333 --> 00:07:30.543
Yeah.
00:07:30.692 --> 00:07:31.262
You've been building.
00:07:31.507 --> 00:07:31.627
Yeah.
00:07:31.927 --> 00:07:34.307
How, how do you get into well building so much?
00:07:34.737 --> 00:07:36.752
Well, I mean, again, building so much by building stuff, right?
00:07:36.872 --> 00:07:40.427
So like, like I said, I, I've been, you know, even cooking is building stuff.
00:07:40.427 --> 00:07:40.788
Mm-hmm.
00:07:40.872 --> 00:07:42.632
Like you want to make dinner.
00:07:43.052 --> 00:07:43.173
Yeah.
00:07:43.202 --> 00:07:44.523
You have to assemble the ingredients.
00:07:44.523 --> 00:07:45.932
You have to have process management.
00:07:45.932 --> 00:07:49.233
You have to understand how to respond to unexpected things that happen.
00:07:49.473 --> 00:07:56.343
And it sounds very trivial, but even making fried rice, the person who knows how to do can produce good fried rice.
00:07:56.343 --> 00:07:58.442
And the person who doesn't will produce something which maybe is edible.
00:07:59.023 --> 00:08:01.423
And even that kind of building is building.
00:08:01.723 --> 00:08:01.963
Mm.
00:08:02.023 --> 00:08:22.812
And so the moment you start to build something, if you are thinking broadly about the act of building, rather than I'm building software, or I'm building a dish of fried rice, or I'm building a chair, uh, the logic of how you build, like, first you have to understand what kinds of outcomes you're looking for, what you're willing to trade off to achieve those outcomes, what materials you have available to you, what are the affordances of those materials.
00:08:23.233 --> 00:08:25.543
Then you can start to build almost anything as long as you have the tool.
00:08:25.557 --> 00:08:31.648
Mm. And I think the, some of the technical stuff that you're talking about that, um, agent Framework on GitHub.
00:08:31.887 --> 00:08:32.067
Mm.
00:08:32.097 --> 00:08:41.217
So it's a free and open source way of designing very, very small disposable agents that are inspectable by humans.
00:08:41.278 --> 00:08:43.837
So a lot of the malicious prompt injection that comes when you get.
00:08:44.472 --> 00:09:01.602
Someone releasing a massive skill that's like hundreds of, uh, lines long, uh, you don't get with agency, which is this, uh, tool that I've built because agency agents are spun up on demand based on what you want them to do with very small primitives that define parts of their roles.
00:09:01.602 --> 00:09:11.232
Just like humans in an organization and the primitives are very short, they're like natural language and you can read them to make sure that they're not trying to tell your agent to do something bad without you knowing about it.
00:09:11.238 --> 00:09:11.488
Mm-hmm.
00:09:11.668 --> 00:09:11.888
So.
00:09:13.302 --> 00:09:14.832
That is a piece of software.
00:09:15.582 --> 00:09:21.913
In a way, it's very technical because I would say even two years ago, I would've had no way of building this on my own.
00:09:21.972 --> 00:09:22.062
Mm-hmm.
00:09:22.423 --> 00:09:30.375
But agency as a framework actually is just the research that I've done for, at this point, uh, since 2008.
00:09:30.822 --> 00:09:34.332
About how you build organizations that have people inside them that do cool things.
00:09:34.663 --> 00:09:38.653
So if I can describe, I can describe very technically and precisely'cause I'm trained to do that.
00:09:38.707 --> 00:09:39.128
Mm-hmm.
00:09:39.207 --> 00:09:46.332
And that just turns out to be the way you write a good product requirements document that you send off to a person in the past, a software engineer to build.
00:09:47.143 --> 00:09:49.783
But it also turns out to be, now you've got agents that build software.
00:09:49.812 --> 00:09:53.202
If you give them a good PRD, they can write something that works as software, right?
00:09:53.202 --> 00:09:58.123
So all code is just an accurate representation of what you want a machine to do.
00:09:58.543 --> 00:10:07.243
And if you can write clearly and unambiguously, then in the past you needed to write in ways that humans would understand who are software developers.
00:10:07.572 --> 00:10:09.582
And today you just have to be unambiguous and clear.
00:10:09.582 --> 00:10:16.572
And it turns out if you've got the right process for managing the product requirements documentation process, you can generate usable software.
00:10:16.993 --> 00:10:18.673
And I can show you some of them later.
00:10:18.673 --> 00:10:26.982
But I think the, the point I wanted to make is how do you get into building, you get into building by building some things and then realizing that actually the act of building.
00:10:27.403 --> 00:10:35.682
Is more or less similar is just using different tools to build and having a mindset of being clear about what you want and what you're willing to give up to get what you want.
00:10:35.863 --> 00:10:36.043
Mm,
00:10:36.133 --> 00:10:39.312
turns out to be actually very good preparation for building tools.
00:10:39.972 --> 00:10:46.423
That are software tools today, because the tools for building software have become so much more accessible to people like me.
00:10:46.423 --> 00:10:48.763
I'm completely untrained in computer science.
00:10:48.793 --> 00:10:48.822
Okay.
00:10:49.062 --> 00:10:51.163
I'm a sociologist and I'm a qualitative sociologist.
00:10:51.163 --> 00:10:51.312
Right?
00:10:51.312 --> 00:10:53.232
I go into the field, I people, I hang with the
00:10:53.232 --> 00:10:53.653
views.
00:10:53.653 --> 00:10:53.802
Yeah.
00:10:53.863 --> 00:11:02.937
I'm a qualitative sociologist who has never built software in the past, ever, and it is possible if you have good expectations that are reasonable about what kind of software you can build.
00:11:03.207 --> 00:11:06.658
And we can talk much more about this later, but I wanted to throw it back to you.
00:11:06.837 --> 00:11:12.597
I would guess that for most people, getting the clarity of thought is the most difficult step, right?
00:11:17.878 --> 00:11:23.548
Actually, to me, I find the clarity of thought is one of the most difficult things to elicit out of other humans.
00:11:23.607 --> 00:11:24.057
Absolutely.
00:11:24.207 --> 00:11:26.217
But, but not just elicit, but also for yourself.
00:11:26.217 --> 00:11:26.307
For
00:11:26.307 --> 00:11:26.697
myself also.
00:11:26.793 --> 00:11:27.033
Yeah.
00:11:27.033 --> 00:11:27.298
Yeah, yeah.
00:11:27.327 --> 00:11:27.793
No, I, I, I,
00:11:27.893 --> 00:11:32.967
but I go through like this whole enter entire mental model framework in my brain when I have a question, right?
00:11:33.057 --> 00:11:33.298
Yeah.
00:11:33.357 --> 00:11:40.048
To make sure that I, I go through every single, every single point, but like quite deep in thought, right?
00:11:40.048 --> 00:11:44.097
So that I don't end up committing all the errors, you know, your type one or type two errors.
00:11:44.097 --> 00:11:44.337
Yeah.
00:11:44.342 --> 00:11:44.817
That kind of stuff.
00:11:44.817 --> 00:11:45.057
Yeah.
00:11:45.477 --> 00:11:45.807
Yeah.
00:11:45.898 --> 00:11:46.077
Okay.
00:11:46.077 --> 00:11:47.038
So, so I, I agree.
00:11:47.038 --> 00:11:57.327
I think, uh, I won't say for everyone, but whenever I talk to people about the things that they're working on or when I'm thinking about the things I'm working on, being clear about what I'm trying to do.
00:11:57.868 --> 00:11:59.067
Is one of the hardest things, right?
00:11:59.067 --> 00:12:03.298
And it is very sort of like, it takes discipline to do it 'cause it's hard.
00:12:03.687 --> 00:12:05.248
'cause you always wanna just like get on with it.
00:12:05.307 --> 00:12:14.067
But actually thinking clearly is a big flex because if you think clearly at the beginning, you would end up doing things faster and you can do things better in the long run.
00:12:14.307 --> 00:12:18.988
And I would say like the clarity of thought, but there's probably a few aspects of clarity of thought, which, you know, you've got a framework.
00:12:19.048 --> 00:12:20.638
I, I also have a framework for thinking about it.
00:12:21.087 --> 00:12:24.057
One of them is you have to know what outcome is a successful outcome.
00:12:24.327 --> 00:12:27.628
And some of them you can define very clearly and some of them you cannot.
00:12:28.288 --> 00:12:28.467
Right?
00:12:28.467 --> 00:12:42.597
And this will be relevant for boring, tiny tools because I think these kinds of tools, the software that we write should only be written for things where the success outcome is very, very clearly definable in advance, fully specifiable, and also stable and not changing, right?
00:12:42.597 --> 00:12:43.258
So that's one aspect.
00:12:43.258 --> 00:12:45.118
You have to be clear about what you're trying to achieve.
00:12:45.807 --> 00:12:49.648
The second aspect is you have to be clear about what you're willing to give up that's valuable to achieve it.
00:12:49.707 --> 00:12:51.327
Like for example, I could spend.
00:12:51.807 --> 00:12:58.107
Three or four months writing a really, really well designed user experience, user interface for some of the boring, tiny tools that I have.
00:12:58.648 --> 00:13:02.908
But I would much rather have something that works fast, we can iterate rather than something that looks nice.
00:13:02.908 --> 00:13:08.998
And when I show you some of the boring, tiny tools, you'll see the user interface is very, very junky, but it works well underneath.
00:13:09.628 --> 00:13:13.288
So the first one is clarity of what outcomes are?
00:13:13.288 --> 00:13:14.187
Success outcomes.
00:13:14.217 --> 00:13:17.518
The second one is clarity about what trade-offs are acceptable and unacceptable.
00:13:18.148 --> 00:13:24.148
Then the third one, which I think goes to this question of like you can build in this vertical, but not in that vertical.
00:13:24.388 --> 00:13:30.028
A lot of clarity of thought is knowing like what are the tools that you have available to you and what are their affordances, right?
00:13:30.087 --> 00:13:31.707
Affordances just means what can the tools do?
00:13:32.307 --> 00:13:45.207
And the reason why software is now available to so many more of us, including people like me who have never built software before, is we used to not have machines that can build software based on natural language inputs as well as they can now.
00:13:45.447 --> 00:13:47.067
And I think this really only happened.
00:13:47.982 --> 00:13:49.332
I don't know, December last year.
00:13:49.332 --> 00:13:49.602
Yep.
00:13:49.692 --> 00:13:49.812
Mm-hmm.
00:13:50.052 --> 00:13:50.293
December.
00:13:50.653 --> 00:13:51.187
That was the 20 point.
00:13:51.187 --> 00:14:01.873
So to me, December is like, there's this like sudden cut point where I was using, I don't know, like Claude Code to write some things between March and December last year.
00:14:01.873 --> 00:14:04.932
And I was like, yeah, it was slow, was not especially good.
00:14:05.383 --> 00:14:09.043
And then suddenly December, there was a phase shift and all of a sudden many, many things became possible.
00:14:09.102 --> 00:14:19.393
And in the months after that, I think we've, we have seen that things are possible now that were not possible before because the tools have become better, the affordances have become much better.
00:14:19.393 --> 00:14:33.732
So if you know what you're trying to achieve in terms of success, you know what trade offs you're willing to make, and you're keeping an eye on the tools and what they can do for you, suddenly when the tools become much better, you can start to do a lot more things as you've got clarity already in what you're trying to achieve and what trade offs you're trying to make.
00:14:34.003 --> 00:14:34.273
Mm.
00:14:34.393 --> 00:14:34.543
Yeah.
00:14:49.581 --> 00:14:51.591
one last thing before we look at your screen.
00:14:51.760 --> 00:14:55.181
why do you find it so important to build boring, tiny tools?
00:14:55.313 --> 00:14:58.614
I mean, it seems like it's your, your thesis for the work that you're doing right now, right?
00:14:58.614 --> 00:14:58.673
Yeah.
00:14:58.703 --> 00:14:59.783
Why, why is that important?
00:14:59.813 --> 00:15:00.413
Why is it okay?
00:15:00.418 --> 00:15:11.183
Yeah, so I think boring, tiny tools, there's like a, in itself, I think boring, tiny tools are important because we use lots and lots of boring things all the time, and they're just very useful.
00:15:11.183 --> 00:15:21.144
So anyone who uses Excel and has formulas that do stuff repeatedly, you have actually built a boring, tiny tool in what is basically a programming platform, right?
00:15:21.144 --> 00:15:22.374
Excel is a programming platform.
00:15:22.374 --> 00:15:27.563
You just like write the formula to transform some cell contents into some other cell contents that you want.
00:15:27.953 --> 00:15:28.673
We do it all the time.
00:15:28.673 --> 00:15:39.714
We do it when we, I don't know, like on the table there, there's a, a teabag on a, on a saucer that is a boring, tiny tool to prevent like the tea from just like spilling all over.
00:15:40.134 --> 00:15:42.923
So boring, tiny tools are important'cause we use them all the time.
00:15:43.014 --> 00:15:50.573
They, they're much, they're actually much more important than the really big, charismatic, sexy tools that we use very infrequently or don't do very much for most people, right?
00:15:51.053 --> 00:15:52.823
So that, that's one reason.
00:15:52.884 --> 00:15:54.293
It's a bit like saying.
00:15:54.849 --> 00:16:07.568
Uh, should we only focus on the very big animals in the African savanna, like the lions or the, or the giraffes or the elephants because they're so big and they're, so, obviously they call them charismatic megafauna 'cause they're really big.
00:16:08.078 --> 00:16:14.318
But actually in almost any ecosystem, the vast majority of the biomass is all the boring small stuff that you can barely even see.
00:16:14.318 --> 00:16:16.208
The fungi, the bacteria, the plants.
00:16:16.599 --> 00:16:18.428
That makes up almost everything.
00:16:18.999 --> 00:16:20.229
So boring, tiny tools are important.
00:16:20.229 --> 00:16:23.408
'cause they are probably going to be the vast majority of the tools that we use.
00:16:23.589 --> 00:16:25.899
And we actually don't make them very well.
00:16:26.259 --> 00:16:26.438
Right?
00:16:26.438 --> 00:16:32.589
Because most of the time the tools that we have that are software tools, they were built under old software economics.
00:16:32.649 --> 00:16:38.019
It used to be the rate limiting factor for building software was smart, expensive software engineers.
