Intent Is the New Code: Artem Koren on Why the Next 100 Years Belong to Communicators, Not Engineers

WEBVTT
00:00:00.077 --> 00:00:14.538
So famously, you know, AI is really great at information things, but whenever there's real world endpoints, like you actually like talk to a customer or you send an email or whatever it is, AI still can't meet you for coffee and all that stuff.
00:00:14.538 --> 00:00:19.018
So there's still things that happen in the real world that you need people for.
00:00:20.358 --> 00:00:22.928
Hello, everyone, and welcome back to the show.
00:00:23.538 --> 00:00:27.347
On this podcast, we are driven by truth, sometimes the hard truth.
00:00:28.007 --> 00:00:32.777
We believe it's imperative to be relentless for results because if you're not, your competitor is.
00:00:33.198 --> 00:00:36.927
We're obsessed with how to be better every day because that's what our customers deserve.
00:00:37.237 --> 00:00:45.487
And if you can set aside your ego, if you can truly be no ego, then we're the show to help you go all in, because good enough isn't.
00:00:46.747 --> 00:00:51.728
Today's guest is Artem Koren, the co-founder and chief product officer of Sembly AI.
00:00:52.148 --> 00:00:59.207
He is a self-possessed writer, builder, teacher, and lifelong student of how people think, create, and make sense of complexity.
00:00:59.838 --> 00:01:07.938
Other fun tidbits include the fact that Artem is a published author and a 1950, I hope I'm saying this right, 1950 rated bullet chess player.
00:01:08.337 --> 00:01:09.507
So a man of many talents.
00:01:09.947 --> 00:01:11.597
Artem, welcome to"Good Enough Isn't."
00:01:12.308 --> 00:01:14.097
Thank you for having me on.
00:01:15.308 --> 00:01:15.647
Yeah.
00:01:16.018 --> 00:01:20.808
So before we get into the company and maybe some chess I'd like to roll back the clock a little bit.
00:01:21.048 --> 00:01:28.418
And I saw in the research that you grew up in New York, and you ended up at a specialized high school whose name I'm sure I'd butcher if I tried to say it.
00:01:28.867 --> 00:01:33.737
But I looked up that school and it only has a 3% acceptance rate, which is interesting.
00:01:33.748 --> 00:01:41.658
So I'm curious to go back to you as that kid, and when you end up at a school like that, were you trying to prove something, and did you?
00:01:43.995 --> 00:01:45.575
What a starter question.
00:01:46.495 --> 00:01:52.099
Taking me all the way back I don't think I was trying to prove anything.
00:01:52.310 --> 00:02:02.519
I think growing up I was naturally had an aptitude for math and tests, I would say in general.
00:02:02.519 --> 00:02:04.730
I was pretty good at studying for tests.
00:02:05.340 --> 00:02:18.740
But there was one distinct event, I think might have happened when I was in like second or third grade, where-- And I, I, you know, we don't remember a lot from our childhood, but there's cert- like little key moments that we do.
00:02:19.129 --> 00:02:39.210
And so I remember this distinctly that my mom or my aunt, or maybe both, woke me up from sleeping in the evening because I didn't finish my homework, and they made me wake up and go finish my homework.
00:02:39.659 --> 00:02:45.349
And ever since then, that instilled a certain kind of, uh, study discipline in me.
00:02:46.370 --> 00:02:50.409
And I was-- I, I, I got pretty good at school.
00:02:50.550 --> 00:02:54.009
And I also am a little bit competitive, I think, naturally.
00:02:54.810 --> 00:02:56.330
So this is, uh...
00:02:56.349 --> 00:03:03.699
The, the-- When those factors come together you get, uh, Stuyvesant High School and...
00:03:03.740 --> 00:03:06.550
Well, if I'm gonna brag, I mean, okay, if we're there- Let's do it.
00:03:06.860 --> 00:03:06.979
Let's do
00:03:07.330 --> 00:03:07.389
it.
00:03:07.409 --> 00:03:07.699
Yeah
00:03:08.020 --> 00:03:11.870
first of all, I got a score of 666 on that test, I remember.
00:03:12.650 --> 00:03:15.110
Which I think is, foreshadowing my future life mostly.
00:03:16.180 --> 00:03:25.430
Um, but a 666 on that test was, I was 33rd in the city out of, at that time it was something like 20,000 ta- test takers.
00:03:25.930 --> 00:03:26.159
Wow.
00:03:26.330 --> 00:03:29.389
And today I think it's like four...
00:03:29.400 --> 00:03:32.909
Like 30 or some ridiculous number of kids take this test today.
00:03:33.789 --> 00:03:37.830
And, uh, and I think the acceptance rate is probably even, even lower than that.
00:03:37.830 --> 00:03:40.460
It's, it's kind of insane, the competition for that school now.
00:03:40.530 --> 00:03:43.000
But yeah was not expecting that question.
00:03:43.780 --> 00:03:46.259
Uh, but thank you for, uh, jogging that memory.
00:03:47.439 --> 00:03:47.750
Yeah.
00:03:47.780 --> 00:03:48.270
All good.
00:03:48.590 --> 00:03:56.560
I wanna go back to Aunt, Mom waking you up for homework too,'cause I think that's interesting, and it says a little bit to what it was probably like growing up in your house as a kid, right?
00:03:56.560 --> 00:04:05.090
If you were gonna get woken up in the middle of the night to finish your homework, there's like a standard of excellence and high expectations that I'm sure was rooted into you that carries forward.
00:04:05.599 --> 00:04:07.789
Um, is that where chess came from?
00:04:07.870 --> 00:04:14.360
Because like most people just like play chess, but whatever a 1950 rating is, it sounds good.
00:04:14.400 --> 00:04:17.779
But that means like you, you don't just play chess, like you, you went in on chess.
00:04:17.810 --> 00:04:21.310
Like you, you said your competitive nature, uh- E- you went for it
00:04:21.980 --> 00:04:23.550
everything starts in childhood.
00:04:23.560 --> 00:04:49.038
My grandfather taught me to play chess when I was maybe four years old, and I started to play competitively when I was seven or eight for a couple of years, and I got to a pretty high rating, but then I stopped And I stopped because, uh, when I was growing up in my very, very early years, my family lived in Belarus, and then we immigrated to U- to the United States, to New York, and I came here when I was nine years old.
00:04:49.478 --> 00:04:56.148
And so when I came to New York, I basically stopped playing completely, and I didn't touch a chess piece for the next 20, 25 years.
00:04:57.028 --> 00:05:00.747
And that randomly one day I was doing management consulting.
00:05:00.747 --> 00:05:09.317
I was in a very boring meeting in a hotel room, and as, you know, everybody else was talking, I needed something to do, so I, I would...
00:05:10.278 --> 00:05:23.637
I just randomly loaded up chess on Yahoo Games and started tinkering, and I, I didn't touch it for decades at that point, but I really got into it and I've been playing chess every day ever since.
00:05:24.108 --> 00:05:31.168
And bullet chess is the kind of game where each side only gets 60 seconds to make all of their moves, so the maximum duration of a game is about two minutes.
00:05:32.197 --> 00:05:34.218
So it's a very, very fast game.
00:05:34.218 --> 00:05:41.708
It's a very-- it's a lot about your instinct, very fast response, and it's a lot about prediction.
00:05:41.807 --> 00:05:45.848
You have to foresee a lot in this game.
00:05:46.398 --> 00:05:53.848
Like, it-- you're basically, you It, it-- There's a concept of pre-moves where, like, you move and then you also move your preset, your next move.
00:05:54.538 --> 00:05:57.637
That's the only way you can be as fast as you need to be in this game.
00:05:57.987 --> 00:06:03.067
So you need to predict what your opponent will do at the same time as think about what you're gonna do.
00:06:03.648 --> 00:06:06.987
And, um, yeah, I, I found, I find it very fascinating.
00:06:06.987 --> 00:06:17.088
I, I, I like the kind of the intense, intense, um, uh, way that it makes you focus because it's a game where you also can't look away for a second, right?
00:06:17.088 --> 00:06:18.867
If you look away for a second, you're, you've lost.
00:06:19.598 --> 00:06:20.817
So I, I like that a lot.
00:06:20.867 --> 00:06:33.798
And I think in some, in some way also it's a distraction from other, like, complex things going on in, in, in work and life because it's something where it just completely absorbs your attention for those couple of minutes that you're playing.
00:06:36.612 --> 00:06:38.601
What, uh, what, what platform are you playing on?
00:06:38.601 --> 00:06:41.302
Just so I know that if I run into you, I should just end the game.
00:06:41.422 --> 00:06:44.192
I play primarily on, uh, Lichess, L-I chess.
00:06:44.661 --> 00:06:45.012
Okay.
00:06:45.052 --> 00:06:45.692
Yeah, yeah,
00:06:45.692 --> 00:06:45.891
yeah.
00:06:45.891 --> 00:06:46.942
Not chess.com.
00:06:47.182 --> 00:06:47.451
Yeah.
00:06:47.901 --> 00:06:48.742
So you're safe- Got it
00:06:49.252 --> 00:06:49.502
on chess.com.
00:06:49.502 --> 00:06:51.182
I'm, I'm safe on chef- chess.com.
00:06:51.182 --> 00:06:51.572
Perfect.
00:06:52.572 --> 00:07:09.161
Um, thinking about chess and you talk about it's like predicting I saw that the first company that you founded Nusana, if I'm saying that correctly, in 2009, it was machine learning focused on medical imaging, and that's like way before machine learning was in the popular vocabulary.
00:07:09.891 --> 00:07:13.262
So in a way, I guess that predicting made its way into business too.
