TRANSCRIÇÃO 🔗
WEBVTT
00:00:00.000 --> 00:00:14.160
I'd become totally convinced that we needed to become AI practitioners fully, to use others' software and hope that they would build in the right AI capabilities for us to be at the forefront.
00:00:14.480 --> 00:00:18.079
It's a little bit too much of a hope and pray type of strategy.
00:00:18.399 --> 00:00:20.079
So we decided to start to build our own.
00:00:20.399 --> 00:00:29.280
So this is an experimental system that we're building it on the fly as we go and we're shipping new features on a weekly or daily basis.
00:00:29.519 --> 00:00:32.960
I want everybody to be 100% committed and have no fear.
00:00:33.119 --> 00:00:38.079
No one is going to lose their jobs as a result of us becoming an AI firm and a tech platform.
00:00:38.320 --> 00:00:44.240
Do you actually see organizations changing the roles that they're hiring for and the whole like job description itself?
00:00:44.479 --> 00:00:50.479
We're seeing a lot more kind of evolvement and new jobs coming up faster than uh we've seen before.
00:00:50.640 --> 00:01:05.519
I become 100% convinced that humans are just we're just too creative, we are too curious, we're too competitive, we're too excited, we're too petty about things, we want to win, we wanna build things, we wanna explore and stuff.
00:01:05.599 --> 00:01:08.560
We there's no way that we're not gonna find ways to be busy.
00:01:08.879 --> 00:01:12.079
Hi everyone, I'm Eric, co-host of the Cotriff Podcast.
00:01:12.239 --> 00:01:14.799
Uh with me, I have my other co-host, Joong.
00:01:14.959 --> 00:01:17.760
And with us, we have a very special guest today, Roger Olofsen.
00:01:17.840 --> 00:01:19.519
Roger, hey, it's great to have you here.
00:01:20.000 --> 00:01:21.680
Great to be here with you guys as well.
00:01:21.840 --> 00:01:22.719
Really good to see you again.
00:01:23.040 --> 00:01:25.760
So we're gonna spend our time mainly in three parts today.
00:01:25.920 --> 00:01:33.840
So first, we will understand more about yourself and and what's happening inside the AI lab of Olofsson.ai.
00:01:34.239 --> 00:01:41.359
Uh second, we'll go a bit more broadly into AI native services and hopefully some demos as well.
00:01:41.760 --> 00:01:45.439
And then lastly, we will go into you know recruitment and the job market.
00:01:45.760 --> 00:01:48.879
Onto our first question, just to understand you a bit more.
00:01:49.120 --> 00:01:54.000
Uh you spent 25 years in recruitment and search, and you went and bet your whole firm on AI.
00:01:54.079 --> 00:01:59.760
Can you just kind of take us to the moment of when you decided to go all in with your firm as well?
00:02:00.159 --> 00:02:01.920
Yeah, yeah, absolutely, Eric.
00:02:02.079 --> 00:02:04.799
And so good to be here and uh catch up with both of you.
00:02:04.879 --> 00:02:16.560
And uh I'm actually to maybe take it back a little bit further, even in a um so man, it's such a cool time to meet now, 3rd of July in three days.
00:02:16.800 --> 00:02:36.479
It's 30 years since I came here with a backpack in a suitcase on the 6th of July um 1996 to um spend 10 months here as an exchange student in NTU to finish my software engineering um fourth year, and and then fell in love with the place and uh never went back.
00:02:36.560 --> 00:02:42.479
So uh, apart from the longest I've been away from Singapore in a stretch is three months over these uh 30 years.
00:02:42.560 --> 00:02:46.240
I've really been based here the whole time, traveled a lot, but based here.
00:02:46.479 --> 00:02:50.879
And um Singapore became home.
00:02:51.120 --> 00:02:57.759
Um talent and top talent became something that I fell in love with very early in my in my career.
00:02:57.919 --> 00:03:32.479
I did work a short stint in uh turnkey software projects we were building, um, and I was sharing with him a project where we were building a um uh a CRM solution before custom-building it before CRM was almost uh a term and we were building it on this super cool language that was like the absolute you know hottest thing in the market called Java, which is now as legacy as you can get, you know, just one step above COBOL kind of thing, right?
00:03:32.639 --> 00:03:38.319
And but at that time it was the coolest thing on the planet because it was developed by Sound Microsystems.
00:03:38.479 --> 00:04:09.120
That's around that time, very early in my career, that the opportunity and appreciation to marry technology or being in the leading edge of technology and um having an working on projects or opportunities that had an impact on business and transformation and change and innovation, and you know, um unlock the human potential in in top talent, right?
00:04:09.199 --> 00:04:19.519
In talent to get that right, to get the right people into the right roles, get a team or or a team of people to work really really well together to produce outsized returns.
00:04:19.759 --> 00:04:23.519
Um became I got exposure to that.
00:04:23.680 --> 00:04:33.920
Uh I joined Robert Walters, which is a big uh big recruitment firm, it's a recruitment consultant number six in Southeast Asia as it was like it's here, even though it was hosted by a publicly listed firm.
00:04:34.240 --> 00:04:38.879
So we had resources, but we had to build from scratch, and uh it was fun.
00:04:39.120 --> 00:04:46.000
And I I I fell in love with recruitment, you know, and uh that this became my passion to be in that intersection, always in technology.
00:04:46.240 --> 00:04:57.279
So it's always been in leading edge technology, working with top talent to work on critical high impact roles in uh in in companies in tech.
00:04:57.759 --> 00:05:02.560
And um I got to work in uh in the tail end of the dot-com boom.
00:05:02.639 --> 00:05:08.160
So I worked with founders in the late 90s, 99, 2000, just before the dot com crash.
00:05:08.319 --> 00:05:26.079
Yeah, and um so I saw the I got to see the exciting things with the boom that happened then, but also the the
Load full transcript
aftermath of a bubble like that, where you know people are struggling then to get uh jobs in tech for a period of time.
00:05:26.399 --> 00:05:42.399
Tech after the dot com uh crash became a bit unsexy, you know, and uh most of my colleagues in recruitment that I was in then they moved to doing front office recruitment for investment banking or maybe finance, yeah, life science.
00:05:42.560 --> 00:05:43.680
Uh yeah, exactly.
00:05:43.839 --> 00:05:57.920
All the and I was dumb or unfortunate for I mean, not actually fortunate now in hindsight, but I was like, oh no, I just I couldn't find anything else that excited me as much as technol technology.
00:05:58.160 --> 00:06:07.040
Yeah, I I was just I tried kind of to take on and manage other teams and stuff, and nothing gave me the excitement and the passion like uh tech did.
00:06:07.120 --> 00:06:27.279
So I just continued to do uh to to to um focus on the tech side, um and of course now in hindsight that ended up being the best move that I could have ever made because now out of luck, sheer luck, more than any intelligent career planning or anything like that.
00:06:27.600 --> 00:06:44.800
Um I now built maybe like 27 years of uh experience in in leading edge tech talent in Singapore and Asia, and uh maybe there is just a handful of us of if even that in this country with that type of experience, right?
00:06:44.959 --> 00:07:11.360
And uh and uh so long story short, eight years plus ago, I finally did what I had wanted to do since I was seven years old, start my own firm um together with my with uh initial investorslash co-founder, angel investor Simon Rowling, that was a um is a successful entrepreneur, serial entrepreneur that has built several businesses in the past and exited.
00:07:11.439 --> 00:07:22.240
And uh he'd been a client, we'd learn to know each other really well, and we joined forces to start all of a sudden company in uh on the on the 8th of January 2018.
00:07:22.720 --> 00:07:27.920
That was the day when we switched the lights on in a serviced office next door to where we are now.
00:07:28.720 --> 00:07:34.319
Um and um tell us more about the the rebrand, I guess, that that recently happened as well.
00:07:34.480 --> 00:07:36.959
Yeah, or if you if you don't call it a rebrand, what what else?
00:07:37.279 --> 00:07:49.759
Yeah, I mean it's actually probably not so much of a I mean like the rebrand is uh again probably sort of the least significant part of what we're doing and who we are now, right?
00:07:49.920 --> 00:08:01.839
So I just had a conversation with our team today, and uh we have our half-year uh say uh team update meeting on Tuesday next week.
00:08:02.000 --> 00:08:10.560
Uh we've just uh wrapped up the most successful six months uh as a business that we've had since we started.
00:08:11.040 --> 00:08:25.759
Um there are some really exciting and interesting data uh around our performance that I'm you know happy to sort of share share a little bit about because
it's very AI related.
00:08:26.000 --> 00:08:32.240
Uh the way that I'm looking at us now is that we're not a recruitment firm, or we're not a traditional recruitment firm for sure.
00:08:32.720 --> 00:08:39.919
We are an a tech firm or we are an AI native firm that happens to be expert experts in recruitment.
00:08:41.120 --> 00:08:44.399
Yeah, so we're experts in recruitment, but that's that's not who we are.
00:08:44.480 --> 00:09:16.879
We are a technology firm, but we're producing recruitment services, and we're doing it uh, you know, with uh utilizing you know an uh an AI native platform, and that's why so we decided so um I probably didn't I didn't have the the fame and the means like Sebastian who started uh Klana, uh the CEO of Klana, which uh you know when he played around with but but we probably had similar type of experiences around generative AI.
