YT YouTube2Text

YouTube2Text → Transcripts → #108 【全編英語回】The Ai Workforce Team Goes Global: Voices from the Engineers Behind Agentic Workflows...

#108 【全編英語回】The Ai Workforce Team Goes Global: Voices from the Engineers Behind Agentic Workflows... — Transcript

by LayerX 公式 · 5,827 words · 871 segments · language ja · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00[Music]
0:03Hi everyone and welcome to Reax. Now
0:07this is our very first episode in
0:09English. So thank you so much for
0:11joining us. I am your host today. My
0:14name is Rya and I am the head of the AI
0:17and department at Leia X. For those of
0:20you who may not be familiar with Leia X,
0:22let me give you a quick introduction.
0:25Leia X is a Tokyo based AI and fintech
0:28startup founded in 2018. We currently
0:31operate three main businesses. First, we
0:35have Bakuraku, a business spend
0:37management service platform that helps
0:39companies streamline their expenses and
0:41operations. Second, we run a joint
0:44venture with a major financial
0:46institutions focusing on asset
0:49management. And third, which is the main
0:52topic of today's episode is our AI and
0:55LM business. In this space, we provide
0:58an enterprisegrade AI platform called AI
1:01workforce, which has already been
1:02adopted by some of Japan's leading
1:04companies, including MUFC and Mitsui.
1:09Today we have two amazing guests from
1:11our team, Isami and Tim. Let's start
1:14with a quick introduction from each of
1:16them. Isami, would you like to go first?
1:20Yes. Uh, hi everyone. Uh, my name is
1:23Isami.
1:25Uh, I'm the tech leader at Leia X. Uh, I
1:30started my career as a software engineer
1:32in
1:332012.
1:35In 2019, I moved to Vancouver and worked
1:39as a founding engineer at a SF based D2C
1:44startup. Later I found fun founded a
1:48startup in San Francisco raised $1
1:51million and they developed SAS for the
1:55logistics industry.
1:59Cool. And you said ramen, right? H Yeah.
2:05So you're a ramen expert? Yes. Yeah.
2:08What's the best ramen place in Tokyo?
2:10Ramen maybe Yamongaya.
2:14I don't know.
2:16Yeah.
2:18By the way, why are you called Isami?
2:21Uh, Isami is a a Japanese restaurant in
2:25Vancouver.
2:27I used to go there.
2:30Cool.
2:31Thank you. Um, Tim, could you introduce
2:34yourself?
2:36Uh, yeah, sure. So, my name is Tim
2:38Mansfield and
2:41uh I've been doing different kinds of uh
2:44software engineering for a long time
2:47really since the original.com
2:50boom long ago. I've worked at um as a
2:54tech lead at Google and YouTube a long
2:57again a long time ago and I've been at
2:59many startups uh since then in different
3:01verticals um ranging from fintech to
3:05edtech uh green tech and um you know
3:10construction software all sorts of stuff
3:13uh so yeah it's been a really
3:15interesting run and uh I have moved to
3:20Japan as of about a year ago and I had
3:25been in Japan um a long time ago as
3:28well. Uh but now that I've moved here um
3:34I have wanted to work with a Japanese
3:37company um and uh so yeah that kind of
3:41brings me to uh here to this podcast.
3:46Thank you. So um you were at the YouTube
3:51in its early days. Uh yeah that's right.
3:54Uh not the very very earliest days but
3:56after it was acquired by Google uh I
3:58joined that team. Did you expect did you
4:01expect you know YouTube to become such a
4:03huge platform? Oh well it was uh pretty
4:06hot uh as of the time of acquisition as
4:08you can imagine which is why it was
4:09acquired. Um, so honestly I'd barely
4:13even heard of it before uh before that
4:15because there were other uh video
4:17startups at that at that time and I
4:19wasn't involved in video at all. Um, but
4:22yeah, having uh joined that team
4:25postacquisition uh by Google uh it was
4:29uh really cool to to see that space. I
4:32see. So I know you speak Japanese very
4:35well. Would you mind speaking a little
4:37Japanese right now?
4:40Um just uh about what?
4:45Okay. So, what's your favorite place in
4:47Nakamero? My favorite place? Oh, you
4:50mean like uh the ramen question? Uh
4:53yeah, sure. Um let's see.