00:07:13.401 --> 00:07:17.322
Like what, what made you go into that as the first venture before most people were thinking about that?
00:07:19.262 --> 00:07:28.372
So in, in just with the many things in life, it's one part luck, one part circumstance, and one part maybe preparation.
00:07:29.432 --> 00:07:34.672
This was a time when I just got disenchanted with investment banking.
00:07:34.672 --> 00:07:47.021
So I, I, at some point I thought,"Okay, I wanna..." I don't know, don't ask why, but I thought I wanted to be an investment banker, and I went to get my MBA, and I did a summer as an summer associate, a banker at Citigroup.
00:07:48.172 --> 00:08:07.552
And that summer I was like,"Okay, this is not what I want to do." Not because it's like not something I couldn't do, it's the certain-- there are cultures that come with every type of job that you decide to do, and there's certain kind of typical standards for how people are in those circles, and it just wasn't my circle of people.
00:08:07.692 --> 00:08:13.942
So after I did that summer, I was looking at, okay I think I want to do something different, maybe a startup.
00:08:13.942 --> 00:08:14.851
That would be interesting.
00:08:14.851 --> 00:08:17.192
I had no idea what kind of startup or anything like that.
00:08:18.451 --> 00:08:27.791
I happened to, uh, have a friend who is a pathologist who was in the same idea space at the point-- at that point.
00:08:27.791 --> 00:08:31.081
He was a little bit bored about being a, just a pathologist.
00:08:31.081 --> 00:08:32.052
He wanted to do something.
00:08:32.111 --> 00:08:33.081
He wasn't sure what.
00:08:33.792 --> 00:08:38.562
But that wasn't yet enough critical mass for a, a startup to happen.
00:08:38.642 --> 00:08:44.782
The critical mass happened when I randomly ran into an old friend of mine from high school, Stuyvesant.
00:08:45.241 --> 00:08:58.022
His name is Yuri Niyazov, and he was doing a PhD at NYU's computer science lab under Yann LeCun, which you-- a name at this point everyone knows right?
00:08:58.022 --> 00:09:00.611
One of the godfathers of modern AI.
00:09:01.282 --> 00:09:06.631
At that point, he was just a little famous, uh, and he was still at NYU.
00:09:07.861 --> 00:09:12.062
Uh, not yet, you know, the, the world worldwide famousness.
00:09:13.511 --> 00:09:21.572
And so, uh, Yuri was working with convolutional neural nets, which was the early image recognition stuff.
00:09:21.572 --> 00:09:26.322
So if you guys remember, finding cats in an image was very difficult back in the day.
00:09:26.971 --> 00:09:37.855
And then the CNNs came in, and Yann LeCun was one of the I guess the first people who got CNNs to do what they're supposed to do.
00:09:38.485 --> 00:09:43.225
And with CNNs, you can suddenly start to find cats everywhere and that was really good.
00:09:43.515 --> 00:09:50.846
So Yuri was in that lab where some of those earliest algorithms were developed, and he put me in touch with someone else from their lab.
00:09:50.846 --> 00:09:53.186
And so at that point, the critical mass was born.
00:09:53.505 --> 00:10:19.696
And there was this really special moment where this other PhD student from the lab that he put me in touch with, I met at a cafe right by NYU, and this kid brings his laptop, pops it open, and shows me like handwritten Python code for matrix math That basically does all the CNN stuff, all the convolutional neural net stuff that you need it to do.
00:10:19.696 --> 00:10:27.005
But at that time, no PyTorch, no- Yeah TensorF- well, like no, you know, nothing.
00:10:27.096 --> 00:10:29.395
It was just the-- You had to do raw math.
00:10:29.456 --> 00:10:33.115
So he had to build the library from scratch to do all that stuff.
00:10:33.505 --> 00:10:43.725
And he was showing me, and then he would show me the practical results where you can do handwriting recognition of numbers and stuff like that.
00:10:43.775 --> 00:10:45.796
And that was the first time I've ever seen anything like it.
00:10:45.796 --> 00:10:47.405
I've never seen anything like that before.
00:10:47.405 --> 00:10:56.615
So that was a magical moment, and who knew that, 15-plus years later we would have, uh, walking, talking AI.
00:10:56.615 --> 00:11:00.755
Uh, this, uh, this kid himself went to work- went on to work at NVIDIA.
00:11:00.755 --> 00:11:02.525
I haven't kept up with him.
00:11:02.525 --> 00:11:04.946
I don't know what he's up to, but I'm sure he's doing fantastic.
00:11:05.395 --> 00:11:07.186
But that was the genesis of Newsana.
00:11:07.196 --> 00:11:08.666
We were a little early.
00:11:08.985 --> 00:11:09.985
We were a little early.
00:11:10.316 --> 00:11:13.535
Things like processing big files was difficult.
00:11:13.546 --> 00:11:15.576
Storing big files was difficult.
00:11:15.826 --> 00:11:18.186
Digitizing biopsies was difficult.
00:11:18.206 --> 00:11:20.676
This technology was up and coming, but it was very early.
00:11:20.995 --> 00:11:27.056
Getting enough samples to train was very, very difficult, which probably was the main blocker.
00:11:27.775 --> 00:11:39.836
Uh, but yeah, I think we-- there were just too many technical challenges that we couldn't overcome, and we basically shut down after a couple of years of tinkering in that space.
00:11:39.875 --> 00:11:41.826
But yeah, that was the first, uh, foray.
00:11:42.416 --> 00:11:51.135
I love those magical moments where you see something for the first time like that, and it's, it just kinda opens your mind to possibility, right?
00:11:51.546 --> 00:12:01.155
Um, I feel like in, you know, from 2000 to 2010, there were many uh, right when, you know, I was coding AI back in 2004, right?
00:12:01.216 --> 00:12:03.686
And, you know, during that time, everything was hard.
00:12:04.985 --> 00:12:07.525
And, and then it starts getting easier and easier.
00:12:07.525 --> 00:12:17.796
And then, 2010 to 2020, I thought there was a, there was a lot of advancements in AI and-- but, again, you had to be super technical, machine learning.
00:12:17.806 --> 00:12:19.416
There were a lot of aha moments.
00:12:19.735 --> 00:12:32.323
The past six years, though, I, I feel like the number of aha moments, the number of times I do something now where I'm like,"That's just magic it's exponentially increasing every single day, right?
00:12:32.734 --> 00:12:48.994
And like to your point, like you start to see that, and then if you're an entrepreneur or, or it's something that you've been dreaming about for 10 years and now it just gets unlocked because the technology exists you know, you can start to, you can start to build these things and you can start to do things that you only imagined.
00:12:49.004 --> 00:12:50.673
But you can be too early, right?
00:12:50.693 --> 00:12:58.163
You can get in there too early like that and, you know, there's blockers still even in, when you're building today.
00:12:58.163 --> 00:13:07.683
So what did that lesson of like getting in, seeing the field, uh, you know, that clearly is something that was necessary, that was possible, it was just hard.
00:13:08.014 --> 00:13:14.394
You know, how did that lesson, you know, go into your, the your next venture, your next work, the next thing that you chose to do?
00:13:14.624 --> 00:13:17.224
And how do you think about it today with how fast things are moving?
00:13:20.206 --> 00:13:20.556
Yeah.
00:13:20.566 --> 00:13:28.485
I, I think my meth- my method, and it w- it's not an intentional method, it's just in retrospect I can say this about myself.
00:13:28.485 --> 00:13:29.836
I couldn't tell you this in the moment.
00:13:29.905 --> 00:13:46.966
But I think my method is to dive in, swim in the water, feel out where things are and how things work in detail, and then think about,"Okay, can I do something useful with this?" And if yes, dive in.
00:13:47.196 --> 00:13:50.316
If not, leave the pool and come back another time.
00:13:51.096 --> 00:14:01.436
And I tried to do AI in Columbia-- So I went for undergrad at Columbia University, and I did computer science there, and I was very j- very interested in AI at the time, very interested.
00:14:02.296 --> 00:14:16.076
And, um, I, I took AI courses, several, and it completely destroyed my AI dreams because everything we did was not...
00:14:16.255 --> 00:14:18.865
I can feel it at my core was not AI.
00:14:18.865 --> 00:14:20.426
Like I, I knew it wasn't AI.
00:14:20.796 --> 00:14:25.816
This was, you know, like tree searches and- Yeah, binary search tree
00:14:26.346 --> 00:14:26.385
things.
00:14:26.385 --> 00:14:26.395
Yeah.
00:14:26.395 --> 00:14:26.596
Yeah.
00:14:26.596 --> 00:14:26.946
Correct.
00:14:26.946 --> 00:14:27.096
Yeah.
00:14:27.755 --> 00:14:28.115
Yeah.
00:14:28.485 --> 00:14:31.855
And I was like,"That's not how brains do things." Yeah.
00:14:31.855 --> 00:14:34.076
So it's, it's not even close.
00:14:34.115 --> 00:14:39.166
This is you're, you're simulating something, but you're-- there's no intelligence here.
00:14:40.046 --> 00:14:45.255
And I thought it was i- disingenuous to call it AI, and I was so disappointed.
00:14:45.836 --> 00:14:55.035
So I, after that swim in the pool, I jumped out and I was like,"Nonsense." And so I didn't touch it for, six whatever years.
00:14:55.655 --> 00:14:58.196
Then the Newsun episode happened.
00:14:58.196 --> 00:15:16.225
I was like,"Okay, what's going on?" And I was like,"Okay, this, this is starting to smell like AI to me." But because there's so many technical things unsolved I was like,"Okay, we're not, we're not ready for, for something big here yet." Like cats are cool, like it's kind of important.