00:09:16.960 --> 00:09:29.120
So when he played around with Chat GPT in the Christ Christmas 22, just after the ChatGPT event in November 22, he realized and had this epiphany, this is world-changing.
00:09:29.519 --> 00:09:30.480
This is world changing.
00:09:30.720 --> 00:09:39.360
We saw the changes ahead of him, and he decided immediately when he came back to office in January to uh reach out to Sam Altman and book a meeting with him.
00:09:39.600 --> 00:09:42.240
And he did the oldest sales trick in the book.
00:09:42.320 --> 00:09:51.440
He said, uh wrote, it's like, hey, I'm gonna be in I'm in San Francisco uh later this month on a trip, and I wonder if I could drop by and meet you.
00:09:51.600 --> 00:09:55.840
He had no trip planned, uh you know, but that's the way you do it in sales, right?
00:09:55.919 --> 00:09:57.279
I'm around the corner, you know.
00:09:57.919 --> 00:09:59.200
Could I could I meet you?
00:09:59.360 --> 00:10:01.519
Yes, rush there as quickly as you can.
00:10:01.759 --> 00:10:11.919
And um and he went and said to them, I want Klana to be your guinea pig in financial service or in fintech.
00:10:12.240 --> 00:10:18.639
Anything that you suggest can be done with generative AI, you can do it with us, you can do it on us.
00:10:19.039 --> 00:10:20.080
No questions asked.
00:10:20.320 --> 00:10:27.919
He gave his full commitment, and he's now turned that into a completely AI native company with uh you know phenomenal results.
00:10:28.080 --> 00:10:36.960
I had a similar type of epiphany, much much smaller impact and smaller scale, but um and it was semi-existential as well.
00:10:37.279 --> 00:10:38.799
Three things happened at the same time.
00:10:39.039 --> 00:10:57.679
One was that I had developed a really strong relationship with uh you know Chris, an AI engineer that we were very aligned in terms of our thinking and the way we looked at uh AI and and business and the talent space, and it became available, so it was one thing.
00:10:57.840 --> 00:11:12.080
Second was that uh I was out running and I uh listened to uh Anton Osica, this co-founder of founder of
Lovable on the Harry Stabbings 20 VC podcast.
00:11:12.559 --> 00:11:31.039
And uh I was blown away by that interview, by what they're building, and because I love the name, I use love in my language a lot, and just uh you know, in everywhere, you know, it's it's it's and and lovable was such a cool name, and then it's Swedish, and the patriotism uh kicked in.
00:11:31.200 --> 00:11:34.720
It's like man, Swedish coding agent.
00:11:35.039 --> 00:11:44.159
So I don't download it the same evening immediately and started playing around with it and realized oh my god, this is this is incredible, you know, this is incredible.
00:11:44.320 --> 00:11:53.440
So there was the second thing that the agentic coding just started to get to the level where it uh started to become really useful.
00:11:53.679 --> 00:11:54.960
It didn't suck as much, exactly.
00:11:55.039 --> 00:11:56.960
It didn't suck as much, exactly, exactly.
00:11:57.120 --> 00:12:00.879
And even though it still sucked, but not as much compared to now, right?
00:12:00.960 --> 00:12:02.080
And then you could see something.
00:12:02.240 --> 00:12:02.960
Yeah, yeah, exactly.
00:12:03.200 --> 00:12:05.919
Yeah, but the third element again, just super quickly.
00:12:06.159 --> 00:12:12.639
So the first one was that uh Chris and I hit it off in terms of sort of someone with AI engineering skills, yeah.
00:12:12.879 --> 00:12:22.480
Um, and then second was that the gentic coding got to a level where it was it started to get really useful, uh, and it was very early days of actually getting useful.
00:12:22.559 --> 00:12:48.799
And then third thing was that I'd become totally convinced that we needed to become AI practitioners fully, you know, and and and we need to be in charge of our own destiny and uh to use others' software and hope that they would build in the right AI capabilities for us to be at the forefront is kind of a uh uh it's a little bit too much of a hope and pray type of strategy.
00:12:49.120 --> 00:12:51.679
You know, I wanted us to be in be in control, right?
00:12:51.919 --> 00:13:05.919
So we decided to start to build our own, and um yeah, and now we become a you know much more of as I said, the uh the uh a tech firm that um is producing recruitment services.
00:13:06.720 --> 00:13:35.200
And um maybe one of the ways, if you want to look at some of the things that we so this is a experimental system that we build, we're building it on the fly as we go and we're shipping shipping new features on a on a weekly or daily basis, and we're testing different things, so that's why it's also looking really, really kind of quite very big, uh very unique, very quite quite exhaustive, very busy.
00:13:36.399 --> 00:13:46.480
This is so this is an internal system that is used by highly technical recruiters that are now become very AI native, very comfortable with the workflows and stuff that we're using.
00:13:46.799 --> 00:14:00.799
Uh, but it gives us it's all built based on the way that we
operate, the way that we work in the markets that we serve, and not uh like an off-the-shelf type of software that has just been configured to to a recruitment firm.
00:14:01.200 --> 00:14:10.960
Um, we have a search capability uh that we've built that uh we are currently rebuilding actually.
00:14:11.120 --> 00:14:21.759
So this one we've we used um we've used it and tested it significantly, but with all models like GPT 4.0, 4.0 mini, uh, etc.
00:14:22.240 --> 00:14:30.720
It's all about also trying to build optimize for both precision as well as cost uh to scale it.
00:14:31.039 --> 00:14:49.360
Um but what this model what this uh user interface enables us to do is to go through tens of thousands of candidates very quickly with an agentic workflow that would run through, for example, 10,000 candidates in like less than 10 minutes.
00:14:50.000 --> 00:15:01.200
And it does it by spinning up in this case we're asking the uh the agent to give us a top 15 candidates.
00:15:02.399 --> 00:15:04.399
How many candidates do you have on your platform?
00:15:04.639 --> 00:15:11.440
It's constantly moving, it's three hundred roughly three hundred and ninety-five thousand candidates now, so we're getting we're approaching half a million, yeah.
00:15:11.679 --> 00:15:25.759
Where maybe about uh around fifty percent is in Singapore and uh about uh thirty-five percent is in the rest of Asia Pacific and 15% in Silicon Valley, North America, Europe.
00:15:26.320 --> 00:15:28.080
So yeah, it's growing.
00:15:28.320 --> 00:15:32.320
So now we have a so we both a couple of things.
00:15:32.480 --> 00:15:35.919
What you see here, for example, EXA is that we have an ability.
00:15:36.159 --> 00:15:38.720
EXA is a$250 million.
00:15:39.759 --> 00:15:46.480
The latest round they got was$250 million Series C led by A16Z about one or two months ago.
00:15:46.799 --> 00:15:54.960
A couple of young super bright founders that uh went to YC and told them we're gonna disrupt Google search.
00:15:55.200 --> 00:16:00.399
And for some reason, crazy reason, YC still gave the money because it sounds like a crazy idea.
00:16:00.799 --> 00:16:03.519
A lot of people have tried and failed, including Microsoft, right?
00:16:03.679 --> 00:16:17.039
Yeah and uh but they didn't give they decided and then they decided to go against YC's advice, they didn't talk to any customers, they've they took the check and they bought GPUs for it and locked themselves into a basement and started to experiment and build.
00:16:17.360 --> 00:16:21.840
And now they built this tool that is agentic AI search for agents.
00:16:22.480 --> 00:16:26.399
So you we are actually so if you're a client of Excel, yeah.
00:16:26.879 --> 00:16:39.759
So we are we are we're doing a POC with them, we built in their solution into our UI where we can then put in the So this will spin off a new web search for exactly.
00:16:39.840 --> 00:16:42.639
So this will I mean I mean roles that you're looking for.
00:16:43.039 --> 00:16:43.519
Exactly.
00:16:43.679 --> 00:16:46.480
So this enables us to cast the net a lot wider.
00:16:46.559 --> 00:16:54.559
So let's say that for example we think that we we've exhausted our 395,000 candidates that we have on our own platform and go wider, yeah.
00:16:54.720 --> 00:17:08.240
We kick off this website search, which can then which is which where we can then put in the um the query or what we're looking for, or uh uh distal it from job descriptions, etc.
00:17:08.640 --> 00:17:17.839
And then uh see if I can find something that we did for um in Japan in particular, we had some good.
00:17:18.400 --> 00:17:19.519
Japan is not so easy, right?
00:17:19.599 --> 00:17:24.079
Because candidates are usually not not on LinkedIn or Japanese, yeah.
00:17:24.319 --> 00:17:24.880
Yeah, exactly.
00:17:25.440 --> 00:17:26.559
So exactly specific.
00:17:26.799 --> 00:17:28.640
You have no one speaking Japanese here, right?
00:17:28.799 --> 00:17:30.079
Uh no, actually, we don't.
00:17:30.240 --> 00:17:34.319
We have a little bit of Korean, we have Chinese, of course, and uh but not Japanese.