00:16:38.198 --> 00:16:38.408
Yep.
00:16:38.798 --> 00:16:48.519
And as a result, if you need smart, expensive software engineers to build software, you tend to only build software that serves a very, very large number of people.
00:16:49.089 --> 00:16:56.379
When you serve a very large number of people with software, it tends to be the case that the software only serves everyone in an average to mediocre way, right?
00:16:56.379 --> 00:16:57.639
The lowest common denominator.
00:16:58.749 --> 00:17:07.659
But then everyone is using it so it makes sense for the software developer to make something that doesn't work great for any particular person, but works kind of okay for millions and millions of people.
00:17:08.019 --> 00:17:09.939
So we don't build boring, tiny tools.
00:17:09.939 --> 00:17:15.038
Now, by default, we build really big charismatic megafauna tools that aren't very good.
00:17:15.608 --> 00:17:29.229
But now that it's possible to build software on your own, some kinds of software and we'll, we'll talk about the kinds of software, some kinds of software on your own, you can start to build software that is really boring, very tiny and completely tailored to how you use it.
00:17:29.588 --> 00:17:41.439
Those boring tiny tools are the most useful ones and, and so I just wanna say before we share screen or whatever that I'm not saying that boring tiny tools are all of the software that we should build.
00:17:41.888 --> 00:17:45.068
And I'm not saying that big, difficult to build.
00:17:45.443 --> 00:18:08.574
Software that implements new mathematics or new algorithms that have never before been developed are things that I, or other so-called small, boring, tiny tool developers should build those things legit, should be built by trained, expensive, really smart mathematicians and software engineers, but really boring tiny things where the, the problem you're trying to solve has been asked many times in the past.
00:18:08.963 --> 00:18:11.124
The answer is convergent.
00:18:11.273 --> 00:18:14.453
Every time you ask the problem, the same solution shows up.
00:18:14.874 --> 00:18:18.773
Those tools can be boring and tiny, and we should be building them ourselves.
00:18:18.834 --> 00:18:18.983
Mm-hmm.
00:18:19.223 --> 00:18:20.574
And then we should tailor them to how we use them.
00:18:21.068 --> 00:18:21.269
Interesting.
00:18:21.653 --> 00:18:22.013
Okay.
00:18:22.074 --> 00:18:23.064
Enough questions.
00:18:23.124 --> 00:18:24.653
Maybe if you could show us.
00:18:24.683 --> 00:18:24.923
Okay.
00:18:25.193 --> 00:18:27.923
I can show you, show you, uh, one, let me see.
00:18:27.963 --> 00:18:36.034
So, uh, one boring, tiny tool is basically, it's a way for us.
00:18:36.094 --> 00:18:39.993
Uh, so this is a tool built for tangent, and tangent is two people.
00:18:40.263 --> 00:18:42.094
Uh, but we run an incubator program.
00:18:42.544 --> 00:18:50.523
So most incubator programs take three to six people between 12 and 18 months to start up, uh, tangent started April.
00:18:50.943 --> 00:18:58.203
We launched our first cohort at the halfway through July, and we did it with two people, right?
00:18:58.203 --> 00:19:14.884
So one of the assumptions that we're making with tangent for its participants and also for ourselves is if we know how to build boring, tiny tools, well we can do many more things with the humans that are the valuable limiting resource than if we don't use boring, tiny tools.
00:19:14.943 --> 00:19:22.564
So some of the things that most small organizations find really difficult is you haven't developed workflow processes.
00:19:23.449 --> 00:19:27.169
And a lot of things have to happen and you've only got a small number of people to do it.
00:19:27.648 --> 00:19:32.419
And all these tasks that show up, they happen in the course of doing the work that you're doing.
00:19:32.419 --> 00:19:41.239
So when you send emails to people, uh, you've gotta follow up on a board meeting and it comes by email or you are messaging your colleague and, oh, we gotta do this thing.
00:19:41.868 --> 00:19:48.739
Not only are you saying, yeah, we gotta do this thing, you also have to remember to put it into an issue tracker so that you remember to do it and you can follow up on it.
00:19:49.249 --> 00:19:59.689
So instead of doing that where you have to remember to do all this stuff, uh, we now have the beginnings of a tool that will automatically look at all of our message channels.
00:19:59.929 --> 00:20:08.659
And when it identifies something which is a potential task, it will collect it and surface it for human review to make sure that it actually is a task.
00:20:08.659 --> 00:20:11.358
And then you can also enrich it, modify it, and change it.
00:20:11.624 --> 00:20:11.743
Hmm.
00:20:11.749 --> 00:20:14.749
So I can show you all what it looks like.
00:20:14.749 --> 00:20:17.628
And this goes down to, uh, that.
00:20:17.689 --> 00:20:28.759
Uh, distinction that I was making before between, uh, knowing what you're trying to achieve and also knowing what valuable outcomes you're willing to give up to get that, uh, the user interface of this is very ugly.
00:20:28.969 --> 00:20:29.058
Mm-hmm.
00:20:29.568 --> 00:20:32.598
And it, so I'll show you what happens.
00:20:32.598 --> 00:20:37.909
So basically we're using WhatsApp, telegram and email to manage this whole program.
00:20:38.449 --> 00:20:43.608
And whenever a message comes into one of those channels, I have a little system that runs on a machine that's here.
00:20:43.608 --> 00:20:50.118
It's just a very old MacBook Pro 'cause we're cheap and it's just listening to all these channels over a matrix bridge.
00:20:50.328 --> 00:21:02.628
And when it sees something that has some action words inside it, it'll pull those action words out and it'll ask us whether or not this is a real task that we need to include into our issue trackers.
00:21:02.628 --> 00:21:12.469
So I'll show you first, uh, the review interface, which is what happens after the system has pulled a potential task out of it.
00:21:14.433 --> 00:21:18.784
So that's, uh, TRK is the tool, uh, review is the state.
00:21:18.963 --> 00:21:21.784
So it's the tracker tool in review state.
00:21:21.814 --> 00:21:24.693
And after role force, the role number four is reviewer.
00:21:25.413 --> 00:21:25.594
Hmm.
00:21:25.624 --> 00:21:38.403
So, uh, right now there's a few things in the pipeline, uh, because I'm the only person using it, I, I'm going to extend it to, uh, Harper, my colleague once I make it bug free for me.
00:21:39.183 --> 00:21:47.314
So what you've got here is you've got, um, five tasks that came out of the last, I think probably two and a half hours since I last cleared this
00:21:47.433 --> 00:21:47.554
Wow.
00:21:47.763 --> 00:21:52.804
Of messages with various people or with Harper or emails that came through.
00:21:53.493 --> 00:22:01.173
And, uh, I triaged this by simply putting in, so for example, I, I, I'm just going to fix it later, but I'll just randomly do stuff.
00:22:01.173 --> 00:22:03.693
If I wanted to accept item one.
00:22:04.368 --> 00:22:05.538
I accept item one.
00:22:05.898 --> 00:22:15.528
I wanna modify like item two, and then I'm going to leave item four pending and I will reject everything else.
00:22:15.858 --> 00:22:16.068
Hmm.
00:22:16.759 --> 00:22:22.669
And then it will just do all this stuff on the backend, and then I get to modify the title.
00:22:22.669 --> 00:22:26.358
So, uh, yeah, quarterly meetings.
00:22:30.138 --> 00:22:30.439
Okay.
00:22:32.808 --> 00:22:33.138
Alright.
00:22:33.409 --> 00:22:34.759
So now all that's done.
00:22:34.818 --> 00:22:42.798
If I wanna see all of the existing, uh, tasks that have been approved, I go to again, tracker.
00:22:43.098 --> 00:22:52.608
Now the view is the selection view, and I can see all of the tasks that have already gone through that review and acceptance, uh, workflow.
00:22:52.999 --> 00:22:54.378
And these are all the things that are not open.
00:22:54.378 --> 00:22:58.398
And I can, I, I won't show you now that, but we can also assign to different people.
00:22:58.459 --> 00:22:59.298
We can change the state.
00:22:59.298 --> 00:23:01.519
This is the state machine for managing all the tasks.
00:23:01.818 --> 00:23:03.169
So the intention behind this is.
00:23:03.723 --> 00:23:23.403
Really automate the part that really, that in the past was very hard to automate, which is ingestion of things that an organization has to do into a pipeline that a human can look at so that the human can decide whether it's a real task or not, and then can act on that task by assigning it to different people acting on it, updating its state, and then clearing it out.
00:23:23.644 --> 00:23:23.763
Mm.
00:23:23.884 --> 00:23:24.094
Right.
00:23:24.094 --> 00:23:34.413
The idea is to make this as friction free as possible so that a small team can work without, for example, executive assistant, without someone managing all this stuff in a very time intensive way.
00:23:34.624 --> 00:23:41.523
Now we can just do it in as automated way as possible, leaving out all the stuff of judgment that, that humans still have to do.
00:23:41.523 --> 00:23:44.074
Mm. So this is one example and then another example,
00:23:44.074 --> 00:23:44.824
which I have a question on this.
00:23:44.973 --> 00:23:45.153
Yeah, go for it.
00:23:45.334 --> 00:23:52.114
So why, why do you decide to use this interface instead of, you know, uh, existing?
00:23:52.788 --> 00:23:55.759
Uh, trackers for teams, right?
00:23:55.848 --> 00:23:56.118
Yeah.
00:23:56.568 --> 00:24:00.979
Uh, that, that also has a kind of similar functions and you build a breach to them instead of uh,
00:24:01.009 --> 00:24:01.548
yeah, for sure.
00:24:01.548 --> 00:24:02.088
Like Slack.
00:24:02.094 --> 00:24:02.449
Slack or something.
00:24:02.808 --> 00:24:03.318
Not Slack.
00:24:03.318 --> 00:24:07.519
Not just Slack, but, you know, like ban board or, and monday.com com.
00:24:07.523 --> 00:24:08.298
So, so right
00:24:08.469 --> 00:24:09.469
what's the thought process behind it.
00:24:09.469 --> 00:24:12.108
So the thought process behind it is, I don't use Kanban.
00:24:13.128 --> 00:24:14.239
Uh, I don't use Monday.
00:24:14.959 --> 00:24:16.368
I tried using Monday once.
00:24:16.368 --> 00:24:16.999
I don't like it.
00:24:17.509 --> 00:24:19.189
So, and I know what I want.
00:24:19.459 --> 00:24:25.038
So this is the thing about boring, tiny tools, like boring, tiny tools are almost by definition not meant to be for enormous numbers of people,
00:24:25.249 --> 00:24:26.358
specific for your needs.
00:24:26.358 --> 00:24:29.298
They're very, so Clay Shirky calls them situated software, right?
00:24:29.298 --> 00:24:39.469
They're software that works just maybe in, in the extreme case, just for you, but almost always just for a very small group of people for whom the other stuff maybe is not so good.
00:24:40.128 --> 00:24:49.368
So, uh, there's no particularly good reason other than I don't like the user experience of using those other things, which is in this case totally legit.
00:24:49.368 --> 00:24:51.828
If I need to build software for 10 million people.
00:24:52.203 --> 00:24:55.324
I have to make sure it's different from all the other software that's out there for 10 million people.
00:24:55.324 --> 00:24:55.683
Mm. Right.
00:24:55.683 --> 00:25:01.653
I have to have a USP when I build boring, tiny tools for myself to run the organization that I run.
00:25:01.894 --> 00:25:03.544
I built them just for what I want.
00:25:03.544 --> 00:25:08.433
And this is the really, really interesting thing about the change in software economics that happened eight months ago.
00:25:08.703 --> 00:25:09.094
Right.
00:25:09.483 --> 00:25:10.054
But, okay.
00:25:10.203 --> 00:25:19.743
The other reason why we do this is because if at some point I want to pipe this into one of those other things, the data structure behind it, I have designed the data structure to be something that you can interact with programmatically.
00:25:19.773 --> 00:25:29.644
So it's easy to, for example, use this backend, which is there's a state machine on the backend that understands how we treat the states of different tasks inside tension.
00:25:30.334 --> 00:25:34.263
And after that, the bit that ingests, I can add other things, right?
00:25:34.263 --> 00:25:36.423
So we have email, we have WhatsApp, we have Telegram.
00:25:36.903 --> 00:25:44.013
Maybe in the future we'll have voice to text that's on plot, and I'll just hook that in maybe later.
00:25:44.134 --> 00:25:45.814
One of my other colleagues who we haven't hired yet.
00:25:46.324 --> 00:25:47.884
They want to use Slack to manage all this stuff.
00:25:47.884 --> 00:25:49.683
I can pipe it into Slack maybe, I'm not sure.
00:25:49.743 --> 00:25:53.973
But the idea is to try and build tools that are extensible and forward compatible if you can.
00:25:54.844 --> 00:25:57.183
But there's no particularly good reason why I don't use man or Kamala.
00:25:57.183 --> 00:25:57.693
I just don't like that.
00:25:57.903 --> 00:26:00.453
It's basically that's a legit reason.
00:26:00.453 --> 00:26:00.634
You know,
00:26:01.818 --> 00:26:05.223
it's, it's a legit reason for this kind of thing, again, it's very important.
00:26:05.223 --> 00:26:09.874
If I'm trying to build big scale software that I'm trying to sell to 10 million people, I cannot say that kind thing, right?
00:26:09.874 --> 00:26:14.554
I have to go do the market research and figure out Monday and Kanban do these things.
00:26:14.554 --> 00:26:15.513
Many people love them.
00:26:15.634 --> 00:26:17.463
What is it that they don't do?
00:26:17.584 --> 00:26:19.294
And do enough People like that thing.
00:26:19.294 --> 00:26:19.953
They don't do that.
00:26:19.953 --> 00:26:21.124
I built that feature.
00:26:21.213 --> 00:26:29.013
So I've actually been doing this on my Claude Code, so I have a to-dos file on my desktop, which I run all my cloud code command from.
00:26:29.733 --> 00:26:33.124
Um, maybe, maybe two observations of what you built.
00:26:33.124 --> 00:26:34.324
One is, is very simple.
00:26:34.413 --> 00:26:34.564
Yeah.
00:26:34.594 --> 00:26:38.344
It's, it looks almost like intentionally, stupidly simple, and.