00:15:16.225 --> 00:15:19.535
I can see something's gonna happen soon, but we're not there yet.
00:15:19.696 --> 00:15:22.225
So I, you know, I did other things for a few years.
00:15:22.735 --> 00:15:30.005
Then I came back in around 2018 And I think that was the beginning of the modern.
00:15:30.135 --> 00:15:47.285
2018 was when you could do fairly easily on your own computer, your own time you can train neural nets to, like, predict which movies you're gonna like and all this kind of stuff.
00:15:47.456 --> 00:15:49.755
Like a bag of words.
00:15:49.865 --> 00:15:51.995
Does anybody remember what that is anymore?
00:15:52.046 --> 00:15:53.686
But bag of words was a big deal.
00:15:54.115 --> 00:16:04.186
That means, you know, you give it a lot of terms, like y- you would ba- basically you collect movie reviews from Rotten Tomatoes, and then you can, like, predict certain things with the movies.
00:16:04.816 --> 00:16:07.596
So that was, I think, the, the beginning of the modern stretch.
00:16:07.596 --> 00:16:15.099
And once I swam in those waters in 2018, I was like smells good." And then I started digging more and more and more, more and more.
00:16:15.339 --> 00:16:25.328
And, um, in 2019 was the first time that I saw a transformer summarize a news article.
00:16:25.879 --> 00:16:29.609
Oh, wow 2019.
00:16:29.629 --> 00:16:33.089
So that w- that was like right when they were being rolled out.
00:16:33.119 --> 00:16:34.078
I mean, that, that is-
00:16:35.009 --> 00:16:36.009
Right, like early
00:16:36.408 --> 00:16:36.869
yeah.
00:16:37.119 --> 00:16:39.339
I had one GPU to my name, mind you.
00:16:39.408 --> 00:16:52.538
But so phys- we, we put togeth- so when we started Assembly we put together a custom server and we bought like an$800 GPU, I forget the model now.
00:16:52.548 --> 00:16:55.698
It was one, an NVIDIA big NVIDIA card.
00:16:56.229 --> 00:17:05.429
We had to build custom fittings for the server box so that it would s- stay in'cause it just it's just this big blob of thing.
00:17:06.358 --> 00:17:16.939
And I put it in my, I put it in my uncle's house, connected up to his Wi-Fi and his power, and I'm like,"Just don't turn this off, okay?" Put it...
00:17:17.118 --> 00:17:20.169
It's loud, so just put it somewhere where it doesn't bother you in there and turn it on.
00:17:20.608 --> 00:17:23.199
That was our development server for a few years.
00:17:23.548 --> 00:17:24.249
Oh, wow.
00:17:24.689 --> 00:17:36.689
And yeah, so we ran the transformer, and I remember to this day, it was a CNN article about like a red Ford truck coming out soon, and maybe it was an electric truck.
00:17:36.699 --> 00:17:54.959
It was something about a red Ford truck on CNN, and I took the article and I fed it to the transformer, and I had it summarize, and it gave me a one-line summary of the article and said, you know,"Red Ford truck something." And at that point I knew basically...
00:17:55.028 --> 00:17:59.118
Listen I didn't, I couldn't exactly tell, okay, there's gonna be OpenAI, there's gonna be a Grok model- Right right?
00:17:59.118 --> 00:18:00.348
I, I couldn't say that.
00:18:00.999 --> 00:18:04.098
But when I saw that, I'm like,"We're building.
00:18:04.348 --> 00:18:07.058
We're building because the tech it's, it worked.
00:18:07.419 --> 00:18:09.548
It w- it summarized a news article.
00:18:09.548 --> 00:18:14.999
I've never seen anything like it in my life." So it was, you know, it was still crappy at the time.
00:18:14.999 --> 00:18:15.568
It wasn't good.
00:18:15.568 --> 00:18:17.269
It was super hallucinaty.
00:18:17.278 --> 00:18:19.058
It was super like wrong, blah, blah, blah.
00:18:19.088 --> 00:18:22.648
Like it would throw in random characters here and there, but it got it.
00:18:22.709 --> 00:18:28.298
Like you give it a page of text and it says,"This is, this is about a red Ford or a red Ford truck." I'm like,"Okay.
00:18:28.909 --> 00:18:29.138
We're-"
00:18:29.229 --> 00:18:29.469
Yeah
00:18:29.469 --> 00:18:29.489
we're going
00:18:31.012 --> 00:18:31.222
Yeah.
00:18:31.222 --> 00:18:38.633
And then, you know, obviously in the lab, you know, OpenAI is rolling out, they probably just rolled out GPT-1, GPT-2 is being built in 2019.
00:18:38.633 --> 00:18:56.772
GPT-3 came out in 2020, and that was my magic moment when I got-- when I played with GP- I think it was like June 2020, I played with GPT-3 in the playground and, uh, and I, I think I probably did something similar where I was like,"Hey, take, take this text," uh, you know.
00:18:56.873 --> 00:19:05.772
Now what we do with every email that we get, uh, you know,"Take this text and summarize it." I'm like,"Oh, wow, this is something different," right?
00:19:06.083 --> 00:19:07.163
This is something...
00:19:07.242 --> 00:19:12.232
You know, generative AI, I always kind of agreed with you, a background in math, you know, comp sci.
00:19:12.232 --> 00:19:16.153
I'm like, ML just feels like deterministic, you know.
00:19:16.153 --> 00:19:19.492
Like yes, it's advanced analysis, but it's still analysis.
00:19:19.492 --> 00:19:27.192
This was the first time for me where I used an AI and it felt creative in a way that I couldn't explain, right?
00:19:27.633 --> 00:19:29.502
And I was like...
00:19:29.502 --> 00:19:34.903
and then, you know, with Level being a marketing agency, I'm like,"Oh man, there's so many options here," right?
00:19:34.932 --> 00:19:41.803
You think about what we do as a company, you think about writing copy, you think about, just knowledge work in general.
00:19:41.803 --> 00:19:50.593
I'm like,"This is going to change everything." And then you s- then you started using GPT-3 and 3-5 and you're like,"Oh, this is terrible." But you could see the field, right?
00:19:50.593 --> 00:19:52.502
You could see what was gonna happen.
00:19:52.692 --> 00:19:59.192
But I don't ma- man, I don't think I thought it was gonna happen this fast, uh, when I saw that six years ago.
00:19:59.292 --> 00:20:05.482
I thought it would get slightly better, not what Fable 5.1 just came out and holy cow, right?
00:20:05.492 --> 00:20:07.232
Or GPT 5.6 Sol.
00:20:07.532 --> 00:20:18.673
Like these are light years ahead of where we were back when you tried that, that transformer in your uncle's house to, to where we are now, and it's instant as well, which is the other crazy part, right?
00:20:20.042 --> 00:20:35.948
It's, you know, we, we keep saying that it's geometric progression and we're just not good at pre- predicting it, and we say it, we know it, we tell ourselves, and then we, we fail anyway because that's just how we are.
00:20:35.948 --> 00:20:36.262
We can't comprehend
00:20:36.262 --> 00:20:36.462
it.
00:20:36.532 --> 00:20:36.843
Yeah.
00:20:36.903 --> 00:20:37.222
Yeah.
00:20:37.222 --> 00:20:38.792
It's super, super fast.
00:20:38.893 --> 00:20:46.126
My, my First moment with modern transformers was early, late'21, early 22.
00:20:46.136 --> 00:20:58.096
And there, I think my exa- example was like I said,"Okay, like my head hurts, like what should I do?" And then they were like-- It was able to complete like take an Advil or whatever it is.
00:20:58.896 --> 00:21:09.176
And I-- when I saw that, I was like,"Okay, we can build things with this now." And, you know, we, we were doing meeting intelligence-- We were focused on meeting intelligence at the time.
00:21:09.817 --> 00:21:17.086
And so I told my team, I'm like,"Okay, let's do like meeting summaries, task, tasks, decisions, events, all this kind of stuff.
00:21:17.116 --> 00:21:29.967
Let's build it with transformer tech." Because we at that, by that, up to that point, we had to build all our own models w- trained on like classic data science stuff, machine learning data science stuff.
00:21:29.967 --> 00:21:35.596
We had to engineer data sets both of like both real and synthetic.
00:21:35.596 --> 00:21:41.846
We had to train all our own custom models and everything, and then run those models on meeting transcriptions.
00:21:42.576 --> 00:21:45.376
And I was like,"Use this tech.
00:21:45.446 --> 00:21:47.527
This is the new tech." And you know what?
00:21:48.366 --> 00:21:50.906
They just couldn't do it.
00:21:51.376 --> 00:21:52.257
My team couldn't do it.
00:21:52.257 --> 00:21:53.057
Hmm.
00:21:53.057 --> 00:22:15.787
They, because they're really smart and really good at engineering, but the kind of engineering that large language models require it's a different mindset completely from a standard, from a classical software engineer, and these guys just couldn't get their heads out of the, out of the code.
00:22:16.616 --> 00:22:23.336
We ended up building something really cool, like we were-- we built the first chaptered summaries in the industry, completely automatic.
00:22:24.126 --> 00:22:38.636
But he did it in some like super complex, brilliant way where he would like calculate gradients of something in, in, in the conversation, but he just couldn't get the, um, the large language model to go.
00:22:38.656 --> 00:22:40.076
Well, I should say he and his team.
00:22:40.207 --> 00:22:43.346
We had a lead and he had a team, just couldn't do it.
