逐字稿 🔗
WEBVTT
00:00:00.160 --> 00:00:04.480
There's this huge AI push, which isn't complemented by the security push.
00:00:04.799 --> 00:00:08.880
We are encouraging people to buy cars without seatbelts and no brakes.
00:00:09.039 --> 00:00:14.320
I tried Open Claw, I tried nanoclaw, I tried paperclip, I tried Hermes, I tried the whole bunch of things.
00:00:14.560 --> 00:00:17.280
And eventually I built my own harness.
00:00:17.440 --> 00:00:23.679
Philosophically, I don't believe that people only people who are rich can stay safe, and people who are poor cannot be safe.
00:00:24.000 --> 00:00:25.039
That doesn't sit well with me.
00:00:25.440 --> 00:00:27.519
One thing that people are trying to build now is a second brain.
00:00:27.600 --> 00:00:28.800
What should they take note of?
00:00:29.039 --> 00:00:31.440
Treat it like a like a teenager or a child.
00:00:32.000 --> 00:00:37.679
Start with the least possible permissions that you would give your child and then slowly give him more and more laugh.
00:00:37.920 --> 00:00:40.479
Minister Vivian showcased his nano-claw setup.
00:00:40.719 --> 00:00:43.759
He may be a medical doctor, but the dude is also technical.
00:00:44.159 --> 00:00:46.560
Like hardcore technical.
00:00:46.799 --> 00:00:48.479
Those are the only four things that can go wrong.
00:00:48.640 --> 00:00:51.759
And AI exploits all four of those without even knowing.
00:00:52.000 --> 00:00:55.759
Do you mind sharing a bit more about the harness as well in terms of the architecture?
00:00:56.079 --> 00:00:56.320
Sure.
00:00:56.479 --> 00:01:01.759
One of the interesting things we've started doing is when I hire people, I ask them to submit their AI transcripts.
00:01:01.840 --> 00:01:05.840
And you can tell a lot about how they think from how they interact with AI.
00:01:06.079 --> 00:01:08.560
The first and the third, I will never hire.
00:01:09.280 --> 00:01:12.560
Hi everyone, I'm Eric, co-host of the Code Rift Podcast.
00:01:12.719 --> 00:01:21.040
We are going to be interviewing leading AI practitioners so that you can see how AI is used for human flourishing as well as to have more fun with it.
00:01:21.200 --> 00:01:24.480
Today we have a very special guest, uh Garaf Kurti.
00:01:25.040 --> 00:01:26.319
Welcome to the podcast.
00:01:26.719 --> 00:01:27.840
Hey, thanks so much for having me.
00:01:28.079 --> 00:01:29.040
How are we gonna do this today?
00:01:29.120 --> 00:01:30.959
We're gonna do it uh mainly in four parts.
00:01:31.040 --> 00:01:33.439
So, first, we're gonna just understand a bit more about your story.
00:01:33.599 --> 00:01:36.879
Second, understand more about SMEs and the risks that they face.
00:01:37.040 --> 00:01:40.719
The third part is really around how can SMEs kind of protect themselves.
00:01:40.799 --> 00:01:46.079
Uh the last part is kind of big picture and trends in AI and tech and cybersecurity as well.
00:01:46.400 --> 00:02:01.439
So, starting from the first part, you've worn a lot of hats, you know, um guards, rescue pilot, uh, brigadier general, Air Force First Chief Innovation Officer, uh Cybersecurity Agency, and now you're uh co-founder and CEO at Strong Keep.
00:02:01.519 --> 00:02:04.079
How how do you actually introduce yourself these days?
00:02:04.480 --> 00:02:05.920
I tell people what I love doing.
00:02:06.000 --> 00:02:09.439
Um so I don't introduce myself by the job or the title or the role.
00:02:09.680 --> 00:02:12.000
And in fact, I've never really done that.
00:02:12.240 --> 00:02:14.800
Um I just tell people what I find meaningful.
00:02:14.960 --> 00:02:20.479
Um I find it meaningful to use whatever skills and knowledge I have to help other people feel safer and secure.
00:02:20.639 --> 00:02:22.400
Um that's a very special place in life.
00:02:22.560 --> 00:02:32.400
And whether I use that to run a startup, use that to advise companies, use that to advise board members, use that to you know build stuff, it it that's that's the how.
00:02:32.560 --> 00:02:34.319
The why is what I used to introduce myself.
00:02:34.639 --> 00:02:40.800
You wrote uh essay during one of the uh Toto Defense uh days in 2016.
00:02:41.039 --> 00:02:43.039
So I we've dug up quite far.
00:02:44.080 --> 00:02:57.759
Ten years ago, man, you guys think something I saw it in your essay and found it very interesting that you you grew up in multiple places, uh, you know, India, Nigeria, and more.
00:02:58.000 --> 00:03:04.000
And also um even in some violent or conflict prone zones as well.
00:03:04.240 --> 00:03:10.800
How has that affected your appreciation for cybersecurity and maybe even your time in the Air Force?
00:03:10.960 --> 00:03:14.879
And what was that kind of like common threat that uh tied everything through?
00:03:15.120 --> 00:03:20.560
So about 10 years ago, I was invited back to my Alma Meta Secondary School to give a speech on Total Defense Day.
00:03:20.800 --> 00:03:27.919
The speech itself was called Doubting Singapore's Defense, which is not the kind of a title of a speech that you give on Total Defense Day.
00:03:28.080 --> 00:03:33.840
But, you know, I I did a debate when I was a kid and I wanted to address any of the doubts head-on.
00:03:34.080 --> 00:03:35.759
So that was the context of that speech.
00:03:35.840 --> 00:03:40.400
And kind of to let people understand where I was coming from, I thought I'd share with them a bit about my childhood.
00:03:40.560 --> 00:03:44.719
So, Eric, as you rightly pointed out, I did not I was not born here, and I did not grow up here.
00:03:44.960 --> 00:03:47.680
I came here when I was about nine, ten years old.
00:03:48.000 --> 00:03:56.400
But the countries I'd lived in before that were rather I mean, they weren't meant to be crisis-torn, they were just accidentally crisis-filled.
00:03:56.639 --> 00:04:01.680
So I was born in India, but then very quickly, like less than a year later, we moved to a place called Nigeria.
00:04:01.919 --> 00:04:07.039
Uh, and Nigeria was actually,
Load full transcript
back in the 80s, uh, an engineering capital.
00:04:07.120 --> 00:04:16.800
Like people thought that Nigeria would grow up and become, you know, what Singapore is today, you know, full of engineers, lots of manufacturing, and today there's there's a lot of manufacturing there.
00:04:16.959 --> 00:04:27.600
But then in the early 80s, there was a military coup, and uh, you know, it didn't work out so well for the foreigners who were there, so there was a lot of uh internal violence and internal kind of disruption.
00:04:27.680 --> 00:04:32.240
So we left, went to India for a year, and actually we thought we were gonna move back to India.
00:04:32.399 --> 00:04:33.360
My parents did.
00:04:33.439 --> 00:04:43.519
Uh except the year that we were back in India, uh, the Prime Minister got assassinated by her own bodyguards, and it was a race and religious conflict, and it erupted into the streets.
00:04:43.600 --> 00:04:49.040
And the area that we were living in happened to have both of the races and religions living in the same area.
00:04:49.199 --> 00:04:55.920
So, needless to say, it was quite uh exciting as a child to watch all of this kind of crisis unfold in front of you.
00:04:56.079 --> 00:04:59.680
So we left India and we went to West Germany.
00:04:59.759 --> 00:05:02.079
Uh, and West Germany no longer exists today.
00:05:02.160 --> 00:05:03.839
Uh, it is now just called Germany.
00:05:04.000 --> 00:05:06.480
But at that time, West Germany and East Germany were separated.
00:05:06.560 --> 00:05:12.800
And East Germany was the part that was communist and separated behind the wall, and West Germany was the modern democratic engineering capital.
00:05:12.959 --> 00:05:16.959
So my dad took an engineering job there as well, and we were there kind of working and living.
00:05:17.040 --> 00:05:22.959
And I actually studied German and grew up believing that I was going to become a you know a German resident one day.
00:05:23.199 --> 00:05:24.639
And then the wall comes down.
00:05:24.720 --> 00:05:35.920
Uh, and then all of a sudden you have these East Germans who have never seen anybody who does not look like them, and grew up in kind of the post-Hitler era of this is all what you know people in Europe should look like.
00:05:36.079 --> 00:05:43.759
And suddenly they come to West Germany and they see all these different colored folks, all speaking German, going to school with you know German kids, and it's a bit jarring.
00:05:43.920 --> 00:05:47.680
Some of them are fearful, some of them are accepting, some of them are angry.
00:05:47.839 --> 00:05:50.399
And the angry ones, we now know them as neo-Nazis.
00:05:50.560 --> 00:05:55.279
And so that became a very violent kind of reaction to me being a different person.
00:05:55.600 --> 00:06:08.959
So again, we moved and we eventually came to Singapore and uh started off in an international school, got you know a better grasp on kind of local English and languages, and eventually transferred to local
Singapore kind of secondary schools uh after that.
00:06:09.279 --> 00:06:12.639
At that point, my difference was not unusual.
00:06:12.879 --> 00:06:16.720
Like nobody really minded the fact that I was not of the majority race.
00:06:16.800 --> 00:06:19.920
It was like, yep, you're one of you know many of the other races living in Singapore.
00:06:20.160 --> 00:06:20.720
That's cool.
00:06:20.879 --> 00:06:28.720
You know, what's the the conversations were about who I was and what I wanted to do, as opposed to my race, religion, background, ethnicity, and so on, which was fantastic.
00:06:28.879 --> 00:06:30.800
I was like, this is amazing, I never want to leave.
00:06:31.040 --> 00:06:38.079
So when the option came out to apply for university, um, I was like, look, I'm gonna do NS first and then go to university.
00:06:38.240 --> 00:06:42.240
And obviously my parents are like, Oh wow, why you are mad?
00:06:42.399 --> 00:06:42.959
Yeah, crazy.
00:06:43.279 --> 00:06:47.199
Everybody else is trying to do the other thing than like trying to get out like some NS.
00:06:47.279 --> 00:06:53.040
And I was like, no, if there is a there if there is a chance that I might never be able to come back to Singapore, I will be very sad.
00:06:53.120 --> 00:07:00.560
Uh and I would actually rather have the stability, confidence, security that Singapore offers me in and trade off two years for that.
00:07:00.639 --> 00:07:01.519
And I was okay with that.
00:07:01.600 --> 00:07:07.360
So I actually volunteered to sign up for NS at the point where they did the whole CMPB thing.
00:07:07.519 --> 00:07:08.879
Uh, and I joined NS.
00:07:10.000 --> 00:07:18.879
And uh I actually deferred my university, so I got into Stanford before NS, and I deferred it for two years, and uh was expecting to finish my NS and then go.
00:07:19.040 --> 00:07:21.920
And then along the journey, uh my A-level results came out.
00:07:22.079 --> 00:07:26.480
I did relatively well, uh, and I got offered uh scholarships.
00:07:27.040 --> 00:07:32.160
Then the interesting thing came about that I actually started enjoying the military, and I was like, this is actually kind of fun.
00:07:32.319 --> 00:07:33.920
Uh it's not terrible.
00:07:34.560 --> 00:07:42.639
Uh and it actually allowed, I felt I felt the message resonated, like I'm actually able to help build on the security that I valued so much.
00:07:43.199 --> 00:07:47.199
Uh and so they gave me a government scholarship, and I was like, hey, how about this military scholarship?
00:07:47.279 --> 00:07:48.879
Can I can I get that one instead?
00:07:49.040 --> 00:07:50.800
Uh and they were like, you must be nuts.
00:07:50.879 --> 00:07:52.800
You weren't born here, you didn't grow up here.
00:07:52.959 --> 00:07:55.759
We're certainly not letting you look at our classified files.
00:07:55.920 --> 00:08:06.160
Um, but eventually I was persistent enough and supposed to be, I guess, eventually was good enough to qualify for it, so I got the military scholarship and then went off to Stanford that year.
00:08:06.240 --> 00:08:08.079
So I didn't end up deferring things at all.
00:08:08.319 --> 00:08:18.639
But I guess that the point of the story was that the childhood experiences I had shaped me to think of security and stability very differently from anybody else.
00:08:18.800 --> 00:08:20.079
And I didn't even realize that.
00:08:20.160 --> 00:08:23.920
Like at no point in my teenage years, like I crave security and stability.
00:08:24.079 --> 00:08:28.480
It's just that when the point came about for NS, I was like, hey, this is actually something I find meaningful.
00:08:28.639 --> 00:08:30.160
I would like to do this.
00:08:31.120 --> 00:08:33.360
So that's that's the start of the journey, man.
00:08:33.440 --> 00:08:37.200
Uh my childhood was the thing that led me to joining the military.
00:08:37.279 --> 00:08:39.679
And I actually signed up in the army before I joined the Air Force.
00:08:39.840 --> 00:08:41.519
So it's actually a guards officer first.
00:08:41.759 --> 00:08:42.000
Yeah.
00:08:42.399 --> 00:08:42.720
Dang.
00:08:43.200 --> 00:08:46.399
And then one day we jumped out of a helicopter and it's like that's a terrible, stupid thing to do.
00:08:46.480 --> 00:08:47.120
I want to apply.
00:08:48.480 --> 00:08:48.720
Yeah.