5:29hinge.
5:32Yeah, it's really good. Okay. Yeah, it
5:35was actually recommended to me by uh
5:37Parisians who said that it was as good
5:39as anything they had in Paris. So that
5:40was a pretty strong recommendation and
5:42it is very good. What's your expert
5:45opinion on?
5:49It's it's okay.
5:56Awesome. Thank you. Um so let's start
5:58our conversation with a simple question.
6:01Why did you decide to join Dax
6:04team? Could you go first? Oh, yeah,
6:07sure. So, um I mentioned that I uh moved
6:11here uh only a year ago, but I do intend
6:15to stay here for a long time. My wife is
6:17Japanese. I have two young uh children.
6:20U the older one is about to turn five
6:23and she speaks Japanese and English uh
6:27you know, fluently on both sides. and we
6:29intend to be here for a long time. As
6:31such, I would like to contribute to the
6:34ongoing stability of the Japanese
6:36economy. And you know, Japanese society
6:38is famous for being safe and uh you
6:41know, all this stuff. It's it's great,
6:43but of course the the nice features come
6:46at a cost. And I think it's important
6:48that Japanese uh business is able to
6:52support through you know profits and
6:55taxes paid on profits is able to support
6:58that that uh the economy and the society
7:00that um I want to be in. So that's kind
7:03of like a real high level thing. And
7:04then otherwise layer specifically um I
7:08really like the uh the global expansion
7:11strategy. I think a lot of Japanese
7:12companies, they they have dreams of
7:15going global, but it's really hard to do
7:17that actually for a typical Japanese
7:19company. And Larex has a very good
7:20strategy for doing that. Um, which I
7:22don't know if we want to get into what
7:24that is, but it it has a it has one that
7:26I believe is actually viable.
7:30Yeah. So, it's considered to be hard for
7:34especially startups. So, let me explain
7:36a bit on the, you know, grow expansion
7:38strategy you mentioned. Um first of all
7:42expanding AI workforce globally has been
7:45our dream and um since many of our
7:49customers are large enterprise in
7:52Japan with global operations. So we have
7:55already started providing air workforce
7:57to their overseas branches. So building
8:00on that track record we are now
8:02preparing to launch our sales and
8:03marketing efforts.
8:06Mhm. Yeah, it makes a lot more sense
8:08than just with no particular uh sort of
8:12advantage, local advantage, just trying
8:14to opening up an office in San Francisco
8:16and then just going for it. That's
8:18probably not as good of a strategy.
8:22Right. Thank you. How about you, Isami?
8:25Uh so yeah, it's it's kind of same, but
8:29yeah. Yeah. My my plan was uh going to
8:33the San Francisco to continue my
8:36previous setup. Oh, really? Yeah. Yeah.
8:40Yeah. Because I I already have have the
8:44H1B visa to grow to work in the States.
8:50Yeah. But the that
8:54the side of it is difficult. So we we
8:58didn't have clients but but I wanted to
9:03develop
9:05uh something using a
9:08but we didn't have just clients. So um
9:13and I I I met Matosang
9:16Mazan CO of the of Lelay X and he
9:21mentioned
9:23uh
9:26uh Lelay X is going to expand. Yeah. So
9:31I I was really excited. So I decide to
9:36come back to Japan. M Yeah. Because
9:39Yeah. There are many good restaurant
9:44ramen restaurant that's ramen.
9:47Yeah.
9:49Yes. So ramen inspired you to work on AI
9:53and make you to move Japan. Yeah.
9:57Yeah. also um
10:00my uh my goal is um I I I I want to uh
10:06work work on the product Japanese
10:10product. So I want Japanese product and
10:13make Japanese product global. So I I
10:17wanted to join the team and that's why I
10:20joined. Yeah, I think you know many
10:23people in my team had a similar reason
10:26to you know come to player X. They have
10:30spent a lot of time to work on foreign
10:33product but they you know at the end of
10:36the day try to do something like you
10:39know Japanese
10:41forecast software. Yeah. Yeah. Also I I
10:43think uh in particular so LM has uh many
10:49potential.
10:51Yeah,
10:53cool.