00:22:43.346 --> 00:22:45.846
They failed to use LLMs that first year.
00:22:46.656 --> 00:22:48.067
And, um- It's really- Mm-hmm.
00:22:48.067 --> 00:22:49.217
It's really interesting, right?
00:22:49.217 --> 00:22:50.547
Like I, I think...
00:22:50.636 --> 00:22:54.547
I'd li- I wanna dive into that because I think that's a, that's an important distinction.
00:22:54.846 --> 00:23:20.497
I have some thoughts on why that, that is the case, but what-- as I've used these models more, what I have found is I'm actually pulling more from my, my experience as a manager and less my experience as a developer as I'm directing the models, especially as, you know, they're getting better at interpreting what I want and memory, uh, you know, memory is getting better at remembering what we've already decided.
00:23:20.967 --> 00:23:23.287
But, I'm curious, like I...
00:23:23.777 --> 00:23:26.596
You know, this use case that you have I think is very telling, right?
00:23:26.606 --> 00:23:29.126
Like you have a group of just brilliant engineers.
00:23:29.567 --> 00:23:33.767
You know, what do you think the root cause was there that, you know, they failed to transform?
00:23:33.767 --> 00:23:42.196
'Cause we're all going through some sort of AI transformation right now, whether you just started using it yesterday or you've been using it like you for 25 years.
00:23:42.207 --> 00:23:45.136
Like we're all going through a transformation.
00:23:45.136 --> 00:23:47.057
So this c- this might be helpful.
00:23:47.057 --> 00:23:50.237
This thought process might be helpful for someone listening going through their transformation.
00:23:50.237 --> 00:23:52.527
So what do you, what do you think was the blocker for those folks?
00:23:53.096 --> 00:23:54.797
I think you hit the nail on the head.
00:23:54.987 --> 00:24:01.997
The last century was driven-- The progress of the last century was driven by STEM.
00:24:02.866 --> 00:24:05.027
It was driven by science, math, engineering.
00:24:05.936 --> 00:24:15.116
Every major leap, any, any technology you point to, information technology, compute nuclear, whatever you want, it's all STEM.
00:24:15.547 --> 00:24:20.836
You had to be an engineer, and then you make that next step or that next leap.
00:24:21.876 --> 00:24:34.767
Um, and so it, it was skills in the hard science and math that moved things forward, and that requires a certain way of thinking a certain kind of interactions with machines.
00:24:34.787 --> 00:24:39.416
Like it takes a special person to be able to code an assembly.
00:24:40.277 --> 00:24:41.076
Let's be real, right?
00:24:41.076 --> 00:24:48.616
Like I've, you know, I've coded plenty in my life but in assembly, I've never made a whole program.
00:24:48.616 --> 00:24:51.176
I might have like tinkered with like something little here and
00:24:51.176 --> 00:24:51.207
there.
00:24:51.207 --> 00:24:52.596
I hated assembly, by the way.
00:24:52.616 --> 00:25:05.376
Like I had to, I ha- I had to take it for two semesters, and it was like-- I'm like,"Why am I doing this?" I mean, it taught me a lot about a computer and machines and how they function and how I never wanna have to handle that or do that ever again.
00:25:05.707 --> 00:25:16.576
But I can't, I-- People that do that, th- I mean, this is, a long time ago, but like people that do that, did that on a regular basis, I had no idea how they could even function like that, right?
00:25:16.586 --> 00:25:24.876
So, so I think for our audience, so, uh, guys, assembly is this kinda thing where we, when we lived in caves and the world was black and white.
00:25:26.307 --> 00:25:31.737
It's basically, it's not coding in ones and zeros, but it's pretty darn close.
00:25:32.267 --> 00:25:39.826
So it was basically working directly with the registers and memory addresses of the CPU.
00:25:40.527 --> 00:25:47.429
So there was no higher level abstractions that modern programming languages give you.
00:25:47.568 --> 00:25:59.898
You basically, an, a- an assembly program looks something like push AX comma eight and push AY comma 12 or whatever, and hex code this, hex code that.
00:26:00.699 --> 00:26:11.118
And somehow at the end of the day, something calculates or appears on your screen, and to any normal human, you just-- there's no way you can figure out why that just happened.
00:26:11.729 --> 00:26:11.838
Yeah.
00:26:11.838 --> 00:26:20.489
So, but the reason I bring that up is you have to be so deep in that code space, it takes a certain mind to do that.
00:26:21.209 --> 00:26:23.259
Now that's the last 100 years.
00:26:24.068 --> 00:26:31.878
Fast-forward to today and you know, I'm convinced that the next 100 years are not about STEM.
00:26:32.078 --> 00:26:39.778
They're about those who can most effectively communicate what they want to accomplish.
00:26:40.538 --> 00:26:43.068
And I have this cliché thing that I came up with.
00:26:43.689 --> 00:26:55.028
I don't know if I love it or hate it, but I keep saying it, which is legacy is information technol- so IT is information technology, but the future is IT intent technology.
00:26:56.108 --> 00:27:10.669
So rather than messing with information, the future is messing with intent, and it all becomes about shaping the intent as sharply as you can to your AI.
00:27:11.628 --> 00:27:15.009
And if your intent is really sharp, the result will be sharp.
00:27:15.459 --> 00:27:19.798
If your intent is loose and unclear, the result will be loose and unclear.
00:27:20.108 --> 00:27:25.608
I just saw a video about what actually happened in the hack of OpenAI to Hugging Face.
00:27:26.429 --> 00:27:38.199
So that famous couple of months, there was a famous episode where an agent broke out of confinement in OpenAI testing space and hacked Hugging Face to get answers to the questions it was being tested on.
00:27:39.068 --> 00:27:44.913
And, uh- What happened was, so the, the actual hack is fascinating.
00:27:44.923 --> 00:27:48.913
I recommend to everyone to Google and to search on YouTube and watch a video.
00:27:48.913 --> 00:28:10.022
But essentially, it's a whole swarm of agents escaped, reverse-engineered a communication-- a secret encrypted communication platform for themselves, organized into a swarm, including kamikaze agents that were required to do certain things, and then that swarm attacked Hugging Face.
00:28:10.583 --> 00:28:28.903
But the problem was that there was a, essentially a looseness of the intent where they were tasked with solving this test, but that's not a sharp enough intent, because the intent wasn't hack your way out of things to-- and do anything you want to solve this test.
00:28:29.303 --> 00:28:35.962
The intent was let's figure out what you know and what you are capable of resolving on your own.
00:28:36.593 --> 00:28:42.752
But because that intent was con- wasn't contoured specifically enough, they interpret it.
00:28:42.813 --> 00:28:46.202
"They" mean being the agent swarm interpreted it their own way.
00:28:46.803 --> 00:28:51.853
So information technology versus intent technology, stems versus communicators going in.
00:28:53.906 --> 00:28:54.767
I think that's super smart.
00:28:54.807 --> 00:28:55.497
I think that's right.
00:28:55.737 --> 00:29:07.346
You know, you even look at, you know, what I've been using a ton of-- Uh, uh, it's been my goal since January, I think, to see how long I can get tasks running to a good outcome.
00:29:07.856 --> 00:29:12.106
And I was just talking about this last night with, w-with one of our technical folks here.
00:29:13.787 --> 00:29:20.247
I'm using goal a lot,/goal in the, in the CLI, and if you haven't used that, if you're not using the CLI, you can ignore me.
00:29:20.247 --> 00:29:23.136
But if you are,/goal is phenomenal.
00:29:24.076 --> 00:29:27.017
And, uh, you give it an intent.
00:29:27.027 --> 00:29:29.856
You say,"This is what I want," but you do have to be specific, right?
00:29:30.207 --> 00:29:33.057
And build a, this is acceptance criteria.
00:29:33.067 --> 00:29:34.497
This is what I'm looking for.
00:29:34.507 --> 00:29:36.896
This is how-- This is what done looks like.
00:29:36.926 --> 00:29:38.106
This is what good looks like.
00:29:38.727 --> 00:29:52.567
And I think that, as, as I use that more and more, I'm writing few-- like less documentation, which is interesting, and more like user stories and more, this is how it should function when you're done.
00:29:52.576 --> 00:30:00.106
Not like, here's the tech stack and here's the test-driven dr- development framework you're gonna use, and here's all the other stuff.
00:30:00.106 --> 00:30:02.636
It's just like, no, I want an app that does this.
00:30:02.636 --> 00:30:05.307
And, you know, when a user logs in, they should be able to do this.
00:30:05.307 --> 00:30:06.846
And when a user logs in, they should do this.
00:30:06.846 --> 00:30:08.846
And if there's a problem, it should, it should fix...
00:30:08.886 --> 00:30:09.626
You know, it should do it.
00:30:09.626 --> 00:30:15.707
And, it's changed even in the past nine months, how I think about development, where it's more intent driven.
00:30:15.737 --> 00:30:17.277
You know, what am I trying to get?
00:30:17.757 --> 00:30:19.656
Um, not what do I want you to do?
00:30:20.007 --> 00:30:26.086
And I don't know if those are-- I don't know if that distinction's clear, but, uh, it, it-- to me it is very different things, right?
00:30:26.086 --> 00:30:37.166
It's, it's talking to you, Myles, and saying,"Hey, I want all of our employees to be happy," versus,"I want all of our employees to log in at nine AM." It's very two, two different things.
00:30:37.166 --> 00:30:41.856
If I say I want all our employees to be happy, you're gonna be like,"Well, what time should they log in in the morning?" You start thinking through it, right?
00:30:42.076 --> 00:30:47.567
Like, so it's it's a very different s-- you know, nuance to how I think you direct these systems.