00:26:38.673 --> 00:26:43.193
Very restricted in a way that it doesn't allow you to kind of do whatever you want with it, right.
00:26:43.374 --> 00:26:45.054
It really boiled it down to the essence.
00:26:45.054 --> 00:26:45.804
So it's very simple.
00:26:45.923 --> 00:26:49.344
And then secondly, I think if you heard of Claude Tag, right?
00:26:49.344 --> 00:26:53.064
That, that sits recently claude launched that, and it sits in your Slack messages.
00:26:53.064 --> 00:26:53.084
It's a
00:26:53.138 --> 00:26:53.979
terrible product by the way.
00:26:55.679 --> 00:26:58.824
Uh, it sits in your messages, it reads everything.
00:26:58.824 --> 00:27:02.064
And then it also kind of, you know, if you ask it, what should I do for the day?
00:27:02.064 --> 00:27:04.854
It can do it, but it's a lot more token intensive, right?
00:27:04.854 --> 00:27:05.003
For sure.
00:27:05.003 --> 00:27:07.584
So simple and very, it looks very token efficient.
00:27:07.673 --> 00:27:08.509
Like, do you mind explain,
00:27:08.828 --> 00:27:10.314
actually, I don't even think there's token involved, right?
00:27:10.314 --> 00:27:21.713
now we can talk about some of the more philosophical aspects of building boring, tiny tools because one thing I also don't wanna be locked into is I don't wanna be locked into the harnesses and harness adjacent components of some of the foundation labs.
00:27:22.104 --> 00:27:23.604
So, as.
00:27:24.159 --> 00:27:28.479
Both Anthropic and also OpenAI and also the Chinese labs.
00:27:28.479 --> 00:27:33.219
Everyone is trying to build better and better user experience skins around their models.
00:27:33.788 --> 00:27:43.328
We are becoming used to the idea that we must interact with all of these software services and machines via an intermediary that is not us.
00:27:43.719 --> 00:27:50.499
And I much prefer to build largely deterministic software using very, very frontier models.
00:27:51.098 --> 00:27:55.058
But then once the software is running, it only touches very stupid models.
00:27:55.058 --> 00:28:00.098
So for example, this, uh, what you're seeing on the screen mostly is deterministic software.
00:28:00.308 --> 00:28:00.489
Mm-hmm.
00:28:00.608 --> 00:28:00.939
Right?
00:28:01.148 --> 00:28:01.298
Yeah.
00:28:01.358 --> 00:28:09.878
Uh, but all of the stuff where, so everything where it looks at particular channels via a matrix bridge absorbs the content.
00:28:10.509 --> 00:28:14.588
That stuff is deterministic, normal software.
00:28:15.128 --> 00:28:21.159
Once it takes the content out of that channel, and it has to look at it to infer whether it's a, it's a task or not a task.
00:28:21.548 --> 00:28:24.354
Then I call very, very dumb models, right?
00:28:24.354 --> 00:28:24.634
Mm-hmm.
00:28:24.713 --> 00:28:25.808
So right now I'm using Haiku.
00:28:25.868 --> 00:28:26.439
It's very cheap.
00:28:27.098 --> 00:28:28.328
I have tried it with Gemini.
00:28:28.358 --> 00:28:29.138
It's even cheaper.
00:28:29.138 --> 00:28:30.278
It works almost as well.
00:28:30.818 --> 00:28:34.118
I can probably try it with some Chinese model that is like even cheaper Yep.
00:28:34.239 --> 00:28:35.919
And faster and can run locally.
00:28:36.098 --> 00:28:36.219
Yep.
00:28:36.278 --> 00:28:39.638
But I, my box is not because, you know, the Mac Mini hasn't arrived, right?
00:28:39.759 --> 00:28:45.489
So like, at some point I will make this entirely local and it will not touch the outside world at all.
00:28:45.519 --> 00:28:45.729
Right?
00:28:45.729 --> 00:28:58.118
Now this is just a, a way it's, it's a step on the way there, but I think the, the key is you can always do it easier with someone else, or you can do it less well with someone else.
00:28:58.118 --> 00:29:06.818
But if you want to do it in a way that is more controllable, where you understand what's going on and you can also tailor it to what you want, you almost have to do it yourself.
00:29:06.848 --> 00:29:10.419
And now you can, uh, for some limited set of problem.
00:29:10.419 --> 00:29:20.858
Like, if the problem is, again, I think it's really important to overemphasize this if the problem has been posed many, many times in the past, and the solution to the problem is always more or less the same solution.
00:29:21.233 --> 00:29:34.884
This is a great domain to get machines to help you write code for if nobody's ever asked this problem before, if the solution requires brand new mathematics, do not try and vibecode your way into that.
00:29:34.884 --> 00:29:35.034
Right?
00:29:35.034 --> 00:29:36.094
Like, that is not for vibecoding.
00:29:36.304 --> 00:29:42.923
vibecoding is for all the really well established problems that have been solved since the 1970s.
00:29:42.983 --> 00:29:44.874
We just keep reimplementing the same libraries.
00:29:45.084 --> 00:29:46.523
That's where we should be working on.
00:29:46.523 --> 00:29:47.933
Those are the boring things.
00:29:48.023 --> 00:29:48.233
Mm,
00:29:48.324 --> 00:29:49.253
boring, tiny tools.
00:29:49.913 --> 00:29:50.003
Yeah.
00:29:50.913 --> 00:29:55.953
the second one is a bit harder to show on screen because it happens on, well, so I, I can show on the phone.
00:29:56.003 --> 00:29:57.084
This one is very, very boring.
00:29:57.084 --> 00:29:57.564
Very tiny.
00:29:57.683 --> 00:29:57.953
Yeah.
00:29:58.044 --> 00:30:00.443
And the problem it's trying to solve is actually very much a personal, Okay.
00:30:00.443 --> 00:30:02.064
So let me just share this.
00:30:02.124 --> 00:30:02.243
Yeah.
00:30:02.243 --> 00:30:07.584
And while I'm starting share, I can explain a bit more about what it is that, uh, we're doing Telegram.
00:30:08.469 --> 00:30:08.618
Okay.
00:30:08.669 --> 00:30:21.388
Alright, so this is a, a new version of a tool that I built probably about a year and a half ago to manage my own, um, receipts and invoices and expenses for the consulting company that I also run.
00:30:22.199 --> 00:30:32.638
And now that I am working for another organization that I'm incurring lots of very small expenses because we were setting up the payment systems, uh, for the last like four or five months.
00:30:33.239 --> 00:30:49.019
Uh, I wanted to have a way to manage all of these small expenses and make sure that I actually claim them without just accumulating a massive stack of receipts and then putting it off until three or four months in and then having to process like hundreds of receipts and make sure that they're all correct.
00:30:49.378 --> 00:30:51.388
And so this is a very, very simple tool.
00:30:52.138 --> 00:30:54.659
Uh, we can, it's actually been working quite well.
00:30:54.659 --> 00:30:59.009
It's quite stable, so I can just use it right now.
00:30:59.999 --> 00:31:12.749
And what it does is it takes a, um, it takes a receipt, a photo of a receipt, a PDF, uh, a scan, and it extracts key receipt information out of it.
00:31:13.499 --> 00:31:17.878
Stuff like what date, who is the merchant, what is the amount?
00:31:18.824 --> 00:31:33.084
Then it infers what the description and justification of this event was based on looking at my calendar and it proposes what the receipt justification should be So normally my poll is every two hours.
00:31:33.183 --> 00:31:40.713
it'll actually come back as a digest that has extracted the receipt, name is extracted, the merchant, all that other stuff.
00:31:40.773 --> 00:31:47.493
And then it lets me just say, approve, confirm, or reply to the message that extracted all that to update some stuff.
00:31:47.499 --> 00:31:47.699
Mm-hmm.
00:31:48.003 --> 00:31:59.013
And then once I approve it, it goes into an approved queue to automatically generate a Zoho expenses, uh, report that I can then check on Zoho and send.
00:31:59.449 --> 00:31:59.719
Hmm.
00:31:59.749 --> 00:31:59.929
Right.
00:31:59.929 --> 00:32:06.769
So what it does is it means that whenever I incur an expense, I can just take a photo of the receipt, send it to Telegram, and I don't have to think about it anymore.
00:32:06.798 --> 00:32:11.868
It will be in a queue for me to manage later, and I don't have to like even keep the receipts anymore.
00:32:12.499 --> 00:32:16.878
Uh, if it comes by email, I can just take the PDF, drop it in Telegram and send it on.
00:32:17.419 --> 00:32:23.449
And it, it is very weird that this boring, tiny tool has changed my life, but it has kind of changed my life.
00:32:23.449 --> 00:32:23.749
It's great.
00:32:23.753 --> 00:32:23.894
No,
00:32:23.959 --> 00:32:24.019
no.
00:32:24.199 --> 00:32:25.308
This, this is fantastic.
00:32:25.338 --> 00:32:25.578
Yeah.
00:32:25.638 --> 00:32:27.469
Actually I've built this as well.
00:32:27.528 --> 00:32:27.648
Yeah.
00:32:27.648 --> 00:32:30.108
But I've done it in a Claude project.
00:32:30.199 --> 00:32:30.409
Okay.
00:32:30.409 --> 00:32:33.769
So I think philosophically there are a bit of different world views.
00:32:33.769 --> 00:32:33.858
Yeah.
00:32:33.858 --> 00:32:35.689
I like to dive a bit deeper into this.
00:32:35.689 --> 00:32:35.929
Okay.
00:32:35.989 --> 00:32:36.348
Yeah, for sure.
00:32:36.348 --> 00:32:39.709
how do you find a discipline to build this?
00:32:39.713 --> 00:32:39.804
Yeah.
00:32:40.098 --> 00:32:43.378
As compared to just making a skill, you know, or like Oh, yeah, yeah.
00:32:43.378 --> 00:32:43.699
My
00:32:43.699 --> 00:32:44.449
least as a skill
00:32:44.808 --> 00:32:44.929
also.
00:32:45.199 --> 00:32:47.239
And I think there's a philosophy that you.
00:32:47.374 --> 00:32:49.144
Um, you don't wanna be locked in.
00:32:49.144 --> 00:32:52.173
You want kind of freedom and, and that's why you do things.
00:32:52.173 --> 00:32:57.663
Deterministically is like a very disciplined way of, it's a very disciplined way of thinking, as in, I I'm just trying to unpack that.
00:32:57.663 --> 00:32:57.784
Yeah,
00:32:57.784 --> 00:32:58.203
possibly.
00:32:58.203 --> 00:32:58.443
Okay.
00:32:58.443 --> 00:33:00.963
So, so the, I think we, we split into two parts, right?
00:33:00.963 --> 00:33:01.054
Yeah.
00:33:01.054 --> 00:33:05.223
Why not just do it as the quick, like if you write a skill, it's very fast.
00:33:05.973 --> 00:33:07.473
Uh, why not just do it that way?
00:33:07.983 --> 00:33:13.233
And one of the explanations is legit, I'm not that interested in continuing to be super locked in.
00:33:13.233 --> 00:33:17.523
I, I use Claude Code mostly, but I do not want to be stuck using Claude Code.
00:33:17.644 --> 00:33:17.973
Right.
00:33:18.483 --> 00:33:26.314
Um, and the more, the more things and even skills are not actually truly Claude Code or anthropic specific.
00:33:26.314 --> 00:33:26.644
Exactly.
00:33:26.644 --> 00:33:26.733
Mm-hmm.
00:33:26.973 --> 00:33:27.034
Yeah.
00:33:27.153 --> 00:33:38.374
But the more I externalize into a particular system where it's not controlled by me in terms of how I, how I design it, the less I am in the habit of thinking about how to do this.
00:33:38.433 --> 00:33:47.719
So the big distinction for me between doing something that's a soft, uh, soft control playing skill and writing a. Like basically this is a weird kind of application.
00:33:48.348 --> 00:33:51.528
Uh, it's a bot application thing.
00:33:52.669 --> 00:33:57.439
The main difference for me is how often will I use it and will I want to extend it to other people?
00:33:57.499 --> 00:34:13.818
Because for this one, if you think about how you would have to do it, if you were using it as a Claude skill, even if you're using the Claude app, I would have to fire up the app, start a new chat, and then invoke the skill and attach, uh, this, you know, receipt.
00:34:14.329 --> 00:34:24.168
And in some ways what happens is you, if you get used to doing that, that's fine, but it's easier for me to imagine a situation where I know I want to do this.
00:34:24.768 --> 00:34:27.588
This is not the first time I've built something like this, but now I know how to build it better.
00:34:28.009 --> 00:34:34.248
And now that I know I build it better, I can just, not one shot, but I know how to describe what I want this to do in terms of success, outcomes.
00:34:34.278 --> 00:34:35.418
I know what I'm willing to give up.
00:34:35.568 --> 00:34:39.289
Obviously I don't care about a fancy user interface for this app, right?
00:34:39.289 --> 00:34:41.208
Like this is the entire user interface for this app.
00:34:41.239 --> 00:34:45.108
Like I send some receipt, it comes back as a record.
00:34:45.528 --> 00:34:53.208
I amend the record in Telegram and then I send it back, and if I approve it in that send back, it will generate a report in Zoho.
00:34:53.869 --> 00:35:05.958
This is the user interface that I want, and I cannot achieve it actually in this way unless IMCP the hell out of record code and I give it access to Zoho, which I am willing to do, but I don't need to for this.
00:35:06.289 --> 00:35:06.528
Right?
00:35:06.528 --> 00:35:08.869
So this, I, it's not even about discipline.
00:35:08.869 --> 00:35:12.289
I think it's just like, okay, I prefer doing this because I know I'll use this a lot.
00:35:12.528 --> 00:35:28.278
And now that I've done this and it works for me, I can extend this to any other employee that I have in the organization because this system is wired into my, uh, Zoho expenses account for the organization, and I don't need to build them an application.
00:35:28.338 --> 00:35:29.059
Everyone's telegram.
00:35:29.914 --> 00:35:35.958
So, so that there's a extensibility, uh, reasoning behind why, why I chose to do it this way.
00:35:36.048 --> 00:35:36.199
Yeah.
00:35:36.438 --> 00:35:37.728
So the discipline, I don't know.
00:35:37.849 --> 00:35:38.748
It makes it sound too good.