10:55Thank you. So
10:57now, so you both of you started to work
11:00in layer X this winter I think. So what
11:05is your first impression of layer X
11:08especially in terms of culture and
11:11technology?
11:13Um Tim Tim, could you go first?
11:16Uh yeah.
11:18So
11:20um I I've talked about this before in
11:22another context, but um uh as I
11:27mentioned, I've been at a lot of
11:29different startups and bigger companies
11:31too. Um not not just Google. And uh uh
11:35so I've seen many different approaches
11:37to managing development at different
11:40stages of the growth life cycle. And I
11:44would say that um in honesty what I've
11:47seen so far for where the AI LLM
11:50business unit um or I guess it's a kind
11:53of a company within the group
11:56uh within LREX where it's at it seems
11:59perfectly normal to me for for uh this
12:03stage of growth. And so there's always a
12:06balance between trying to um move
12:08quickly and prototype and get things out
12:10there and sort of approach things with
12:13uh a lean startup approach or whatever
12:16um so that you get validation from
12:19contact with reality as quickly as
12:20possible. So there's all that but then
12:23you know this is actually serving some
12:25really big customers. some of the
12:26biggest companies in Japan um are using
12:29Layer X's software and so you you have
12:32to kind of balance the move fast and
12:34break things so to speak with not
12:37actually breaking things. Um so that's a
12:40difficult balance to strike and I think
12:42you know Larrex has been doing fine as
12:44far as I can tell. Um, and
12:48uh, yeah, I mean I haven't been here for
12:51very long as you as you just mentioned.
12:53So I'm still I think getting used to
12:56everything and this is also the first
12:57Japanese company that I've joined. So
13:00um, you know, there's also that. But
13:02yeah, it's been a great learning
13:04experience so far. So we sometimes talk
13:06about the stereotype of Japanese
13:10software company. So what is the
13:12definition of stereotype with Japanese
13:14software company and what do you think
13:16the difference between us and them?
13:19Ah yeah well the ultimate stereotype of
13:22a Japanese company is the um you know
13:25all the so-called like sire
13:29kind of companies the system integrators
13:31where it's like they originally were
13:33part of the big company groups that
13:36themselves used to be zybatsu or
13:38something and so they were kind of
13:39there's these ecosystems where Japanese
13:42software uh companies are kind of
13:44protected from too much competition
13:47because they end up serving the group
13:49other group companies and so they have
13:51automatic business that whole thing um
13:54that that phenomenon I think really
13:56stifled Japanese uh software development
13:59um that's not the whole picture because
14:01there's amazing stuff going on in other
14:03areas uh you know Japan's famous for
14:06uh cutting edge gaming and so on and so
14:08forth lots of different areas
14:11but when it comes to regular retail
14:13software development I think that whole
14:15sire thing really pulled Japan back Um
14:20so yeah by contrast with that of course
14:23there are it's not just layer X there
14:25are there are you know many companies in
14:27Japan that aim consciously to have a
14:31modern approach to things and not like
14:33that the opposite of that and lyrics is
14:36definitely one of them so um yeah that's
14:39how I think I would put the contrast is
14:42s on the one spec end of the spectrum
14:44and then really being uh essentially the
14:48same as uh you know American or European
14:50startups on the other end of the
14:51spectrum and layer X is on that end.
14:53Yeah. Yeah. Mhm. I think we can learn a
14:57lot from SI as well but we have to you
15:00know be different from them as well.
15:02Yeah. I don't honestly I don't know much
15:04about it. It's just reputation and rumor
15:07um and so forth. So yeah, I I've never
15:11worked uh for or with uh those
15:14companies. So uh I could be completely
15:16wrong.
15:17I I don't mean to uh badmouth people who
15:20are probably working very hard um and uh
15:23are just as smart as anybody else.
15:25There's just something systemic about
15:27that part of uh Japanese uh software
15:30industry that I've heard is uh yeah is
15:34maybe slower to adopt things and it's
15:37you know kind of harder to work with
15:38somehow. Yeah.
15:41Yeah. So, Isami, you started your career
15:45in Japanese large intent companies. Yes.
15:49From that experience, what is your first
15:51impression with Leax?
15:54Yes.