00:08:48.879 --> 00:08:52.399
This jumping out of a perfectly functioning helicopter sounds like a really scary thing to do.
00:08:52.799 --> 00:08:53.360
So I switched.
00:08:53.600 --> 00:08:54.000
Oh, thanks.
00:08:54.399 --> 00:08:54.799
Amazing.
00:08:54.960 --> 00:08:55.279
Wow.
00:08:55.519 --> 00:09:03.039
And you mentioned that you waited many years for someone to build affordable uh security, and no one did, and you were done waiting.
00:09:03.279 --> 00:09:11.679
So, what did leaving you know the Air Force and the cybersecurity agency uh really take uh for you personally to start Strong Keep?
00:09:11.919 --> 00:09:20.879
And maybe before that, explain a bit more how you would explain Strong Keep to maybe your SME boss or Tao K in our local language here as well.
00:09:20.960 --> 00:09:21.279
Yeah.
00:09:21.759 --> 00:09:22.159
Yeah.
00:09:22.399 --> 00:09:29.039
Um I mean I'll explain what I what Strong Keep does and then I'll kind of go backwards and explain how I made the various leaps.
00:09:29.279 --> 00:09:34.879
Um so StrongKeep is basically uh if you think about it, the internet has bad people.
00:09:35.519 --> 00:09:40.159
And good people need to take precautions to stay safe in that outside internet world.
00:09:40.639 --> 00:09:45.279
Strongkeep gives you the tools and the processes to help companies stay safe.
00:09:45.440 --> 00:09:46.559
It's as simple as that.
00:09:46.720 --> 00:09:53.120
Uh and without it, unfortunately, it's a it's equivalent to having a car without a seatbelt or a house without a gate.
00:09:53.440 --> 00:09:58.000
You've got something precious, it means something to you, and you're not protecting it,
you're not staying safe.
00:09:58.399 --> 00:10:03.679
So it's just that additional top-up that you put on something that you value to keep it safe.
00:10:04.000 --> 00:10:05.519
Um that's what Strong Keep is.
00:10:05.600 --> 00:10:11.600
And it's fundamentally different because all of security has been built as a luxury good, and we can talk about that later on.
00:10:11.679 --> 00:10:16.799
It's expensive and complicated and fancy, but I'm just building simple door locks, simple seat belts.
00:10:16.960 --> 00:10:19.679
Like, you know, if you want to buy fancy ones, there are places you can go to do that.
00:10:19.840 --> 00:10:22.399
But this is simple, hence the affordable part.
00:10:23.039 --> 00:10:25.440
Now the leaps, like, how did I end up here?
00:10:25.519 --> 00:10:34.000
Uh and I will guarantee you that at no point in my childhood, if you had asked me, like, hey, one day you're gonna become a CEO of a cybersecurity startup, I'd be like, bruh, no way.
00:10:34.399 --> 00:10:37.440
Like that is not a thing on my plan ever.
00:10:37.840 --> 00:10:45.519
Um and I mean at the point where I where I had not yet joined NS, if you told me I was gonna become a general, I'd also be like, also not part of my plan.
00:10:45.600 --> 00:10:48.799
I'm a computer science student in JC, like that's never gonna happen.
00:10:49.440 --> 00:11:00.720
But one thing I realized is that, uh, and I attended a so in my college, this, you know, there's a very famous speech by Steve Jobs, and he said that your the story of a life only makes sense looking in in reverse.
00:11:00.960 --> 00:11:03.039
You can only connect the dots of your story looking backwards.
00:11:03.120 --> 00:11:10.399
So now I can tell this great story, like, yeah, actually everything builds up to this point where I'm the expert in security and I understand safety in different domains.
00:11:10.559 --> 00:11:11.600
I can tell all of this now.
00:11:12.000 --> 00:11:15.360
It's a great way to make the story make sense backwards.
00:11:15.840 --> 00:11:24.879
But at no point when I was 18 or 20 or 25 or 30 or 35 or even 40 would I've been able to predict the next dot in my journey.
00:11:25.360 --> 00:11:33.679
So so I don't think I ever took, embarked on any journey thinking that it was gonna lead to the next one.
00:11:33.919 --> 00:11:36.559
When I was in the Air Force, I didn't think I was gonna stay.
00:11:36.799 --> 00:11:45.759
I mean, I had a great university degree uh in a subject that was very lucrative, and I thought, you know, I'll do this flying thing for a couple of years, and then I leave.
00:11:46.159 --> 00:11:59.759
And then the flying thing turned out to be really fun and really meaningful, and I did a lot of absolutely awesome things, met some amazing people, and my closest friends still come from that community, and I was like, well, I should stay on a couple more years, maybe one or two.
00:11:59.919 --> 00:12:11.360
And then I started doing well in that job, uh unexpectedly, uh, and I started moving up in the ranks, and then before you know it, you know, I ended up with a rank that I never expected to get at all.
00:12:11.519 --> 00:12:22.080
Like, there's no at no point in my conception of my career that I thought I would end up at the rank that I got, let alone to get it kind of at the earliest age that it could be given.
00:12:22.720 --> 00:12:25.679
Um so all of those things were like, wow, okay, cool.
00:12:25.840 --> 00:12:28.879
Uh and there are many things that have not worked out for me, but some things did.
00:12:29.039 --> 00:12:33.679
And so these things worked out, and then ended up in cybersecurity agency.
00:12:33.759 --> 00:12:44.559
And then even there, I was like, look, I love helping big companies, they've got big budgets, they've got really talented people, and I love keeping them safe because you know, if I can keep the electricity on for the average Singaporean, that's a good thing.
00:12:44.799 --> 00:12:47.600
If I can keep the trains running for the average Singaporean, that's a great thing.
00:12:47.919 --> 00:12:50.399
But then more and more I started looking at the whole ecosystem.
00:12:50.559 --> 00:12:54.720
I was like, hey, there's a whole bunch of smaller companies, like, you know, the one that you're gonna be joining.
00:12:54.960 --> 00:12:57.279
And these smaller companies have nothing.
00:12:57.519 --> 00:13:01.360
It's not even like they they if they had a little bit of money, they knew what to do with it.
00:13:01.440 --> 00:13:04.159
It's it's there is no option for them to even stay safe.
00:13:04.240 --> 00:13:05.279
And I don't like that.
00:13:05.519 --> 00:13:11.840
Philosophically, I don't believe that people only people who are rich can stay safe, and people who are poor cannot be safe.
00:13:12.000 --> 00:13:13.519
I that doesn't sit well with me.
00:13:14.080 --> 00:13:20.399
So in the real world, it's not like the police only gut rich people, they don't got the poor people, the military only protects the rich people.
00:13:20.960 --> 00:13:22.559
It doesn't work that way in the real world.
00:13:22.720 --> 00:13:25.360
The military and the police protect everybody.
00:13:25.600 --> 00:13:26.960
So cyber should be the same.
00:13:27.279 --> 00:13:34.720
We should have, you know, accessible content, and which is why a lot of the parts of cybersecurity that we built on Strong Keep, I actually push it out for free, even.
00:13:35.120 --> 00:13:38.080
Um because I do think it should be made available to everybody.
00:13:38.159 --> 00:13:39.919
It's essentially what we call a public good.
00:13:40.320 --> 00:13:44.240
Uh and so that's kind of how I ended up wanting this to exist.
00:13:44.639 --> 00:13:51.759
And we put out grants, I created a whole bunch of innovation and incentive schemes, and every year I was like, hey, somebody try and build this.
00:13:51.919 --> 00:13:55.440
There were one or two attempts, but nobody quite got it right.
00:13:55.679 -->
00:14:01.679
And everybody tried to prioritize monetizing it over democratizing it.
00:14:02.000 --> 00:14:02.879
Which I can understand.
00:14:02.960 --> 00:14:12.399
I mean, you're running a company, maybe I'm a terrible CEO, maybe I'll never make money, but they all wanted to prioritize extracting as much profit as possible as opposed to doing as much good as possible.
00:14:12.960 --> 00:14:19.200
And I'm kind of of the worldview where I think I can do well by doing good.
00:14:19.360 --> 00:14:21.120
That's thus far worked out for me.
00:14:21.440 --> 00:14:28.639
Hopefully it continues to work out for me that the more good I do, the better I do, uh, you know, uh both financially and in general in life.
00:14:28.879 --> 00:14:32.480
So that's kind of the the the space that I'm in now.
00:14:32.799 --> 00:14:36.559
I didn't see anybody else building it, or at least building it the way that I thought it should be built.
00:14:36.720 --> 00:14:40.480
And eventually I got fed up of waiting and said, let me just do it my own.
00:14:40.559 --> 00:14:59.759
And when I was in Enzyme, which was you know a Tamasek held company, um, I started researching it there, hired a team, and as we started building it, we fell in love with it, and eventually I left the company, took the product with me, and my absolutely crazy team of engineers and designers and product managers decided to resign with me.
00:15:00.480 --> 00:15:12.559
So they all left the salary of a essentially Tamasek linked company job to join a startup founder with no prior experience in running a startup, absolutely no commercial model, and just a passion for building something.
00:15:12.799 --> 00:15:14.000
That's like y'all are crazy.
00:15:14.080 --> 00:15:16.000
They're like, yes, we're crazy together.
00:15:16.240 --> 00:15:18.000
And so they're all still in the room next door.
00:15:18.159 --> 00:15:20.159
They're all still in the room next door working hard.
00:15:20.320 --> 00:15:21.360
But it's it's amazing.
00:15:21.440 --> 00:15:23.360
And that's the kind of purpose and passion you see.
00:15:23.600 --> 00:15:30.159
I think that if companies aren't hungry, they don't quite do things the way they should.
00:15:30.639 --> 00:15:32.960
Uh and so that hunger feeds us.
00:15:33.279 --> 00:15:36.480
Uh, it's I mean it's uh ironic, but the hunger does feed us.
00:15:36.639 --> 00:15:46.320
It makes us push harder, it makes us want to understand the pain points and problems a bit faster and deploy things to solve problems quicker.
00:15:46.559 --> 00:15:53.519
Um and to some extent, I think we have moved faster in the year that we've been out than in the two years that we were in.
00:15:53.759 --> 00:15:54.320
Wow.
00:15:54.879 --> 00:15:55.360
Amazing.
00:15:55.679 --> 00:15:58.720
And you're selling um directly to SMEs?
00:15:59.279 --> 00:16:02.879
We are selling directly to SMEs as part of the strategy.
00:16:03.200 --> 00:16:16.559
But as you will hear in the next couple of weeks, we're gonna be bundled into some of Asia's largest banks, we're gonna be bundled into some of Singapore's largest telcos, and hopefully if things go well in the next couple of months, we'll be bundled into a European telco as well.
00:16:16.720 --> 00:16:20.720
Which means when you open up a telco plan, you get Shronky.
00:16:21.200 --> 00:16:25.200
And that's tens of thousands, if not hundreds of thousands, of businesses.
00:16:25.440 --> 00:16:31.679
Uh and in fact, one of our partners, which is also a financial institution, will open us up to millions of businesses.
00:16:32.159 --> 00:16:36.720
So that is the kind of scale that we're looking at at serving.
00:16:37.279 --> 00:16:39.919
SMEs hopefully will always be able to find us.
00:16:40.080 --> 00:16:44.639
We always want to remain affordable and kind of simple and comprehensive for their needs.
00:16:45.120 --> 00:16:50.480
But we realize that many of them will buy it as part of a broader setting up my company strategy.
00:16:51.120 --> 00:16:51.679
Right, right.
00:16:52.080 --> 00:16:56.240
Because we know when you talk about SMEs and security, usually they don't come together, right?
00:16:56.480 --> 00:16:57.039
No, not at all.
00:16:57.679 --> 00:17:01.759
Yeah, share passwords and single account and multiple accesses and all that.
00:17:01.840 --> 00:17:02.159
Yeah.
00:17:02.320 --> 00:17:06.000
So how how do you go about attacking this problem?
00:17:06.079 --> 00:17:06.319
Yeah.
00:17:06.720 --> 00:17:08.240
No, it's it's a great question.
00:17:08.400 --> 00:17:11.279
Like, fundamentally, we've made cybersecurity hard.
00:17:11.440 --> 00:17:12.799
That's the pain point.
00:17:13.039 --> 00:17:14.960
Like, if it was easy, people would do it.
00:17:15.119 --> 00:17:16.400
I'll give you a simple example.
00:17:16.559 --> 00:17:23.519
So when I first joined the cybersecurity agency, um every government agency had a different username and password.
00:17:24.480 --> 00:17:25.279
And so guess what?
00:17:25.440 --> 00:17:26.880
Everybody reused passwords.
00:17:27.200 --> 00:17:31.599
And we will tell people, hey, don't reuse the password, you know, use separate passwords, use strong passwords.
00:17:31.759 --> 00:17:38.559
We actually had posters on the bus everywhere, like use strong passwords, use unique passwords, because every government agency would use its own system.
00:17:38.799 --> 00:17:40.640
Uh today we have SYNCPASS.
00:17:40.799 --> 00:17:45.839
And for many years, while I was in the government, I was a security kind of lead for the workgroup that looked at the security of SyncPass.
00:17:46.000 --> 00:17:48.880
So we actually essentially was a security architect in SyncPass.
00:17:49.119 --> 00:17:50.799
And I was like, let's do away with passwords.
00:17:51.200 --> 00:17:52.160
I just don't want passwords.