00:35:38.989 --> 00:35:41.929
This is just like, I'm very, I'm very.
00:35:42.588 --> 00:35:44.179
Uh, stubborn about particular things.
00:35:44.239 --> 00:35:44.329
Yeah.
00:35:44.329 --> 00:35:44.478
And
00:35:44.659 --> 00:35:47.869
if I'm gonna use it a lot and I know how to do it, I'm not gonna do the soft version first.
00:35:48.048 --> 00:35:48.199
Yeah.
00:35:48.228 --> 00:35:50.929
Because I've already done, I did the soft version two years ago and I kept iterating on that.
00:35:51.108 --> 00:35:51.289
Right.
00:35:51.289 --> 00:35:54.018
But if I was doing this for the first time, a hundred percent, I would first do it as a skill.
00:35:54.139 --> 00:35:54.318
Yeah.
00:35:54.378 --> 00:35:55.878
I would understand how to do the extractions.
00:35:55.998 --> 00:36:16.969
Like even sort of the logic of extracting from the receipt and then comparing with calendar and having, so right now the logic behind this is extract from the corporate calendar and identify any calendar event that's within minus three hours of the receipt time and date, and use that as the top inference for what this receipt justification should be.
00:36:17.119 --> 00:36:17.208
Mm-hmm.
00:36:17.449 --> 00:36:26.838
That logic came over about two years of, you know, the LLM extraction just randomly makes something up, but it doesn't have any anchor.
00:36:26.838 --> 00:36:27.108
Right.
00:36:27.108 --> 00:36:27.168
Yeah.
00:36:27.378 --> 00:36:32.088
So now that I know how to do all that stuff, because I've spent two years figuring it out, I don't have to do the soft skill anymore.
00:36:32.088 --> 00:36:33.079
I'm just gonna go straight and, yeah.
00:36:33.079 --> 00:36:33.528
Makes sense.
00:36:33.588 --> 00:36:33.739
Yeah.
00:36:34.039 --> 00:36:35.869
But this is also very, very soft control app.
00:36:36.018 --> 00:36:36.259
Yeah.
00:36:36.259 --> 00:36:38.719
I built this in four and a half hours.
00:36:38.898 --> 00:36:39.259
Yeah.
00:36:39.349 --> 00:36:39.559
Yeah.
00:36:40.534 --> 00:36:40.744
Hmm.
00:36:53.840 --> 00:36:58.880
I mean, in the past, this was a whole class of applications that you had to download.
00:36:58.880 --> 00:36:59.181
Right?
00:36:59.331 --> 00:37:00.501
Hundred companies were running on this.
00:37:00.561 --> 00:37:00.681
Yeah.
00:37:01.041 --> 00:37:03.800
To submit expenses for large groups of people.
00:37:03.800 --> 00:37:03.951
Yeah,
00:37:03.951 --> 00:37:04.280
for sure.
00:37:04.490 --> 00:37:04.731
Yeah.
00:37:04.911 --> 00:37:06.201
But now you don't need to.
00:37:06.291 --> 00:37:06.590
Correct.
00:37:06.590 --> 00:37:08.751
Have someone who's a mobile app developer.
00:37:08.751 --> 00:37:09.501
Developer, correct.
00:37:09.501 --> 00:37:11.210
After you discuss, that's expensive.
00:37:11.391 --> 00:37:12.170
This one.
00:37:12.530 --> 00:37:14.735
This is actually incredibly soft and light.
00:37:15.380 --> 00:37:15.681
Hmm.
00:37:15.860 --> 00:37:18.351
It took so little time that if tomorrow we decide.
00:37:18.351 --> 00:37:19.090
Not useful, throw.
00:37:19.221 --> 00:37:19.545
Yeah, that's true.
00:37:19.545 --> 00:37:19.706
Yeah,
00:37:20.001 --> 00:37:20.300
exactly.
00:37:20.400 --> 00:37:22.860
Okay, this weekend I know what I'm gonna do with my favorable credits.
00:37:23.041 --> 00:37:23.221
Yes.
00:37:24.780 --> 00:37:25.231
This one.
00:37:25.231 --> 00:37:25.240
Okay.
00:37:25.240 --> 00:37:25.831
This one don't even need fable.
00:37:25.831 --> 00:37:26.190
So
00:37:26.190 --> 00:37:26.371
No, no.
00:37:26.371 --> 00:37:27.215
To build, to build
00:37:27.420 --> 00:37:27.931
the collections.
00:37:27.931 --> 00:37:29.655
No, but you don't even need favorable to build this.
00:37:29.820 --> 00:37:37.471
This one was built with, so one of the other things which I hope we get to talk about, which is what agency enables is when you decompose a task, enough release, stupid models can do it.
00:37:37.471 --> 00:37:48.751
So this was actually built with Haiku, but the PRD was converted into a wave plan for implementation using, not even Fable, actually, it was using Opus.
00:37:49.050 --> 00:37:49.710
Hmm, okay.
00:37:49.831 --> 00:37:50.070
Yeah.
00:37:50.130 --> 00:37:53.641
But if you decompose enough, which you is partly about the.
00:37:54.541 --> 00:37:59.340
Quality of the model, doing the decomposition, but also partly about how you describe what it is that you want the model to build.
00:37:59.431 --> 00:37:59.670
Hmm.
00:37:59.880 --> 00:38:10.081
When you decompose enough, I think you can use very, very, very non frontier models to do almost exactly as good a thing as if you try one shot a very high quality.
00:38:10.081 --> 00:38:13.141
So actually what you see of this receipt that I just sent.
00:38:13.431 --> 00:38:15.141
this is actually the extraction of that.
00:38:15.380 --> 00:38:17.061
Oh, so this is from Playmore Connect.
00:38:17.300 --> 00:38:17.481
Okay.
00:38:17.481 --> 00:38:20.181
It comes back with date merchant amount.
00:38:20.510 --> 00:38:22.190
So it says 5 28 USD.
00:38:22.190 --> 00:38:23.240
It got it wrong, right?
00:38:23.240 --> 00:38:23.300
Yeah.
00:38:23.331 --> 00:38:29.530
So when I correct it, I will just say, so the way I correct it is I might reply Let's see.
00:38:29.530 --> 00:38:29.920
That works.
00:38:29.981 --> 00:38:30.221
Okay.
00:38:30.221 --> 00:38:32.710
You, I realize you forced yourself to be in the loop.
00:38:33.161 --> 00:38:33.400
Yeah, for sure.
00:38:33.400 --> 00:38:33.460
Is
00:38:33.460 --> 00:38:33.851
that, uh,
00:38:34.061 --> 00:38:34.826
what do you usually a hundred percent
00:38:34.885 --> 00:38:36.280
How you usually design your tools?
00:38:36.820 --> 00:38:41.891
You, as in, I force myself to be in the loop for the things that I think require judgment.
00:38:41.920 --> 00:38:42.010
Yeah.
00:38:42.010 --> 00:38:42.190
Right.
00:38:42.190 --> 00:38:45.820
So whatever can be made deterministic, I will make deterministic.
00:38:45.820 --> 00:38:45.971
Yeah.
00:38:46.271 --> 00:38:47.351
And that one shouldn't require me.
00:38:47.351 --> 00:38:47.530
Right.
00:38:47.590 --> 00:38:52.900
If I understand how to do the thing well enough that I can turn it into a rule that never changes, that's something a machine should do.
00:38:53.771 --> 00:39:06.041
If there's something that requires judgment from me that I don't know how to do until I see it and I make a decision, or it's a taste thing or something, that's something which should be given to only a human, whether it's me or somebody else.
00:39:06.070 --> 00:39:06.161
Mm-hmm.
00:39:06.550 --> 00:39:11.590
And then there's going to be some things where I think it requires judgment, but it's not very important.
00:39:12.581 --> 00:39:19.221
And within that very limited space, for example, like extracting stuff from a receipt, if I'm gonna review it, that one I, I give to a stupid llm.
00:39:20.110 --> 00:39:23.590
So keeping yourself in the loop is not always in the loop for everything.
00:39:23.590 --> 00:39:25.931
It's like deciding what stuff should be in the loop for you.
00:39:25.960 --> 00:39:26.681
Mm-hmm.
00:39:26.920 --> 00:39:34.326
And making sure that when you design the system that allocates work between a human and a machine, the machine gets the work that the machine should be getting, which is.
00:39:35.036 --> 00:39:37.525
Highly deterministic, fully specifiable in advance.
00:39:37.585 --> 00:39:39.385
That's what machines should be doing.
00:39:39.445 --> 00:39:39.536
Mm-hmm.
00:39:40.436 --> 00:39:45.775
And then the human gets the stuff that human must do, which is stuff that is not deterministic, requires judgment, requires taste.
00:39:45.835 --> 00:39:46.016
Mm.
00:39:46.135 --> 00:39:47.606
Or in some cases requires morality.
00:39:47.860 --> 00:39:49.675
Mm. Oh, okay.
00:39:49.675 --> 00:39:49.916
Yeah.
00:39:50.005 --> 00:39:51.085
Is that, is that all for this?
00:39:51.175 --> 00:39:51.505
Yeah, for sure.
00:39:51.505 --> 00:39:51.925
Thank you.
00:39:51.925 --> 00:39:52.706
Thank you for sharing.
00:39:53.077 --> 00:39:55.717
Segue to, uh, the next section on tangent.
00:39:56.286 --> 00:39:56.737
Alright, sure.
00:39:56.827 --> 00:39:57.097
Yeah.
00:39:57.097 --> 00:39:58.206
Since you're talking about business.
00:39:58.536 --> 00:39:58.806
Yeah.
00:39:58.996 --> 00:40:03.556
because, because I'm, I'm gonna join, um, my family, SME, who is
00:40:03.766 --> 00:40:04.036
Oh,
00:40:04.036 --> 00:40:05.027
a very traditional,
00:40:05.086 --> 00:40:05.356
you should
00:40:05.356 --> 00:40:05.717
definitely be.
00:40:05.717 --> 00:40:08.327
Um, and I'm gonna be leading tech and, and bd, so you should
00:40:08.327 --> 00:40:09.827
definitely come hang out with tangent
00:40:09.827 --> 00:40:09.916
and
00:40:10.336 --> 00:40:11.206
we can talk more about tangent.
00:40:11.211 --> 00:40:11.402
Yeah.
00:40:11.541 --> 00:40:12.081
You want, because
00:40:12.481 --> 00:40:12.601
I
00:40:12.601 --> 00:40:13.581
feel like I was wanna hang out with you guys.
00:40:13.632 --> 00:40:13.922
Yeah.
00:40:13.922 --> 00:40:14.041
Yeah.
00:40:14.512 --> 00:40:14.802
Okay.
00:40:15.117 --> 00:40:15.476
You launched
00:40:15.476 --> 00:40:16.706
almost a one month anniversary now.
00:40:16.827 --> 00:40:17.306
Wow.
00:40:17.306 --> 00:40:17.666
Yeah.
00:40:17.666 --> 00:40:19.586
The final day will be on, on August 7th.
00:40:19.632 --> 00:40:19.661
Okay.
00:40:19.931 --> 00:40:20.322
Which is
00:40:20.322 --> 00:40:20.621
on Friday.
00:40:20.742 --> 00:40:21.101
Oh, on Friday.
00:40:21.132 --> 00:40:21.431
Wow.
00:40:21.431 --> 00:40:21.702
Okay.
00:40:21.702 --> 00:40:25.751
So yeah, A tangent is an incubator that backs small ventures.
00:40:25.811 --> 00:40:26.141
Yes.
00:40:26.172 --> 00:40:28.452
Um, please, please explain it in your own words.
00:40:28.541 --> 00:40:30.552
Uh, what, what are you doing with, uh, tangent?
00:40:30.641 --> 00:40:30.731
Yeah.
00:40:30.731 --> 00:40:30.851
Okay.
00:40:30.851 --> 00:40:34.711
So it's an incubator that has multiple interconnected theses.
00:40:35.021 --> 00:40:35.141
Yeah.
00:40:35.141 --> 00:40:44.742
But the simplest way to put it, which is not totally accurate, but the simplest way to put it is we are, to our knowledge, the world's only zero to one private equity incubator.
00:40:45.461 --> 00:40:45.612
Right?
00:40:45.612 --> 00:40:53.592
So we're trying to bring people from idea to, it is making enough money to pay for itself zero to one, but not for venture capital.
00:40:54.731 --> 00:41:00.072
And that that will link into some of the other aspects of the thesis.
00:41:00.371 --> 00:41:09.371
So one aspect of the thesis is if you are trying to build a venture capital business, it must be a business that has one or more of two things, right?
00:41:09.402 --> 00:41:11.862
Must have a very large, total addressable market.
00:41:11.952 --> 00:41:12.972
Must be enormous.
00:41:13.302 --> 00:41:14.291
Otherwise, it's just not worth it.
00:41:14.952 --> 00:41:18.371
And this partly comes because of the economics of venture capital itself.
00:41:18.431 --> 00:41:21.222
Yeah, you put in some money for some equity.
00:41:21.882 --> 00:41:27.612
The money you put in is usually quite a lot of money, and you have to spend a lot of money to get to the point where you get any money back.
00:41:27.762 --> 00:41:27.972
Right?
00:41:27.972 --> 00:41:32.382
Because the unit economics in venture capital are usually not great until they become amazing.
00:41:32.652 --> 00:41:32.771
Hmm.
00:41:33.461 --> 00:41:40.661
What this means is the way you make money in venture capitals, you put money in for equity and you must make the equity increase in value by a massive multiple.
00:41:41.262 --> 00:41:43.302
And to do that, there's only two possibilities.
00:41:43.302 --> 00:42:10.422
One is you find an enormous market that nobody has found before you create it, or you discover some brand new technology that nobody has discovered before, and you use it to generate that massive increase in enterprise value that allows you to come out as a VC with your investment having made back all of itself plus more because brand new markets are very hard to build, and brand new technology is very, very uncertain to build.
00:42:10.811 --> 00:42:16.452
Most venture capital funded businesses are forced to try and do things that are destined to fail.
00:42:17.007 --> 00:42:18.146
And the statistics actually bear out.
00:42:18.297 --> 00:42:18.387
Mm-hmm.