00:30:47.567 --> 00:30:53.376
I think more people are going to start doing that in the future, where it's like,"Hey, this is what I'm looking for.
00:30:53.376 --> 00:31:01.237
Go do it." And then-- And it's gonna be so fast and so good that we're not gonna have to direct it from an engineering aspect at all.
00:31:02.875 --> 00:31:05.585
And I think you're gonna see this trend.
00:31:06.144 --> 00:31:17.674
So you're talking about it in, in the, in what I would call the micro or the tactical, where you're on the ground with the agent, and that's, you know, you're physically giving the intent to the agent.
00:31:18.194 --> 00:31:21.575
But we can actually cast this in a wi- much wider way.
00:31:22.224 --> 00:31:30.855
So think about how software has been used by people up to now.
00:31:31.625 --> 00:31:36.384
It was very much let's say, you know, let's take it, into my world.
00:31:36.654 --> 00:31:39.634
So let's say like I wanted to build a, a slide deck.
00:31:40.384 --> 00:31:42.365
I have to know how to use PowerPoint.
00:31:42.744 --> 00:31:49.545
So I have to-- It's, whatever my intent is, PowerPoint don't care, long hair don't care.
00:31:50.194 --> 00:31:53.085
I have to be a, a pretty pro PowerPoint user.
00:31:53.085 --> 00:32:03.085
I have to like know how to create all the elements, organize, arrange, blah, blah, blah, all the really useful things we learn in investment banking, which is by the way, what you learn in investment banking.
00:32:03.095 --> 00:32:04.045
You don't learn about deals.
00:32:04.105 --> 00:32:05.325
You learn about how to do PowerPoints.
00:32:05.865 --> 00:32:08.234
That's one big misconception about investment banking.
00:32:08.535 --> 00:32:13.634
But so it, it m- makes you a really pro PowerPoint user and a really pro Excel, uh, user.
00:32:13.644 --> 00:32:15.194
They're really good at PowerPoints though.
00:32:15.194 --> 00:32:19.974
I mean, to be fair, like they're, like IBs are some of the best that I've found.
00:32:20.105 --> 00:32:21.865
Like it's, it's pretty, pretty impressive.
00:32:23.234 --> 00:32:23.545
Yeah.
00:32:23.644 --> 00:32:26.775
They kinda have to be because that's their...
00:32:27.134 --> 00:32:28.494
Oh my God, I'm going to.
00:32:28.875 --> 00:32:29.184
Okay.
00:32:29.394 --> 00:32:31.315
Let's- Well, let's make it spicy, right?
00:32:31.345 --> 00:32:32.375
What, you know, we're here.
00:32:32.384 --> 00:32:32.474
Let's
00:32:32.474 --> 00:32:32.835
do it.
00:32:33.154 --> 00:32:34.055
Let's rock and roll.
00:32:34.065 --> 00:33:01.431
But that's a-- The, because their PowerPoint, their PowerPoints have to be flawless because that's how they, uh, signal the quality of their deal-making operation, and it's meant as a proxy for how quality of a process they have to execute deals.
00:33:01.701 --> 00:33:14.520
And so because the deal is worth a billion dollars, you have to be able to look at a PowerPoint and say,"Oh, this came from an organization that can handle a billion dollars well."
00:33:14.520 --> 00:33:15.211
Yeah.
00:33:15.211 --> 00:33:15.371
Yeah.
00:33:15.471 --> 00:33:17.080
And so that, that's the main reason.
00:33:17.080 --> 00:33:46.080
So the, the perfect PowerPoint was how, you know, the, the top consulting companies of the world and the investment banks of the world and the top legal firms of the world signal to you that this is their quality standard, and hope that you extrapolate date, extrapolate that to say,"Oh, okay, if their PowerPoints are, like, s- spotless, everything else that happens is spotless." That's the thinking, and that's why- Yeah the PowerPoints have to be perfect.
00:33:47.101 --> 00:33:55.800
What are these guys going to do when they can go now into my product and put your intent?
00:33:56.661 --> 00:34:28.900
Like, I can, I can go in Assembly and I can say,"Create a flawless executive PowerPoint to pitch this product to my, to this client," and Assembly generates it on brand, brings in, like, complex ROI, like, calculation slides, beautiful styling, everything's perfect, everything is grounded and sourced and referenced.
00:34:29.240 --> 00:34:31.731
There are no spelling mistakes.
00:34:32.130 --> 00:34:36.110
The arc, the story arc is really strong.
00:34:36.610 --> 00:34:39.400
The slides are sequenced in a good way.
00:34:39.440 --> 00:34:46.590
They're worded in a tonality that's going to be impactful for the client you're presenting to.
00:34:46.590 --> 00:34:49.701
If it's a legal firm, there's gonna be a legal tonality.
00:34:50.061 --> 00:34:53.510
If it's a creative firm, there's gonna be a creative tonality.
00:34:54.331 --> 00:34:56.121
What are th- what are they gonna do?
00:34:56.170 --> 00:34:58.280
'Cause that's what you can do with Assembly today.
00:34:58.291 --> 00:34:58.311
Yeah.
00:34:58.690 --> 00:35:08.291
And Assembly will pull from your meetings, your documents, your CRMs, whatever you want, and figure out on its own what that deck should look like and build it end to end.
00:35:08.311 --> 00:35:11.210
It derives all the branding stuff auto- automatically.
00:35:11.690 --> 00:35:19.340
So you literally put, you really put your intent into Assembly and you get back a slide deck versus, you know, having to monkey with PowerPoint.
00:35:20.271 --> 00:35:24.170
So what, h- how do they communicate their quality now is my question.
00:35:25.001 --> 00:35:31.197
Yeah, I mean, I think it I, I think it raises the floor, right?
00:35:31.387 --> 00:35:40.137
BCG, McKinsey, whatever, they could afford to spend that much time to make a PowerPoint amazing.
00:35:40.146 --> 00:35:42.836
They could hire the best people to be able to do it.
00:35:42.847 --> 00:35:52.847
They could do all of that because they were winning the billion-dollar projects, and if you were only winning$10,000 projects, you probably weren't gonna spend$30,000 to put together a PowerPoint presentation, right?
00:35:53.367 --> 00:35:58.856
And so the quality level would be different or, you know, the custom, the custom nature of those would be different.
00:35:59.226 --> 00:36:08.126
I think, you know, now with what you're describing, I wanna dive more into Sembly'cause it sounds like your entire career has been to the moment of building this, this product.
00:36:08.427 --> 00:36:16.936
But, uh, you know, I think now a$10,000 project, a$5,000 project, I can give you what looks like a$30,000 presentation, right?
00:36:17.246 --> 00:36:27.822
And I think we all have to-- the signal to your point that this means quality Doesn't exist anymore.
00:36:28.393 --> 00:36:31.632
And I think that is the thing that's scary, right?
00:36:32.012 --> 00:36:46.273
When you have spent your en- entire career looking for those signals to be able to make decisions, and now I can get something from you, I can get someone from-- something from Myles, and I can't tell the difference between which one's actually gonna give me a better outcome.
00:36:46.543 --> 00:36:59.862
I think we need to probably think about different ways that we are showing those signals to our clients and to our prospects and to our employees and to our family and all of those things of, how do you build that trust now?
00:37:00.373 --> 00:37:07.693
You know, we've started at Level, we've started to do more kind of upfront like,"Hey, we're not gonna give you a PowerPoint presentation on a new website.
00:37:07.762 --> 00:37:11.682
We're just gonna build a website for you." And that's the pitch.
00:37:12.472 --> 00:37:13.782
Like,"Here's the website.
00:37:13.782 --> 00:37:15.862
We bu- we built it for you in the last week.
00:37:16.063 --> 00:37:19.552
Take it, throw it away." But, you know, who would you rather work with?
00:37:19.583 --> 00:37:26.083
The company that put it all together for you and you can iterate on, or the company that gives you a perfect PowerPoint that they put together?
00:37:26.083 --> 00:37:43.373
And I think-- But then I will say on the flip side of that, communication, and I'm, I'm dialoguing here, but I apologize, but, like, the communication has, has-- it's almost never been more important to have a tool, I think, that can help you communicate ideas.
00:37:43.693 --> 00:37:54.523
Because, like, the last thing I wanna read right now is a 75-page output from someone's Claude that they cut and paste into Google Docs and sent me, and there's no intent behind it.
00:37:54.532 --> 00:37:55.802
There's no story behind it.
00:37:55.813 --> 00:37:57.273
There's no audience thought behind it.
00:37:57.273 --> 00:38:00.402
It's just,"Hey, this was interesting," and I read the first page and they send it to me.
00:38:00.702 --> 00:38:03.063
So I think, like, we're in this-- We're-- it's a dual thing.
00:38:03.063 --> 00:38:11.862
We need to figure out different ways to send those signals, but also, we need to make it easier and better for people to be able to build those, those thoughts and communicate better.
00:38:11.862 --> 00:38:15.822
It sounds like some of that is the reason why Sembly exists today.
00:38:16.523 --> 00:38:17.043
Is that true?
00:38:18.413 --> 00:38:28.135
We originally built Sembly with the idea that it would be very powerful for technology to understand what happens when your team meets there was nothing like that at the time.
00:38:28.304 --> 00:38:34.144
So there was a lot of technology that helped you do meetings, but there was nothing that actually understood what was being said in the meeting.
00:38:34.594 --> 00:38:42.425
And having a lot of management consulting background, so my co-founder, Gil Makleff spent, you know, his whole career in management consulting.