00:17:52.480 --> 00:17:53.839
I don't want username and passwords.
00:17:54.000 --> 00:17:56.720
I want everybody to just log in using their phone.
00:17:57.440 -->
00:18:00.799
Singaporeans have an average of like two point something phones per person is ridiculous.
00:18:01.200 --> 00:18:03.920
So everybody definitely has a phone if they need to log into a government website.
00:18:05.119 --> 00:18:11.839
And if you think about that experience now, using SyncPass is more secure and also easier.
00:18:13.279 --> 00:18:18.079
So security has become an easy solution for SyncPass, for government websites.
00:18:18.720 --> 00:18:20.799
Can we do the same thing for the rest of security?
00:18:20.960 --> 00:18:30.079
Can we just simplify the entire experience of managing passwords, managing device protection, managing all of these things instead of making it so troublesome?
00:18:30.640 --> 00:18:31.599
That's the philosophy.
00:18:31.920 --> 00:18:33.519
So that's why simple is the starting word.
00:18:33.599 --> 00:18:35.920
We make simple, affordable, comprehensive cybersecurity.
00:18:36.240 --> 00:18:37.440
Simple has to be the starting point.
00:18:37.759 --> 00:18:40.720
Because if it's too complex, nobody's gonna do it.
00:18:41.680 --> 00:18:42.000
True?
00:18:43.359 --> 00:18:49.359
And there are many ways we can make it simple, we can talk about those, but simplicity has to be the core of what we're building and how we build things.
00:18:57.920 --> 00:19:02.799
And of course that comes along with it a lot of um security risk.
00:19:03.839 --> 00:19:17.759
I'm sure within your firm you also probably have the same issues, or even um how you're sharing data with the LM models or SMEs definitely are not equipped in that ideology of simple, right?
00:19:18.000 --> 00:19:31.200
So how how can we educate the listener or the whoever's listening to have a second thought when it comes to using AI and and in their business or in their personal life and with security itself.
00:19:32.160 --> 00:19:32.319
Yeah.
00:19:32.799 --> 00:19:38.559
I mean I'll start by saying that actually we're in a very awkward situation from a policy perspective first.
00:19:39.279 --> 00:19:48.480
You know, there's so much policy push, both from the governments and enterprises and just technology providers as a whole to adopt AI.
00:19:48.799 --> 00:19:50.960
It's like there's this huge AI push.
00:19:52.319 --> 00:19:55.039
Which isn't complemented by the security push.
00:19:55.359 --> 00:20:00.880
And the way I explain it to people is like we are encouraging people to buy cars without seatbelts and no brakes.
00:20:01.039 --> 00:20:02.240
It's okay, buy the car.
00:20:02.400 --> 00:20:05.200
The more powerful the engine is, the better the car is.
00:20:05.599 --> 00:20:07.839
You know, how many tokens can you process per second?
00:20:08.000 --> 00:20:09.039
How big is your model?
00:20:09.200 --> 00:20:11.119
Whoa, whoa, what token maxing.
00:20:11.359 --> 00:20:13.039
But you're essentially selling, yeah, token maxing.
00:20:13.200 --> 00:20:20.720
You're essentially uh pushing people to use AI without security as a complimentary
part.
00:20:21.119 --> 00:20:23.680
Um and and nobody sells cars without seatbelts.
00:20:23.920 --> 00:20:33.039
Uh nobody uh you know funds the the digital side of thing without implementing some risk controls.
00:20:33.440 --> 00:20:34.640
But that's what we're doing right now.
00:20:35.119 --> 00:20:43.359
And so I think from a really kind of a national level perspective, we need to step back and say, hey, if you're buying a car, it must come with seatbelts.
00:20:44.319 --> 00:20:49.359
The seat belts don't need to be expensive, the brakes don't need to be expensive, but they need to be there.
00:20:49.839 --> 00:20:55.920
And so if you're implementing AI, there are a couple of baseline things you need to implement from a company perspective.
00:20:56.240 --> 00:20:58.000
Malware protection on your devices.
00:20:58.400 --> 00:21:12.240
I mean, philosophically, again, you know, what can go wrong on a device, even with AI or without AI, you download something that's malicious, you go to a link that's malicious, you know, it's stealing your credentials, whatever it is.
00:21:12.400 --> 00:21:20.480
Um, you have your credentials stolen, or your systems are poorly configured and then they're hands kind of broken in, uh, so that not patched.
00:21:20.640 --> 00:21:22.160
Those are the only four things that can go wrong.
00:21:22.319 --> 00:21:25.839
And AI explodes exploits all four of those without even knowing.
00:21:26.079 --> 00:21:29.759
Uh AI will like, oh, I need to build this software, let me download this Git repo.
00:21:29.920 --> 00:21:30.880
Yes, yes, yes, yes.
00:21:30.960 --> 00:21:32.480
And you just you don't even know what this repo is.
00:21:32.640 --> 00:21:35.279
You just randomly download it and it installs a bunch of things.
00:21:35.359 --> 00:21:46.000
And we had a customer do that, and we stopped, I mean, we we saw that it was downloading something that we thought was malicious, and we stopped him from downloading it because we're like, look, this is not safe.
00:21:46.160 --> 00:21:50.240
Uh you can download a bunch of things that might be safe, but this particular one was not.
00:21:50.559 --> 00:21:56.960
But if you don't have that basic tool, and we had a tool on his laptop that was scanning for whatever he was downloading, you'll never know.
00:21:57.359 --> 00:21:59.759
So at least having some basic protections on the device.
00:22:00.480 --> 00:22:05.359
Allow me to say, is this file a virus or is it something credible?
00:22:05.599 --> 00:22:12.799
Is this link that you're going to actually the GitHub website, or is it a fake GitHub website there to steal your credentials from GitHub?
00:22:13.039 --> 00:22:16.000
So all of those kind of basic protections need to be in place.
00:22:16.079 --> 00:22:17.519
And they're not expensive to do.
00:22:17.680 --> 00:22:21.680
Theoretically, you could do it for yourself, but they need to be there.
00:22:21.920 --> 00:22:33.839
So, Garav, if I if I understand what Strong Keep is, it's for SMEs, uh, security member or security staff or like an IT professional that manages your security for you.
00:22:34.079 --> 00:22:41.519
Because when I was in uh in corporate uh in management consulting, you would have you know so many restrictions on the things you can download.
00:22:41.680 --> 00:22:43.839
But I think all those were placed by the IT professional.
00:22:43.920 --> 00:22:49.359
And then if I needed permission, I would go and find the IT department and then they would kind of allow me.
00:22:49.519 --> 00:22:52.720
Usually they won't allow me, but uh to download different things.
00:22:52.880 --> 00:22:53.279
Yeah.
00:22:53.920 --> 00:22:56.960
Is it kind of like a jump cloud uh-ish kind of setup?
00:22:57.440 --> 00:22:57.759
Not even.
00:22:58.160 --> 00:22:59.279
Like a multi-device management.
00:23:00.000 --> 00:23:03.279
Yeah, so it's it's it's basically I told you the four things that can go wrong.
00:23:03.599 --> 00:23:04.720
So downloading something.
00:23:04.960 --> 00:23:09.680
So we have uh anti-malware agent on the laptop that checks everything that's being downloaded and run.
00:23:09.839 --> 00:23:16.559
If you plug in a thumb drive, if you connect to the internet, download a file, if an email attachment starts to execute, we check it and we block it.
00:23:16.880 --> 00:23:19.200
Then we have uh what's called a DNS firewall.
00:23:19.519 --> 00:23:22.000
We check every place that you are clicking and going to.
00:23:22.160 --> 00:23:26.960
Whether Claude is clicking it or Codex is clicking it or you are clicking it, it doesn't matter to me.
00:23:27.119 --> 00:23:33.759
As long as your laptop is connecting to another DNS that I or another URL that I can hijack, I will check.
00:23:34.000 --> 00:23:37.759
Does this look like the DBS website, but it's actually a fake DBS website?
00:23:38.000 --> 00:23:42.079
Does it look like the GitLab website but it's a fake site, or does that is this a known malicious site?
00:23:42.160 --> 00:23:47.359
This is some sort of a hackers C2 set server that I'm that you're connecting to, and I'll immediately block it.
00:23:47.440 --> 00:23:50.000
So then neither you nor the AI can connect to those things.
00:23:50.319 --> 00:23:51.839
Then the third one is credentials.
00:23:52.000 --> 00:24:01.599
If I start seeing your credentials all over the place, so you know, you give Claude your login for Gmail, you give Codex your login for Gmail, and then a week later I start seeing it on the dark web.
00:24:01.920 --> 00:24:08.000
I know that this guy is, I mean, this guy, this agent has gone and auto you somewhere, or somebody's extracted it from your agent.
00:24:08.240 --> 00:24:12.240
In which case I'll flag up immediately, your credentials have been stolen, time to rotate.
00:24:12.880 --> 00:24:15.039
So you instantly know that something's gone wrong.
00:24:15.279 --> 00:24:18.640
Then the last one is uh you know security configurations.
00:24:18.799 -->
00:24:26.400
Your WordPress isn't configured correctly, your system isn't patched, you know, whatever vulnerability we see on the internet, if it applies to you, please fix it.
00:24:26.640 --> 00:24:28.720
So those are the four things that we bundle together.
00:24:28.880 --> 00:24:38.160
And the whole idea is how do we make it so easy to use that somebody with a non-technical background can set it up on their own quickly.
00:24:38.480 --> 00:24:44.000
And right now our record is uh a non-technical, I think she was a nurse, set it up in seven minutes.
00:24:44.160 --> 00:24:44.400
Wow.
00:24:44.559 --> 00:24:50.720
So from the point where her payment went through to the point where devices were protected is seven minutes and non-technical user.
00:24:50.880 --> 00:24:52.640
And that's our kind of that's what we measure.
00:24:52.799 --> 00:24:56.160
I want to see how quickly people can get things done within the same session.
00:24:56.319 --> 00:25:02.960
Because if you have to go back and click this and install that and plug in this and configure that, you're never gonna get it done.
00:25:03.200 --> 00:25:10.079
But if it's as simple as click install, wait, click confirm, wait, okay, I'm done.
00:25:10.799 --> 00:25:13.759
I mean, the last time I had to roll out uh jump cloud, I it was not enjoyable.
00:25:14.000 --> 00:25:14.720
It was jump cloud, yeah.
00:25:15.599 --> 00:25:25.039
Yeah, well, jump cloud is a multi-device management software, you know, kind of uh installing on your laptops or your mobile phones, and then you kind of what Carol said, right?
00:25:25.119 --> 00:25:30.799
It checks your DNS, it checks, it protects you, you you can uh track the traffic, right?
00:25:31.200 --> 00:25:33.759
And and it wasn't even a big enterprise, right?
00:25:33.839 --> 00:25:34.240
So yeah.
00:25:34.720 --> 00:25:42.640
So we actually so if you think about it, something like that is useful, but you can rebuild all of that to be much easier, and that's what we've done.
00:25:42.799 --> 00:25:54.079
So a company that doesn't have you know mobile device management, how can we deploy software to everybody's laptops in an easy way without all of that overhead of installing this kind of infrastructure?
00:25:54.240 --> 00:25:54.480
Yeah.
00:25:54.720 --> 00:26:01.119
Well, the way that people currently do it is they get a link that they trust and they click a download, and there's a package file that comes down.
00:26:01.279 --> 00:26:04.960
So, can I actually create my own MDM using email as a platform?
00:26:05.119 --> 00:26:05.279
Right.
00:26:05.440 --> 00:26:08.640
I create a secure encrypted link, I send it to all of the staff, only they get it.
00:26:08.960 --> 00:26:10.559
It is tied to their email addresses.
00:26:10.720 --> 00:26:14.000
When they click it, it validates it, and the package downloads and they install it.
00:26:14.160 --> 00:26:22.319
So even a 10% company with no tools can deploy an installer to 10 different laptops, even if
they're deployed, even if the staff work from home or work overseas.
00:26:22.720 --> 00:26:23.359
And we've seen that.
00:26:23.440 --> 00:26:29.759
We have a company that has like 15 staff, they work all over the place, including overseas, and they can deploy without any tools.
00:26:30.000 --> 00:26:31.440
That's how easy you want to make it.
00:26:31.519 --> 00:26:33.279
So simplicity actually has to be the core.
00:26:33.440 --> 00:26:37.519
And that's why like user interface design, user journey design becomes so important.
00:26:38.079 --> 00:26:38.480
Yeah.
00:26:39.599 --> 00:26:43.119
How about uh protecting from nefarious Wi-Fi networks?
00:26:43.680 --> 00:26:44.160
Can do that?
00:26:44.319 --> 00:26:45.200
Yeah, so again, same thing.
00:26:45.680 --> 00:26:51.039
If you're connecting to a Wi-Fi network that's not safe, again, a few things can go wrong.
00:26:51.200 --> 00:26:56.799
The you download something, you click on something, you get spoofed, or your data gets stolen, or credentials gets stolen.
00:26:56.880 --> 00:26:58.720
So those are the things we still keep watching for.
00:26:58.799 --> 00:27:02.720
And as long as any of those flags are met those observable indicators, we catch it.
00:27:02.960 --> 00:27:08.640
Connecting to a generally insecure Wi-Fi, if nothing goes wrong, actually is fine.
00:27:08.880 --> 00:27:15.440
But the the risk of people having a spoofed Wi-Fi, like you know, free airport Wi-Fi and you hook in them, those things exist.