00:42:18.387 --> 00:42:26.186
The vast majority of venture capital funded businesses don't work because the economics of venture capital force them to do things that are inherently unlikely to work.
00:42:26.757 --> 00:42:32.726
So we don't say, because I don't think it's true that venture capital is bad, we just say it's not for us.
00:42:33.027 --> 00:42:33.117
Mm-hmm.
00:42:33.146 --> 00:42:33.387
Right.
00:42:33.387 --> 00:42:43.827
So for us, we would much rather incubate companies that aren't trying to be enormous TAM, total adjustable market, aren't trying to build deep tech.
00:42:44.097 --> 00:42:52.646
Instead, they're trying to build things that use existing technology to solve well understood problems, ideally in small, well understandable markets.
00:42:53.067 --> 00:42:55.226
And this happens to be boring, tiny tools, right?
00:42:55.231 --> 00:42:55.512
Mm-hmm.
00:42:55.597 --> 00:43:07.556
And also happens to be the, the well understood technology is perfect for our current moment where writing software for well understood problems for boring tiny tools is actually super easy to do.
00:43:07.646 --> 00:43:07.856
Mm-hmm.
00:43:07.916 --> 00:43:08.157
Right?
00:43:08.157 --> 00:43:09.777
And it's also not bad anymore.
00:43:09.836 --> 00:43:10.467
It used to be bad.
00:43:10.467 --> 00:43:14.936
And then December last year became, okay, so part of the thesis behind tension is.
00:43:15.027 --> 00:43:15.056
Okay.
00:43:15.867 --> 00:43:22.586
If VC funded entrepreneurship is so likely to fail, why only focus on doing that?
00:43:22.586 --> 00:43:25.197
If you're a tech entrepreneur, why not also do this other kind?
00:43:25.797 --> 00:43:34.976
And especially when, right now, you know, we're talking about tools and being aware of their affordances when tools now's not like this has been the case for the last four or five years.
00:43:34.976 --> 00:43:41.666
It's that at this very moment, we are eight months into a moment when all of a sudden this thesis could actually make sense.
00:43:41.697 --> 00:43:41.786
Mm-hmm.
00:43:42.027 --> 00:43:44.936
So that's one other part of the, of the thesis.
00:43:45.717 --> 00:43:50.367
The third part of the thesis, I think, is, um, AI is here, right?
00:43:50.367 --> 00:43:51.476
Like, we can't deny it's around.
00:43:51.476 --> 00:43:58.737
I, I think there's many aspects of what people say AI is gonna be that are very overblown, but there's some aspects that are totally not overblown.
00:43:58.947 --> 00:44:00.717
Like we just cannot deny that it's here.
00:44:00.717 --> 00:44:02.697
So that's one part that's not, that's not overblown.
00:44:03.206 --> 00:44:10.947
And I think there's probably at least four, but certainly there are these four ways of thinking about how AI affects people doing business, right?
00:44:10.947 --> 00:44:11.487
One way.
00:44:12.146 --> 00:44:13.467
You build the AI itself.
00:44:13.887 --> 00:44:20.876
This one is like super CapEx intensive, like foundation labs, chip foundries, build data centers.
00:44:21.117 --> 00:44:24.027
For me, and certainly for tangent, we are not interested in this.
00:44:24.027 --> 00:44:25.436
It's way too capital intensive.
00:44:25.827 --> 00:44:30.027
And also if you think about it like this is a thesis about Southeast Asia and Singapore.
00:44:30.686 --> 00:44:35.427
There is no geography in Southeast Asia that is venture capital scale, right?
00:44:35.427 --> 00:44:35.516
Mm-hmm.
00:44:35.757 --> 00:44:45.266
No market is big enough to sustain a VC level investment here, which is why I think it's very hard to run VC entrepreneurship in Singapore or Southeast Asia.
00:44:46.376 --> 00:44:54.927
So, uh, if you are in this situation, you can't do the first type of AI business, which is to build the AI itself.
00:44:55.286 --> 00:44:58.646
There are still three other kinds of AI business that I think actually are quite attractive.
00:44:58.706 --> 00:45:02.456
One of them is to use AI in your product, right?
00:45:02.456 --> 00:45:03.806
So like, um.
00:45:04.452 --> 00:45:12.521
At one point, people were making these tools for people who do advertising copy, where you put in some information by your product and then it writes copy for using an LLM.
00:45:12.911 --> 00:45:22.601
To me, this is now like I think that market for products using LLMs in that way as part of the product itself that runs at runtime, that has some way to go still.
00:45:22.601 --> 00:45:27.072
Because you can still imagine, you know, like you programmatically generate video that does some cool stuff.
00:45:27.072 --> 00:45:29.382
Not for advertising, but for some other thing like for training.
00:45:29.952 --> 00:45:35.442
Uh, I think that second category build products that use AI at Runtime has a ways to go.
00:45:35.802 --> 00:45:36.791
That's one other category.
00:45:37.391 --> 00:45:39.942
A third category is one that's much more interesting to me.
00:45:40.001 --> 00:45:41.711
Use AI to build deterministic software.
00:45:41.771 --> 00:45:43.152
These are boring, tiny tools, right?
00:45:43.362 --> 00:45:46.572
So I know this workflow, it's super deterministic.
00:45:46.811 --> 00:45:48.222
My staff hate doing it.
00:45:48.222 --> 00:45:54.461
They're super bored and also they're kind of lazy, so they're not very good at it, but they would be much happier if you didn't make them do that.
00:45:54.461 --> 00:45:59.472
And they get to do other fun stuff that is prime territory to turn into a boring, tiny tool.
00:45:59.802 --> 00:46:00.672
And now we've got.
00:46:01.226 --> 00:46:08.186
AI coding tools that help you do that, but only if it's truly a boring, tiny tool that you understand really well, fully deterministic, completely specifiable.
00:46:08.786 --> 00:46:10.106
So that's category three, right?
00:46:10.347 --> 00:46:13.467
So category one was build the AI itself, I think, no, let's not do that.
00:46:13.856 --> 00:46:16.706
Category two is build products that use the AI at run time.
00:46:16.737 --> 00:46:17.666
Maybe we can do that.
00:46:18.117 --> 00:46:22.556
Category three is built deterministic software that doesn't use AI at runtime, but use AI to build it.
00:46:22.617 --> 00:46:24.117
We should a hundred percent do this.
00:46:24.356 --> 00:46:25.556
And the fourth one I think is super fun.
00:46:25.947 --> 00:46:28.797
The fourth one is now the AI is everywhere super ubiquitous.
00:46:29.097 --> 00:46:33.476
What are the products and services that are newly or much more important?
00:46:33.476 --> 00:46:39.447
Because now there's ai and I think the creative industries, if we know how to build the businesses around them, they will come back.
00:46:39.836 --> 00:46:50.726
So the way to build something like a, for example, fashion label that really addresses the question of, I want to express personal taste at a time when everything appears to be AI slot.
00:46:51.027 --> 00:46:56.487
You can build a fashion label now with just one person if you run your business using boring, tiny tools, right?
00:46:56.487 --> 00:47:00.056
So I think one of the deep underlying thesis for tension is.
00:47:00.461 --> 00:47:11.501
Make small companies that are much more likely to succeed and fail by not trying to build massive companies that are likely to fail because this is just the logic of VC versus pe.
00:47:13.061 --> 00:47:19.601
Understand all the different ways AI can affect the business model and learn how to use them both for the product you're building and for yourself.
00:47:20.112 --> 00:47:25.181
And also when you're doing that, you happen to be building things that there's enormous market for, right?
00:47:25.181 --> 00:47:32.262
Like you talk about SMEs, the vast majority of all companies in any country, anywhere in the world are not big, right?
00:47:32.262 --> 00:47:42.036
So in Singapore, I think it's like it's either 90% or more than 90% are SMEs And mostly because for us, our definition of SME is actually, the mediums size is still quite small.
00:47:42.041 --> 00:47:42.186
Quite, quite.
00:47:42.186 --> 00:47:42.246
It
00:47:42.862 --> 00:47:43.327
is quite small.
00:47:43.387 --> 00:47:46.266
There's some countries where the medium size is actually much larger.
00:47:47.436 --> 00:47:49.637
But when you have this enormous market of.
00:47:50.302 --> 00:47:51.202
SMEs, right?
00:47:51.231 --> 00:47:52.882
These are not big companies are SMEs.
00:47:53.242 --> 00:47:55.222
The vast majority of them are not tech SMEs.
00:47:55.222 --> 00:47:56.902
These are all industries that pre-existed tech.
00:47:56.902 --> 00:47:59.152
They do construction, they do retail, they do food.
00:47:59.601 --> 00:48:02.061
All of them are trying to do digitalization things, right?
00:48:02.061 --> 00:48:04.251
They're trying to transform their business so that they can work better.
00:48:04.311 --> 00:48:09.862
In a world where digital transformation is everywhere, SMEs have found it really difficult to do it.
00:48:11.001 --> 00:48:11.391
Boring.
00:48:11.391 --> 00:48:13.012
Tiny tools are the answer, right?
00:48:13.282 --> 00:48:21.411
If there are processes that are well understood enough that you can really say what they do and they never change, you should not be having people do them.
00:48:21.442 --> 00:48:24.112
'cause people are bad at doing them and people just, they get bored, right?
00:48:24.262 --> 00:48:26.152
Why not let them do better things?
00:48:26.541 --> 00:48:36.501
They'll have more fun, and you'll have a better organization if you get someone who is good at building boring, tiny tools to come in and help you figure out how to build boring, tiny tools for your business.
00:48:36.771 --> 00:48:39.831
Mm. So in a way, you would say that it's a kind of bespoke, right?
00:48:40.282 --> 00:48:40.342
Yeah.
00:48:40.342 --> 00:48:42.291
Plus every, every organization is different.
00:48:42.382 --> 00:48:42.711
Yeah.
00:48:42.771 --> 00:48:44.391
Their processes will be different.
00:48:44.391 --> 00:48:44.402
Yes.
00:48:44.481 --> 00:48:44.782
Yeah.
00:48:44.931 --> 00:48:47.961
So I would say it is bespoke, but at the same time.
00:48:49.146 --> 00:48:55.137
It's also potentially quite interesting to think about a situation where you are in a particular vertical.
00:48:55.197 --> 00:49:04.117
Let's say you are in shipping and many people in shipping, if you're doing the last mile stuff, you have the problem of someone comes in with a way bill.
00:49:05.197 --> 00:49:14.797
Most of the time if you are shipping a lot of stuff, the way bill is machine printed, so you can ocr r very easily, uh, or you can just automatically, it's automatically transmitted.
00:49:15.186 --> 00:49:21.456
But sometimes you have way bills that are written in, you know, and then your OC not very good right now.
00:49:21.817 --> 00:49:24.217
OCR obviously is way better than even two years ago.
00:49:24.786 --> 00:49:35.766
And if you are at the point of receipt, if you need to type the way bill into your system, which many, uh, last mile logistics people still do that is not good, but.
00:49:36.097 --> 00:49:48.157
Almost every person who receives a way bill at a last mile logistic organization will probably want to use something like use cell phone that every employee has, take photo, automatically extract structured data put into database, right?
00:49:48.161 --> 00:49:48.351
Mm-hmm.
00:49:48.516 --> 00:49:57.336
So it is very custom in a sense, but I think you will also start to find, when we start to investigate boring, tiny tools at the vertical level, we'll find that there will be consistency within verticals.
00:49:57.726 --> 00:50:07.442
And if you're building, you know, earlier we were talking about build the spine and then the ingestion, is like a state machine or it's like a database that is like structural in a particular way.
00:50:07.711 --> 00:50:13.081
And then what you customize is you customize the ingestion and you customize maybe the, the output, right?
00:50:13.322 --> 00:50:21.391
So really thinking about how do you build boring, tiny tools in a more scalable way, not by making the tool itself enormous in market.
00:50:21.706 --> 00:50:30.677
But by building the tool architecturally so that you can customize it very easily for people who are actually trying to do the same thing, but may to see it in different ways.
00:50:30.777 --> 00:50:44.012
so, so I, I think the, the maybe interesting thing about this is if you go down to the ground, like people actually talking to SMEs or working in SMEs, you find that this problem, everyone knows we have it.
00:50:44.431 --> 00:50:47.757
And yet, if you think about everyone who thinks about tech entrepre.
00:50:48.376 --> 00:50:54.016
All they want is they want to go to San Francisco, be invested in by VC, and 95% of the time fail.
00:50:54.706 --> 00:50:54.976
Right.
00:50:55.277 --> 00:50:57.856
This makes no sense to me.
00:50:58.186 --> 00:51:08.597
I don't know why we continue to, in, like from a policy level, we make it super easy for venture capitalists to come to Singapore because the government will likely be an LP in your company, right?
00:51:09.077 --> 00:51:20.956
We make it super easy to become a person who wants to build a unicorn, which means we are generally pushing people to do things that are likely to fail in a country which is not venture capital scale.
00:51:21.706 --> 00:51:35.387
Why are we not encouraging more people to do the unsexy, but actually potentially quite enjoyable and certainly much more likely to be lucrative task of building boring, tiny tools for the enormous number of SMEs that exist in this country.
00:51:35.686 --> 00:51:36.376
Makes no sense to me.
00:51:36.376 --> 00:51:39.467
So instead of big, risky bets, it's like small and stable bets.
00:51:40.907 --> 00:51:45.047
No, actually, I totally agree you, because I, I spend the last 10 over years of my life, right In.
00:51:45.452 --> 00:51:49.052
Exactly that condition we're talking about, uh, startup life.
00:51:49.141 --> 00:51:49.922
VC life.
00:51:50.132 --> 00:51:50.282
Sure.
00:51:50.492 --> 00:51:55.112
And here I am, you know, trying to help SMEs to build boring tools.
00:51:55.666 --> 00:51:56.067
I, you know,
00:51:56.126 --> 00:51:58.742
so, so, you know, I, I totally see where, where you're going.
00:51:58.742 --> 00:52:00.121
And we are quite aligned on that.
00:52:00.751 --> 00:52:00.931
Cool.
00:52:01.112 --> 00:52:01.786
Um, I, I,
00:52:01.791 --> 00:52:03.981
I think for one more tangent for me on tangent, right.
00:52:03.981 --> 00:52:07.132
It's more about the economics portion of it, right.