15:56I feel it's it's kind
16:00flexible.
16:02Yeah. Also, yeah, I I feel
16:07that we we want to improve our product.
16:11Mhm. So we we we focus on pro product
16:14and client
16:16customers. Maybe that's the different
16:19difference between their first company
16:21and lay x uh focusing on client client.
16:26Yeah. And product sometimes Japanese you
16:29know intent company driven by sales
16:32department. Yes. Yes. Revenue. Yeah. I
16:35think that part of the reason why it is
16:38is you know read the shape of AI X is
16:40mostly their back background is
16:42engineering. Oh it's like it's like
16:45Google right? Yeah that's right. The um
16:48the leaders were generally uh engineers
16:53a couple of exceptions but um it's very
16:55engineering at its core. So for example
16:57um on the uh YouTube team that I was the
17:02um tech lead of, we had 15 engineers and
17:06uh we only had part of a product manager
17:10like one and basically we made the
17:14decisions for what to do with some
17:16oversight
17:18uh from the product management team but
17:19they were very very small in ratio to
17:21the all the engineers. Mhm. So I would
17:26say that there's something similar at uh
17:28the especially the earlier part of the
17:31AI LLM business.
17:34Um wouldn't you say that's true
17:36Nakamura? Yeah. So what do what do you
17:40think is the actual effect of having
17:43your engineer in the leadership team?
17:46People say you know it is a good thing
17:48but there's less discussion about why it
17:51is.
17:54Well, it's not perfect. It's probably
17:56better to have uh you know, if you can
17:59if you have enough good people uh at the
18:03very beginning of a startup, then it's
18:05good to have a mix. Um but yeah, I guess
18:09one thing is um we could say is uh and
18:13of course we're biased because the three
18:15of us are all engineers so um we're
18:18patting ourselves on the back but having
18:20a very systematic approach I think um
18:24that's one plus I think u and generally
18:28being uh you know quai scientific
18:32approach means that you're often looking
18:34to falsify your own um you're looking to
18:38challenge your own uh statements and
18:41other people's statements and so that
18:43can be a little bit annoying going back
18:45and forth and challenging each other
18:46like that. But I think we are seeking um
18:50a lot of the time if we're good we're
18:53not just trying to get our own way but
18:55we're trying to really find uh optimal
18:58solutions.
19:00Um, so that's kind of like the
19:01engineering approach to things. And you
19:03have to make a lot of trade-offs too, I
19:05would say as well. Uh, you know what I
19:08mean? It's not like u pure research in
19:10that way. Like you you have to deal with
19:12constraints and so we're pretty good at
19:14dealing with that. So it kind of maps
19:16well to a to a business um space as
19:21well.
19:23Cool. Yeah. So before we move on to the
19:28next topic, we have a big announcement.
19:31So actually the very reason we are
19:33recording this episode in English is
19:36because we recently started hiring
19:38English speaking team members. Yay. Yes.
19:41So this is a big big shift for us. We
19:46have decided to open up engineering
19:48positions to candidates who don't speak
19:50Japanese. Um the goal isn't just to
19:53strengthen our development team but also
19:56to lay the foundation for the future
19:59global expansion of our business as I
20:01explained. So it's a bold step for
20:04Japanese startup. So I'm super excited
20:06about it. So Isami, you are the one who
20:10advanced this shift in in my team.
20:13What's your thoughts on this change?
20:16Uh yeah, I really I'm really excited
20:19about it. Excited about it. Yeah. Also,
20:23yeah, I I think it's it's really
20:26difficult decision because
20:29Japanese
20:31speaks only Japanese, right? So, so
20:34yeah, there is a huge language language
20:38barrier. Mhm.
20:40So maybe so so m maybe we we need to
20:45update many many things for example
20:48document
20:50meeting or something
20:52yeah that's what you know I'm curious on
20:56the current situation of engineering
20:58team um team how English and Japanese
21:02are used in our team right now
21:05uh yeah so the uh I would say that the
21:09ratio of language use uh is pretty
21:12dependent on the context.