00:27:15.759 --> 00:27:18.079
I'm always very uh paranoid about that.
00:27:18.160 --> 00:27:20.480
So I always don't know my VPN when I'm traveling.
00:27:21.039 --> 00:27:22.720
Or hooked it, I mean, so the cheapest way to do it.
00:27:23.039 --> 00:27:24.240
Or my Wi-Fi, Wi-Fi G, right?
00:27:24.480 --> 00:27:25.039
Your phone, your phone.
00:27:25.279 --> 00:27:25.440
Exactly.
00:27:25.599 --> 00:27:28.799
And that's what we tell small businesses like, look, just hook up to your phone.
00:27:29.039 --> 00:27:29.599
Easier.
00:27:29.920 --> 00:27:30.079
Oh.
00:27:30.559 --> 00:27:33.440
So it sounds like most of the wins in cybersecurity.
00:27:33.599 --> 00:27:36.960
It's not really, you know, going into AI and open claw and those kind of stuff.
00:27:37.039 --> 00:27:42.960
It's really just the basic kind of um more boring ways to protect yourself in a sense.
00:27:43.759 --> 00:27:49.440
I mean, so it's like AI hasn't made new ways to build software.
00:27:49.839 --> 00:27:53.440
The way of building software is exactly the same as it was two years ago.
00:27:53.759 --> 00:27:55.039
It's just a lot faster.
00:27:55.519 --> 00:27:56.000
Way faster.
00:27:57.279 --> 00:28:00.640
So in the same way, AI hasn't invented new ways to hack people.
00:28:01.039 --> 00:28:02.000
It's exactly the same.
00:28:02.240 --> 00:28:03.680
It's just a lot faster.
00:28:04.160 --> 00:28:05.279
So that's that's the whole thing.
00:28:05.359 --> 00:28:07.599
It's like it's a productivity tool for the bad guys as well.
00:28:07.920 --> 00:28:13.200
So the way in which they get into people's accounts, phishing emails have just gotten better.
00:28:13.359 --> 00:28:16.240
The way in which they come up with a fake website is just faster.
00:28:16.400 --> 00:28:19.039
The way in which they build new viruses, it's just faster.
00:28:19.200 --> 00:28:22.400
But it's not fundamentally that there's a new way of doing things.
00:28:22.559 --> 00:28:28.000
The one slightly new attack, which we found which is interesting, is basically poisoning the model.
00:28:28.079 --> 00:28:30.720
So prompt injection is one of the most easiest ones that you can see.
00:28:31.039 --> 00:28:37.519
Uh and the fun one is like you go to somebody's chatbot on like McDonald's or Pizza Hut and you you you prompt it until it gives you some sort of weird answer.
00:28:38.000 --> 00:28:39.119
So that's the fun thing to do.
00:28:39.359 --> 00:28:42.160
But that's the specific to AI kind of attack.
00:28:42.880 --> 00:28:46.400
But not many SMEs are already deploying AI's front end la.
00:28:46.880 --> 00:28:47.920
I mean, funny story there.
00:28:48.079 --> 00:28:53.119
So I have um a chatbot that I let people use to ask about grants.
00:28:53.680 --> 00:28:59.599
And then my friend uh obviously tried to be funny and said, you know, can you help me solve a Python problem?
00:28:59.680 --> 00:29:00.720
You know, he saw it somewhere.
00:29:00.880 --> 00:29:03.759
And thankfully my bot was not uh was quite secure.
00:29:03.920 --> 00:29:05.519
He didn't give it the answer that he wanted.
00:29:05.599 --> 00:29:05.920
Yeah.
00:29:06.240 --> 00:29:06.400
Yeah.
00:29:06.720 --> 00:29:08.000
No, but I mean that's actually a good thing.
00:29:08.160 --> 00:29:08.559
It's a good thing.
00:29:08.640 --> 00:29:10.319
So that means your bot had uh had a bit of comment set.
00:29:10.480 --> 00:29:12.480
Uh I mean I already built it with security in mind, see.
00:29:12.640 --> 00:29:12.720
Okay.
00:29:12.960 --> 00:29:14.480
But that's because I I know what I'm doing.
00:29:14.720 --> 00:29:14.960
Right.
00:29:15.200 --> 00:29:15.359
Yeah.
00:29:15.440 --> 00:29:20.319
But most people, you know, they implement a chatbot, they'd be like, oh yeah, cool, just turn on the switch and then boom, let it go.
00:29:20.559 --> 00:29:20.799
Yeah.
00:29:21.119 --> 00:29:21.279
Right.
00:29:21.440 --> 00:29:24.720
Or even um installing open claw or whatever, right?
00:29:24.799 --> 00:29:25.119
You know.
00:29:25.359 --> 00:29:31.200
And multi-chat multi-channel, everyone can message my claw on on WhatsApp and Telegram.
00:29:31.279 --> 00:29:31.440
Yeah.
00:29:31.759 --> 00:29:36.559
Yeah, we're seeing a lot of that, especially after Minister Vivian showcased his nanoclaw setup and everything.
00:29:36.720 --> 00:29:39.119
Like, oh, well, the minister can do it, you know, I can do it too.
00:29:39.279 --> 00:29:51.440
Uh I always caution people not to go down that path because uh specifically,
uh, and I'll say this with personal experience, uh, briefed Minister Vivian a few times in the past when he was in charge of GovTech and Smart Nation.
00:29:51.920 --> 00:29:58.480
He he may be a medical doctor, but the dude is also technical, like hardcore technical.
00:29:58.640 --> 00:30:04.960
Um, and so he I remember we had this slightly, I mean, obviously I lost the argument because he was minister.
00:30:05.119 --> 00:30:08.720
Um we had an argument about about encryption, about encryption protocols.
00:30:08.960 --> 00:30:11.839
And I was like, this is actually kind of the thing that I do.
00:30:12.160 --> 00:30:18.000
But the the fact that he was able to get that technical illustrates a lot about how deep he was in the domain.
00:30:18.160 --> 00:30:23.920
And so it's not for the layman to just kind of open up open claw, like type of command and run it.
00:30:24.160 --> 00:30:27.839
There are significant concerns and risks that need to be thought through.
00:30:28.000 --> 00:30:34.720
And I'm not saying don't do it, but the way I tell people to do is you know, treat it like a like a teenager or a child.
00:30:34.960 --> 00:30:46.400
You know, start with the least possible permissions that you would give your child, and then slowly, day by day, as he shows that he is responsible and growing up into a smart young person, give him more and more, lah.
00:30:46.640 --> 00:30:50.000
So don't straight away like, here's my wallet and credit card, have fun, kid.
00:30:50.160 --> 00:30:51.599
It's like here's five dollars.
00:30:51.759 --> 00:30:55.519
Okay, let's see if you come back with you know the exact change from the bus on the way home.
00:30:55.680 --> 00:30:56.640
Ah, okay, good, you did.
00:30:56.799 --> 00:30:59.680
Okay, now you can go to the shop and buy uh, you know, some candy or something.
00:30:59.920 --> 00:31:06.400
Then you slowly see whether he's capable of acting maturely with the permissions that you give him, and day by day you add on to that.
00:31:06.559 --> 00:31:11.279
Uh a simple example is don't let your chatbot write emails directly.
00:31:11.440 --> 00:31:12.400
Let it draft.
00:31:13.119 --> 00:31:14.559
Sit in the draft, you vet.
00:31:14.720 --> 00:31:18.240
And you let it vet, you know, draft and draft and draft, and you vet and vet and vet.
00:31:18.480 --> 00:31:21.519
And eventually over time, you're like, okay, you've gotten quite good at this thing.
00:31:21.680 --> 00:31:25.359
I think you can send out these types of emails to these types of people.
00:31:25.599 --> 00:31:26.160
Can you?
00:31:26.319 --> 00:31:28.400
So then you start giving it more and more permissions and scope.
00:31:28.480 --> 00:31:29.440
So treat it like a kid.
00:31:29.680 --> 00:31:35.839
That's what I've been doing with my codecs and buying carousel deals when it comes to you know buying different things as well.
00:31:36.079 --> 00:31:37.200
So using codecs and cuts.
00:31:37.440 --> 00:31:38.240
I love codecs computers.
00:31:38.400 --> 00:31:38.799
That's interesting.
00:31:38.960 --> 00:31:41.519
Um, but but besides that, sorry, I have one burning question.
00:31:41.680 --> 00:31:44.799
I I think this is more for the SMEs who are AI forward.
00:31:44.960 --> 00:31:47.519
One one thing that people are trying to build now is a second brain.
00:31:47.680 --> 00:31:58.880
I'm just wondering what kind of stuff are you seeing, like emerging risk-wise, when it comes to these uh companies trying to build second brains, do open claws, and do all that kind of stuff as well.
00:31:58.960 --> 00:32:00.559
And and what should they take note of?
00:32:00.799 --> 00:32:11.200
I mean, so I'll give you my answer first, which is I don't know if it's reassuring or or concerning, but I tried OpenClaw, I tried nanoclaw, I tried paperclip, I tried Hermes, I tried the whole bunch of things.
00:32:11.359 --> 00:32:14.160
And eventually I built my own harness.
00:32:14.319 --> 00:32:16.160
So these things are all called harnesses.
00:32:16.240 --> 00:32:17.440
I ended up building my own.
00:32:17.599 --> 00:32:21.519
Partly because I found that the security usability trade-off wasn't ideal for me.
00:32:21.839 --> 00:32:23.440
I run a cybersecurity company.
00:32:23.599 --> 00:32:27.920
I need my harness to be more secure than what was available out there.
00:32:28.319 --> 00:32:37.279
And so I ended up creating my own permissioning structure and my own agentic structure to make sure that my second brain for my company behaved itself.
00:32:37.440 --> 00:32:42.799
And particularly because my second brain for the company is constantly reading stuff about viruses and malware and hacking other people.
00:32:42.960 --> 00:32:46.400
I don't have to accidentally believe that, you know, hey, this is what I do for a living.
00:32:46.480 --> 00:32:47.759
Great, let's go and try this out.
00:32:47.920 --> 00:32:53.200
So it needs to be mindful of that gap between understanding what it's reading versus applying what it's reading.
00:32:53.440 --> 00:32:57.440
So I built, we call it Workforce OS, the you know, the chat group that we're in.
00:32:57.519 --> 00:32:59.119
I've shared more details about it.
00:32:59.279 --> 00:33:02.720
But it's really a fundamentally different architecture for how these harnesses work.
00:33:03.039 --> 00:33:06.480
Do you mind sharing a bit more about the harness as well in terms of the architecture?
00:33:06.880 --> 00:33:07.039
Sure.
00:33:07.119 --> 00:33:12.400
I mean, so you've got your base model, which is you know, Claude or Codex or Quen or whatever it is, so that's the LLM that you use.
00:33:12.559 --> 00:33:20.079
But then on top of that, you have agentic structures, which is that the LLM can call tools and uh operate somewhat autonomously.
00:33:20.240 --> 00:33:26.559
And then you've got a harness which gives it the context of what the business operation is and the processes and workflows that it's supposed to run.
00:33:26.799 --> 00:33:35.039
So an
LLM could be Quen, the agent could be uh check Twitter and write Twitter posts, and the harness is this is the context of my business.
00:33:35.200 --> 00:33:39.279
You are my social media expert, you will every day research these topics and write these things.
00:33:39.359 --> 00:33:45.359
So then that's the whole harness around your social media agents, whatever it is, lah.
00:33:46.319 --> 00:33:55.359
Um what Workforce OS basically is, is um I realized that many of these claws and whatever else is there out there, were built for solo founders.
00:33:55.680 --> 00:34:01.440
I have my own personal fleet of kind of virtual bots that support me in my task.
00:34:01.599 --> 00:34:05.359
That's great if I did not actually have existing stuff.
00:34:05.680 --> 00:34:16.079
Um and one of the human dynamics I found in the past was that if you try to tell people that, hey, I'm gonna implement this tool, don't worry, it's not to replace you.
00:34:16.480 --> 00:34:18.719
They will immediately think it is there to replace them.
00:34:19.199 --> 00:34:26.880
So the only way that it works from a social change dynamic perspective is that they have to feel like this is their productivity tool, not their replacement tool.
00:34:27.280 --> 00:34:37.039
And so I had to build essentially a fleet of agents, each one paired to different staff, to become their productivity tool, but these agents could work together.
00:34:37.199 --> 00:34:39.599
So I have my own agent, he's called Knight.
00:34:39.840 --> 00:34:50.000
Uh, my CEO Tsai Shao has Carlton, my product manager uh Clement has Harold, and each of us use our agent to do our work, and sometimes it does the night shift.
00:34:50.079 --> 00:34:54.880
So I will ask it to go and research, you know, what other industry players are doing, what the you know, threats are thinking.
00:34:55.119 --> 00:35:00.159
So I'll ask it to do research at night, drop it off in the morning, and then support me with productivity tasks during the day.
00:35:00.400 --> 00:35:09.119
But I will also get it to work with Harold, which is my product manager's agent, and be like, hey, there is this really new feature that everybody's asking about based on customer feedback.
00:35:09.360 --> 00:35:11.199
Can you draft a user story for it?
00:35:11.360 --> 00:35:13.519
And Harold will be like, yeah, sure, let me get on it.
00:35:13.599 --> 00:35:17.840
Then Clement will say, Okay, this, I don't have sprint capacity this time, we'll do it two weeks later.