00:17:34.480 --> 00:17:46.240
Okay, and um so this was this is a bit sensitive though because it shows trying to inform but uh it was particularly effective for us in Japan, you know, because LinkedIn is not very strong there, yeah.
00:17:46.480 --> 00:17:50.480
And uh here we were able to extract really, really good information.
00:17:50.720 --> 00:18:02.559
We've also done work on AI researchers and AI scientists across the region, which includes Singapore, of course, China, because that has the largest population of AI scientists outside of the United States, probably bigger.
00:18:02.880 --> 00:18:14.720
And um, then what we can do here, the good thing with using a tool like EXI is that it's it gives us an ability to search across anything on the public uh internet and in hyper speed.
00:18:15.279 --> 00:18:18.799
Um so this goes off and do it asynchronously.
00:18:18.880 --> 00:18:29.119
So you know we just kick off the query, the search, we avoid the complete with our work, and then it just flags it to us when it's uh when it's ready and it's up and running.
00:18:29.920 --> 00:18:33.519
Yeah, we can we can look at some of the the the search results and stuff.
00:18:33.599 --> 00:19:21.839
So what we do is um everything that we do, yeah, whether it's an exa search or uh a SQL search, you know, just a Boolean search on our uh database or an agency search where we either using this um UI that I just showed you, or we AI source search the UI that we used to use in the past that we're now rebuilding, or we go into using uh just a terminal or clawed code on a terminal where we built again a whole library or catalogue of different skills to enable with including self-learning skills as well, where these are continuously getting modified based on feedback from our consultants.
00:19:22.160 --> 00:19:29.119
All of the results go into a page called
precision, and we cut we call this precision search.
00:19:29.279 --> 00:19:40.160
Um our working name now for the system internally for the fun of it has been mankind so far, but we are looking for a different name for if we're gonna bring it out to the market or productize it.
00:19:40.240 --> 00:19:42.240
So it's like yeah, just our own internal code.
00:19:42.720 --> 00:19:49.519
So here, for example, we could show we could look at when we have brought a client that we're working with is looking for account executives, yeah.
00:19:49.680 --> 00:20:44.079
Then we have done an agentic search on the all the 395,000 candidates with various with the where we give the job description and we we give the consultant comments based on the client meeting or even the client notes from a transcript, putting it into in this case, currently in the new version that we're working on, we we put it into the terminal, and uh the skills that we built on the uh on uh cloud code then goes off to work and connects to or searches the database based on and we've then we've done a cascading type of agent where we've trained it with context because in itself it's not necessarily it can be even though it's super intelligent, it can be quite uh clumsy in terms of how it would think about in our case structuring a search, but it doesn't have the experience that we have, right?
00:20:44.480 --> 00:20:46.640
Based on 25 plus years of doing this.
00:20:46.880 --> 00:20:53.200
So we've then trained it with context to say, okay, initially you have 395,000 candidates.
00:20:53.680 --> 00:21:02.480
What are the things that you can do with with SQL, with Boolean search statements and stuff to filter this down to a more manageable number, like 10,000?
00:21:02.720 --> 00:21:35.599
So if you go to 10,000 or below, then we start to work with in this case because we're using clawed code in this current prototype, yeah, haiku to do kind of low weight, low lifting volume tasks on these 10,000 candidates because there are patterns that we can look for, they're pretty straightforward basically and um um but um even uh so some of these things like years of experience and stuff is already indexed on the system, so you can uh you know extract that straight away.
00:21:35.759 --> 00:21:51.200
But let's say that you're looking at uh a different a certain type of let's say they're looking for uh candidates that have uh studied at uh Ivy League or Ivy League equivalent universities across the world.
00:21:51.440 --> 00:21:55.039
Very difficult query to do on LinkedIn, for example, in the old days.
00:21:55.440 --> 00:21:59.519
You had to list every bloody name of uh every university that you think.
00:22:03.839 --> 00:22:08.480
Pay for sales navigator and you can kind of filter for the the specific schools they're looking for.
00:22:08.799 --> 00:22:08.960
Yeah.
00:22:09.759 --> 00:22:36.480
So what the the system is able to do is to go and do searches on these 395,000 candidates, complement it
with uh candidates from anywhere in the world with tools like EXA in particular, then grab it all, put it all together, create a data pipeline for the consultants uh that it produces, that it then presents on a UI that uh again maybe looks busy to the untrained eye.
00:22:36.640 --> 00:22:43.039
But for us, we've kind of optimized this for technical recruiters that are really good at what they're doing, they want to move fast.
00:22:43.119 --> 00:23:02.720
So they get a lot of data in front of them without having to scroll and look at the one or two things to get to see all the ideas that are important to them, they can very quickly go in and look at the candidate information as well as the this is critical, the commentary, the reason why the agent is selected.
00:23:03.119 --> 00:23:08.559
And you actually have a score there for exactly what is candidate specific to that role they're hiring for.
00:23:08.720 --> 00:23:08.960
Exactly.
00:23:09.119 --> 00:23:10.079
So it's not a generic score.
00:23:10.240 --> 00:23:10.799
No, exactly.
00:23:10.880 --> 00:23:21.599
It's spot on, specific to that, and then you see these green dots and the red dots are very significant because this is where human judgment is agreeing with the with the agent.
00:23:21.920 --> 00:23:37.279
So any every candidate that has been that has been presented by the model, selected by the model or the agent, and then Sean, in this case, our expert recruiter in this uh for this role, she says, Yes, this is a good candidate tick uh pipeline.
00:23:37.440 --> 00:23:38.640
Yes, this is a good candidate.
00:23:38.799 --> 00:23:42.480
No, this is not good, then exclude it with a red tick.
00:23:42.880 --> 00:23:45.920
And at times we say you don't have to do it all the time.
00:23:46.079 --> 00:23:50.559
But every now and then, if something stands out, put a comment in, put some feedback in as well.
00:23:50.720 --> 00:23:52.640
And then this I can train on this.
00:23:52.880 --> 00:23:53.279
Exactly.
00:23:53.440 --> 00:24:00.880
So the next uh iteration of search that the agent goes off and do, then it automatically goes back and looks at this whole list.
00:24:01.119 --> 00:24:14.240
Any candidates been screened, and it looks at the green ones as good examples, red ones as bad examples, and any commentary as well to then tune the rubric, the decision-making criteria for the next round, and it just gets better and better and better.
00:24:14.640 --> 00:24:16.319
And new new candidates that show up.
00:24:16.400 --> 00:24:19.200
Um, they can actually use a rubric that you use here.
00:24:19.599 --> 00:24:20.559
Exactly, exactly.
00:24:20.960 --> 00:24:29.920
And this, I would say, this is part of the secret source that is making us super fast now compared to attracting the manual way exactly.
00:24:30.079 --> 00:24:31.839
And this is finally right candidates.
00:24:32.079 --> 00:24:39.440
We're automating a lot of other things as well, like uh proposals, uh uh CV formatting uh is completely automated.
00:24:39.920 --> 00:24:46.720
Um uh write-ups and summaries and all these kinds of things, uh, outreach more and more and more.
00:24:46.880 --> 00:24:52.960
Um, but the one that is the biggest crux for us in the value creation is the matching of the right.
00:24:53.359 --> 00:25:02.960
So, Will Brook in uh EXA says that they are inventing perfect search because they felt Google failed or didn't get it all the way.
00:25:03.759 --> 00:25:06.799
So they're trying to not just do okay search but perfect search.
00:25:07.359 --> 00:25:10.480
And my response to him was that that's super cool.
00:25:10.559 --> 00:25:20.240
We should work closer together because we are doing perfect matching, we're building perfect matching of the right candidate with the right job, and that is super complicated.
00:25:20.640 --> 00:25:38.160
Actually, advanced models, frontier models is the only thing so far over 27 years of experience that I've seen that has come close to being able to touch you know, being a solution for this problem because a human being is so complex.
00:25:38.480 --> 00:26:04.160
So if you look at this, this bar here shows us that out of uh a total of uh uh 193 candidates that were uh produced for Shan, she's gone through 61 and she has said she's selected, she's agreed with the agent 45 out of those 61 times, 73.8% agreement between the agent and expert recruiter.
00:26:04.880 --> 00:26:10.799
And if you do what we're trying to avoid is what we call the doom scrolling, right?
00:26:10.880 --> 00:26:14.079
The doom scrolling on on Instagram and that sort of stuff, right?
00:26:14.240 --> 00:26:20.640
The equivalent for a recruiter is to go through page after page of LinkedIn, it is soul destroying.
00:26:20.960 --> 00:26:23.440
You know, I I I joked with Chris when we started building this.
00:26:23.519 --> 00:26:25.440
Do you know how good this feels for me?
00:26:25.759 --> 00:26:32.480
I probably wasted about eight years of my life total time looking at wrong candidates on LinkedIn.
00:26:32.799 --> 00:26:33.359
How does it feel?
00:26:33.440 --> 00:26:34.079
How does it feel?
00:26:34.640 --> 00:26:44.480
That feels terrible to think about how many years I've wasted, but it feels so exciting about not having to waste any more of that time, me, yeah, and saving it for other fellow recruiters as well, right?
00:26:44.799 --> 00:26:47.119
And just producing better results much, much quicker.
00:26:47.440 --> 00:26:48.319
Match more people today.