21:14So um of course each person has a
21:18different mix of uh you know skill
21:21levels or comfort levels when it comes
21:23to speaking uh reading writing and
21:26listening. And uh I think in my
21:30experience so far the team members vary
21:33but um generally hearing and reading are
21:39pretty good and then the speaking part
21:42is a little bit harder. um depending on
21:44the person uh but and then writing is
21:48sort of in between but depending on who
21:52the participants are I have not found
21:55any real difficulties because um we can
21:58accommodate each other with different
22:01tools and different techniques for
22:04making that accommodation it's it's
22:06worked out fine so far
22:09I think most of the Japanese member talk
22:12to you in Japanese, but you talk to them
22:15in English. Yeah, I'll I'll speak in a
22:17mix of Japanese and English. Um, I'm
22:21definitely obviously way more
22:22comfortable uh reading, writing, and
22:26speaking um English. My listening of
22:30Japanese is is pretty good. Uh so
22:33similar to, you know, you guys when you
22:35hear English, you're you're very good.
22:37Um but then when I try to speak, it's
22:40just way slower. um and uh roundabout
22:44like Ma Koi. Um compared to you guys, I
22:47mean, I'm just blown away at how uh you
22:50know, brilliant and fast and uh precise
22:54uh you guys are at communicating. Um and
22:57it's not just the fact that it's your
22:58native language. I think there's a
23:00different approach to communication. Um
23:03and you guys have uh some ways that I
23:06can describe later or whatever if it
23:08matters. But um but yeah, it's pretty
23:11it's pretty good stuff. So um yeah,
23:14anyway, we we find ways to uh get around
23:18the language barrier that Isaman you
23:21were talking about. Um however, it does
23:24it admittedly depending on the
23:27communication route, we do it does slow
23:29things down a little bit. Um but I think
23:31there's enough gain to be had um that it
23:35it makes it worth it.
23:38Um yeah so for example uh we use a lot
23:42of uh AI support. Yeah there was there
23:46was a little bit of processing time to
23:47applying the AI support. Uh so that
23:51slows down document creation a little
23:53bit but um yeah we added a translation b
23:56in what is called ling ling
24:00uh lingx or something ling. Yeah I think
24:03that is so cool. Yeah. Yeah. We um I
24:07think it was Inawasan that investigated
24:08that and brought it in. So one of the
24:11product managers uh you know brought
24:13that in and so a few different people
24:16have been exploring different ways to
24:17make it easier. So it's not all on for
24:20example all on me or all on Isaman or
24:22whatever to figure it all out. So the
24:24whole team is um trying to make it
24:26easier and accommodate. So I appreciate
24:28that um as the first native English
24:31speaker I think on this team. Um yeah
24:35and we also I mentioned the AI support
24:38in notion. We use notion AI. Uh this was
24:41actually Isaman's uh contribution. He
24:44figured out that we could use that to um
24:46automatically create dual language
24:49versions of our documents. So you can
24:51just uh drag select some text and run
24:54the an AI prompt on it and it will um
24:57create side by side translations which
25:00makes it really easy for uh either um
25:03party to skim and pick up the gist so
25:06that we can make it intelligent comments
25:08about it um without you know that extra
25:12overhead of having to uh run the
25:15translation yourself over and over and
25:16over. Yeah. Yeah. And just to throw out
25:18one more thing, uh let's see. Uh why
25:22Matsusan
25:24uh he invested in a Japanese startup
25:27that has some pretty amazingly fast uh
25:30real-time translation technology. So I
25:33installed that on my phone as well and
25:36it it really is incredibly good. Yeah.
25:39Yeah. I wouldn't be surprised if they
25:40get acquired like like next week. Yeah.
25:43Maybe you can use that in the next
25:45offsite, you know, for documents. It's
25:47easy to to have both.
25:50For presentation, you have to pick
25:52either Japanese or English in general.
25:54So, Right. Right. Yeah.
25:58Yep. Yep. So, now let's talk about
26:01timing. I want to ask both of you what's
26:05exciting or interesting about joining,
26:08you know, the AI space or X division
26:11right now. Um, how about you Isami?
26:14Yeah,
26:16maybe the answer is
26:19just AI is hot hot topic.
26:24It's really really fun to develop also.