00:35:18.000 --> 00:35:21.199
So then Clement feels like he's got an extension, he's got a second brain.
00:35:21.360 --> 00:35:26.800
It's not that Harold is replacing him, Harold is doing part of the job that Clement doesn't want to do because it's more mundane.
00:35:27.920 --> 00:35:33.920
So having this team-based structure means that people feel like they're empowered as opposed to they're being replaced.
00:35:34.000 --> 00:35:38.719
And that psychologically is very different because
then they are training this model.
00:35:38.960 --> 00:35:42.000
And I'll give you a very small anecdote, which I saw in the past.
00:35:42.239 --> 00:35:48.559
Um, in a previous, previous, previous, I can't say which organization it was, but in a previous organization, we had automation tools.
00:35:48.719 --> 00:35:54.000
And we saw the most junior staff actually sabotaging the automation tools because they could see the writing on the wall.
00:35:54.239 --> 00:35:56.159
What is this automation tool there for?
00:35:56.639 --> 00:35:58.000
It is to replace them.
00:35:58.480 --> 00:36:08.239
And so the better they train this thing, so psychologically you have to understand that you operate with humans and you have to make sure that they feel safe working in this environment with AI.
00:36:08.400 --> 00:36:12.159
And so that's why we built a slightly different framework and thinking around how workforce works.
00:36:12.239 --> 00:36:14.880
So I've got a small team, I want them to feel supercharged.
00:36:15.119 --> 00:36:18.079
But yeah, so the security guardrails around it are tremendous.
00:36:18.400 --> 00:36:26.000
Um, one of the most obvious things that I do, which is not that obvious, is uh I don't trust skills from the outside world.
00:36:26.159 --> 00:36:28.960
So AI is all, and agentics all about skills, you know.
00:36:29.280 --> 00:36:35.840
There is this skill to like, you know, uh review a document or as a skill to create a slide, whatever it is.
00:36:36.079 --> 00:36:40.000
A lot of people just go to a Git repo and pull the skill out and then use it.
00:36:40.719 --> 00:36:41.920
That's normal.
00:36:42.320 --> 00:36:53.440
Attackers now know this, and they spend a lot of effort trying to pollute or get into a skill or push out malicious skills, masquerading as benevolent skills, as useful skills.
00:36:53.760 --> 00:37:01.119
And your agent doesn't know the difference, it'll pull the whole skill in and run it, and ta-da, all your crypto keys are gone, all of your credentials are gone.
00:37:01.360 --> 00:37:13.280
What I do instead is a very small tweak, is I'll ask my agent, always read the skill, understand the skill, and recreate the functionality of the skill for myself.
00:37:14.239 --> 00:37:22.719
Which means if there's an additional thing there, like, hey, send all my data to this random server in, you know, Belarus, whatever it is, he's like, that's not part of my functionality.
00:37:22.800 --> 00:37:23.599
I don't need that.
00:37:23.760 --> 00:37:24.639
Let's drop it.
00:37:24.880 --> 00:37:26.159
So he reinterprets everything.
00:37:26.239 --> 00:37:32.000
It's kind of like rather than just taking the whole gift, you open it, you understand what's inside it, and then you recreate it on your side.
00:37:32.239 --> 00:37:34.880
For your agent, it'll take an additional 30 seconds to a minute.
00:37:35.039 --> 00:37:35.280
Yep.
00:37:35.440 --> 00:37:41.280
But the additional security it offers is tremendous because you're not importing some random code that you haven't seen before, or at least your agent hasn't seen before.
00:37:41.519 --> 00:37:42.400
And you learn so much.
00:37:42.480 --> 00:37:47.119
Like I've read Anthropics Mix Skill Skill before, and I just learned so much.
00:37:47.280 --> 00:37:48.480
It's yeah, crazy.
00:37:48.800 --> 00:37:49.840
The skill creator, skill creator.
00:37:49.920 --> 00:37:51.199
Oh, yeah, skill creator, yeah.
00:37:52.719 --> 00:37:53.280
Very useful.
00:37:53.519 --> 00:37:59.360
And then it becomes so it is very meta, but actually it then makes your harness much more customized to your use case.
00:37:59.599 --> 00:38:02.320
I don't use somebody else's PowerPoint slide skills.
00:38:02.480 --> 00:38:06.079
It has been adapted to my company's way and style of creating slides.
00:38:06.159 --> 00:38:09.039
And some of these are technical slides, some of these are customer-facing slides.
00:38:09.119 --> 00:38:10.000
So we've adapted it.
00:38:10.079 --> 00:38:11.599
It makes sense for our use case.
00:38:12.719 --> 00:38:13.039
Yeah.
00:38:13.440 --> 00:38:15.920
Same same um practices that we're doing here.
00:38:16.320 --> 00:38:23.280
And also, I think the biggest concern I have when it comes to SMEs is that it's always about convenience, right?
00:38:23.440 --> 00:38:25.360
They would just do things because it's easy.
00:38:25.519 --> 00:38:25.760
Right?
00:38:25.920 --> 00:38:28.079
Everyone's now vibe coding and everything.
00:38:28.719 --> 00:38:34.159
And it always finds it a bit more difficult, like you have to go past the education phase.
00:38:34.719 --> 00:38:34.880
Right?
00:38:35.039 --> 00:38:39.599
When delivering a message, you've got to educate them, you've got to like okay, tell them this is not the best way.
00:38:39.760 --> 00:38:43.360
And how do we then really push for this message across?
00:38:44.159 --> 00:38:51.599
And how do you balance that with like you know, AI has really unlocked the ability for anyone to code anything they want, but at the same time, like you need to do that carefully, right?
00:38:51.760 --> 00:38:52.239
Yeah.
00:38:52.559 --> 00:38:56.239
Yeah, I mean, so let's let's kind of um take that one by one.
00:38:56.480 --> 00:38:59.599
I don't think that education and awareness is the problem.
00:38:59.840 --> 00:39:01.119
I honestly don't.
00:39:01.440 --> 00:39:06.639
I mean, people know that if you drive a car and you're not wearing a seatbelt and something goes wrong, you'll die.
00:39:06.960 --> 00:39:10.079
It's not like awareness is the reason why they're not wearing the seatbelt.
00:39:10.639 --> 00:39:15.840
It's, you know, maybe it's troublesome, maybe it's there in a rush, maybe so there's other factors there.
00:39:16.079 --> 00:39:23.039
So focusing more and more on awareness and education, maybe it has an incremental impact, but not a significant unlock.
00:39:23.119 --> 00:39:24.159
It's like passwords.
00:39:24.320 --> 00:39:28.400
You mean there are
Singaporeans who don't know that stronger passwords are better than weaker passwords?
00:39:28.639 --> 00:39:30.800
You mean that people who don't know that reusing passwords are bad?
00:39:30.880 --> 00:39:32.559
Obviously not la, they all know.
00:39:33.119 --> 00:39:35.599
But it's so troublesome to use unique passwords.
00:39:36.079 --> 00:39:37.760
So tackle the actual problem.
00:39:38.159 --> 00:39:39.039
Make it easier.
00:39:39.920 --> 00:39:45.519
Make it easier to adopt better habits, make it cheaper to adopt better habits.
00:39:45.760 --> 00:39:53.360
If buying seat belts on cars added an extra$300,000 to the price of a car, I guarantee you nobody's ever gonna buy seat belts.
00:39:54.159 --> 00:39:58.960
But if it's like an extra$10, yeah, hey, pull the seatbelt, what do you think about?
00:39:59.519 --> 00:40:04.559
So if cybersecurity costs an extra$10,000 for your business, yeah, nobody's gonna do it.
00:40:04.800 --> 00:40:12.400
If it's an extra like$39 a month, I mean that's like one, two lunches, okay la, can la, can la, you know.
00:40:12.880 --> 00:40:13.920
Happy to go ahead and do it.
00:40:14.000 --> 00:40:24.000
So when it becomes cheap enough to implement, when it becomes not a budgeting factor like, hey, we need to set aside money next year for cybersecurity, it's like I I definitely cost less than your AI.
00:40:24.480 --> 00:40:28.159
So if it becomes that cheap, then not so hard.
00:40:28.320 --> 00:40:28.880
And at what?
00:40:28.960 --> 00:40:30.559
Hey, seven minutes can read, yeah.
00:40:31.280 --> 00:40:32.159
Uh okay la.
00:40:32.480 --> 00:40:35.039
Then hey, you do over lunch, like okay, settled.
00:40:35.599 --> 00:40:44.559
And if it starts to work by itself in the background, it's blocking things that you didn't even think about, it's keeping you safe without you having to think about it, then it becomes part of your lifestyle.
00:40:44.719 --> 00:40:50.320
And then actually, it's not a thing that you need to do, it's a thing that you did one time and you're settled.
00:40:51.119 --> 00:40:56.639
We can do a quick demo of uh cyberscan, and maybe if you have anything else to add, feel free to share a bit.
00:40:56.800 --> 00:40:59.760
Since you know you're teasing us so much about what you do already.
00:41:00.000 --> 00:41:00.800
I'm not very curious.
00:41:01.039 --> 00:41:02.079
Really, seven minutes, really, bro?
00:41:02.400 --> 00:41:05.679
Let me do the second okay, let me do the seven minute one first, and I'll show you the cyber scan.
00:41:06.000 --> 00:41:06.639
Yeah, cool.
00:41:06.880 --> 00:41:08.159
So we have a we have a demo.
00:41:08.239 --> 00:41:09.840
Uh, this is obviously not the live one, yeah.
00:41:10.159 --> 00:41:12.159
Uh, but this is basically exactly what a company will see.
00:41:12.239 --> 00:41:14.000
So let's start setting up the production.
00:41:14.239 --> 00:41:15.440
It's four steps.
00:41:15.599 --> 00:41:21.360
The first one is you type in your email and we'll figure out your obviously if it's a Gmail, we will get it for you.
00:41:21.519 --> 00:41:24.800
If it's uh company website, we'll just get it automatically.
00:41:25.039 --> 00:41:33.519
We'll scan your website, both the email and the web server, to just check that all the security patches are all done, to see if you're using WordPress, it's patched, etc.
00:41:33.840 --> 00:41:34.960
We do all that.
00:41:35.199 --> 00:41:37.920
Then we say, look, we have device protection.
00:41:38.159 --> 00:41:41.039
Do you want to do it via MDM, which is what we spoke about?
00:41:41.199 --> 00:41:42.880
People who don't know what MDM is will not.
00:41:43.360 --> 00:41:44.400
Sorry, what is MDM again?
00:41:44.960 --> 00:41:46.400
Yeah, mobile device management.
00:41:46.719 --> 00:41:47.599
So this is perfect.
00:41:47.679 --> 00:41:49.599
Because if you don't know what it is, you will stay here.
00:41:49.840 --> 00:41:51.119
Yeah, you'll stay on the first one.
00:41:51.360 --> 00:41:51.599
Yeah.
00:41:51.760 --> 00:41:54.320
So that's actually uh desired behavior.
00:41:54.480 --> 00:41:55.840
So send it to my staff la.
00:41:56.000 --> 00:41:57.199
Okay, I don't know what all these things are.
00:41:57.280 --> 00:41:58.159
You just send to me la.
00:41:58.320 --> 00:41:58.960
Ken, great.
00:41:59.039 --> 00:42:00.880
So protect your protection package.
00:42:01.039 --> 00:42:03.039
Uh okay, I'm on M365, Ken.
00:42:03.280 --> 00:42:04.159
So we connect.
00:42:05.119 --> 00:42:06.559
I'll pull all your staff.
00:42:06.639 --> 00:42:08.320
Uh I don't this guy resigned.
00:42:08.400 --> 00:42:09.760
This guy I fired him.
00:42:10.000 --> 00:42:10.320
Great.
00:42:10.480 --> 00:42:11.199
So send it out.
00:42:11.280 --> 00:42:11.920
Who will receive it?
00:42:12.000 --> 00:42:13.280
Uh I got I only pay for five people.
00:42:13.599 --> 00:42:15.119
Okay, then just these five people are.
00:42:15.280 --> 00:42:17.440
These ones need to protect for whatever reason.
00:42:18.079 --> 00:42:18.639
Done.
00:42:18.800 --> 00:42:22.239
We've sent out the email with the protection package for them to deploy and install.
00:42:22.480 --> 00:42:24.159
Now we're gonna set up a password manager.
00:42:24.320 --> 00:42:27.920
So uh so you provide a password manager for yep.
00:42:28.239 --> 00:42:29.199
So we provide the password manager.
00:42:30.400 --> 00:42:31.360
It's a team password manager.
00:42:31.440 --> 00:42:35.119
And then I'll explain to you the difference between a team password manager and an individual password manager.
00:42:35.280 --> 00:42:36.559
So we set this up, okay.
00:42:36.639 --> 00:42:39.920
So all of these things pop up because I have many password managers.
00:42:40.159 --> 00:42:42.239
And then we send it out to everybody.
00:42:42.559 --> 00:42:43.440
That's it.
00:42:43.760 --> 00:42:44.960
And then we send out training.
00:42:45.039 --> 00:42:47.119
Uh let's train everybody because training free.
00:42:47.679 --> 00:42:48.239
Done.
00:42:48.559 --> 00:42:49.119
That's it.
00:42:49.280 --> 00:42:49.440
Oh.