00:26:48.640 --> 00:26:48.960
Exactly.
00:26:49.279 --> 00:26:53.519
It's kind of being uh a forerunner or like the first one to do a certain thing, right?
00:26:53.599 --> 00:26:56.400
And you're kind of innovating on the model of recruitment, right?
00:26:57.119 --> 00:26:59.200
True, true, but excited, yeah, true.
00:26:59.440 --> 00:27:00.319
That's true, Eric.
00:27:00.480 --> 00:27:23.279
But still, expertise, uh, intention, uh authenticity, the desire to deliver a great
service and get it right between clients and candidates, um makes a difference, and I think that's gonna continue to make a difference, even if we have uh a similar type of tool set but just more advanced, right?
00:27:23.440 --> 00:28:00.319
It's just moving the expectation frontier of what we can do, you know, instead of taking two months to make a placement, maybe we do it in two weeks, um but then uh we're we're just gonna be able to, and hopefully, instead of getting it right maybe 70% of the time on average in the industry or whatever it is, maybe we start to get it right 98% of the time, which is then adding value to the world, you know, making yeah, so one of the things, one of the things that I a little bit cheesy at times say as a tagline is that uh we make the world a happier place one placement at a time.
00:28:01.200 --> 00:28:07.200
I mean if you have tried recruiting before, you would appreciate um a good recruiter, right?
00:28:07.440 --> 00:28:08.960
Yeah, and that's so exciting.
00:28:09.039 --> 00:28:10.640
And and Yaoung, you're right about that.
00:28:10.799 --> 00:28:12.000
That's another way that we're looking at it.
00:28:12.079 --> 00:28:27.680
It's like if you think about if you have ever recruited someone, or if you have been looking for a job yourself and you have had the fortunate experience of working with a really good recruitment consultant that knows their stuff and they have their best interest at heart, it's a great experience.
00:28:27.839 --> 00:28:33.839
You would you would never want to go through a recruitment process without that person, right?
00:28:33.920 --> 00:28:37.279
Yeah, but you wouldn't use it all the time because it's expensive, right?
00:28:37.359 --> 00:28:39.279
You know, it's prohibitive in that way.
00:28:39.680 --> 00:29:04.960
What if you could democratize this by making it so so AI supercharged and supported so that I as a recruiter can do 10 times the work that I could do manually traditionally, then I could do it arguably at fifth of the cost and make twice as much money, or do it as at a tenth of the cost and still making as much money, which means everybody could afford it.
00:29:05.599 --> 00:29:06.559
Everybody could afford it.
00:29:06.880 --> 00:29:14.000
Suddenly it's not something that you just use for the most difficult roles or the uh the most senior roles, right?
00:29:14.240 --> 00:29:24.480
So maybe I would add to that is that I mean in for a recruiter, having knowing the scores or having the candidates is actually not not the secret source, right?
00:29:24.559 --> 00:29:32.000
It's actually you knowing the candidate personally, the relationship that makes you a better recruiter in a way.
00:29:32.400 --> 00:29:32.559
Yeah.
00:29:32.880 --> 00:29:35.039
Because I mean I'm I'm looking for my own experience, right?
00:29:35.119 --> 00:29:44.079
Like in the past when I've known quite a few recruiters, and there are some I'm like, okay, this guy is just here to kind of uh just
for this role, and then that's it, right?
00:29:44.160 --> 00:29:49.200
But there's some that follow my my career and they would reach out when whenever something pops up in the radar.
00:29:49.440 --> 00:29:49.519
Yeah.
00:29:49.839 --> 00:29:50.160
Exactly.
00:29:50.240 --> 00:30:36.799
And I think that comes to uh uh you know in and uh you know something that we uh we we I think we've touched on before and that we uh that we that um um would be that we probably that the you know something that really uh important that at the core uh of this whole AI transformation is what's what's gonna be left for in our industry or criminal consultant to do what's the uniquely human and I think Yao you you you you hit the nail on the head that's the it's still that relationship, that trust, that understanding of another human being, and having their best interest at heart that I think uh is uh still gonna continue to be the secret source and the reason why our roles will continue to exist as well.
00:30:36.880 --> 00:30:46.079
I think that the you know, like we said when we started this journey with the team, I told them listen, I want everybody to be 100% committed and have no fear.
00:30:46.319 --> 00:30:51.759
We're not no one is gonna lose their jobs as a result of us becoming an AI firm and a tech uh platform.
00:30:52.000 --> 00:30:52.880
So you made that promise.
00:30:53.119 --> 00:30:54.079
I made a promise, yes.
00:30:54.240 --> 00:30:55.119
Everybody's jobs are safe.
00:30:55.200 --> 00:30:56.559
We just need a human there.
00:30:56.720 --> 00:30:56.960
Yeah, right.
00:30:57.279 --> 00:30:57.680
Exactly.
00:30:57.920 --> 00:31:00.960
We're just gonna build our in two things we're gonna do with AI.
00:31:01.039 --> 00:31:13.519
One is we're gonna supercharge you, you're gonna be like a superhuman recruitment consultant with powers that you never had before, and you're gonna be able to move a lot faster and deliver much better uh results to your clients and candidates.
00:31:13.759 --> 00:31:24.720
And second is that the cool thing is that for most of you, or probably all of you, a lot of the things that you don't enjoy doing so much in recruitment is gonna be taken away.
00:31:25.039 --> 00:31:25.759
You're not gonna have to do that.
00:31:26.000 --> 00:31:27.279
You don't have to scroll LinkedIn anymore.
00:31:27.440 --> 00:31:27.680
Exactly.
00:31:27.759 --> 00:31:29.920
You don't have to scroll doom scroll LinkedIn anymore.
00:31:30.160 --> 00:31:31.359
Exactly, exactly.
00:31:31.519 --> 00:31:35.200
And um and do all the admin work and all that sort of stuff, right?
00:31:35.440 --> 00:31:52.880
And um, so you're gonna that's gonna be taken care of by AI and agentic AI, and then you're gonna be left to do all the cool and fun stuff, talking to super smart people, negotiating, you know, coming to you know, I didn't you know, coming to a good deal that works for both sides and uh have impact, right?
00:31:53.119 --> 00:31:56.960
So it's uh in that from that perspective, it's all good news, right?
00:31:57.200 --> 00:32:05.519
So I'm just curious, did did that message resonate with people that you you hired, or were there any pushback, you know, when you made that statement as well?
00:32:05.680 --> 00:32:07.839
Because I'm sure some people might not trust that fully, right?
00:32:07.920 --> 00:32:09.680
Like how how how do I take your word for it?
00:32:10.000 --> 00:32:11.519
Yeah, it's a good point.
00:32:11.680 --> 00:32:17.039
I think we have a uh I think we're fortunate enough to have a very high trust culture.
00:32:17.359 --> 00:32:24.480
And the uh so the people that uh came on board, I mean we're a small team, we're just eight people in the business, right?
00:32:24.559 --> 00:32:25.920
So it's easier that way.
00:32:26.079 --> 00:32:36.720
But uh the people that came on board, like Sean, that we're seeing her name here on the system and stuff, she she's the last batch, and she's very she feels very happy about it.
00:32:36.799 --> 00:32:42.720
She's the last batch that actually saw the old All of Sudden company, and it's been through the whole transition.
00:32:42.799 --> 00:32:51.039
So the later, because as they've joined lately, they've only seen us as a as our own tech, having our own tech platform and being an AI native firm.
00:32:51.200 --> 00:32:58.240
But she's been through the whole transition, she's appreciating what it used to be like and what it is like now, uh, which is which is pretty cool.
00:32:58.400 --> 00:33:02.640
And uh, but no, it hasn't been for us, it hasn't been any pushback.
00:33:02.720 --> 00:33:28.880
People have been really AI forward, but I think you have to in order for that to happen, I think there is a response you have the responsibility as a leader to work really hard at uh building that trust, and that's similar to what you said as well, Yahong, about the in recruitment to have uh to work with a recruiter that you can feel has your best interest at heart, either as a client or as a candidate.
00:33:29.119 --> 00:33:45.920
Because of the way that the commercials are structured in our business model, in our industry, it's not obvious, it's not obvious or intuitively intuitively true for a client or a candidate that me as a recruiter has your best interest or your best interest at heart.
00:33:46.160 --> 00:33:48.640
Our fee is a percentage on salary.
00:33:49.519 --> 00:33:54.880
So, from a client perspective, it looks like we probably would always want to try to push salaries up because then we make more money.
00:33:55.039 --> 00:34:08.159
Likewise, for a candidate, if you think about it, when we are trying to guide and advise you as a candidate towards the end, and we're trying to maybe give you the reasons why this could be a phenomenal career move for you to take this opportunity as a candidate.
00:34:08.400 --> 00:34:21.039
And the way that I always explain it to candidates is that most recruitment firms, whether you know it or not,
they would celebrate the placement where you sign the offer and they would be super happy.
00:34:21.440 --> 00:34:35.280
Or they would celebrate the uh deal when you're starting the first day when you're starting, or they would celebrate when you pass your probate your guarantee period of three months, and if you resign now, we don't have to pay anything back.