00:38:42.425 --> 00:38:45.105
He was the CEO of UMT Consulting Group, got acquired by EY.
00:38:45.105 --> 00:38:46.125
He was a partner at EY.
00:38:46.885 --> 00:38:48.724
You know, management consulting is his world.
00:38:48.755 --> 00:38:51.824
I spent, in and out, about a decade in management consulting.
00:38:51.905 --> 00:38:55.764
Also, like I was a senior-- When we got acquired, I was a senior manager at EY.
00:38:56.704 --> 00:38:59.855
And, um, meeting, meetings are life.
00:38:59.905 --> 00:39:00.965
Meetings are your whole life.
00:39:00.965 --> 00:39:03.014
It's o-over 80% of what you do.
00:39:03.204 --> 00:39:12.184
So we knew that if there's a way for tech to understand the meeting, there could be a lot of valuable things that happen after, and that was always the idea for Sembly.
00:39:12.235 --> 00:39:19.914
It's using what's going on in your workflow, in your business to cr- to give you results afterwards.
00:39:20.244 --> 00:39:24.224
And we were one of the founding companies in the meeting intelligence domain.
00:39:24.275 --> 00:39:27.644
We were one of the first very, very few companies to attempt to do it.
00:39:28.204 --> 00:39:32.304
Uh, and we had to solve a lot of complex technical challenges to get that done.
00:39:32.824 --> 00:39:38.144
Today, most of the complex challenges are solved one way or another through different components.
00:39:38.144 --> 00:39:41.525
There's Recall AI, there's transcribers, there's whatever have you.
00:39:42.364 --> 00:39:47.514
And so now you can string these components together and you can get, you know, decent quality, let's say, meeting notes.
00:39:47.864 --> 00:39:50.094
And so that was the initial focus of our product.
00:39:50.545 --> 00:39:56.635
But meeting notes and kind of searchability meetings and even just like task extraction of meetings, that was never the end goal.
00:39:56.644 --> 00:40:05.414
The end goal was how far c- how much of your work can we front based on what we understand about what's going on in your environment?
00:40:05.505 --> 00:40:20.304
And so today, we can pull all of that information, so like your meetings with your client, your conversations with your team, the things you're updating in your CRM, like the documents across your, your organization.
00:40:21.085 --> 00:40:28.074
And then it's just about you coming in and giving your intent about what you want to do with that.
00:40:28.094 --> 00:40:30.045
So let's say your intent is to sell.
00:40:30.045 --> 00:40:36.014
You literally would go in at Sembly and say,"I want this client to purchase Sembly licenses.
00:40:37.135 --> 00:40:40.175
Go." And it does research.
00:40:40.364 --> 00:40:43.195
It brings together all the information that it needs.
00:40:43.724 --> 00:40:50.295
It builds something that we call a knowledge deck, where it shows you the facts that it found, the gaps, the clarifications.
00:40:50.304 --> 00:40:53.675
You can say which facts you want to use, which ones you don't.
00:40:53.764 --> 00:40:56.675
You can just let it go, like you don't have to reply to everything.
00:40:56.954 --> 00:40:57.985
You can say,"Yeah, good.
00:40:58.034 --> 00:41:07.641
Use your research." But you can also go in and steer it to say,"Okay," like, you know- It, present to the leadership team, not to the actual CEO or whatever, right?
00:41:07.690 --> 00:41:09.791
So we'll ask you all of these questions.
00:41:10.121 --> 00:41:13.731
Then it builds a story arc, and you can accept it or give feedback on it.
00:41:14.150 --> 00:41:16.311
And then it builds the whole deck for you.
00:41:16.971 --> 00:41:22.210
And what that does is it allows you to create material.
00:41:22.661 --> 00:41:28.090
So first of all it completely abstracts your information tools.
00:41:28.090 --> 00:41:33.003
Like you don't have to touch PowerPoint You don't have to touch Microsoft Word.
00:41:33.103 --> 00:41:36.393
Like you can build a PowerPoint and you can also build a Word document in our platform.
00:41:37.023 --> 00:41:39.222
But you don't have to touch those tools anymore.
00:41:39.333 --> 00:41:41.882
Forget about being an expert, you don't have to know that it exists.
00:41:42.353 --> 00:41:44.193
It will build the thing for you.
00:41:45.213 --> 00:41:47.913
Um, you can pro- you can view it, you can provide feedback.
00:41:48.083 --> 00:41:55.262
The end result is a PowerPoint document, so if you really love going into PowerPoints and mangling with stuff, you can after it's generated.
00:41:55.322 --> 00:41:57.913
Like all the elements are completely editable obviously.
00:41:58.623 --> 00:41:59.592
But you don't have to.
00:41:59.893 --> 00:42:07.632
And so it becomes more of a partner relationship with technology rather than a tool relationship with the technology.
00:42:07.893 --> 00:42:23.592
So you have a partner that's-- You can imagine it's almost like, uh, you have a team of designer, researcher, s- y- a Po- PowerPoint creator, associate that brings it all together, and that whole team is contained inside Assembly.
00:42:24.253 --> 00:42:28.893
And you just ask it like, we want a presentation." You get back a full result.
00:42:29.202 --> 00:42:32.422
That result is bespoke for the situation you're in.
00:42:32.463 --> 00:42:41.992
So whatever client you're trying to sell to, it's going to pull information that's going to be the most effectual for that client in that situation.
00:42:41.992 --> 00:42:59.402
So it's gonna find like, oh, they're really strong in Australia, or, you know, they've been having problems in this kind of operations or whatever, or they have this size of a team and that has this implication, or they're very globally distributed and that has that implication as far as what your service can provide for them.
00:42:59.902 --> 00:43:03.282
It will bake that in into the presentation and story arc.
00:43:03.282 --> 00:43:08.853
So everyone gets a completely personalized sales pitch, if you will, on your product.
00:43:09.043 --> 00:43:14.762
If sales pitch is your goal, obviously there's many, many ways to use this, but it becomes hyper-personalized.
00:43:14.762 --> 00:43:26.463
And so one use case I like to share that we're experimenting with now is there, there are tools out there that let you de- anonymize who goes to your website.
00:43:26.463 --> 00:43:35.322
So like anyone, any time someone touches a website, you can figure out, not always, but often you can figure out who that was, which organization they work for, and things like that.
00:43:36.132 --> 00:44:04.268
And then you can use that information and you can pl- well, plug that into Assembly to generate a customized pitch of Assembly product to that organization and to that person, and then you can email them or contact them on LinkedIn with a completely custom pitch deck or, or completely custom one-pager about your product that's, that could be in their brand And all of that can be done end-to-end automatic.
00:44:05.099 --> 00:44:21.748
So anytime somebody touches your website, like in an hour or in, in a few minutes maybe, I don't know, they would get a custom one-pager that's like exactly for their situation and shows you how your product can help them.
00:44:22.369 --> 00:44:23.219
It's kind of magic.
00:44:24.199 --> 00:44:25.418
Yeah, that's super powerful.
00:44:26.628 --> 00:44:29.048
And all that has me going back to a couple things you said.
00:44:29.518 --> 00:44:40.648
The idea of intent being the star of the show now, and then also the story you shared about the super smart developers who, like, couldn't do it the one way at the very beginning of things, right?
00:44:42.088 --> 00:44:55.338
If the tools are doing all of that now, and your tool is, so much of intent comes from experience in doing it the hard way, like building that deck from scratch to know the best way to do it moving forward.
00:44:56.009 --> 00:45:03.489
And if we remove all of that, what are you seeing and where do you think it goes with like how intent gets developed?
00:45:03.719 --> 00:45:14.869
Because like essentially we've created an AI version of that, like a s- entry level associate at the investment banking firm who did all that grunt work and had to get to do all that to get to be the senior manager.
00:45:16.108 --> 00:45:23.639
Now that gets replaced, so how does one get to be a senior manager and learn that intent in this new world that's being created?
00:45:25.759 --> 00:45:28.108
Great, complicated question.
00:45:28.869 --> 00:45:31.268
Uh, I have a perspective on this.
00:45:31.298 --> 00:45:36.409
I don't know if I'm right, but I've thought about this and this is the best I've come up with.
00:45:36.418 --> 00:45:54.199
So I think first of all, what's happening is the pyramid is getting-- The, the career pyramid, which is like, you know, you start like entry level, like intern, and then maybe analyst and da, da, da, you move up and the, the pyramid is pretty, you know, equilateral, let's say.
00:45:55.128 --> 00:46:06.119
The pyramid is getting warped, and the way it's getting warped is the middle part is becoming skinnier, so it's kind of like becoming more of an Eiffel Tower rather than a pyramid.
00:46:07.099 --> 00:46:14.588
And what it means is that it's a lot s- more similar to what happens in the world of medicine.
00:46:15.338 --> 00:46:29.699
So if you think about how doctors work, they're in school for like 10 years, then they're in practicum for like another, whatever, four or five years, and then they make their first salary when they're like 67, uh, like whatever, 35, I don't know, right?
00:46:29.739 --> 00:46:38.309
But and all of that is because there's a lot, a lot of training until someone lets them touch a human being for medical purposes.
00:46:39.329 --> 00:46:50.568
And I think a lot of careers are starting to take that shift where you're basically not very useful until you're senior
00:46:51.568 --> 00:46:54.608
That's a really, that's a really, really interesting thought.
00:46:54.878 --> 00:47:03.309
You know, I've heard, I've heard people say the pyramid is in-- for those of you who are listening and can't see, we're making hand gestures of pyramids on the, on the video right now.