00:52:07.132 --> 00:52:08.842
How does it work for listeners, right?
00:52:09.231 --> 00:52:10.427
I would, I would love to talk about that.
00:52:10.456 --> 00:52:10.606
Okay.
00:52:10.606 --> 00:52:16.126
So a tangent is a four week incubator that is completely philanthropic, right?
00:52:16.547 --> 00:52:24.407
And one of the reasons we wanna do that is a lot of the incubator programs and entrepreneurship programs in Singapore, especially once you're an adult.
00:52:25.112 --> 00:52:28.202
They are only accessible to people who are very conventional, right?
00:52:28.231 --> 00:52:33.961
Get good grades, go to good school, and generally are trying to start companies that are venture capital style companies.
00:52:34.322 --> 00:52:39.242
So basically it's an extension of the problem that we are focusing on the wrong thing.
00:52:39.242 --> 00:52:53.731
When we're trying to get people to be entrepreneurs, especially tech entrepreneurs, we're trying to push people who are very conventional in how they think about what success looks like in the direction of doing things that are very, very likely to fail and actually will probably not work in Singapore even if they did succeed.
00:52:53.731 --> 00:52:56.641
So naturally we're pushing people to go overseas, right?
00:52:56.641 --> 00:53:05.791
Because if you want to build something that is successfully venture capital scale, the moment you hit that scale, you realize Southeast Asia is probably not the right place for you.
00:53:05.942 --> 00:53:06.692
Where must you go?
00:53:06.692 --> 00:53:07.447
You must go to either China.
00:53:08.327 --> 00:53:09.257
Or the US at the moment.
00:53:09.467 --> 00:53:14.666
With Tangent, if you come in, you spend four weeks and, uh, later you can go talk to some of the people.
00:53:14.672 --> 00:53:14.882
Mm-hmm.
00:53:15.262 --> 00:53:21.956
Uh, it turns out that people actually work super hard, but not because we give them a very rigorous schedule where every day is programmed.
00:53:21.956 --> 00:53:23.817
In fact, most days we don't have a program.
00:53:24.657 --> 00:53:32.246
Uh, there's a program on the first day, and then after that, the cohorts participants will figure out what they want.
00:53:32.847 --> 00:53:36.806
Like, for example, they want a session on how to do cold calling of SMEs, right?
00:53:36.806 --> 00:53:40.706
So that they can meet people who have a problem and find out more about a problem.
00:53:41.007 --> 00:53:43.347
One person will be like, I want a cold calling.
00:53:43.737 --> 00:53:46.286
Like session, they'll go ask other people in the cohort.
00:53:46.286 --> 00:53:54.567
We have 17 people in this cohort, and if enough people want a cold calling session, we will schedule one and then whoever can deliver or deliver, so often I will deliver.
00:53:55.106 --> 00:54:01.677
But if they want something that I can't deliver, like some people are asking for legal clinic, for startups, when should I incorporate what the structure will be?
00:54:01.916 --> 00:54:02.996
I'm not qualified to do that.
00:54:02.996 --> 00:54:04.436
Then we bring in someone who's like a lawyer.
00:54:04.586 --> 00:54:04.856
Mm.
00:54:04.976 --> 00:54:09.206
So the programming is not compulsory, highly structured in advance.
00:54:09.416 --> 00:54:15.657
It is structured, but emergent and responsive, and it is not mandatory.
00:54:16.317 --> 00:54:25.467
But it turns out many people in the office anyway because they are motivated to do it, which I think is very cool that like I was worried that people would just come and they're not show up, but it's nice that they're showing up.
00:54:25.887 --> 00:54:34.286
So the way the program works is you work really hard in these four weeks to figure out, am I even interested in asking the question of an SME?
00:54:34.527 --> 00:54:37.646
What is the boring tiny problem that you want me to solve with a boring tiny.
00:54:37.646 --> 00:54:41.097
To, there are going to be some people for whom they realize, actually this is not for me.
00:54:41.097 --> 00:54:43.947
I want to, I want to do a big idea, and that's okay also.
00:54:44.067 --> 00:54:50.246
But then you have to understand, like to do a big idea, you must go overseas, probably going to fail, and that's fine.
00:54:50.306 --> 00:54:54.447
If that's the set of trade-offs that you're willing to make strategically, then yes, you should do that.
00:54:54.536 --> 00:55:02.817
The problem will be, if you don't realize vc, very cool, very sexy, but 95 to 98% probability of failure.
00:55:03.416 --> 00:55:06.237
If you don't realize that and you decide you wanna move VC anyway, that's bad.
00:55:06.536 --> 00:55:13.556
Once you realize that the probability of failure is high and you have to do all these other things that you don't want, and you still decide you wanna do it, that's okay.
00:55:13.947 --> 00:55:14.186
Right?
00:55:14.186 --> 00:55:21.746
So one outcome of our four eight program, this is the reason why we don't take equity, is like it's meant to be a very wide funnel where it's okay.
00:55:22.166 --> 00:55:24.086
To have both false positives and false negatives.
00:55:24.237 --> 00:55:24.327
Mm-hmm.
00:55:24.567 --> 00:55:24.867
Right?
00:55:25.047 --> 00:55:29.637
So that what falls through the funnel at the end is where we look for the next step and the next step.
00:55:29.637 --> 00:55:37.016
We also don't know how it's gonna work, because what we're trying to do with tangent is we're trying to get people as quickly as possible to actually meet people who want to buy the thing that they want to build.
00:55:37.047 --> 00:55:37.257
Yep.
00:55:37.496 --> 00:55:38.577
So they have to get the cash flow.
00:55:38.577 --> 00:55:43.047
ASAP, if they can get the cash flow within four weeks is great as soon as possible.
00:55:43.047 --> 00:55:46.226
After that, we want 'em to get to accounting profitability, right?
00:55:46.286 --> 00:55:52.047
Where you can magic the numbers a little bit and the money that's coming in is leading you to a paper profit.
00:55:52.706 --> 00:55:54.086
That is a good benchmark.
00:55:54.146 --> 00:55:59.907
But actually where we really want to go is we want to get people as soon as possible to free cash flow to become default alive.
00:56:00.327 --> 00:56:03.927
And if we can get that in 12 months, we will be better than almost any grader anywhere.
00:56:04.077 --> 00:56:04.166
Mm-hmm.
00:56:04.556 --> 00:56:10.856
But some of these people are already, like, they're out there aggressively selling stuff so far, they're not succeeding at selling stuff.
00:56:10.856 --> 00:56:13.016
But first you have to try, then you, then you get it, right?
00:56:13.021 --> 00:56:13.052
Yeah.
00:56:13.351 --> 00:56:17.577
And so the longer term goal for tension, the economics come in where.
00:56:18.101 --> 00:56:23.831
After this four week program, we see who the people are who are going through it and we know who are the people that we would put bets on.
00:56:24.192 --> 00:56:28.121
Not because the product is already there, but because they are the people who want to build something like this.
00:56:28.121 --> 00:56:28.271
Right.
00:56:28.271 --> 00:56:30.731
You have to want to build something small, something boring.
00:56:31.032 --> 00:56:32.501
You have to want to work with SMEs.
00:56:32.507 --> 00:56:34.001
'cause it's working with SMEs is very different.
00:56:34.152 --> 00:56:34.422
Right.
00:56:34.512 --> 00:56:35.802
You know it 'cause you're workings.
00:56:36.731 --> 00:56:41.561
So if we know that they want to do this and they are, everyone says grit, but it's actually different.
00:56:41.561 --> 00:56:44.831
It's like you just have to be willing to tog this particular phase.
00:56:44.831 --> 00:56:45.012
Yeah.
00:56:45.402 --> 00:56:50.771
And most people who want to do VC type businesses, they are willing to tong a different kind of problem.
00:56:50.771 --> 00:56:51.552
This one is a different
00:56:51.552 --> 00:56:51.702
thing.
00:56:51.702 --> 00:56:51.762
Yeah.
00:56:51.762 --> 00:56:52.871
It's a different, different pain.
00:56:52.871 --> 00:56:53.021
Yeah.
00:56:53.081 --> 00:56:53.621
But pain.
00:56:53.681 --> 00:56:56.081
So, so we, we want to watch them in this four week program.
00:56:56.086 --> 00:56:56.226
Mm-hmm.
00:56:56.322 --> 00:57:02.411
And then the people who come out the other end who look like they are able to do that, we want to be able to invest.
00:57:02.862 --> 00:57:07.842
But if we really do get them to free cash flow fast, the logic of normal fund economics doesn't work anymore.
00:57:07.842 --> 00:57:08.052
Right.
00:57:08.351 --> 00:57:08.411
Yeah.
00:57:08.411 --> 00:57:15.251
The logic is you as the startup are spending all this time with negative or just bad unit economics.
00:57:15.251 --> 00:57:16.572
That's why you need more capital injection.
00:57:17.141 --> 00:57:24.911
That's why you're giving away equity and the capital injection and the bad unit economics is to increase enterprise value so that the equity you gave away eventually becomes worth a lot.
00:57:25.242 --> 00:57:29.711
But if you get people to be free cashflow positive super fast, that math doesn't work anymore.
00:57:29.742 --> 00:57:31.032
Because also there's no liquidity event.
00:57:31.362 --> 00:57:31.601
Exactly.
00:57:31.601 --> 00:57:32.652
Free cashflow positive really.
00:57:32.922 --> 00:57:37.871
So we don't understand how to do this yet, but uh, we're working with some very smart people to try and figure it out.
00:57:38.231 --> 00:57:46.121
And some of the possibilities might include things like, uh, we give you, not necessarily money, but we give you strategic insight, which is what we try and do in the four weeks.
00:57:46.121 --> 00:57:49.061
But we can continue, we give you connections to people in SMEs.
00:57:49.061 --> 00:57:57.552
Like we, for every one of the participants in tangent, if they come to us and they say, we want to meet someone in, for example, semiconductor industry learning and development.
00:57:57.552 --> 00:58:00.641
'cause we're building a training tool for semiconductor engineers.
00:58:01.092 --> 00:58:04.632
Mm. We will try and find through our networks people who are willing to talk to them and we'll connect them.
00:58:05.141 --> 00:58:08.052
So I think those are the things, not the capital, right?
00:58:08.141 --> 00:58:09.192
Strategic insight.
00:58:09.581 --> 00:58:14.231
A very different way of thinking about how to build products and connections in local industry.
00:58:14.577 --> 00:58:15.836
Will be worth something.
00:58:16.197 --> 00:58:25.947
And what they're worth is probably not equity, but probably worth a stream of economic considerations like rev share for some determinate period, profit share for some determinate period.
00:58:26.277 --> 00:58:30.387
Something we don't know yet, because there's no other thing like tangent in the
00:58:30.387 --> 00:58:30.597
world.
00:58:30.621 --> 00:58:40.827
In, in a way, is a incubator by, its, uh, is it a profit for-profit, um, entity or is it a Phil philanthropy
00:58:40.856 --> 00:58:41.277
tangent.
00:58:41.427 --> 00:58:41.666
Yeah.
00:58:41.666 --> 00:58:42.416
Maybe, maybe.
00:58:42.416 --> 00:58:45.407
Um, you, you wrote about how Singapore, it's a tech mittelstand.
00:58:46.137 --> 00:58:46.467
Yeah, for sure.
00:58:46.771 --> 00:58:47.521
why is it a non-profit?
00:58:47.521 --> 00:58:49.231
Is this like, are you serving the nation or,
00:58:49.952 --> 00:58:52.922
well, okay, so we, it's a nonprofit because, okay.
00:58:53.161 --> 00:58:53.402
Yeah.
00:58:53.402 --> 00:58:55.681
The honest answer, which I'm very happy to make public, right?
00:58:55.862 --> 00:58:56.041
Yeah.
00:58:56.192 --> 00:58:58.561
Is we would love for this to make money.
00:58:59.222 --> 00:59:03.431
But the problem that we raised already, which is if we are successful at doing what we do.
00:59:04.442 --> 00:59:06.541
It will be very difficult for us to make money.
00:59:06.541 --> 00:59:08.851
Like the normal PE model won't work for this.
00:59:08.851 --> 00:59:09.032
Yeah.
00:59:09.032 --> 00:59:10.532
The rollout model doesn't work for this.
00:59:10.652 --> 00:59:12.032
The search fund model doesn't work for this.
00:59:12.032 --> 00:59:13.351
The VC model doesn't work for this.
00:59:13.411 --> 00:59:13.592
Right.
00:59:13.592 --> 00:59:15.242
The venture debt model doesn't work for this.
00:59:15.362 --> 00:59:16.411
We have to build something new.
00:59:17.101 --> 00:59:17.822
We don't know what it is.
00:59:17.822 --> 00:59:18.181
Yeah.
00:59:18.391 --> 00:59:28.472
So your question about whether it's a non-profit whatever, go Tangent Limited, which is the entity that runs the incubator that's four weeks long, is a company limited by guarantee.
00:59:29.041 --> 00:59:29.222
Right?
00:59:29.222 --> 00:59:32.972
So the main subscriber is Majurity Trust, uh, but it.
00:59:33.467 --> 00:59:37.847
Receives money from a funder and then we spend it, we are very, very cheap to run.
00:59:38.266 --> 00:59:38.536
Right?
00:59:38.896 --> 00:59:40.246
Like, we're also fast.
00:59:40.336 --> 00:59:50.086
Like if it normally takes you 12 to 18 months to start your first cohort and three to six people and we're doing it in four months with two people, we are better use of money in this way.
00:59:50.092 --> 00:59:50.422
Mm-hmm.
00:59:50.626 --> 00:59:56.806
And if anyone in government is thinking about entrepreneurship, encouragement for SMEs in Singapore, you should spend your money with us.
00:59:58.097 --> 01:00:02.297
But, um, that's, that's as far as the sort of structure of it goes.
01:00:02.626 --> 01:00:08.956
The reason why it's a not-for-profit is because I cannot figure out how to appropriate the value of this particular part of it.
01:00:09.496 --> 01:00:11.927
It's, and so if you can't prove it, don't try.
01:00:11.927 --> 01:00:12.197
Right.
01:00:12.197 --> 01:00:21.976
Like, uh, if we figure out how to do it, we will do it because I'm pretty sure we're creating value by helping people understand that there's a different way of thinking about entrepreneurship that isn't necessarily just vc.