00:34:37.199 --> 00:34:49.840
We do not we celebrate a year after you've joined, but we're meeting for a lunch or a drink or a cup of coffee, and you're telling us, Roger, I'm so happy this is the best career move I've ever made in my life.
00:34:50.239 --> 00:34:53.039
That's when we know that we did the right thing.
00:34:53.440 --> 00:34:54.400
You measured it.
00:34:55.039 --> 00:34:57.360
So yeah, yeah, yeah, we we do, we do.
00:34:58.239 --> 00:35:04.400
And uh yeah, and uh and uh our percentage success rate is really, really high.
00:35:04.559 --> 00:35:10.719
You know, it's in the 90% plus and the scope level.
00:35:33.840 --> 00:35:35.599
Two observations from your platform, right?
00:35:35.760 --> 00:35:38.719
One is that it seems like the human is usually in the loop.
00:35:38.880 --> 00:35:45.119
And the second one is it seems like you're also training the system to do better than it was with the upvotes and downvotes.
00:35:45.280 --> 00:35:48.960
Can you explain a bit more about the philosophy around that as well?
00:35:49.280 --> 00:35:53.280
Yes, we do like to have uh human in the loop.
00:35:53.519 --> 00:36:01.440
We really believe in that, that that human judgment is so important uh, particularly in this area because we're dealing we're only dealing with humans.
00:36:01.519 --> 00:36:10.719
And to think that something unhuman would be able to miss ironically, but it's all all robotics inside, yeah.
00:36:10.960 --> 00:36:12.000
Exactly, exactly.
00:36:12.239 --> 00:36:23.599
And um, so um, but we are also having kind of uh now built abilities for the agent to do quite relatively long horizon tasks.
00:36:23.760 --> 00:36:31.519
So as an example, I was meeting with a uh I bumped into a climb downstairs with both runners, so he's a marathon runner as well.
00:36:31.679 --> 00:36:39.760
And so and then they are an AI company and like super uh um advanced, and um, we work very closely with them.
00:36:39.920 --> 00:36:44.400
And then and and I just said, like man, I've yeah, I'm so excited.
00:36:44.480 --> 00:37:08.159
I'm just gonna go up now to the office and I'm gonna see what uh uh what results uh has showed up on my computer, uh my agent, because uh one of our consultants have just started a search for an account executive for a data um integration platform company in Hong Kong, and uh it's slightly different role than we've done before, so we kind of extracted new data and run the agent.
00:37:08.239 --> 00:37:26.000
So the agent has been running throughout when I'm out running, so uh, and then now the the I will have a hopefully a short list of or long list of 50 really good candidates for my consultant to start looking at, and they don't have to do any of the search, and uh and I've been able to be out doing a run.
00:37:26.079 --> 00:37:38.880
And he's like he he was he was so excited because he said, like, that's exactly you know, with uh with AI, it's not about just uh improving or chipping away of your processes and making them a little bit faster, a little bit better.
00:37:39.039 --> 00:37:50.320
It's about reimagining the way that you do your work and to do it in a completely different way that you have uh you know you've never you haven't had the new uh discovering new capabilities, if you like.
00:37:50.639 --> 00:37:54.239
And uh, so that's a lot what we're doing now.
00:37:54.400 --> 00:38:19.920
We have those uh those so we can we can actually we can kick off and run many searches in parallel that goes off and does work and across different countries whilst we go off and do other things, whether we are doing other work or we're in an interview or we go out for a run or go out for lunch or whatever it is, and then we come back and then put the human in the loop touch in terms of uh still when it comes to then the screening and decision making around which one are the right candidates.
00:38:20.079 --> 00:38:23.199
We don't know what we don't want to automate the whole process and to end.
00:38:23.679 --> 00:39:11.760
How much of the long-running tasks capability has been due to like the frontier models improving versus like you guys training the model with all your upvotes, downvotes and comments and eval and evaluations and so the I mean the long the cap capacity to do more you know better instruction following and longer horizon tasks that comes almost entirely from the improvement of the models and the frontier models, and that's one of the reasons why we feel um we have an opportunity in uh in uh our business being a startup and being a tech platform company now and sort of an AI lab in the space of human or talent matching, if you like, right?
00:39:12.000 --> 00:39:19.119
We're able to constantly use the latest models and then test them and tune more with context engineering.
00:39:19.199 --> 00:39:24.079
And we're actually following again, maybe by like if you look at you could do it in different ways.
00:39:24.239 --> 00:39:32.239
Like if you look at uh, for example, in the agentic AI for law, yeah, you have Harvey out of the US and you have Legora out of Sweden.
00:39:32.400 --> 00:39:35.039
And again, I've happened to be siding more with the Swedish.
00:39:35.360 --> 00:39:38.000
Forgive me for that, being a little bit patriotic again.
00:39:38.159 --> 00:39:44.719
Um and Harvey has heavily post-trained and fine-tuned models for their systems.
00:39:45.119 --> 00:39:48.239
Legora has done zero, zero fine-tuning models.
00:39:48.400 --> 00:39:53.199
But I think Harvey has shifted to use more of the models from their models instead.
00:39:53.360 --> 00:39:53.440
Yeah.
00:39:53.760 -->
00:39:56.000
So they realize that there was no point in overtraining it.
00:39:56.239 --> 00:39:56.559
Exactly.
00:39:56.719 --> 00:40:00.800
So they exactly so they've kind of admitted defeat in terms of the strategy in a way, right?
00:40:01.280 --> 00:40:09.039
Because Legora is eating their lunch in Europe, they're taking over Europe, and now they're taking the battle to the US, and both are fantastic companies.
00:40:09.199 --> 00:40:11.280
But again, and that's the timing.
00:40:11.440 --> 00:40:19.280
So Harvey started earlier, so they didn't know how advanced the models were gonna get then, and they also got a lot of funding.
00:40:19.840 --> 00:40:22.400
Legora started much later, two years ago.
00:40:22.719 --> 00:40:29.760
Models were a lot more advanced, and also they only had$50,000 worth of funding, so they couldn't buy GPUs, they had no choice.
00:40:30.079 --> 00:40:36.639
So they had to play the only choice was to let's bet on the models become developing so fast, yeah.
00:40:36.719 --> 00:40:40.000
Yeah, and now so they build so we want to be the same for recruitment in a way.
00:40:40.079 --> 00:40:46.880
We want to build the most advanced harness, the most advanced agentic framework that constantly plugs in place the best models.
00:40:47.039 --> 00:40:51.920
So the first time reasoning models came out, 4.0, we were working with that.
00:40:52.400 --> 00:41:09.679
And then as soon as Gemini 2.5 Pro came out in April, about one year ago, one about 14 months ago, the experiment, the preview version, with 1 million token context window, the first one that had 1 million token contacts and reasoning, it was the best.
00:41:10.000 --> 00:41:11.760
That was a game changer for us.
00:41:12.239 --> 00:41:31.920
I still meet constantly or often I meet with a candidate that I placed that I we're joking about all the time that he might be having a badge of honor, unofficially at least, to be possibly the first candidate that was selected by Gemini 2.5 Pro for a job matching exercise in the world.
00:41:32.320 --> 00:41:33.840
Not just in Singapore, but in the world.
00:41:34.559 --> 00:41:43.679
And he was one out of three candidates that Gemini 2.5 Pro uh shortlisted for a job when we put it into production when it was brand new.
00:41:43.840 --> 00:41:48.079
No one, I guarantee you, we were the first for a recruitment-specific use case.
00:41:48.159 --> 00:41:50.559
Yeah, we must have been the first in Singapore for sure, yeah.
00:41:50.639 --> 00:41:52.159
Yeah, to use that model.
00:41:52.480 --> 00:41:53.840
And he was one of the three.
00:41:53.920 --> 00:42:02.559
I couldn't believe the the precision that that model had with the tuning and the context tuning that we brought to it.
00:42:02.960 --> 00:42:04.239
And he got the job.
00:42:04.639 --> 00:42:16.880
And um and then so then we had moved from GPT to Gemini, and then now these last six months we've been very uh entropic or cloud-centric, right?
00:42:17.360 --> 00:42:21.039
Um since 4.5 came out, right?
00:42:21.119 --> 00:42:28.480
There was the shift where the sound the reasoning capability, of course, on coding, but not just coding, went to a new level, right?
00:42:28.559 --> 00:42:30.960
Yeah, and then who knows what's gonna be next.
00:42:31.039 --> 00:42:42.000
And we are relatively we are compl in some ways, but we are we are completely or actually not in some way, we're completely model neutral in a way.
00:42:42.320 --> 00:42:56.880
And very similar to how Max sees it in Lagora is that for their clients, and they obviously built the software, we're not we are an AI native services company currently, uh AI powered recruitment platform or tech company.
00:42:57.519 --> 00:42:59.519
One last thing before I move off the platform that you built.
00:42:59.760 --> 00:43:07.440
You also mentioned that now given that you know Cloud Code and all is improving, um, you guys are shifting a bit more off your your own platform.
00:43:07.519 --> 00:43:10.159
You're kind of rediscovering new kind of workflows.
00:43:10.400 --> 00:43:22.079
Um can can do you do you mind sharing a bit more about what drove that decision when it comes to you know um like going back to the terminal, going back to Clot Code or the Cloud app, right?