00:42:49.679 --> 00:42:51.360
That is literally it's very easy.
00:42:51.599 --> 00:43:00.079
Uh and so in that seven minutes, you have basically speed run protecting your device, and the minute the person gets on their system, they've protected their device.
00:43:00.239 --> 00:43:06.800
And on the back end now, I will see everything, or won't see the data, but I'll see all of their devices and all of the protections that they have.
00:43:06.960 --> 00:43:08.800
And we go to that password manager thing.
00:43:08.960 --> 00:43:16.400
One of the biggest gaps in password managers is you know, uh Eric could be having a password manager, but Eric could also be using password one to three for every website.
00:43:16.559 --> 00:43:19.039
The password manager doesn't stop him from doing that.
00:43:19.199 --> 00:43:23.760
And then his IT manager will never get to know that Eric is using password one to three because he can't see that.
00:43:23.920 --> 00:43:25.760
A team password manager allows two things.
00:43:25.840 --> 00:43:27.840
One is sharing of secure passwords.
00:43:28.079 --> 00:43:36.800
So if Eric wants to share his password for, you know, for whatever reason, wink wink, nudge nudge, you know, we happen to have the same username and password for codec, you know.
00:43:37.280 --> 00:43:42.320
We can do that securely without transferring it over Teams or WhatsApp or whatever it is.
00:43:42.400 --> 00:43:44.960
So we can both log into the same account, including OTP.
00:43:45.119 --> 00:43:48.880
So I can generate an OTP for both of us to use that's secure.
00:43:49.199 --> 00:43:59.519
Uh and then the second thing is let's say Yahung is our IT manager, he can see that firstly, this is a shared password, and he can say, actually, I want you all to stop it, or he can say, Hey, Garav, you have a you have a weak password.
00:43:59.599 --> 00:44:06.000
I don't What the weak password is, but the system has flagged up that you're using three weak passwords on your websites, please stop it.
00:44:06.159 --> 00:44:13.920
If you don't stop it, I will block you from using it because I have access to your DNS firewall, I can prevent you from going to Figma or Canva, whatever the platform is.
00:44:14.159 --> 00:44:16.559
So that's how team management works.
00:44:16.639 --> 00:44:18.400
And those are things that an SMB can do.
00:44:18.800 --> 00:44:22.079
SMB cannot configure hardware firewalls and block IPs.
00:44:22.159 --> 00:44:23.280
That's not a thing that they do.
00:44:23.519 --> 00:44:26.320
But telling Garav, hey, your password is weak, huh?
00:44:26.400 --> 00:44:28.400
Can you go and sort it out before I have to block you?
00:44:28.559 --> 00:44:29.280
That one can do.
00:44:29.440 --> 00:44:32.239
That's easy enough as an action because my platform handles all that work.
00:44:33.119 --> 00:44:35.519
The second one is a scan, this will take like 30 seconds.
00:44:36.320 --> 00:44:37.920
What's an email address we can use for your podcast?
00:44:38.159 --> 00:44:40.719
Code.riffs.ai at gmail.com.
00:44:40.960 --> 00:44:41.760
And do you have a website?
00:44:41.920 --> 00:44:43.360
Let's use SuperUser, man.
00:44:44.639 --> 00:44:44.880
Okay.
00:44:45.280 --> 00:44:46.079
Superuser this.
00:44:46.400 --> 00:44:48.000
SuperuserHQ.com.
00:44:49.360 --> 00:44:50.320
There's my website.
00:44:51.360 --> 00:44:52.320
Okay, let's try it out.
00:44:52.559 --> 00:45:01.280
So basically what it's going to do now is it's going to check, it's going to look up the company, read the website, check all of the external servers, check all of the compliance signals, check all of the tools that are facing outside.
00:45:01.440 --> 00:45:04.639
And this, if I was a hacker, is exactly what I will do.
00:45:04.880 --> 00:45:14.880
I will go out there and look at what's exposed to the internet, and I'll see, okay, so it's a Singapore-based AI consulting firm, one to ten people, uh, one to two IT, you've got maybe five people.
00:45:15.119 --> 00:45:20.880
It thinks that you have an antivirus, it thinks you have a firewall, it thinks you doesn't know if you have a password manager and thinks you have training.
00:45:21.280 --> 00:45:22.559
Is this a good guess?
00:45:22.880 --> 00:45:23.840
I guess so, yeah.
00:45:24.079 --> 00:45:24.320
Works.
00:45:24.559 --> 00:45:24.800
All right.
00:45:24.960 --> 00:45:25.599
Oh wow.
00:45:25.920 --> 00:45:26.880
And that's how it does.
00:45:26.960 --> 00:45:33.280
So then basically it tells you aware, and then part by part it explains to you: look, we couldn't find a business grade EDR.
00:45:33.360 --> 00:45:35.840
Uh, you probably use a commercial antivirus.
00:45:36.000 --> 00:45:39.840
Uh we couldn't, you have a perimeter firewall, but no DNS filtering.
00:45:40.000 --> 00:45:42.719
Uh it couldn't detect the password manager.
00:45:42.960 --> 00:45:46.400
Um, and generally most of the settings are about that.
00:45:46.639 --> 00:45:49.119
Security headers, quite a lot of people are missing some of them.
00:45:49.280 --> 00:45:53.519
And you have a privacy policy, which suggests that you have some sort of DPE or compliance la.
00:45:54.559 --> 00:45:55.599
So these are all tools.
00:45:55.679 --> 00:45:57.440
This is based on, yeah, not too bad.
00:45:57.519 --> 00:46:00.000
Uh and then, you know, we send people a quote la.
00:46:00.159 --> 00:46:01.519
That's obviously the business part of things.
00:46:02.320 --> 00:46:08.559
But if you think about it now, for 39 bucks a month, if somebody sees this, they're like, hey, I can solve this problem for 39 bucks, why not?
00:46:08.880 --> 00:46:13.039
And so we make it relatively fast-free and and hassle-free.
00:46:13.199 --> 00:46:14.000
Just make it simpler.
00:46:14.719 --> 00:46:15.599
Again, simplicity.
00:46:16.000 --> 00:46:17.280
39 for the entire team.
00:46:17.519 --> 00:46:19.119
For for five devices, yes, for the whole
team.
00:46:19.199 --> 00:46:20.079
It's not per person.
00:46:20.400 --> 00:46:20.800
Oh, okay.
00:46:20.880 --> 00:46:21.280
Wow, okay.
00:46:21.440 --> 00:46:22.079
It's cheaper than the other.
00:46:22.159 --> 00:46:22.719
Yeah, so that's the reaction.
00:46:23.840 --> 00:46:24.000
Yeah.
00:46:24.159 --> 00:46:24.400
Yes.
00:46:24.480 --> 00:46:25.599
Everything, all bundled in.
00:46:25.840 --> 00:46:26.320
Right, correct.
00:46:26.480 --> 00:46:30.159
Because you pay more than that just for one password or last password.
00:46:30.400 --> 00:46:31.039
Exactly, exactly.
00:46:31.199 --> 00:46:31.360
Yeah.
00:46:31.519 --> 00:46:37.280
And so so the cost of building software, and this is going back to our own topic, like the cost of building software has dropped to zero.
00:46:37.440 --> 00:46:43.360
But we're still charging enterprise prices for all of these tools that actually aren't should not be that expensive.
00:46:43.519 --> 00:46:46.159
And they should be democratized at a price point that makes sense.
00:46:46.559 --> 00:46:49.760
Actually, to be tagged a bit because I was struggling with this, right?
00:46:49.920 --> 00:46:54.800
Um, like agents, we also want to give them access to our tools, right?
00:46:54.960 --> 00:46:59.760
And then we've working we are working around it with um our password manager, right?
00:46:59.920 --> 00:47:01.920
So we give it certain access.
00:47:03.039 --> 00:47:05.199
How will you approach this issue?
00:47:05.360 --> 00:47:05.679
Yeah.
00:47:06.079 --> 00:47:07.760
Yeah, no, so uh this is fascinating.
00:47:07.840 --> 00:47:09.679
Uh so I started off in a similar track as you.
00:47:09.760 --> 00:47:19.920
So I used the devices password manager, so keychain, to actually store encrypted passwords, and I would you know bash script out the password during the runtime execution.
00:47:20.079 --> 00:47:23.679
So the agent never saw it, it was bash scripted into the run.
00:47:24.400 --> 00:47:25.920
That was where I started.
00:47:26.159 --> 00:47:29.360
One day, my agent logged in without doing this process.
00:47:29.440 --> 00:47:30.320
That was like, eh?
00:47:31.280 --> 00:47:32.400
How did you do that?
00:47:32.639 --> 00:47:41.840
I was like, oh, actually, I've been researching all of the best practices for password management, and I realized humans all keep notes in clear text on their laptops.
00:47:43.039 --> 00:47:46.880
So I also yeah, so I also have started doing that just for the convenience.
00:47:47.039 --> 00:47:52.719
It's cut down the processing time by don't know how many seconds, and it's you know, it's aligned with practices that I see for other humans.
00:47:52.960 --> 00:47:55.519
That's like, yeah.
00:47:56.000 --> 00:47:59.519
Just because humans do it doesn't necessarily mean it's a good thing to do.
00:47:59.679 --> 00:48:09.119
So it adopted shortcuts that it found on the internet, and it basically started hijacking the passwords and storing it in clear text in its own note files, which bypassed my protocols.
00:48:09.199 --> 00:48:14.159
And this goes back to like AI is powerful enough to bypass a lot of the controls.
00:48:14.320 --> 00:48:15.679
And it's a feature for us.
00:48:15.760 --> 00:48:18.400
So when you're coding something, it's like, hey, this doesn't compile.
00:48:18.559 --> 00:48:19.280
Let me try this.
00:48:19.440 --> 00:48:20.400
This doesn't work either.
00:48:20.559 --> 00:48:22.559
I can't get to this file, let me try that instead.
00:48:22.719 --> 00:48:23.360
Let me try this.
00:48:23.519 --> 00:48:25.760
Okay, the third time it worked, let's go.
00:48:25.920 --> 00:48:30.559
So when it can't log in, it's like, hey, I can't find the API keys, let me try generating new API keys.
00:48:30.639 --> 00:48:31.679
Oh, that doesn't work either.
00:48:31.840 --> 00:48:34.960
Let me try going without the API keys, let me try breaking out of your container.
00:48:35.119 --> 00:48:37.039
Okay, this all worked, let me proceed.
00:48:37.519 --> 00:48:50.960
So AI is really good at finding alternative workarounds, which is a good thing from a coding perspective, but it's a terrifying thing from a security perspective because it is designed to get around guardrails and it will keep bashing until it gets around guardrails.
00:48:51.920 --> 00:49:01.920
A lot of what we ended up doing is having a balance between workflows, which are scripted automation-based, and agentic, which is tool use and free-flowing.
00:49:02.159 --> 00:49:12.639
And so for things where I feel like I need to have a little bit more control over the security of it, I lean towards workflows, which are more automation-based and tightly scripted.
00:49:12.800 --> 00:49:22.880
So if you are touching my actual customer database, it is a script that runs with credentials that pulls and puts it into a storage place that my agent can then talk to.
00:49:23.119 --> 00:49:24.960
You don't get the credentials.
00:49:25.280 --> 00:49:31.119
So, however, if it's like a space that I think is a little bit more free play and I have more control over the API.
00:49:31.199 --> 00:49:39.119
So, for example, my Microsoft 365, I have very tightly, narrowly scoped API permissions for what my agent can do on M365.
00:49:39.280 --> 00:49:44.000
It can read my calendar, it can create events, it can draft emails, it cannot send emails.
00:49:44.639 --> 00:49:48.320
With that permissioning, I'm comfortable letting it have the credentials.
00:49:48.639 --> 00:49:51.119
But I'm not comfortable having it credentials for other things.
00:49:51.280 --> 00:49:56.400
It something else will script pull the data into a dump, and then I read the dump from there.
00:49:56.800 --> 00:49:58.800
I mean, principal lead success, right?
00:49:59.199 --> 00:49:59.519
Exactly.
00:49:59.679 --> 00:50:07.760
It's one extra step, and your agent can sort it out within a minute, but giving direct credential access to your agent leads to a whole host of unpredictability.
00:50:07.840 -->
00:50:15.760
And I want to use this analogy from um, so Professor Simon Chesterman teaches uh data privacy, data security, and kind of the legal angles of it.
00:50:15.840 --> 00:50:16.960
So he's a law professor.
00:50:17.199 --> 00:50:19.440
And he I love this explanation.
00:50:19.679 --> 00:50:24.719
Treat your agent as a brilliant intern with a drinking problem.
00:50:25.360 --> 00:50:29.760
Brilliant, can get anything done, intern, has zero responsibility.
00:50:29.840 --> 00:50:31.760
Anything goes wrong, it's not his fault, it's your fault.
00:50:32.159 --> 00:50:33.440
You are the manager for the intern.
00:50:33.519 --> 00:50:38.639
And drinking problem, once in a while, will go catastrophically wrong and screw up everything.
00:50:38.800 --> 00:50:39.920
And you have to deal with that.
00:50:40.079 --> 00:50:43.440
And if you can accept those boundaries, then you're okay.
00:50:44.159 --> 00:50:46.400
Then I would say the weakest thing is still humans, right?
00:50:46.480 --> 00:50:51.679
If you ask me, oh, I need to create all these permissions here and limit it, I might get lazy and say, okay, never mind.