00:47:03.719 --> 00:47:10.398
But the pyramid, with a wide base is turning into a, a smaller base, uh, which is opening up more senior level jobs up top.
00:47:10.929 --> 00:47:24.161
But what you're describing, I think, is is different than that, and I also think it's probably more accurate this idea of like and we've thought about it here.
00:47:24.661 --> 00:47:27.260
You know, you think about we've been doing this for 16 years.
00:47:27.260 --> 00:47:32.940
You know, you would hire a, a specialist entry-level, they would jump right in and do the work.
00:47:33.061 --> 00:47:35.510
You know, maybe there was some training that would-- that happened.
00:47:35.561 --> 00:47:41.340
You know, there's some oversight, but in general, they're doing they're doing the work at a more junior level.
00:47:41.550 --> 00:47:47.561
And then it kind of ladders up, and then you have, you know, it's a traditional kind of knowledge professional services firm.
00:47:49.351 --> 00:47:52.590
When you think about a doctor, you know what?
00:47:52.590 --> 00:48:02.690
They're at a practice, you may have a few doctors, a handful of them, and then, you know, you have a team of people that are, that are executing against that.
00:48:03.201 --> 00:48:11.130
And a-- and then this period of time where you're thinking about a 10 or 15 year period of time from school all the way to apprenticeship to whatever it might be.
00:48:11.130 --> 00:48:19.701
We've, we've thought about adding apprenticeship here for entry level where you just sit and shadow someone for a year, uh, before you do anything.
00:48:19.981 --> 00:48:20.831
It would be expensive.
00:48:20.831 --> 00:48:21.510
It would be hard.
00:48:21.960 --> 00:48:27.911
And the, the, the society is not ready for it yet, but I do feel kind of feel like that's where we're going.
00:48:27.911 --> 00:48:33.501
It's a very interesting-- I've never heard anyone describe it as a, a, with the analog of a doctor.
00:48:33.501 --> 00:48:35.190
I think that's, I think that's exactly right.
00:48:35.570 --> 00:48:41.710
I think the bottom of the pyramid is, is still wide in the sense that there's always grunt work to do.
00:48:42.920 --> 00:48:44.161
You know, you always need scrubs.
00:48:44.251 --> 00:48:47.710
You always need like just grunt work things.
00:48:48.510 --> 00:48:58.471
And whenever you touch reality, it's going to be a very, very long time until everything can be a-absorbed by AI.
00:48:59.081 --> 00:49:16.971
So famously, you know, AI is really great at information things, but whenever there's physical like and u- real world endpoints like you actually like talk to a customer or you send an email or whatever it is, uh, AI still can't meet you for coffee and all that stuff.
00:49:16.971 --> 00:49:21.521
So there's still things that happen in the real world that you need people for.
00:49:22.161 --> 00:49:24.400
And there's still grunt work that needs to be done.
00:49:24.431 --> 00:49:27.621
I call it grunt work, but it's really just junior work that needs to be done.
00:49:27.621 --> 00:49:28.891
And I think that's not going away.
00:49:28.911 --> 00:49:31.820
Like you'll always need like junior things done.
00:49:32.280 --> 00:49:37.710
And the cool thing now is that the juniors can do a lot more junior things well.
00:49:39.101 --> 00:50:03.530
The problem is when you, when it comes to like that important customer that you need to make a sales to, sale to or that important architecture decision or, you know, that product inception and product shape, that m- go to market perspective You need to have experience for that, and you need to have that, that acquired taste.
00:50:03.630 --> 00:50:11.851
Taste, I think taste is something that AI has to an extent, but will have a lot more of.
00:50:13.130 --> 00:50:45.063
But I think there there's th- there there's a, there's an element of subjectivity and I'm sure AI will-- there will be AI out there that can come up with good taste as well, but I think that's an area that's a little bit more chaotic that, will need that experience of having gone through things and understanding how things work in a very, in a very broad way, but also in a very, like, zeitgeist way.
00:50:45.503 --> 00:50:51.682
Because, at least today, one big limitation of AI is that its zeitgeist is always a little bit outdated.
00:50:52.762 --> 00:50:55.353
Um, it's trained on a training set that has a date.
00:50:55.742 --> 00:51:09.563
Now, of course, it can do research, it can figure things out, but it's-- what it knows, it's still, uh, grounded in its original training data, which is, which is aged the day after it's trained.
00:51:10.043 --> 00:51:11.253
So there's still that gap.
00:51:11.932 --> 00:51:25.043
But yeah, I think you'll need experience and seniority to make those important decisions, and you need time to be able to rise in your career ladder, um, and have experiences before you can do that
00:51:25.092 --> 00:51:34.873
I got a, got into an argument with someone, uh, you know, where I was talking about, the irreplaceably part-- the irreplaceable human part of what we do.
00:51:35.143 --> 00:51:45.193
And, you know, this, uh, very smart guy, CEO of a, of a software company, he's like,"What do, what, what do human what do humans need to do?" And so I started going.
00:51:45.193 --> 00:51:50.702
He's like,"Taste?""No." He's like,"Today." I'm like, he, I'm like...
00:51:50.762 --> 00:51:55.782
He's like,"Toda-today, but in a year it will have better taste than us.
00:51:55.822 --> 00:52:04.103
Judgment?""Nope.""Creativity?""Nope.""Curiosity?""Nope." And I s- you know, so I started going through all of the traditional things we think of.
00:52:04.492 --> 00:52:06.242
And so it, it's gonna be an interesting...
00:52:06.242 --> 00:52:07.742
I mean, obviously no one knows for sure.
00:52:08.083 --> 00:52:10.242
You know, no one is is seeing what we're doing.
00:52:10.282 --> 00:52:19.463
It th-there is one thing for certain: it will come fast faster than we expect probably, and it will be exciting along the way for sure.
00:52:19.632 --> 00:52:23.143
So as Patrick just said, it's probably gonna come faster than we think.
00:52:23.652 --> 00:52:30.532
So as your team thinks about that and on continuing to iterate and innovate in the market, what's next for Sembly AI?
00:52:30.532 --> 00:52:35.663
What things are you doing today to prep for what's coming in the, in the market in the future of AI tomorrow?
00:52:35.972 --> 00:52:47.112
So we ha- we have a very strong meeting intelligence offering from the past, and in that offering, every year we've had to innovate because the market is moving so fast.
00:52:47.733 --> 00:53:00.972
And a couple of years ago, we built this thing called Semblyin, which is, you can go into Sembly and you can ask anything about all the meetings you've had, and you can even ask it to create you a document based on those meetings.
00:53:01.052 --> 00:53:10.822
And so, Semblyin could create, you know, a statement of work, it could create a proposal, it could create an agreement, but it's always, it's not branded.
00:53:11.402 --> 00:53:16.603
It's very good from a content perspec-perspective, but it's not styled in any way.
00:53:17.012 --> 00:53:19.713
And so as an internal document, it's very good.
00:53:20.402 --> 00:53:27.012
As an externally shared document, you'll need to spend some time making it look like the way you want it to look.
00:53:28.358 --> 00:53:39.369
But that part of the product received great attention from our users and a lot of applause because it-- that's the piece of the product that saves them the most time.
00:53:40.329 --> 00:53:43.559
And so we took that idea and we really ran with it.
00:53:44.389 --> 00:53:59.329
And so this year we introduced what we're calling Sembly and 3.- sem- sorry, Sembly 3.0, and we added a new area to our product called Studio, which is now the focus of our product.
00:53:59.748 --> 00:54:05.668
So whereas historically the focus was meetings, we now think of meetings as a data source.
00:54:05.998 --> 00:54:10.818
So our entire historical product has become a feature- Hmm.
00:54:10.829 --> 00:54:13.449
Mm-hmm in our new generation of platform.
00:54:14.088 --> 00:54:28.168
And what Sembly 3.0 is about, it's about you being able to go in and putting your intent into what you want the material to do, and that material could be a PowerPoint or it could be a Word document.
00:54:29.429 --> 00:55:01.268
And Sembly will do-- will connec- will connect all the context that you have will do the research, will allow you to steer if you want to steer, and then will produce the end result that's on brand, very impactfully organized, beautiful styling interestingly visual, pr- interesting visual presentation, and ultimately effective at communicating to that client what you want to accomplish.
00:55:01.509 --> 00:55:08.259
And so in Sembly 3.0, we have an area, for example, of it understands clients and understands client context.
00:55:08.259 --> 00:55:13.088
So you, you can es- effectively create buckets for each one of your clients.
00:55:13.108 --> 00:55:16.878
You can create new clients and put materials related to those clients there.
00:55:17.588 --> 00:55:21.259
You can derive branding from just a website.
00:55:21.278 --> 00:55:29.088
So if a client has a website, you could just hit derive, and it will pull the branding information from that client.
00:55:29.099 --> 00:55:32.579
You can also upload materials from the client, and it will pull information from that.
00:55:33.498 --> 00:55:35.259
And it's not just visual stuff.
00:55:35.259 --> 00:55:39.599
Of course, it's colors and, and logos and fonts, but it's also tonality.
00:55:39.599 --> 00:55:42.309
It's, you know, how that client communicates.
00:55:42.309 --> 00:55:43.378
Are they more casual?
00:55:43.389 --> 00:55:44.289
Are they more serious?
00:55:44.338 --> 00:55:45.048
Et cetera, et cetera.
00:55:45.048 --> 00:55:56.764
So it's a-- It builds what we call a branding package And then when you're creating a new presentation, you can say it's for this client, and here's the branding package I'd like to use.
00:55:57.355 --> 00:56:02.275
And when you say it's for this client, it's going to pull all the context about that client that you've built over time.