26:28Yeah, I think many many engineers want
26:31want to uh involve in developing AI
26:37service. So you're leader of our agentic
26:41workflow project. I think that is you
26:44know huge thing in my team
26:48and you know in two years later the
26:51ideal architecture will answer you'll be
26:56you know there in next two years but
26:58right now everyone in the world
27:00struggling what is agentic workflow so
27:02you are one of that
27:05yeah it is uh it's interesting to create
27:09uh uh something. Yeah. We without
27:13without an answer. Yeah. We Yeah. We
27:17We're creating a new new thing. Yeah.
27:20What is the challenge to work on at
27:23work?
27:25Yeah. The challenge is maybe Yeah.
27:30There nobody has answers. Yeah. To
27:35develop a right way. Some team you are
27:38also working on aic workflow.
27:41That's right. So based on your know long
27:44experience in engineering, what is your
27:47new challenge and
27:51things you think is not challenged? Hm.
27:56Uh yeah. Well,
28:00to begin with the not challenged part,
28:02there's something that's kind of funny
28:03about um agents, which is uh that in a
28:08sense they're as software components,
28:12they're actually extremely simple
28:14things. Yeah. Do you know we all know
28:17what I mean by that? And I think anyone
28:19who's listening to this probably also
28:20knows what I mean. But that's kind of
28:22funny because uh you know it's like oh
28:24agentic blah blah blah but the you know
28:26the real uh crazy magic stuff is
28:29happening back in the LLMs whereas the
28:30agent part is like in and of itself it's
28:33simpler. However, um when you make uh an
28:37agentic workflow, the agentic part, it
28:40starts to get more and more um rapidly
28:43become more and more interesting. Uh
28:46because
28:48there's um kind of a different style of
28:51building that you you have to embrace.
28:55It's the pro and con of having the
28:59probabilistic like non-determinism of um
29:02an LLM based yeah processing flow. It's
29:06it's it allows you to do things that are
29:10not possible any other way really. Um
29:12but of course you have to accept the
29:14uncertainties that come with it and the
29:15the part only partial reliability.
29:18Um but uh what we're thinking about now
29:23um internal to our work on the uh
29:27platform that Isaman and I are working
29:30on together is ways in which um we can
29:33eventually build self-healing uh
29:36workflows where you have problems arise
29:39and you can see this in GPT or whatever
29:42like uh especially if you use a a coding
29:45uh support if you use it for coding port
29:48in the context like cursor IDE or
29:50something like that. But you can see
29:52even in regular GPT it will try to make
29:56something for you and then it will look
29:58at it and will without any prompting
30:01from you further prompting from you it
30:02will try to fix problems that it
30:05perceives and that's unusual for systems
30:09like usually you have to as a programmer
30:11as we we have to manually put in all
30:14sorts of checks and retries and things
30:16like that. This is
30:19an interesting way to build where you
30:22know eventually the systems become more
30:24and more uh reliable ironically even
30:28though they start from a place of
30:29unreliability.
30:32Do you think we are reinventing many
30:35things in AI because apparently concepts
30:38like agents or work for engine existed
30:41before?
30:43Yeah. Um, I think it's probably less at
30:46the level of individual uh pieces of
30:50software like software components that
30:52themselves are amazing. But as we were
30:54alluding to before, even if one agent by
30:58itself is a relatively simple software
31:00component, in concert with all the rest
31:02of the software components, there's
31:05almost endless complexity that we can
31:07build up. And in a way, uh, this is
31:10maybe a strange abstract thing to say,
31:12but when when you think about, um, how
31:16how much of what we're able to do as
31:19humans is itself built on language and
31:22language itself is this endlessly
31:24recursive thing, right? So when we uh
31:28take uh when we introduce
31:31um system components that are themselves
31:35language aware and and built around
31:37language then you can start to build
31:38endlessly recursive solutions
31:41uh in the same way that human cognition
31:43is able to support endless problem
31:45solving. So it's really exciting to be
31:47honest. I'm I'm just thrilled to be
31:50working on this kind of stuff.
31:53Right. Cool. So, what's your challenge
31:55part? Oh, what's my what? Sorry.