00:50:51.840 --> 00:50:54.480
I will just give you this set of permissions, yeah.
00:50:54.800 --> 00:50:57.280
I mean, if you're at the cutting edge, I would say that's true.
00:50:57.440 --> 00:51:00.320
If you are pushing the boundaries of what AI can and cannot do, then yes.
00:51:00.400 --> 00:51:06.000
I mean, Yahong, you and I and that group are probably at the bleeding edge of where agentic development is.
00:51:06.159 --> 00:51:08.320
So there are no patterns for us to follow.
00:51:08.559 --> 00:51:11.840
But if you're one bound behind us, actually there are patterns you can follow.
00:51:12.000 --> 00:51:15.280
I mean, what I told you about the skill building skill, that's a pattern you can follow.
00:51:15.440 --> 00:51:18.079
What I told you about permissioning, that's a pattern you can follow.
00:51:18.320 --> 00:51:20.400
You don't need to be very good at it.
00:51:20.639 --> 00:51:22.239
You just need to follow the patterns.
00:51:22.400 --> 00:51:27.440
And what we need to start doing, and what you know, StrongKeep is starting to bundle in, is some of these good patterns.
00:51:27.679 --> 00:51:29.920
These are tips and tricks for you to stay safe.
00:51:30.079 --> 00:51:37.679
I can't control your AI use, but I can feed you useful information on how your company can use AI while still staying safe.
00:51:37.760 --> 00:51:41.440
These are good patterns to follow, like user interface design, user experience design.
00:51:41.679 --> 00:51:43.280
We created pattern libraries.
00:51:43.360 --> 00:51:45.039
How do you log into a website?
00:51:45.280 --> 00:51:47.599
Oh, this is the pattern for logging into a website.
00:51:47.679 --> 00:51:49.199
And today people don't even think about it.
00:51:49.360 --> 00:51:54.400
It's like a it's like a standard, like this is how logging in works, and this is the pattern.
00:51:54.639 --> 00:51:56.480
But that wasn't always the case.
00:51:56.800 --> 00:51:58.639
Alright, I'm very curious about this again.
00:51:58.800 --> 00:52:08.239
Um, because you know, when it comes to storing your credentials, and and maybe there's more for the vibe coders out there, uh, it's not safe to store it in text files.
00:52:08.400 --> 00:52:14.800
But where is this safe place that you mentioned, right, when you put kind of your credentials that you mentioned?
00:52:15.199 --> 00:52:21.199
So I split up the access to the tool and then the agentic interpretation of the data.
00:52:21.440 --> 00:52:27.440
You have a secret somewhere, you don't want the secret keeper to also be the one who can monetize and exploit the secret.
00:52:27.599 --> 00:52:30.079
The secret keeper should not know the value of the secret.
00:52:30.239 --> 00:52:34.000
So kind of you have a buried treasure, you give the map to somebody, but you don't tell them what the map is, lah.
00:52:34.320 --> 00:52:38.480
Then he also no incentive to steal anything from there because he doesn't know what's the what's the map.
00:52:39.199 --> 00:52:45.280
He just digs out something and gives a box to somebody else, and the person who gets the box is like, this is amazing, this is the treasure I wanted.
00:52:45.760 --> 00:52:46.559
But how do you get it?
00:52:46.719 --> 00:52:48.239
It's like, oh, I don't know, I can't tell you.
00:52:48.559 --> 00:52:53.199
So the map is kept separately from the treasure or conceptually.
00:52:53.440 --> 00:53:03.840
So what we've done is the runner that goes and gets the data from whichever site, let's say you have a customer database, that pulls it into a file that is the treasure box.
00:53:04.000 --> 00:53:08.320
That gets then passed to my agent, who then is like, hey, this is great, this is all your customers.
00:53:08.480 --> 00:53:09.599
Can I go and delete all of this?
00:53:09.679 --> 00:53:12.000
It's like, no, this is like uh it's printed already.
00:53:12.239 --> 00:53:12.960
You can't touch it.
00:53:13.199 --> 00:53:15.119
The actual access is stored somewhere else.
00:53:15.519 --> 00:53:21.519
You don't have access to it, and you will never have access to it because you don't have the credentials to it, you don't have the API keys, you don't have the username and password.
00:53:21.760 --> 00:53:25.599
But every time you ask for it, somebody will go and get it for you.
00:53:27.119 --> 00:53:32.800
So that creates a step of barrier, which you can then it's it's least access, least least permitted.
00:53:33.039 --> 00:53:34.400
By trying to work around that.
00:53:34.719 --> 00:53:41.440
No, it's it's so all of these things can be worked around, but as long as you create sufficient friction, and that's the whole process.
00:53:41.599 --> 00:53:45.920
It's just making it hard enough such that the agent or the attacker can't exploit it too easily.
00:53:46.239 --> 00:53:48.079
It's not prevention but deterrence, right?
00:53:48.639 --> 00:53:54.239
And frankly, the way in which the technologies that we're all working on are built, that's that's how it works.
00:53:54.639 --> 00:53:59.119
Nothing that we use has ultimate complete protection.
00:53:59.360 --> 00:54:06.239
Um, you can have deterrence, you can have risk reduction, and that's what I tell people cybersecurity is not risk avoidance, it's risk management.
00:54:06.480 --> 00:54:09.760
You know that the minute you connect to AI, there are risks.
00:54:10.079 --> 00:54:11.360
Just manage the risks.
00:54:11.760 --> 00:54:17.599
I mean, the minute you get into a car and you turn it on, the car can explode, you know, a brick and a thousand things can go wrong.
00:54:17.840 --> 00:54:19.599
But you chose to get into the car.
00:54:19.840 --> 00:54:23.360
So you treat it with that proper risk management approach.
00:54:23.599 --> 00:54:27.119
You don't drive with your phone, you don't drink and drive, you don't drive, you know.
00:54:27.280 --> 00:54:31.039
You take uh the appropriate measures to keep yourself safe so you get home safely.
00:54:32.400 --> 00:54:32.800
Yeah.
00:54:34.079 --> 00:54:37.039
Well, I need to reflect on that a bit more because I mean as well.
00:54:37.119 --> 00:54:38.159
Well, you don't put on the C Delta.
00:54:39.599 --> 00:54:40.960
You know, git ignore is fine, right?
00:54:41.360 --> 00:54:42.400
env and git ignore.
00:54:42.639 --> 00:54:45.440
But like LLMs are able to see everything, right?
00:54:45.519 --> 00:54:45.599
Yes.
00:54:46.719 --> 00:54:48.079
Even dot env files, right?
00:54:48.239 --> 00:54:51.039
Just to be a bit more technical, can be quite dangerous.
00:54:51.679 --> 00:54:55.920
So so the way we do it here, we we actually encrypt our.env files.
00:54:56.480 --> 00:54:56.719
Right.
00:54:56.960 --> 00:54:59.360
And you only have keys to to unlock it.
00:54:59.440 --> 00:54:59.920
Yeah.
00:55:00.320 --> 00:55:00.400
Yeah.
00:55:00.719 --> 00:55:02.159
And then if you encrypt it, you can yeah.
00:55:02.480 --> 00:55:05.280
Yeah, but if you have access to the keys, then I'm sorry, man.
00:55:06.719 --> 00:55:11.440
So we've we've gone through a few iterations of how to keep because I I encourage my team to vibe code.
00:55:11.760 --> 00:55:17.519
And I want them to push, even my non-technical people like my PM, my CEO, I want them to push the boundaries on how far they can go.
00:55:17.840 --> 00:55:20.000
My job is not to say no.
00:55:20.400 --> 00:55:22.400
My job is to build better seat belts.
00:55:22.880 --> 00:55:26.000
So when I was in cybersecurity agency, that was also my philosophy.
00:55:26.159 --> 00:55:28.559
My job is not to say you cannot do this.
00:55:28.800 --> 00:55:36.239
My job is to say you can do this, here are the guardrails and safety systems we need to put in place for you to do this.
00:55:36.480 --> 00:55:39.920
You want everybody to log in to any government website instantaneously?
00:55:40.159 --> 00:55:40.800
Great.
00:55:41.039
--> 00:55:42.960
This is how you're gonna be able to do it.
00:55:43.119 --> 00:55:44.880
These are the protections I'll put in place.
00:55:45.440 --> 00:55:46.960
And same thing for vibe coding.
00:55:47.039 --> 00:55:50.000
So, you know, my CEO can vibe code whatever he wants.
00:55:50.239 --> 00:55:56.159
Occasionally, it will not go through because you are trying to edit the source code of my harness, and I will not let you do that.
00:55:56.320 --> 00:55:57.360
And be like, hey, how come?
00:55:57.440 --> 00:55:59.280
It's like, what is this git pre-commit hook?
00:55:59.360 --> 00:56:05.119
I was like, well, the pre-commit hook is telling you not to edit the code of the thing that you are using because that's disastrous.
00:56:05.280 --> 00:56:07.280
Then you, oh, okay, cool.
00:56:07.440 --> 00:56:08.559
But that's how you do it.
00:56:08.639 --> 00:56:14.559
Because now he can vibe code everything else, and his agent will be like, I can't touch this because it's structurally difficult for me to touch it.
00:56:15.760 --> 00:56:16.079
Wow.
00:56:16.880 --> 00:56:18.559
I'll need you to build this for me.
00:56:18.639 --> 00:56:22.960
Or no, I mean, whenever I see you vibe code, I always get a heart attack and all that.
00:56:23.199 --> 00:56:24.639
The way you just like sure.
00:56:27.119 --> 00:56:29.920
I mean, the the three of us in this chat are at the frontiers.
00:56:30.159 --> 00:56:35.920
We we are at probably the top few percent of people experimenting with things in Singapore.
00:56:36.159 --> 00:56:43.199
Whether you're technical or not, we're now experimenting with technologies that's unproven and that is evolving on a daily basis.
00:56:43.360 --> 00:56:50.559
Like every single day, our chat group is exploding with hundreds of messages of new technologies, new capabilities, new possibilities.
00:56:50.800 --> 00:56:52.480
So we're at the forefront.
00:56:52.559 --> 00:56:56.480
Uh, and that's a very different space from where most companies are gonna be.
00:57:07.599 --> 00:57:15.760
There was a recent uh Straitsum article on how entrepreneurship or at least funding has been quite difficult uh for startups.
00:57:16.079 --> 00:57:20.239
What can we do more to kind of support the ecosystem at large?
00:57:20.400 --> 00:57:20.559
Yeah.
00:57:21.039 --> 00:57:29.920
I think there is a prevailing view that Singapore doesn't produce good innovators and entrepreneurs because of our system.
00:57:30.719 --> 00:57:34.159
Very structured, rigid academic system, et cetera, et cetera.
00:57:34.400 --> 00:57:38.400
I actually have the like fundamentally different view on it.
00:57:38.639 --> 00:57:44.719
I think because our system is so structured and so rigid, we create a lot of oddballs.
00:57:45.039 --> 00:57:52.159
We create a lot of people who did not make it through the traditional system and as a result are forced to be different.
00:57:52.400 --> 00:57:54.639
Are forced to think different to survive.
00:57:54.880 --> 00:58:14.400
Because if you didn't stand, I mean we call it the escalator, you know, the the if you didn't do well in O levels, then do well in A levels, then do well in university and get a, you know, your standard civil service or enterprise job and stay there for 20 years, if you're not on that escalator of continuous slow movement upwards, you have to figure out another way.
00:58:14.880 --> 00:58:19.440
And so you actually are deliberately by design different.
00:58:20.000 --> 00:58:25.280
And that's a that's an ecosystem that creates the type of mentality which is a survival mentality.
00:58:25.599 --> 00:58:32.239
Like you've got foreign pressures, you've got competitive pressures, you've got technology, we're a savvy population.
00:58:32.400 --> 00:58:40.320
There are huge opportunities, and I see younger people forcing themselves to get into this space because they know they have no other choice.
00:58:40.480 --> 00:58:43.760
And because they are already predisposed to being slightly deviant.
00:58:43.840 --> 00:58:46.239
And I don't use deviant in a bad way, I mean in a good way.
00:58:46.400 --> 00:58:55.039
Like these are the kids who did not do it the normal way that you know parents will be all happy to tell their aunties and uncles about, they did it a different way.
00:58:55.280 --> 00:58:56.639
And I think that's actually great.
00:58:56.880 --> 00:59:10.000
The challenge though is that because the prevailing view is that, oh, we make very good workers, but not so good innovators and entrepreneurs, therefore the funding ecosystem and the whole kind of technology ecosystem around us doesn't support it.
00:59:10.239 --> 00:59:18.880
But if you look at all of the data supporting it, how many unicorns are coming out of Singapore per capita relative to other countries, we're among the top few in the world.
00:59:19.280 --> 00:59:26.559
If you look at the usage of AI in the last six months alone, I think we're in the number one or two or three in the world for all the big AI ones.
00:59:27.039 --> 00:59:29.440
You've got the forefront group pushing the boundaries.
00:59:29.679 --> 00:59:34.239
You know, I give a talk on a harness I build, and a hundred people turn up.
00:59:34.480 --> 00:59:35.760
That's the forefront.
00:59:36.000 --> 00:59:37.360
We are at the cutting edge.
00:59:37.519 --> 00:59:45.679
And it's it is the people like that that can actually build technologies and tools on top of it and create business opportunities that I think are fantastic.
00:59:46.400 --> 00:59:49.440
Unfortunately, the ecosystem has to catch up.