00:43:22.159 --> 00:43:23.679
Like what what drove that decision?
00:43:23.920 --> 00:43:29.039
Other than API costs, I guess APIs are I mean token costs are quite expensive as well.
00:43:29.119 --> 00:43:32.400
But but maybe beyond that, like how how do you think about using that?
00:43:32.719 --> 00:43:35.039
So yes, really good question, Eric.
00:43:35.119 --> 00:43:41.440
And I think I think that the so we have a our team has become like uh semi technical, right?
00:43:41.519 --> 00:43:42.719
You know, even though they're recruiters.
00:43:42.800 --> 00:43:47.519
So they're all now we've trained them on using uh the CLI, the terminal.
00:43:47.920 --> 00:43:50.239
Do they have like a 5x or 20x plan or something?
00:43:50.800 --> 00:43:53.760
So uh yeah, so we have we have a combination of them.
00:43:53.840 --> 00:43:59.760
We have uh we have the the top end max plans and then uh you know mid level, the the hundred you know, hundred dollar time.
00:44:01.280 --> 00:44:02.639
I think that's the 5x, right?
00:44:03.199 --> 00:44:20.719
And um and uh and then we because it's like uh so now then everybody has access to clawed code and that has access to our you know the the platform, the agentic platform that we built, and that works in synergy with each other.
00:44:20.800 --> 00:44:32.559
So you can suddenly now build custom build small pieces of software if you want to, or just uh do any tasks on the system directly from the terminal, and everybody has become comfortable with it.
00:44:32.639 --> 00:44:42.719
And in some ways, because things are so fast moving, it it's there's a reason why you know Entropic itself choose the terminal as kind of the main interface, right?
00:44:42.800 --> 00:44:51.280
Because it's just so it is uh very basic, but it's also then super flexible and versatile, exactly.
00:44:51.440 --> 00:44:54.320
So now we have uh we have all kinds of capabilities.
00:44:54.400 --> 00:45:04.880
If I wanna most of the time these days I don't touch the UI of the system that much anymore, I just tell Claude Code to do this for me, do that, update these things, book this, book that.
00:45:05.039 --> 00:45:12.400
Uh, if it is something that is not this outside the standard kind of outreach and the interview booking or the the search.
00:45:12.880 --> 00:45:18.000
So you just use like MCPs and all to connect it to different apps, exactly.
00:45:18.079 --> 00:45:49.440
And then if for example, like now on um Monday mornings we have our team sales meeting, and I just have a um uh job then that goes, or I can either do it manually, uh ask Claude Code to produce a report of all the important activities that happened last week and then send, you know, just update it or package it in a in a pre-sort of packaged reporting format and uh or and even just use its own intelligence to flag anything that I need to be aware of and stuff, and then just send it to me uh in an email to me.
00:45:49.679 --> 00:45:54.480
And uh so we can even do you know, as we uh are you showing something?
00:45:54.880 --> 00:45:56.639
Send yeah, yeah, I could do that.
00:45:56.880 --> 00:45:57.039
Okay.
00:45:57.679 --> 00:46:02.320
Alright, so for now, Roger has put send me a report on this week's activities on email.
00:46:02.480 --> 00:46:02.719
All right.
00:46:02.880 --> 00:46:03.119
Exactly.
00:46:03.280 --> 00:46:07.440
Yeah, so I can see it's calling Superbase, is looking through your meetings, I guess.
00:46:07.679 --> 00:46:08.480
Exactly, exactly.
00:46:08.559 --> 00:46:13.119
So it just goes through all the activities for the whole team, all the consultants across the week.
00:46:13.280 --> 00:46:20.480
So it picks the dates and stuff and then uh runs through and then Superbase, I guess, is the one that's powering your platform right now.
00:46:20.719 --> 00:46:21.440
Yeah, exactly.
00:46:21.599 --> 00:46:35.599
So it's a Postgres SQL database on Superbase, hosted on Versell, and then uh everything is uh mainly React TypeScript built in terms of the the the front end and the logic and stuff.
00:46:35.840 --> 00:46:44.639
And uh so then it just we can leave that for now and then it will uh hopefully turn up as an email in my inbox uh in a few minutes' time.
00:46:44.960 --> 00:46:45.519
We'll get back to that.
00:46:45.840 --> 00:46:55.199
I guess uh the biggest question that I have right now is that from a recruiter's point of view, do you see the market changing, right, in terms of the roles that people are hiring for?
00:46:55.440 --> 00:46:56.800
Before this whole wave, right?
00:46:56.880 --> 00:47:06.000
Like five years ago, full stack engineer was a thing, front-end engineer was a thing, but now if I can a backend engineer can actually
do front-end work, then what makes him a back-end engineer, you know?
00:47:06.239 --> 00:47:07.599
Yeah, yeah, yeah.
00:47:07.920 --> 00:47:18.079
Um definitely this is this is the most transformative time that we have experienced uh in probably all of human history, to be honest.
00:47:18.320 --> 00:47:44.559
It's uh you know, I think that um uh Demi Sasabis has a very, very cool framework for this, which is like a 10x 10x kind of uh mind map of it that is where he says that the the AI revolution is gonna have 10 times the impact of the industrial revolution, but at a tenth of the time, 10 times speed.
00:47:44.800 --> 00:47:46.480
It's gonna be 10 times faster.
00:47:46.800 --> 00:47:58.159
So the industrial revolution took maybe about 150 years, so in 15 years we're gonna have 10x as much change, which you can imagine like that's hundred times the experience, right?
00:47:58.320 --> 00:48:10.079
Hundred times delta compared to what we experienced in the industrial revolution, and I think that's probably true, and that's the way it feels like it's moving really, really fast, and uh and everything is almost kind of changing by the day.
00:48:10.320 --> 00:48:13.280
Yeah, how how does that I I guess how do you combat that feeling?
00:48:13.360 --> 00:48:15.519
Because it it always feels quite fearful, right?
00:48:15.679 --> 00:48:20.960
For for engineers to for people in tech to think about, and and how do you kind of how are you address that as well?
00:48:21.199 --> 00:48:27.920
I would encourage everyone, this is really, really hard, right, to do, but I would encourage everyone to try not to be scared.
00:48:28.239 --> 00:48:29.360
It's not easy, right?
00:48:29.519 --> 00:48:39.840
But because fear is a very destructive emotion, there's kind of limits, suboptimizes our ability to maximize our our our our experience in life in a way, right?
00:48:40.239 --> 00:48:43.679
And uh and uh I would say so.
00:48:43.760 --> 00:49:02.800
For example, now I've done I've always done a lot of career talks, both to experienced uh people, professionals, leaders, uh at times, and also uh fresh grads or pe or uh students coming out of university, etc., and getting ready for the for work life or MBA students, etc.
00:49:03.199 --> 00:49:17.760
So across you so a lot of career talks, and um and for my whole career, there's all I've always been very certain about, for example, this job category or to become a software engineer will be always good.
00:49:17.840 --> 00:49:19.360
It's always you need more of them, right?
00:49:20.480 --> 00:49:21.360
Except now.
00:49:21.599 --> 00:49:29.280
Now I I've started to say to people honestly, I don't really care what you study and what you want to do in life.
00:49:29.440 --> 00:49:39.360
All I really care about is that you figure out how to use utilize AI to do whatever you love doing or want to do better and faster.
00:49:39.920 --> 00:49:42.400
That's that's it, that's the most important thing.
00:49:42.719 --> 00:49:50.880
Because I don't really it's very hard to predict what type of jobs that's gonna be still there and it's gonna be relevant.
00:49:51.119 --> 00:50:11.920
Um but I think the signal what we can predict predict on is that every job is gonna become almost every single job on the planet is gonna be done better by a human that is assisted by or knows how to use AI tools in a really, really useful and powerful way.
00:50:12.480 --> 00:50:27.519
And uh I really I really believe that it's true that what I've heard a bunch of people say in the market that the competition is not gonna be between humans and AI, it's gonna be t between humans not using AI and humans that use AI.
00:50:28.000 --> 00:50:32.960
And uh I and actually I think I actually really I'm very positive, uh optimistic about this.
00:50:33.119 --> 00:50:50.800
I think that uh I I I do think that we're gonna live in a world where there is gonna be coding is almost coding in software is gonna become and software as a service in a way, more so than the traditional SaaS side.
00:50:50.960 --> 00:50:56.719
Uh software on the fly, maybe as a service, is gonna become so much part of what we do.
00:50:57.119 --> 00:51:01.599
We're just gonna build and throw and build and throw and build and rebuild and build on top of, etc.
00:51:01.840 --> 00:51:11.280
And everybody's gonna do that almost, and uh in a way so that uh I think that the we're not gonna need less software engineering capability.
00:51:11.440 --> 00:51:33.360
Most of us are gonna be able to do it uh on the fly for basic things, our sm our private needs, or as a small business, or whatever, but then it's still gonna be needed to do it to be able to understand at a very large, complex scaled architectural level, or from a security standpoint, how to get these pieces together to actually work reliably and safely.
00:51:33.519 --> 00:51:37.440
And that's still gonna need engineering skills, architectural skills.