00:56:02.864 --> 00:56:08.554
And when you pick the branding package, you can select your own branding, you can select the client's branding or whatever you want.
00:56:08.934 --> 00:56:12.715
It will br- it will build it in that brand that you select.
00:56:13.215 --> 00:56:24.954
So it's really a turnkey way to create brilliant presentations and agreements and documents that are client-ready.
00:56:25.034 --> 00:56:26.514
They're ready to be shared.
00:56:26.664 --> 00:56:31.175
Now, again, you can open them in PowerPoint, you can open in a Word, you can monkey with it afterwards if you want.
00:56:31.695 --> 00:56:34.324
You can actually interact with Sembly about it.
00:56:34.324 --> 00:56:36.034
So once it's generated, you can talk to Sembly.
00:56:36.034 --> 00:56:40.344
You can say,"Actually, could you add one more slide that has this, and that?" And it will do that for you as well.
00:56:41.105 --> 00:56:42.505
So it's hyper-specialized.
00:56:42.894 --> 00:56:51.184
We're now rather-- So historically, we were positioned against companies like Fireflies and Fathom and those kinds of products that are very meeting notes focused.
00:56:51.775 --> 00:57:03.454
This year, we have jumped into a different competitive matrix, and now we're positioned against companies like Gamma, GenSpark, to some extent against Claude, ChatGPT, and Copilot.
00:57:03.965 --> 00:57:08.105
The big difference with us is that, first of all, we never start from scratch.
00:57:08.105 --> 00:57:11.344
So if you go into Gamma, if you go into Claude, you're starting from scratch.
00:57:11.344 --> 00:57:13.094
You need to build that context up.
00:57:13.454 --> 00:57:16.204
Gamma doesn't have a great way of dealing with context at all.
00:57:16.545 --> 00:57:21.635
Claude can deal with context, but every time you're starting new, so you b- e-every presentation is a project.
00:57:22.105 --> 00:57:23.795
With us, you're never starting from scratch.
00:57:23.804 --> 00:57:29.945
So we know about all your meetings, we know about the client material, so we already have that when you're building the presentation.
00:57:30.224 --> 00:57:31.755
The second thing is branding.
00:57:32.135 --> 00:57:39.635
All the products in the market today are-- So Gamma is great, but it's very-- it's got a very creative tilt to it.
00:57:40.005 --> 00:57:50.784
So it's good for creatives, but it's not really great, I personal opinion, for professional services p- who have consistent branding requirements and very strict branding requirements.
00:57:51.155 --> 00:57:53.764
Gamma has a tough time with things like that.
00:57:54.144 --> 00:57:57.355
Claude has a very difficult time, surprisingly, with things like that.
00:57:57.405 --> 00:58:01.844
I mean, he can try to give you branded things, but it's gonna be different every time.
00:58:01.844 --> 00:58:07.744
It's unreliable, and sometimes it just can't get there, and you have to sit there and prompt and prompt until you're blue in the face.
00:58:08.175 --> 00:58:10.445
With Sembly, you just get it turnkey for free.
00:58:10.445 --> 00:58:11.804
It's, it's there with you.
00:58:12.215 --> 00:58:18.385
And the coolest thing is we were able to do all of this and offer all this without changing our price.
00:58:18.494 --> 00:58:25.795
So we've kept all of our standard pricing plans that you are, that you find in all the common meeting intelligence products.
00:58:26.264 --> 00:58:42.630
But now you get meeting-- all of end-to-end best-in-class meeting intelligence, AKA like your meetings from Google Meet, Zoom, Teams, WebEx, all the meeting notes that you want, the tasks, all of that is still there But that's just a little feature in our product now.
00:58:42.760 --> 00:58:47.951
Whereas the main thing is you go in and you go into the studio area and you can just say,"Create new presentation.
00:58:48.400 --> 00:58:56.871
I wanna sell my product to Acme Co." Click go, and in a couple of minutes you get beautiful slides.
00:58:57.570 --> 00:58:59.771
So that's, that, that's where we are today.
00:58:59.780 --> 00:59:07.811
We're really embracing this idea of intent technology, and as we go forward, we will make it bigger, better, faster.
00:59:07.811 --> 00:59:15.791
So it's gonna be even more visually beautiful, more temp- more c- give you more variability of templates and things like that you might wanna use.
00:59:16.141 --> 00:59:24.070
But I think as we go into next year, we're going to start to see more proactive things, so where you don't even need to go in and ask.
00:59:24.451 --> 00:59:32.271
So for example, Sembly could notice in a meeting that a client interaction happened and it can suggest,"Hey, you just had this meeting.
00:59:32.860 --> 00:59:40.521
Here's a, you know, here's a proposal you can send to the client as a result of this meeting," and it will pre-gener- it will just generate it for you in the client's branding.
00:59:40.590 --> 00:59:41.391
It's gonna be crazy.
00:59:42.190 --> 00:59:43.561
So that's where we are today.
00:59:43.561 --> 01:00:05.431
We're very excited about it, and we really think it's in vogue with where the software space is heading, which is away from tinkering in PowerPoint and more about what do I want to accomplish and what's, you know, what kind of materials do I need to get there?
01:00:05.960 --> 01:00:09.291
And, uh, Sembly is really positioning into that space
01:00:11.317 --> 01:00:12.536
Yeah, that's super interesting.
01:00:13.016 --> 01:00:16.126
It's really a great, like to name drop our own podcast here.
01:00:16.126 --> 01:00:18.067
It's a great"Good Enough Isn't" story, right?
01:00:18.157 --> 01:00:24.157
You could have stayed with what you were doing just fine and competed in that space, but you made the choice to say,"That's not good enough.
01:00:24.556 --> 01:00:31.907
We're gonna add on and continue to reinvent ourself," which you've done throughout the history of, sounds like of yourself as an individual, but then also with Sembly as well.
01:00:31.996 --> 01:00:36.427
Every time the market has shifted, you've shifted right alongside of it, which is really impressive.
01:00:37.817 --> 01:00:38.186
We try.
01:00:38.197 --> 01:00:46.197
So for people listening to this podcast that want to find out more about you or find out more about Sembly, what, what links should we give them in the show notes?
01:00:46.197 --> 01:00:47.197
Where do you want us to send people?
01:00:48.246 --> 01:00:49.577
The website is the main thing.
01:00:49.577 --> 01:00:52.367
It's www.sembly.ai.
01:00:52.376 --> 01:00:55.016
That's S-E-M-B-L-Y.A-I.
01:00:55.847 --> 01:00:58.007
All the latest and greatest is there.
01:00:58.737 --> 01:01:01.067
On- if you want to reach out to me directly, it's LinkedIn, Artem Koren.
01:01:01.067 --> 01:01:01.916
That's A-R-T-E-M K-O-R-E-N.
01:01:01.916 --> 01:01:05.967
Uh, and that's probably the best way to reach to me directly.
01:01:06.277 --> 01:01:08.277
There's a free trial available on the site.
01:01:08.317 --> 01:01:14.996
You get a bunch of credits to play with when you sign on, and, uh, encourage you to give it a, to give it a go.
01:01:14.996 --> 01:01:19.427
I think you're, you'll be very impressed when you do.
01:01:19.456 --> 01:01:28.996
It's, it w- it's interesting because when we were rolling out this new area of product we did a closed beta, and the response was so consistent.
01:01:29.007 --> 01:01:32.847
So it was an invite-only beta, and the respon- was- response was so consistent.
01:01:33.797 --> 01:01:38.036
Everyone was like,"When can I get my hands on this for real?" That's awesome.
01:01:38.036 --> 01:01:42.036
And that give us, that gave us a good sense that I think, we have something special here.
01:01:44.150 --> 01:01:44.490
That's great.
01:01:44.561 --> 01:01:45.181
You heard the man.
01:01:45.201 --> 01:01:48.661
Go check out the links in the, in the show notes and try it for yourself.
01:01:48.701 --> 01:01:49.891
And Artem, thanks for your time.
01:01:49.891 --> 01:01:52.610
It's been great hearing your story and learning about the company.
01:01:52.800 --> 01:01:55.130
Very interesting conversation, so appreciate you being here with us today.
01:01:56.221 --> 01:01:56.641
Thank you.
01:01:57.090 --> 01:01:58.641
Yeah, super, super fascinating, man.
01:01:58.641 --> 01:01:59.340
Thank you.
01:01:59.391 --> 01:02:02.581
And just, uh, I'm gonna, I'm gonna stay away from all the online chess.
01:02:02.751 --> 01:02:12.570
I feel like you- you probably are also playing on chess.com just under a, a pseu- uh, a, a pseudonym or something- An alias, yeah so an alias, so, uh, I'm just gonna stay away from all online chess for a while, I think.
01:02:13.021 --> 01:02:13.840
No, you're fine.
01:02:13.840 --> 01:02:14.360
You're fine.
01:02:14.431 --> 01:02:15.251
Listen, it's Bullet.
01:02:15.271 --> 01:02:16.831
Anything can happen in Bullet, so.
01:02:16.911 --> 01:02:17.460
There you go.
01:02:17.610 --> 01:02:18.061
There you go.
01:02:18.181 --> 01:02:18.840
Thank you so much.
01:02:18.840 --> 01:02:20.130
Super fascinating conversation.
01:02:21.050 --> 01:02:21.721
All right, everyone.
01:02:21.780 --> 01:02:23.090
That's a wrap for this episode.
01:02:23.110 --> 01:02:27.661
Be sure to hit the subscribe button so that you never miss an episode of Good Enough Isn't, and we'll see y'all next time.