31:58Challenge. What is the challenge? The
32:01challenge? Oh, the challenge is to make
32:04it all work because um we are currently
32:07early in that curve. So, we get lots of
32:10the unreliability and not much of the
32:13self-healing yet. So uh our goal is to
32:16push as fast as possible to where we
32:19have the sort of amazing emergent
32:22properties of agentic systems. But right
32:24now we're working still working out lots
32:26of basic stuff which honestly as uh
32:29Isaman mentioned earlier and we've heard
32:32from other people who are very well
32:34positioned in the industry to talk about
32:36this stuff. uh everyone's wrestling with
32:39the same problems of uh quality control
32:42uh reliability this and that. So um it's
32:46not to say that we have this horribly
32:47unreliable uh system. It's not that bad,
32:50but it's also not up to our normal
32:53levels of you know customers expect
32:55extreme reliability from modern
32:58platforms and we're working on that um
33:01actively.
33:04So that's the challenge. Yeah, we have
33:06many discussions in the whiteboard
33:08recently. Yeah, we have lots of crazy
33:11whiteboards. Yeah, I think you spend
33:14most of your day in front of whiteboard
33:16discussing with other people writing
33:18some crazy graphs recently. There's been
33:21a lot of whiteboarding. Yeah, but it's
33:23actually been very fruitful. Um, yeah,
33:26Isaman flew down from Okaido. Uh, and we
33:29had a great time twice in a month
33:32recently.
33:39Yep. Yep. So, looking ahead, what do you
33:43personally want to do at least
33:473 years?
33:50How about your team?
33:52Oh, well, um it's I think it's an
33:54extension of what we've been talking
33:56about. Uh so the two threads are um well
33:59three threads is that I think we're all
34:02excited about the possibility of uh you
34:05know this LLM flavor of AI mixed
34:09together with other more traditional
34:11flavors of AI and the possibilities of
34:15that are um they're they're huge. So
34:19that's exciting. And then merging that
34:21with the ample opportunities in Japan
34:25itself. Uh which you know is on the
34:28leading edge of um you know demographic
34:31challenges and other things that uh you
34:34know make it this test bed for how will
34:36we as uh all around the world how will
34:40society deal with you know what's going
34:43on demographically and AI is definitely
34:46going to be a part of that. So I'm
34:47excited about that.
34:50um in the broad sense and then we've
34:53talked about earlier about expanding
34:56internationally and
34:59um
35:02the prospects for that seem very
35:04realistic. So um I'm looking forward to
35:07contributing in that area as well. So
35:11yeah, I think the confluence of those
35:13things
35:14um at a company that's proven that it
35:16can execute I think is
35:19Uh yeah, pretty enticing. So that's the
35:23near to medium term looks like for me.
35:25Right.
35:28In terms of global expansion, which
35:30country are you interested in other than
35:33the US and Japan?
35:36Other than the US? For me, it's like
35:39Germany or Singapore.
35:42Huh. Uh I I'm very US- ccentric um
35:46because I'm American. So, uh I I
35:50honestly hadn't thought too much about
35:51uh where else
35:53um it's just such a huge market, but
35:56yeah, I mean uh tell me more why why uh
36:01why Singapore and Germany? Why those
36:02two? Germany for you know manufacturing
36:05and Singapore for finance.
36:08I see.
36:10Okay. Yeah, I guess that makes sense.
36:14Isan, are you looking at Canada again?
36:18Yeah, kind of. Okay. I've never been to
36:21Canada. Canada, honestly, so I don't
36:24know much about industry there.
36:28Yeah.
36:30Yeah. I I feel Canada is almost
36:33America. Yeah.
36:36Well, don't tell that to Canadian.
36:42But they'll be very nice about it.
36:46Yeah. How about you, Visami? What do you
36:48want to do in the next two or three
36:49years in Le?
36:52Uh, I want to X
36:58uh continue to bet AI because you know
37:04we can we have bet AI in our philosophy.
37:06So yeah. Yes. So actually yeah we we
37:11already can use CL cloud max or we can
37:15we can uh
37:18yeah not get but we can use many many uh
37:23AI stuff to tooling. Yeah. So well
37:28that's true. We're not just building AI
37:30platform. We're of course we're doing it
37:32using AI tools and we get to play with
37:36uh well I mean actually used for work
37:38but it feels also we can be um advanced
37:42team or sophisticated team right? Yeah.