00:51:37.760 --> 00:51:45.679
And uh so I'm I'm very optimistic that uh there's gonna be a place for software engineers in the future as well.
00:51:46.079 --> 00:51:50.960
Yeah, I I think um software engineering as a as a job, right?
00:51:51.119 --> 00:51:55.280
The coding is actually not the main part actually, right?
00:51:55.360 --> 00:51:56.559
It's the problem solving, right?
00:51:56.639 --> 00:51:56.880
Yeah.
00:51:57.039 --> 00:52:03.199
Uh trying to figure out what you're trying to solve and to solve it in a sometimes efficient manner or cost-efficient manner, right?
00:52:03.599 --> 00:52:05.199
Maybe one last point on this one.
00:52:05.280 --> 00:52:12.079
Do you actually see organizations changing the roles that they're hiring for and the the whole like job description itself?
00:52:12.159 --> 00:52:13.599
Like, what is that evolving into?
00:52:14.079 --> 00:52:19.920
Yeah, uh I think job descriptions are still sort of job descriptions, but um you know, I don't I don't think it's a lot of people.
00:52:21.519 --> 00:52:25.199
Yeah, the skills that's required are different, and it's uh and that is changing.
00:52:25.280 --> 00:52:33.920
And I've not seen we're seeing a lot more kind of evolvement and new jobs coming up uh uh faster than uh we've seen before.
00:52:34.079 --> 00:52:54.000
Like uh I mean obviously we had prompt engineers that started the started, and then yeah, and then context engineers, and then uh and then we had uh you know now earlier this year or late last year, late last year we started working on uh eight agenc product managers for the first time.
00:52:54.320 --> 00:52:57.599
We'd only worked on product managers before that, and that's actually different.
00:52:57.760 --> 00:53:08.960
So it's in the easiest way to kind of explain it is the perfect agency product manager is the merging or the breed of a McKinsey consultant with uh with like a Google product manager, right?
00:53:09.039 --> 00:53:11.519
Because Google kind of invented modern product management, right?
00:53:11.679 --> 00:53:21.039
Yeah, uh the uh what's her name, the the the lady that became the CEO of Yahoo was the the one of the sort of first product management leads.
00:53:22.079 --> 00:53:24.480
Yeah, Marissa, Marissa Mayer, Mayer Mayers.
00:53:24.800 --> 00:53:30.159
They really that what that that was where modern tech product management was invented, right?
00:53:30.400 --> 00:53:31.599
Uh in that team.
00:53:32.320 --> 00:54:12.719
And so if you combine like a strategy consultant from McKinsey that knows how to engage with clients and solve problems, yeah, uh, together with like a Google product manager, if you like, and and uh that's the the perfect sort of amalgamation for what would be an agentic product manager now and in the future that then uh builds these agent solutions on site with customers based on the agentic framework uh harness platform that could be a Sierra or a uh uh Ligora or Harvey or uh um anyone else of the new platforms that are coming in, right?
00:54:12.800 --> 00:54:17.679
So that's a new role that wasn't there just uh maybe 18 months ago.
00:54:17.840 --> 00:54:18.400
Yeah, yeah, yeah.
00:54:18.960 --> 00:54:23.280
Because maybe Sierra came up with it and they're only two years old as a company two years plus.
00:54:23.679 --> 00:54:53.440
And um and then we have uh started to work on another another one that is really cool, is that in the customer service space, so the so one of our clients wants to work on a building a program with the with the Singapore government to be able to actually to take customer service uh team leads uh customer support team leads and uh retrain them to become uh agent orchestrators.
00:54:55.360 --> 00:54:56.000
Interesting.
00:54:56.400 --> 00:55:04.159
Because then now they're gonna manage mostly a team of agents because the the responses will actually be agentic by in nature, exactly.
00:55:04.480 --> 00:55:09.599
And this people's responsibility make sure that the agents don't give you
weird responses.
00:55:09.920 --> 00:55:10.320
Exactly.
00:55:10.400 --> 00:55:11.679
And they and they then gonna work.
00:55:11.920 --> 00:55:25.199
But even the agents were the motion that they follow when they actually then deploy a solution is more similar to onboarding a person or people rather than uh implementing and configuring software.
00:55:25.599 --> 00:56:00.800
So they bring their they they come in with their agent product managers of four and slash or and/or slash forward deployed engineers, yeah, and then they start to actually onboard this software, this agent rather than implement it, and then train it together, and then that's where you need the agent orchestrator on the client side, the person that knows the domain of customer service, and then train and onboard and give the instructions and the material for these agents to be able to then get up to speed and start to be able to deal with the different uh uh customer service inquiries and edge cases and these kind of things, right?
00:56:01.039 --> 00:56:01.519
So cool.
00:56:01.679 --> 00:56:02.960
Yeah, more jobs, please.
00:56:03.199 --> 00:56:03.920
Yeah, yeah, yeah.
00:56:04.159 --> 00:56:06.239
No, this is cool because I job jobs will change.
00:56:06.400 --> 00:56:14.559
No, no, I I no, I think that's that's that's really good because right now you just hear about FDEs and everyone's talking about FDEs, but I mean this had a lot of meat.
00:56:14.960 --> 00:56:16.480
Exactly, exactly what they're gonna look like.
00:56:16.719 --> 00:56:17.119
Exactly.
00:56:17.199 --> 00:56:28.800
And honestly, I think so so three years ago after the ChatGPT event, uh I was uh I had an extension existential crisis almost, and I was really scared about we're not gonna have jobs as humans anymore.
00:56:28.960 --> 00:56:34.719
I was kind of a little bit negative there for a period of time, and I've gone through that now into becoming super positive.
00:56:34.880 --> 00:56:51.039
I'm like, I become 100% convinced that we are just humans, are just we're just too creative, we are too curious, we're too competitive, we're too excited, we're too petty about things, we want to win, we want to build things, we want to explore and stuff.
00:56:51.199 --> 00:56:54.159
We there's no way that we're not gonna find ways to be busy.
00:56:54.480 --> 00:57:04.159
And honestly, like me and everyone I speak to that is in the AI world, even though I get maybe 10 times as much done today, uh in a way, sometimes 10 times more busy.
00:57:04.320 --> 00:57:09.599
Yeah, I'm actually I probably don't get maybe I get twice as much done and I'm 10 times as busy in real time.
00:57:09.760 --> 00:57:11.519
I hardly have time to sleep, you know?
00:57:11.760 --> 00:57:18.719
And uh so and that's most of my friends in the industry and stuff uh report the same experience, right?
00:57:19.039 --> 00:57:23.519
So I think it is uh I think I think this is uh a very exciting time.
00:57:23.599 --> 00:57:30.719
I think there's gonna be huge opportunities for people uh in new types of jobs and new challenges.
00:57:31.039 --> 00:57:36.639
And um I think uh and another thing also that is more localized.
00:57:36.880 --> 00:57:49.599
I was a bit all at the same time, I was also worried about Singapore's place on the planet because we have been able to be a really good subcontract manufacturer in the early days in the 60s and 70s.
00:57:49.679 --> 00:58:00.079
That's how we kind of started to develop into a developed nation, and then we became a really good financial services hub in the late 80s and 90s, early 2000s.
00:58:00.239 --> 00:58:04.000
Then we became an innovation and life science hub and etc.
00:58:04.239 --> 00:58:18.639
Right, and we've been able to find our relevance in the world over and over again because, quite frankly, I mean, our government executes extremely well, it comes up with really well thought-through strategies and executes consistently on it better than maybe any other government in the world.
00:58:18.800 --> 00:58:38.800
But I was worried about this time we might not be able to find our place, we might not be able to get it right, it might not even be possible to get it right because it's so large-scale, it is so centralized to Silicon Valley and then a black box of China in a way, and those are the two power centers, and there's nothing in between anymore.
00:58:39.199 --> 00:58:43.760
And I can say, so three years ago I was really worried, two and a half years ago.
00:58:44.000 --> 00:59:03.599
Now I am so excited and so happy about Singapore's position in the new world of AI, in the world order, and how well, how much of a center in Asia, and one of the centers or hubs in the world that I've seen that we've started to become.
00:59:03.760 --> 00:59:22.400
We have OpenAI here, we have Google DeepMind, we have X.ai, we have several early stage startups, series A to C with hundreds of millions of dollars worth of funding that are committed and built, that are building AI labs in Singapore, that are building AI engineering hubs for distributed computing, etc.
00:59:22.800 --> 00:59:26.079
And uh it's uh it's it's a good time.
00:59:26.239 --> 00:59:32.079
It's a good time to be a human being, it's a good time to be a builder, and it's a good time to be in Singapore.
00:59:33.599 --> 00:59:35.039
Very positive end to that.
00:59:36.159 --> 00:59:36.800
Perfect.
00:59:38.000 --> 00:59:38.960
Love that, love that.
00:59:39.039 --> 00:59:42.719
Well, because I mean most of the time when you hear about yeah, it's all my jobs are gonna be gone.
00:59:42.960 --> 00:59:45.440
Yeah, it's gonna take over my lunch, blah, blah, blah.
00:59:45.519 --> 00:59:45.920
You know, yeah.
00:59:46.480 --> 00:59:46.639
Yeah.