01:00:22.367 --> 01:00:24.947
It is something which is much higher likelihood of success.
01:00:25.186 --> 01:00:25.876
It is smaller.
01:00:26.431 --> 01:00:28.231
It is much more doable and faster.
01:00:28.532 --> 01:00:35.581
And also it serves the vast majority of businesses in the country rather than only serving a very, very small crust of enormous companies.
01:00:35.972 --> 01:00:37.561
So there's gonna be value created.
01:00:37.891 --> 01:00:42.601
How we appropriate the value, take it out of where we created it, we don't understand yet.
01:00:43.052 --> 01:00:45.902
So we, we don't want to bake it into the structure of this.
01:00:45.961 --> 01:00:57.001
And at the moment it is inexpensive enough to run that the funder is willing to try to see whether or not something like this will even do what we're hoping it to do.
01:00:57.007 --> 01:00:57.257
Mm-hmm.
01:00:57.356 --> 01:01:03.061
And the first step of hope of hoping right, is the next step is make it capital recycle and make money.
01:01:03.331 --> 01:01:05.402
But before you get there first you have to make sure it works right?
01:01:05.402 --> 01:01:05.461
Yeah.
01:01:05.461 --> 01:01:08.101
Like you have to find the people.
01:01:08.251 --> 01:01:08.431
Okay.
01:01:08.431 --> 01:01:10.652
So we have a, maybe we found some people.
01:01:11.492 --> 01:01:18.541
Once you find the people, you have to make sure that the program that you run, the way you run, it helps them do something that they wouldn't otherwise have been able to do without the program.
01:01:18.661 --> 01:01:21.061
I think we are, we will obviously be learning.
01:01:21.152 --> 01:01:22.831
Like I, I won't pretend that we know how to do it yet.
01:01:23.461 --> 01:01:28.112
Some of the things that we are seeing from the first cohort make me think that that's happening.
01:01:28.742 --> 01:01:37.922
And then after that, if those two first steps already are crossed, then you have to get to the next one, which is how do you appropriate the value out so you can capital recycle.
01:01:38.822 --> 01:01:40.081
We don't know how to do the third part.
01:01:40.231 --> 01:01:44.942
We may know a bit more about the first and the second, and I won't pretend to know all three are.
01:01:45.211 --> 01:01:59.777
So we'll just keep learning and then at some point if we figure out how to appropriate value, that will become some kind of thing that is not a fund, but it may be some kind of network of people who are SMEs that put sweat equity money, who knows, into other SMEs.
01:02:00.121 --> 01:02:00.391
That could
01:02:00.391 --> 01:02:02.177
be a new form of organization, could
01:02:02.177 --> 01:02:02.976
be a new formation.
01:02:03.297 --> 01:02:03.336
Interest.
01:02:03.336 --> 01:02:04.657
I'm an organizational sociologist.
01:02:04.766 --> 01:02:05.056
Yeah.
01:02:05.521 --> 01:02:07.891
Kind of like the, you know, like the Japanese models, right?
01:02:07.891 --> 01:02:08.251
Where
01:02:08.492 --> 01:02:08.851
the resu
01:02:09.097 --> 01:02:10.652
resu each kind of way.
01:02:10.742 --> 01:02:11.851
Kind of like everybody.
01:02:11.851 --> 01:02:12.072
Early, early
01:02:12.257 --> 01:02:12.416
Resu.
01:02:12.722 --> 01:02:12.992
Early ke.
01:02:12.992 --> 01:02:13.081
But
01:02:13.262 --> 01:02:14.101
not modern resu.
01:02:14.552 --> 01:02:15.061
Oh no, no, no.
01:02:15.061 --> 01:02:16.172
I'm talking about that old.
01:02:16.411 --> 01:02:16.831
Yeah.
01:02:16.836 --> 01:02:21.507
But the other things might be also like, uh, and this model is less well known, but you know, Mondragón.
01:02:22.547 --> 01:02:25.547
A true worker and organizational cooperative in Spain.
01:02:26.117 --> 01:02:26.206
Mm-hmm.
01:02:26.836 --> 01:02:34.067
So I'm not saying that's a model, I'm just saying that there's lots and lots of different models that are not very conventional, not very popular, and maybe this will have to be a new model.
01:02:34.186 --> 01:02:34.367
So
01:02:34.456 --> 01:02:36.556
yeah, this, this sounds like we need to do coffee after this.
01:02:37.936 --> 01:02:39.561
By the way, learn this here.
01:02:39.742 --> 01:02:56.266
The big ask, if you will, for your listeners is if you know someone who has, who wants to be an entrepreneur, but has never found the kind of incubator or support that lets them do the small, boring entrepreneurship, everyone's always pushing 'em to be more ambitious.
01:02:56.266 --> 01:02:59.717
Do the bigger idea, get vc, send them to us, right?
01:02:59.777 --> 01:03:03.856
'cause we are very happy to talk to them and see whether they're a good fit for us and whether we are a good fit for them.
01:03:04.516 --> 01:03:12.976
The second thing is, if you wanna be a guide, if you work with SMEs, or if you are in an SME, or if you're just interested in helping more of these small and medium sized enterprises, go from zero to one.
01:03:14.206 --> 01:03:17.416
Then you should also talk to us because we want to find more of these people and the connect.
01:03:17.422 --> 01:03:17.722
Mm-hmm.
01:03:17.722 --> 01:03:17.961
Mm-hmm.
01:03:18.407 --> 01:03:18.737
Yeah.
01:03:18.827 --> 01:03:18.976
Yeah.
01:03:19.036 --> 01:03:19.456
Thanks for that.
01:03:19.487 --> 01:03:25.492
'cause I've been actually trying to build microsaas tools, which are similar, but maybe, I dunno, a bit, but bigger.
01:03:25.492 --> 01:03:26.061
A bit bigger.
01:03:26.061 --> 01:03:26.331
Yeah.
01:03:26.336 --> 01:03:30.742
It's like, it's like yours is really, I think, one step smaller than that, but yeah.
01:03:30.831 --> 01:03:31.132
So.
01:03:31.282 --> 01:03:31.641
Okay.
01:03:31.641 --> 01:03:32.362
That's very interesting.
01:03:32.362 --> 01:03:41.632
And also thinking how does innovation work with like, you know, so many different parts of knowledge are locked into different industries Yes.
01:03:41.632 --> 01:03:44.181
That no one is, you know, bringing value out of.
01:03:44.211 --> 01:03:45.472
So I think that's really interesting.
01:03:45.472 --> 01:03:46.702
I'm, I'm really looking forward to see.
01:03:47.092 --> 01:03:49.492
Where tangent continues to go, but maybe I
01:03:49.492 --> 01:03:49.672
am too.
01:03:49.911 --> 01:03:51.652
Uh, we'll just move into the very, very cool.
01:03:51.862 --> 01:03:53.541
The final part on, on meaning making
01:03:53.601 --> 01:03:53.782
sure,
01:03:53.871 --> 01:03:58.492
uh, you've written quite a bit about meaning making as a decision we make about the subjective value of things.
01:03:58.521 --> 01:04:05.271
I think we've kind of covered that in how you design your own software as well, when you, you don't wanna outsource certain things.
01:04:05.481 --> 01:04:05.572
Yes.
01:04:05.961 --> 01:04:08.632
Um, so maybe just one last thing to wrap it up.
01:04:08.632 --> 01:04:12.141
Like, where is the humans place as we continue interacting with AI systems?
01:04:12.141 --> 01:04:12.291
As a
01:04:12.291 --> 01:04:13.072
sociologist?
01:04:13.161 --> 01:04:13.612
Yeah.
01:04:13.612 --> 01:04:13.672
Yeah.
01:04:13.672 --> 01:04:13.882
I mean,
01:04:14.121 --> 01:04:16.222
also as just like a human, I'm a human also, right?
01:04:17.061 --> 01:04:18.262
I think that's actually maybe more useful.
01:04:18.476 --> 01:04:18.766
Okay.
01:04:18.987 --> 01:04:23.211
So, so I think, um, like maybe first just to say like what meaning making is right.
01:04:23.211 --> 01:04:32.481
So meaning making is a word that I think everyone, sort of, the words are familiar, but the specific meaning is, uh, for me is a very specific definition.
01:04:32.512 --> 01:04:35.541
It's when you decide something is.
01:04:36.262 --> 01:04:37.192
Good or bad.
01:04:37.251 --> 01:04:39.561
That's one kind of meaning making right, morally good, morally bad.
01:04:39.561 --> 01:04:41.572
Like blood diamonds, are they good or bad?
01:04:41.601 --> 01:04:43.282
Some people say they're good, some people say they're bad.
01:04:43.342 --> 01:04:43.521
Right?
01:04:43.521 --> 01:04:44.751
There's no objective truth about it.
01:04:44.751 --> 01:04:48.382
Like blood diamonds, meaning diamonds that are mined in war zones.
01:04:48.382 --> 01:04:48.472
Mm-hmm.
01:04:48.862 --> 01:04:52.731
Where they use the proceeds from selling those diamonds to buy weapons and like do more war.
01:04:53.092 --> 01:04:56.121
Some people say, no war blood diamonds are bad.
01:04:56.541 --> 01:04:58.431
I'm not gonna buy blood diamonds because they're bad.
01:04:59.452 --> 01:05:03.891
Then that's one kind of meaning making another kind of meaning making where you assign value is you say I like or not.
01:05:03.952 --> 01:05:04.192
Right.
01:05:04.342 --> 01:05:05.121
Is there utility?
01:05:05.121 --> 01:05:07.731
So again, the same blood diamond is very beautiful.
01:05:07.731 --> 01:05:11.422
I really like it, but I think it's bad, like good, bad.
01:05:11.422 --> 01:05:14.692
The morality kind of meaning making is different from the utility, kind of meaning making.
01:05:14.692 --> 01:05:16.072
They may intersect, they may not.
01:05:17.121 --> 01:05:18.172
So those are two kinds of meaning.
01:05:18.172 --> 01:05:18.472
Making.
01:05:18.472 --> 01:05:27.262
The third kind of meaning making is when someone tells you or they asks you, um, do you prefer this?
01:05:27.262 --> 01:05:27.981
To that, to that.
01:05:28.012 --> 01:05:31.371
Like, do you like apples more than oranges, more than grapefruit?
01:05:31.641 --> 01:05:34.086
And you say, yeah, actually I like apples more than oranges, more than grape fruit.
01:05:34.657 --> 01:05:42.967
There's no objectively correct ordering the rank ordering of how much you like something either in terms of utility or morality, is also a kind of meaning making, right?
01:05:43.206 --> 01:06:02.797
So you can go to the museum and you can say, I love the paintings of Picasso, but I like them less than, I like the paintings of, I don't know, Richter or Van Gogh, and anyone who you ask could come up with that, meaning making hierarchy of value, but many people don't even think about it.
01:06:02.856 --> 01:06:15.936
Then the fourth kind of meaning making is when someone gives you a hierarchy, they say, you know, uh, for example, if you want to be an entrepreneur, being a venture capital funder entrepreneur is better than being a small entrepreneur doing unambitious things.
01:06:16.507 --> 01:06:19.327
And then you can say, yes, I accept that hierarchy of value.
01:06:19.387 --> 01:06:22.536
Or you can do your own meaning making and say, no, I think that's bullshit, right?
01:06:22.896 --> 01:06:31.507
I think instead it's better to do boring, tiny tools and make a small tech company than to have a almost 0% chance of success building a venture capital funded company.
01:06:31.507 --> 01:06:39.306
So those four kinds of meaning making where we decide what is valuable in the world and how much they are valuable relative to other things, I think that's something that machines can't do.
01:06:39.726 --> 01:06:47.706
Machines look like they can do it because you can ask a machine, uh, apple's better than, orange is better than pears, and an answer will come back.
01:06:47.976 --> 01:06:55.177
But the answer is actually, if it's an LLM, it's a probabilistic reaction to what it was trained on, right?
01:06:55.177 --> 01:06:59.467
It's not actually an entity that has flavor preferences like we are.
01:06:59.677 --> 01:07:03.876
If I ask you what fruit you like and I ask you what fruit you like, you will have an answer.
01:07:03.952 --> 01:07:09.681
Yeah, but the answer isn't because you were trained on a data set of other human answers in the past.
01:07:09.981 --> 01:07:14.001
But that, but that's the fourth kind of mini people tell you this is good and you say, no, it's not good.
01:07:14.001 --> 01:07:14.992
I'm gonna do something else instead.
01:07:14.992 --> 01:07:15.172
Yeah.
01:07:15.411 --> 01:07:16.072
So machines, we have more
01:07:16.297 --> 01:07:16.536
compliant
01:07:16.657 --> 01:07:18.112
machines for now.
01:07:18.771 --> 01:07:20.722
We, I cannot see how they can do that.
01:07:20.722 --> 01:07:23.391
Even conceptually sold that category of me.
01:07:23.391 --> 01:07:27.681
All these categories of meaning making, they actually go from super trivial, like, I like fruit, I don't like fruit.
01:07:27.922 --> 01:07:35.001
All the way through to things like, you know, is this person who comes before me in this court of law, the plaintiff?
01:07:35.692 --> 01:07:38.271
Should we fine for the plaintiff or should we fine for the defendant?
01:07:39.112 --> 01:07:41.032
If it goes one level up Supreme Court.
01:07:41.467 --> 01:07:45.577
Should this precedent on which so much law is based, be overturned or upheld.
01:07:45.967 --> 01:07:46.956
All these are meaning making.
01:07:46.956 --> 01:07:48.036
We should never let machines do it.
01:07:48.456 --> 01:07:54.606
Not until we build machines that actually can do meaning making, but everything that as at least I know as a qualitative sociologist, huh?
01:07:54.606 --> 01:08:02.887
I'm not a computer scientist, but everything I know about how we've built the particular machines that are very popular right now, LLMs, they don't do meaning making at all.
01:08:02.976 --> 01:08:06.456
They simply produce outputs that look like the outputs of people that do do meaning making.
01:08:06.456 --> 01:08:07.416
And that is not the same thing.
01:08:07.806 --> 01:08:07.927
Mm-hmm.