37:47I actually heard um from friends of mine
37:50who uh work with
37:53um more sort of regular Japanese
37:55software companies. I heard that uh they
37:58don't get to use AI very much like they
38:01have company policies that prevent them
38:04from introducing things and it just
38:06sounds so sad like you know to to not
38:09get to use it for your work. Um so I'm
38:14very glad that you know we're not like
38:16that. Yeah. Yeah. And by the way it's
38:19it's like pretty expensive. It's a
38:21considerable investment that the company
38:23puts into these tools because we're
38:25using the really expensive versions of
38:27this stuff, right? As you mentioned, uh
38:30like Max level of Claude, that's the
38:34like like wo kind of prices. So, um
38:39yeah, it's not a it's not a trivial
38:40thing that the company is making this
38:42investment. So, I appreciate that. Mhm.
38:44So any reason to use so
38:50AI to tools? Yeah. So we think you know
38:54we prioritize the performance of
38:55engineering team the first. So we will
38:59spend as much money if you know engineer
39:02gets happy. So this is why so I think we
39:04published a blog post about purchasing C
39:08marks by card yesterday. Anyway, so
39:12yeah, we'll use a lot of tool in the
39:13future. So I think we are running out of
39:17time. So I'll wrap up this episode with
39:20your message to our listeners, including
39:23potential future teammates who might be
39:25considering joining REA X. So Isami,
39:30your final message? Uh yeah. So yeah, we
39:34are we're creating uh global team and we
39:39are we creating a global product not not
39:44only Japan. Yep. Yeah. So I'm I'm
39:48looking forward to join you. Yeah.
39:53Yes. So team, can you give us a perfect
39:57closing line? Oh uh maybe not one line
40:00but um as you were talking you guys were
40:03talking I remembered something that um
40:05when I was uh originally talking to
40:09Larrex
40:10um originally talking actually to um
40:12Wimat's son and he said something that I
40:16thought was really interesting which is
40:18that uh he said that there maybe like 20
40:21people or so um attached to the team at
40:25the time we were talking. Yeah, it's
40:27grown. I think it's grown since then,
40:29but 28 people at the time. And he said,
40:32uh, you know, we'd really like for you
40:33to join and sort of seed our
40:36internationalization,
40:39like the internalization of the team
40:40itself because
40:43uh if we become so we're 20 Japanese
40:46people and then if we grow to 50
40:48Japanese people, then there'll be no co
40:51it'll be really hard to go back like
40:52we'll just be another Japanese company.
40:55So, it's really important that we shift
40:57the direction um and we invest like
41:01we're investing in those AI tools and
41:03anything else. This is an important
41:05investment. Um even though there's some
41:07overhead to it, we really really want to
41:10um absorb
41:13and join forces with um international
41:16team members so that we can really
41:19realize that vision. Otherwise, it's
41:22just going to be talk and we can put
41:24like powerpoints and stuff and talk
41:25about it's our mission and stuff and
41:27it's not going to be real. So again,
41:30there's actual investment in a direction
41:34that makes sense kind of like the
41:37general growth strategy uh makes sense.
41:40Um, so I think my closing line would be
41:43for anyone who's listening to this, you
41:46know, in English and if you're native
41:48language or whatever, if you're at all
41:50like me, then I want to invite you to
41:53have a casual conversation with us about
41:55this stuff because um, you know, I chose
41:59this place for I what I think are pretty
42:02good reasons and um, I think they're
42:04pretty general reasons, too. I think it
42:07probably would apply to uh you as well.
42:11Um you know, maybe not, but uh maybe we
42:14could have a chat about it. So that's my
42:16my closing line. Yeah, I recently found
42:20that there are huge international
42:23engineering community in Tokyo, but
42:25unfortunately I think there's a little
42:28limited opportunity for them to work in
42:29Tokyo. So I want to provide more
42:31opportunities to them by relax. All
42:35right. Thank you for listening and don't
42:37forget to check out the next episode.
42:40See you.

About this transcript

This page contains the full transcript of #108 【全編英語回】The Ai Workforce Team Goes Global: Voices from the Engineers Behind Agentic Workflows... by LayerX 公式, generated from the public captions YouTube serves with the video. The transcript has 5,827 words across 871 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.