00:59:46.880 --> 00:59:51.840
And maybe with that, uh, we've come to the we've come to our very exciting lightning run.
00:59:52.079 --> 00:59:52.639
Yeah, yes.
00:59:52.800 --> 00:59:53.760
Are you ready, Roger?
00:59:54.000 -->
00:59:55.039
Okay, yes, yes, yes.
00:59:55.519 --> 00:59:55.679
Yes.
00:59:56.000 --> 00:59:58.960
Um, what's two or three books that you often recommend?
00:59:59.679 --> 01:00:10.079
Oh, one that comes immediately to mind is uh Hard The Hard Things About Hard Things by uh uh Ben Horowitz in A16Z.
01:00:10.719 --> 01:00:11.920
It is phenomenal.
01:00:12.079 --> 01:00:23.119
One of if you're gonna read one, if you're an entrepreneur, early stage startup builder, and you're just gonna read one book, then that's the book to read, particularly if you're going through difficult times.
01:00:23.280 --> 01:00:25.760
I read it the first time, I read it twice now.
01:00:25.920 --> 01:00:36.960
I read the first time when we were going through before we bet the life of the company several times over these last eight years, but we're going through a real difficult time.
01:00:37.599 --> 01:00:51.599
And that was the first time in a while that I was laughing because when he was going through the stuff that he had gone through, Ben Horowitz, in his first startup or second startup, um it made me feel like man, that what I'm going through is nothing.
01:00:51.679 --> 01:01:12.239
And it was just uh I was out running listening to it, and uh it's a that has so many practical um learnings or or teachings about how to get through anything from the greatest challenges to the biggest opportunities uh to capitalize on that.
01:01:12.400 --> 01:01:33.599
Another one is Peter Thiel's zero to one brilliant as well, and I love how Peter Thiel is so focused on both top talent but also getting the best getting making sure that people are people that love working together work together.
01:01:33.920 --> 01:01:46.719
He does not underrate the importance of you know relationships and liking each other, knowing each other, trusting each other, and those are two phenomenal books.
01:01:47.039 --> 01:01:51.840
Okay, uh a favorite uh kind of content or show or film that you enjoy.
01:01:53.039 --> 01:01:55.440
Favourite UFC podcast.
01:01:58.079 --> 01:02:00.960
Oh man, yeah, podcast is also good.
01:02:01.440 --> 01:02:04.239
Yeah, 20 VC, Harry Stabbings 20 VC.
01:02:04.400 --> 01:02:34.800
I think that guy is doing a phenomenal job getting some of the best VCs and startup guys or founders on the podcast and asking just brilliant questions, getting the best out of them, and UFC because there is something extremely uh honorable and even spiritual, and then at the leading edge of at the bleeding edge of competition in uh something as kind of sort of seemingly barbaric as mixed martial arts and cage fighting.
01:02:35.119 --> 01:02:38.320
What was uh a place that in Singapore that is special to you?
01:02:42.400 --> 01:02:51.039
It's a uh awesome uh halal buffet dinner place in Raffin City in uh Fairmount Hotel.
01:02:51.119 --> 01:03:00.320
We spent many family dinners there with my twin daughters and their birthdays, my wife's birthday, my mother-in-law's birthday, etc.
01:03:00.480 --> 01:03:02.079
So it brings very fond memories to me.
01:03:02.719 --> 01:03:03.840
Thanks for sharing that place.
01:03:05.039 --> 01:03:08.400
What what's the line or philosophy you go back to?
01:03:09.039 --> 01:03:17.599
Never make a decision that might make you feel like you're winning in the short term at the sacrifice of long-term relationships.
01:03:18.079 --> 01:03:18.800
Never do that.
01:03:18.960 --> 01:03:20.800
Never do it, it's never worth it.
01:03:21.199 --> 01:03:30.880
Even if you think I could gain a little bit from this, but if that at all is any risk of sacrificing or damaging long-term relationships, the answer is simple.
01:03:31.039 --> 01:03:31.760
Never do it.
01:03:32.400 --> 01:03:42.559
And if you live by that principle, you become someone that people will want to build and maintain long-term relationships with.
01:03:43.360 --> 01:03:48.159
Well, what's a crazy fitness goal you've done of a crazy fitness thing that you've done?
01:03:48.480 --> 01:03:53.039
The current goal that I have is to to to complete and I've run 10 marathons.
01:03:53.119 --> 01:03:54.719
So I've run six marathons so far.
01:03:55.519 --> 01:04:00.480
So initially I did half marathons and then the goal became No, no, they can be over any time.
01:04:00.559 --> 01:04:02.159
But maybe I should do it in a year, right?
01:04:02.480 --> 01:04:03.599
That might be the next one.
01:04:03.840 --> 01:04:11.119
The thing, the scary thing now, though, is that because initially when I did my first marathon, it was about will I be able to do it or not, right?
01:04:11.280 --> 01:04:15.519
But now it's not a question about that anymore because I've done six, so I know I can do it.
01:04:15.679 --> 01:04:28.320
So now the scary thing is that some of my founder and VC friends and stuff, and tech friends, they're into ultra marathons, and I'm getting really scared that I am I might have to start to do that.
01:04:30.639 --> 01:04:34.960
Last one, uh, where can people find you and how can listeners be helpful to you?
01:04:35.360 --> 01:04:38.400
So I am uh probably easiest to find on LinkedIn.
01:04:38.559 --> 01:04:42.480
That's probably the way the the the social platform that I use the most.
01:04:42.559 --> 01:04:59.840
Um I'm uh and um and I I would be and otherwise in any uh often in meetup groups in uh agentic coding AI type of activities and stuff.
01:05:00.159 --> 01:05:06.559
Um if you are around Raffles Place, you know, you're probably likely to bump into me in at some point of time.
01:05:06.880 --> 01:05:08.880
But um yeah, thanks so much, Roger.
01:05:09.119 --> 01:05:09.920
Thanks for sticking with us.
01:05:10.239 --> 01:05:15.119
Thank you guys, it's been amazing, it's been awesome, uh really, really fun, really enjoyable.
01:05:15.440 --> 01:05:17.360
Thank you so much for uh setting this up.
01:05:17.679 --> 01:05:20.239
Yeah, thank you so much for listening to our episode.
01:05:20.320 --> 01:05:23.199
We hope you enjoyed it and learned as much as we did.
01:05:23.440 --> 01:05:25.360
You can check out our sub stack.
01:05:25.519 --> 01:05:36.880
And if you liked what you heard, please feel free to subscribe, like, and share on your favorite channels on YouTube, Spotify, Apple Podcasts, and more.
01:05:37.039 --> 01:05:39.679
And we hope to see you in the next episode.
01:05:42.800 --> 01:05:44.239
Post credit scene.
01:05:44.400 --> 01:05:46.079
What came into your email?
01:05:46.480 --> 01:05:48.400
Oh yes, that is good.
01:05:48.559 --> 01:05:49.280
Love it, love it.
01:05:49.440 --> 01:05:50.960
It's pretty nice.
01:05:51.280 --> 01:05:54.239
So during the two total actions, yeah, 47, 20.
01:05:57.760 --> 01:05:59.440
I'm not doing I'm not doing anything.
01:05:59.519 --> 01:06:01.440
I'm hanging out doing podcasts and stuff.
01:06:03.280 --> 01:06:03.920
Excited.
01:06:04.159 --> 01:06:16.159
So we see the activities across different consultants, Sean, what she's done, her pipeline, uh, what she's pipeline excluded, client meetings, too, uh the jobs, the projects she's working on.
01:06:16.320 --> 01:06:16.800
Yeah.
01:06:17.039 --> 01:06:24.239
Um the delivery activities across those core activity levels that are critical to our type of business.
01:06:24.639 --> 01:06:31.599
Um daily activities, and then you see it across for Clarence, day.
01:06:32.480 --> 01:06:33.280
Very good.
01:06:34.639 --> 01:06:35.840
And you Roger.
01:06:36.159 --> 01:06:36.639
And me.
01:06:36.800 --> 01:06:36.960
Yeah.
01:06:37.360 --> 01:06:39.199
Because the metrics for you are different from the rest.
01:06:39.599 --> 01:06:40.000
Look at that.
01:06:40.079 --> 01:06:40.559
That is tough.
01:06:40.719 --> 01:06:41.360
Yeah, exciting.
01:06:41.440 --> 01:06:45.199
I clearly hopefully I'm doing this.
01:06:45.280 --> 01:06:50.079
I can also perform on the team.
01:06:50.400 --> 01:06:51.360
Exactly, exactly.
01:06:51.519 --> 01:06:53.039
Always the case, always the case.
01:06:53.599 --> 01:06:56.960
You know, they're carrying me and uh yeah, so that's it.
01:06:57.039 --> 01:06:57.360
That's it.
01:06:57.840 --> 01:06:58.079
Cool, man.
01:06:58.239 --> 01:06:58.320
Yeah.
01:06:59.599 --> 01:06:59.679
Cool.
01:06:59.840 --> 01:07:00.320
Thanks, Eric.
01:07:01.679 --> 01:07:02.719
That's been fantastic, man.
01:07:02.800 --> 01:07:03.840
That was really that was that was really