01:08:08.166 --> 01:08:17.587
So output indistinguishability, if you look at all of the benchmarks, the tests for models, they only deal with outputs, not with the things that generate outputs when humans do them.
01:08:17.917 --> 01:08:24.127
And the act of meaning making is something that, as far as I can tell right now, must be done by humans.
01:08:24.157 --> 01:08:25.447
'cause only humans can do them.
01:08:25.452 --> 01:08:28.027
Mm. And that feeds into how we design work, right?
01:08:28.027 --> 01:08:34.537
Like when you think about work that's being done, if it's completely specifiable in advance, super deterministic, it requires no judgment.
01:08:34.837 --> 01:08:35.947
They won't make a human do it.
01:08:36.067 --> 01:08:36.817
They're bad at it.
01:08:36.817 --> 01:08:37.657
They get bored by it.
01:08:37.806 --> 01:08:40.176
Give them something more fun, more meaning making to do.
01:08:40.567 --> 01:08:40.777
Mm.
01:08:41.077 --> 01:08:46.087
If something requires judgment, there's no objectively correct test for whether or not it's done the right way and how to do it.
01:08:46.417 --> 01:08:47.527
Don't let a machine do it.
01:08:47.617 --> 01:08:48.577
Mm. Get a human to do it.
01:08:48.697 --> 01:08:50.646
They have discretion, but then you have to train discretion.
01:08:50.646 --> 01:08:53.947
So we have to also train humans to be better at meaning making.
01:08:53.947 --> 01:08:54.606
We don't do that.
01:08:54.606 --> 01:08:55.027
Mm.
01:08:55.296 --> 01:08:55.867
So, yes.
01:08:55.896 --> 01:08:56.077
Okay.
01:08:56.287 --> 01:08:56.466
Yeah.
01:08:56.466 --> 01:09:07.027
Will you, will you say then that, you know, um, art, the art scene or people making art is a form of meaning that you cannot really attribute to machines or Yeah, yeah, for
01:09:07.027 --> 01:09:07.296
sure.
01:09:07.657 --> 01:09:15.606
So in, I hope that in every cohort of tangent we will have one or more artists, but we actually have artists in this cohort.
01:09:15.667 --> 01:09:16.957
They're doing very interesting stuff.
01:09:17.436 --> 01:09:29.077
And next year, uh, I can't announce exactly who it is yet, but we're working with another organization here to do maybe one cohort, which is just for art and design people to do things that may become viable businesses themselves.
01:09:29.077 --> 01:09:29.287
Right.
01:09:29.287 --> 01:09:34.207
So never sort of like we bring in a cohort of art people just to sort of like.
01:09:34.207 --> 01:09:37.867
Like, you know, in a patronizing way, like we have to do an artist cohort.
01:09:37.896 --> 01:09:44.676
It's because I actually think, you know, that fourth category of ai business art is one of those things that becomes relatively more important when AI slot is everywhere.
01:09:44.707 --> 01:09:44.947
Exactly.
01:09:44.947 --> 01:09:45.006
Yeah.
01:09:45.006 --> 01:09:47.167
Like have really, really good artists who are thinking in different ways.
01:09:47.497 --> 01:09:53.377
They are simply not replicable by AI because they have a meaning making drive inside them.
01:09:53.917 --> 01:09:57.306
And we a hundred percent are trying to do this at Tangent also.
01:09:57.787 --> 01:09:58.567
That's the fourth category
01:09:59.167 --> 01:09:59.256
of ai.
01:09:59.261 --> 01:09:59.787
So, All right.
01:09:59.817 --> 01:10:01.032
Um, let's move to the quickfire.
01:10:01.032 --> 01:10:01.192
Yeah.
01:10:01.197 --> 01:10:01.646
Quick fire.
01:10:02.367 --> 01:10:03.806
Are you ready for the quick fire round?
01:10:03.811 --> 01:10:04.032
Sure.
01:10:05.756 --> 01:10:08.127
Uh, two to three books you recommend to others.
01:10:08.186 --> 01:10:10.737
Uh, one book, I was prepped for this, so I, I know.
01:10:11.216 --> 01:10:17.726
So one book is seeing is forgetting the name of the thing one Sees, uh, it's a intellectual biography of the artist, Robert Erwin.
01:10:17.731 --> 01:10:17.872
Mm-hmm.
01:10:18.417 --> 01:10:19.556
By Lawrence Wessler.
01:10:19.676 --> 01:10:20.277
Great book.
01:10:20.516 --> 01:10:21.327
Um, go look it up.
01:10:22.167 --> 01:10:24.537
Uh, the second book is For the time being.
01:10:25.436 --> 01:10:25.707
For the
01:10:25.707 --> 01:10:26.936
Time Being by, yeah.
01:10:27.027 --> 01:10:27.777
By Annie Dillard.
01:10:28.377 --> 01:10:29.516
And they're both very weird books.
01:10:29.516 --> 01:10:30.506
They're very difficult to describe.
01:10:30.506 --> 01:10:39.266
One of them is an intellectual biography of an artist that keeps changing what he does in art because he exhausts the possibility of one medium, its tools and affordances.
01:10:39.657 --> 01:10:44.667
And realizes to answer the question he has in his mind, he has to just learn how to use a different material.
01:10:45.087 --> 01:10:49.587
So I went to see a retrospective of his work, um, a few years ago, and it's all new work.
01:10:49.587 --> 01:10:52.917
He's the only artist I've ever seen where the retrospective is only new stuff.
01:10:53.367 --> 01:10:57.716
'cause now he knows how to answer the question better by having gone through a whole bunch of different tools.
01:10:58.407 --> 01:11:02.966
Uh, so that's seeing is forgetting the name of the thing One Sees by, uh, Lawrence Wesler.
01:11:03.567 --> 01:11:09.747
Uh, the second book is for the Time Being by Annie dillard, and it's about why bad things still happen to good people.
01:11:09.747 --> 01:11:12.117
But it's an amazing book, so it's impossible to describe.
01:11:12.117 --> 01:11:17.907
It's about all sorts of weird stuff like archeology, like priests.
01:11:18.087 --> 01:11:19.587
But yeah, it, it's, it's good.
01:11:19.612 --> 01:11:22.136
I, I it's literally undescribable, so I won't try
01:11:22.167 --> 01:11:23.877
annie dillard list definitely on my list.
01:11:24.087 --> 01:11:24.476
So, yeah.
01:11:24.627 --> 01:11:27.957
Uh, a favorite show, film or content that you are into
01:11:28.247 --> 01:11:33.032
so I don't really consume like, like TV or films.
01:11:33.082 --> 01:11:39.021
So I think the only content I consume is I have background YouTube music.
01:11:39.171 --> 01:11:39.801
Oh.
01:11:40.377 --> 01:11:42.337
I like listening to old school country music.
01:11:42.636 --> 01:11:42.856
Wow.
01:11:42.957 --> 01:11:43.896
Old school country music.
01:11:43.942 --> 01:11:48.202
So like, no, but it's, it's, you know, nice, good melody.
01:11:48.591 --> 01:11:50.061
You can always understand the words that they're saying.
01:11:50.662 --> 01:11:51.577
And there's always a story, or
01:11:51.577 --> 01:11:52.162
there's always a story.
01:11:52.162 --> 01:11:54.771
It's always like, my dog died, then my left, my
01:11:54.771 --> 01:11:55.221
wife left
01:11:55.221 --> 01:11:55.282
me.
01:11:55.282 --> 01:11:55.641
Exactly.
01:11:56.721 --> 01:11:57.681
Someone stole my wife.
01:11:57.742 --> 01:12:00.742
I have a, I have an extensive collection of good old school country
01:12:00.742 --> 01:12:00.981
music.
01:12:00.981 --> 01:12:01.341
Nice, nice.
01:12:01.341 --> 01:12:03.652
A place in Singapore that is special to you?
01:12:04.131 --> 01:12:08.002
Uh, I have my favorite part home store in bukit batok where I'm not gonna tell people.
01:12:09.127 --> 01:12:09.417
Okay.
01:12:09.471 --> 01:12:10.042
Keep nice.
01:12:10.731 --> 01:12:11.061
Exactly.
01:12:11.542 --> 01:12:11.872
Yeah.
01:12:11.931 --> 01:12:14.152
Uh, favorite restaurants in Singapore.
01:12:14.152 --> 01:12:14.841
Oh, okay.
01:12:15.351 --> 01:12:15.681
Since you
01:12:15.681 --> 01:12:16.942
are foodie, I mean, I dunno if
01:12:16.942 --> 01:12:17.212
you're,
01:12:17.212 --> 01:12:17.542
but Yeah.
01:12:17.721 --> 01:12:25.896
So I, I actually have a semi policy about not going to fancy restaurants in Singapore because I don't think they're our strong suit.
01:12:26.837 --> 01:12:27.056
Um.
01:12:27.891 --> 01:12:46.851
If I go to a fancy restaurant, it will only be to a Chinese restaurant because I think the execution quality is very high and the amount of money you pay is a lot less than for generally lower quality over overall execution in a non-Asian cuisine fine dining restaurant.
01:12:46.912 --> 01:12:47.551
Mm. Okay.
01:12:47.551 --> 01:12:49.681
A tool or product that you got excited about?
01:12:49.931 --> 01:12:52.181
So years ago, I, I used to make furniture, right?
01:12:52.242 --> 01:12:52.391
Yeah.
01:12:52.421 --> 01:12:52.452
Uhhuh
01:12:52.752 --> 01:12:54.011
and one.
01:12:54.011 --> 01:12:57.582
Okay, so this is actually one of the beginnings of boring tiny tools, like, but I didn't realize it back then.
01:12:57.851 --> 01:12:59.261
So there's a tool called Saw Stop.
01:12:59.622 --> 01:13:00.542
Have you ever used a saw?
01:13:00.612 --> 01:13:00.881
Stop
01:13:01.212 --> 01:13:01.662
Saw,
01:13:01.752 --> 01:13:03.072
saw, stop that.
01:13:03.072 --> 01:13:03.792
That's just the name of the tool.
01:13:03.792 --> 01:13:03.971
No.
01:13:04.301 --> 01:13:07.481
So, you know, the table saw there's a big spinning blade.
01:13:07.542 --> 01:13:11.891
Mm. And you push the wood through it and often people will like accidentally cut off their thumb Yeah.
01:13:11.891 --> 01:13:13.091
As they're pushing the, the wood through it.
01:13:13.301 --> 01:13:15.101
So the saw stop is a really interesting thing.
01:13:15.101 --> 01:13:26.051
It's just a table saw, but there is a electrical circuit that when it's completed by touching something which is conductive, like flesh, it automatically throws.
01:13:26.051 --> 01:13:29.351
I saw that a piece of shape aluminum into the blade and it stops in a fraction a second.
01:13:29.351 --> 01:13:38.502
So I've actually tested it before where you just touch a spinning blade and you barely get scratched and the sauce stop is such a great solution to a problem that everyone knows is there.
01:13:38.742 --> 01:13:43.631
That actually within probably a few years of me first encountering the sauce stop.
01:13:43.811 --> 01:13:47.591
It was in almost every teaching workshop I, I used to help run a teaching workshop.
01:13:48.192 --> 01:13:51.731
It was like horrible to watch people using table sauce unsafely.
01:13:52.061 --> 01:13:54.492
Then the moment the sauce up shows up, you don't even have to worry anymore.
01:13:54.551 --> 01:14:03.942
It's just like, not exactly let them do whatever the hell they want, but you don't have to be watching to make sure that they're not doing something really stupid, because if they do something really stupid, the sauce stop will probably be okay.
01:14:04.301 --> 01:14:04.692
Nice.
01:14:04.692 --> 01:14:06.162
That's a, it's an amazing tool.
01:14:06.252 --> 01:14:06.582
Yeah.
01:14:06.582 --> 01:14:07.466
Okay, one last one.
01:14:07.466 --> 01:14:08.801
Where can people find you?
01:14:09.042 --> 01:14:11.082
Okay, and, and how can they be helpful?
01:14:11.082 --> 01:14:12.431
I think you mentioned already, but Yeah,
01:14:12.431 --> 01:14:13.301
but I can say it again.
01:14:13.301 --> 01:14:13.481
Yeah.
01:14:13.542 --> 01:14:16.452
So, um, you can find me on my website, v tan.org.
01:14:16.577 --> 01:14:22.181
ORG, uh, Vaughn spelled only with one A-V-A-U-G-H-N-T-A n.org.
01:14:22.181 --> 01:14:22.256
RG.
01:14:22.877 --> 01:14:28.511
Um, a lot of my writing is there, uh, for tangent, which I hope you go look at a lot.
01:14:29.082 --> 01:14:37.752
Uh, it's tangent.org sg and the kind of help I would like is if you know, or you yourself are someone who is trying to start a business like the kind that we're trying to incubate.
01:14:38.487 --> 01:14:41.606
Uh, let them know or you come find us.
01:14:41.636 --> 01:14:41.846
Right.
01:14:41.846 --> 01:14:44.247
We're very easy to find and we'll talk to everyone.
01:14:44.966 --> 01:14:53.306
And if you wanna be a guide, uh, if you work in an SME or you want to just help these kinds of small companies start and then grow, uh, also get in touch.
01:14:53.306 --> 01:14:57.747
We want to build a network of people that are interested in this stuff so that we can keep cycling people through.
01:14:57.792 --> 01:14:58.212
Mm-hmm.
01:14:58.587 --> 01:14:58.827
Yeah.
01:14:58.827 --> 01:14:59.907
Thanks so much.
01:15:00.027 --> 01:15:00.777
Thanks for the chat.
01:15:01.016 --> 01:15:01.226
Great.
01:15:01.277 --> 01:15:02.197
That was a lot of fun.
01:15:02.247 --> 01:15:02.546
Yeah.
01:15:02.546 --> 01:15:03.056
Thanks, Juan.
01:15:03.327 --> 01:15:03.476
Thanks.
01:15:04.077 --> 01:15:06.716
Thank you so much for tuning into the Code Riff podcast.
01:15:06.957 --> 01:15:09.447
We hope you learned and enjoyed the conversation as much as we did.
01:15:09.837 --> 01:15:17.277
If you'd like to support us, please consider subscribing, leaving a comment like or review on wherever you listen to your podcast, and we hope to see you in the next episode.
01:15:17.546 --> 01:15:17.907
Bye.