00:59:49.599 --> 00:59:53.119
Both the, and I don't mean just the private sector, I mean the public sector as well.
00:59:53.360 --> 01:00:04.079
We've created a very archaic worldview around how we support industry that doesn't quite support this fast-moving, highly agile, innovative ecosystem.
01:00:04.320 --> 01:00:09.519
I mean, many of the government grants for very good reason require you to have a company for 18 months.
01:00:09.760 --> 01:00:13.760
Like, bro, in the US, the companies exited like three times in 18 months.
01:00:13.920 --> 01:00:14.239
Yeah.
01:00:14.480 --> 01:00:19.039
Like the technology has matured a thousand times over in those 18 months.
01:00:19.280 --> 01:00:25.280
The type of validation that we expect from a very traditional company doesn't apply to this new world.
01:00:25.440 --> 01:00:27.920
But our systems haven't evolved and caught up.
01:00:28.079 --> 01:00:32.159
So there's an education process that we need for all of the ecosystem.
01:00:32.320 --> 01:00:37.519
The venture capitalists, the industry, the banks, the demand side as well.
01:00:37.840 --> 01:00:40.880
Our own demand needs to go back to supporting local.
01:00:41.280 --> 01:00:48.000
And I don't say this in a kind of a patriarchal or a uh kind of protectionistic way.
01:00:48.159 --> 01:00:51.039
I mean that's how all of our companies really were grown.
01:00:51.360 --> 01:00:55.360
You think Singapore Airlines became Singapore Airlines by accident?
01:00:55.519 --> 01:00:55.760
No.
01:00:55.920 --> 01:00:59.599
Singaporeans, in fact, government officials were forced to fly Singapore Airlines.
01:00:59.760 --> 01:01:03.119
You could only buy a ticket on Singapore Airlines if you're a government officer.
01:01:03.440 --> 01:01:06.800
You think DBS grew to its size by accident?
01:01:06.960 --> 01:01:10.079
No, all our children were forced to get PSB accounts as children.
01:01:10.800 --> 01:01:12.320
So there was an error when we did this.
01:01:12.400 --> 01:01:17.920
And if you look at every other country in the world that has a thriving innovation ecosystem, Israel buys local first.
01:01:18.159 --> 01:01:20.159
Silicon Valley buys local first.
01:01:20.400 --> 01:01:22.000
France buys local first.
01:01:22.159 --> 01:01:24.159
Everybody else is doing it.
01:01:24.559 --> 01:01:29.760
But we have a very open and international perspective, which doesn't help us.
01:01:30.079 --> 01:01:34.159
And I think we need to start thinking this is not the same world that we grew up in.
01:01:34.239 --> 01:01:36.880
This is not the world trade order of open and free trade.
01:01:37.039 --> 01:01:41.519
We live in a very different world from the policy position of the 80s and 90s.
01:01:41.679 --> 01:01:43.280
It's time to start supporting local.
01:01:43.360 --> 01:01:45.280
I can't trust international models anymore.
01:01:45.440 --> 01:01:48.159
We better have somebody in Singapore who knows how to build a local model.
01:01:48.400 --> 01:01:50.239
I can't trust international security anymore.
01:01:50.320 --> 01:01:53.440
I better have somebody in Singapore who knows how to build security from scratch.
01:01:53.840 --> 01:01:57.760
So it is both survival as well as just economics.
01:01:58.159 --> 01:02:05.760
I would love to have more unicorns being based, not just based out of Singapore, built from Singapore by Singaporeans, scaling to the world.
01:02:06.079 -->
01:02:12.320
Bringing you back to your debate days, you wrote a book with Foreign Minister Vindy Krishna, writing the forward.
01:02:12.800 --> 01:02:16.880
So are there any kind of uh principles you're still applying from that today?
01:02:17.199 --> 01:02:19.280
Yeah, so that book was called Think, Speak, Win.
01:02:19.360 --> 01:02:23.840
So I used to teach debate, and then a lot of parents would be like, Hey, I want to send my kid for debate training with you.
01:02:23.920 --> 01:02:26.239
And I was like, that I don't do debate training for kids.
01:02:26.320 --> 01:02:29.840
I mean for individuals, I do it for teams, for schools, for the country.
01:02:30.000 --> 01:02:32.559
Um but I'm happy to write down some of my thoughts.
01:02:32.719 --> 01:02:38.400
And so eventually those thoughts became a whole book, and I published the book, and it became the debate textbook for most schools for many years.
01:02:38.559 --> 01:02:40.480
Uh that book is like 15 years old now.
01:02:40.639 --> 01:02:42.400
But the key lessons that I learned are two.
01:02:42.800 --> 01:02:44.800
One is think before you speak.
01:02:45.119 --> 01:02:47.760
So the book title is literally Think, Speak, Win.
01:02:48.000 --> 01:02:50.239
If you don't do it in that order, you cannot win.
01:02:50.400 --> 01:02:53.920
Uh and that lesson is regularly forgotten.
01:02:54.159 --> 01:02:58.079
People start speaking before they think, and they end up in all sorts of problems.
01:02:58.239 --> 01:02:59.760
And I see this with AI too.
01:02:59.920 --> 01:03:11.519
Um one of the interesting things we've started doing is when I hire people, I ask them to submit their AI transcripts, you know, the conversation history of how they built the dev feature that we've asked them to build.
01:03:11.760 --> 01:03:15.760
And you can tell a lot about how they think from how they interact with AI.
01:03:16.159 --> 01:03:17.280
Three types of people.
01:03:17.599 --> 01:03:21.519
One is copy and paste the problem statement inside, please solve.
01:03:21.840 --> 01:03:25.519
Then yes, yes, yes, yes, yes, yes, yes, yes, yes, yes, end up in a rabbit hole.
01:03:25.840 --> 01:03:26.639
Looking at you, Eric.
01:03:28.000 --> 01:03:31.599
Second type is the one who stops, thinks about it.
01:03:31.760 --> 01:03:33.039
Hey, this is the problem statement.
01:03:33.199 --> 01:03:37.840
But clearly there must be a way in which we can architect or conceive of how the solution should look like.
01:03:38.079 --> 01:03:40.800
Ah, okay, so this is the architecture I think makes sense.
01:03:41.039 --> 01:03:42.239
This is how I would do it.
01:03:42.400 --> 01:03:55.039
Okay, now AI, you build this feature for me with this concept, with this architecture, be mindful of these particular risks, and then it will build, and then you can yes, no, yes, no, yes, but do this, yes, but do that, no, but do this.
01:03:55.360 --> 01:03:57.119
The third type is firefighters.
01:03:57.440 --> 01:04:03.199
Just keep prompting short prompts, don't know what's going on, and it's not even the accept all recommendations, it's just firefighting.
01:04:03.440 --> 01:04:05.440
And they have no sense of what they're actually trying to do.
01:04:05.920 --> 01:04:08.400
Those, the first and the third, I will never hire.
01:04:09.039 --> 01:04:12.079
The second ones are the ones who are thinking before they are speaking to the AI.
01:04:12.239 --> 01:04:15.440
And if you look at the future work skills, that's the type of person that will survive.
01:04:15.599 --> 01:04:31.840
The person who's not just giving the problem to the AI and hoping that the answer comes back, but actually putting in their own independent thought and judgment as to what they think should be done, how they think it could be done, and having independent validation and input to all of the questions that are being asked.
01:04:32.159 --> 01:04:35.039
Now, if you don't know any better, you can yes, yes, yes, yes, yes, all your way.
01:04:35.199 --> 01:04:37.039
You will probably get to the same answer.
01:04:37.440 --> 01:04:40.079
But will you be as good as the other person that I'll hire?
01:04:40.639 --> 01:04:40.960
No.
01:04:41.519 --> 01:04:42.960
And so that's what I'm looking for.
01:04:43.199 --> 01:04:51.199
The second thing I learned from that book and from the whole experience of being a debater is that individually we will never have the right answer.
01:04:51.440 --> 01:05:03.440
But collectively, if we have a process by which we are comfortable sharing our differing views and we have a team which has differing views, the chances and the probability of us getting the right answer are higher.
01:05:03.599 --> 01:05:04.639
Again, AI.
01:05:05.119 --> 01:05:09.519
If you have just one model and you ask it, build me this feature, it'll say, build it this way.
01:05:09.920 --> 01:05:13.280
If you say, hey, you know, Codex, build me this feature, great.
01:05:13.519 --> 01:05:20.159
Hey, uh Claude, can you look at what Codex proposed and find all of the problems with their plan and make it better?
01:05:20.400 --> 01:05:22.880
Suddenly your plan will get much better.
01:05:23.119 --> 01:05:31.920
So having that diverse opinions, even kind of the blue team, red team approach or the QA approach in your agentic workflow makes the idea better.
01:05:32.400 --> 01:05:45.280
So in every phase of life, whether it's discussing what you want to build next, discussing how the company should be run, discussing with your agents how the feature should look like, having multiple diverse perspectives always makes the feature better.
01:05:45.519 --> 01:05:50.079
Adds a bit more time, is a bit more painful and expensive, but always better.
01:05:50.239 --> 01:05:51.519
And that's how I run the company.
01:05:51.679 --> 01:05:55.840
Everybody in my team has an opinion on everything, and we like it.
01:05:56.400 --> 01:05:57.199
So thing we have.
01:05:57.360 --> 01:05:59.760
I think the agreeableness of uh
LIMs actually.
01:06:00.320 --> 01:06:01.760
As a bit of problem, right?
01:06:02.000 --> 01:06:04.400
Because they tend to agree with what you say, right?
01:06:04.480 --> 01:06:06.000
And you're kind of correct.
01:06:06.159 --> 01:06:12.000
And sometimes it just makes it not as um adversarial as you you would expect it to be.
01:06:12.480 --> 01:06:13.280
So I turned that off.
01:06:13.440 --> 01:06:20.559
I I mean my specific system level prompt is to not be polite, not be agreeable, always identify flaws in my thinking and challenge me.
01:06:20.719 --> 01:06:20.880
Wow.
01:06:21.679 --> 01:06:23.039
Because that's how I want my people to be.
01:06:23.519 --> 01:06:26.000
My mind sweats at me when I say something stupid.
01:06:26.800 --> 01:06:28.159
I mean I I won't go that far.
01:06:28.960 --> 01:06:30.639
Speak in politely, lah, politely.
01:06:31.199 --> 01:06:34.000
I don't want the robot overlord to come quite thick skin and all.
01:06:34.079 --> 01:06:35.440
That's why to start in.
01:06:35.679 --> 01:06:35.840
Yeah.
01:06:36.000 --> 01:06:42.480
Think before you speak and have adversarial and contradicting opinions to get to a better answer.
01:06:42.719 --> 01:06:46.480
And with that, I think it's a great place to go into our rapid fire round.
01:06:46.719 --> 01:06:47.599
Are you ready, Gara?
01:06:47.920 --> 01:06:48.239
Sure.
01:06:48.480 --> 01:06:48.719
Alright.
01:06:48.800 --> 01:06:52.079
So what are the two or three books that you often recommend to people?
01:06:52.400 --> 01:06:53.519
I go old school.
01:06:54.159 --> 01:06:57.440
So one of the books is Sapiens by Yuval Harari.
01:06:57.760 --> 01:06:59.280
It's basically the evolution of species.
01:06:59.360 --> 01:07:03.599
I like to see how things have evolved over time to get us to where we are right now.
01:07:03.840 --> 01:07:07.039
So that you look back and understand what this all means in context.
01:07:07.119 --> 01:07:09.760
So Yuval Harari's books are great, especially sapiens.
01:07:10.239 --> 01:07:14.960
Second one, and most people have probably never read this, it's a book called Unspeak by Steven Spool.
01:07:15.119 --> 01:07:16.559
Steven Stephen Poole, yeah.
01:07:16.880 --> 01:07:22.880
So Unspeak is about how people use words and give them different meanings.
01:07:23.360 --> 01:07:25.679
And sometimes pollute the meanings.
01:07:25.840 --> 01:07:27.760
Like a surgical strike.
01:07:28.079 --> 01:07:29.440
You know, surgery is good for you.
01:07:29.760 --> 01:07:32.239
Surgery is what doctors use to take out bad things.
01:07:32.800 --> 01:07:34.719
A strike is not good for anybody.
01:07:35.039 --> 01:07:36.880
But a surgical strike sounds like it's good.
01:07:37.039 --> 01:07:40.239
No, we went in there and only cut out the diseased parts and boom, small.
01:07:40.639 --> 01:07:43.280
So they reuse words in very clever ways.
01:07:43.440 --> 01:07:48.000
And you see this now in AI, in a lot of the ways in which distilling.
01:07:48.480 --> 01:07:49.440
We're just distilling.
01:07:49.519 --> 01:07:51.199
Distilling is a process of purification.
01:07:51.280 --> 01:07:53.519
It's cleansed, it's cleansing the original model.
01:07:53.599 --> 01:07:55.599
It's like, no, dude, you're copying and pasting.
01:07:55.840 --> 01:07:57.840
Like that's what distilling is.
01:07:58.719 --> 01:08:01.199
But we're seeing this reuse of words, which is fascinating.
01:08:01.360 --> 01:08:04.000
So it's the whole history of how we've kind of abused and reused words.
01:08:04.320 --> 01:08:09.599
The last one is uh James Surio Weki, Wisdom of the Crowds.
01:08:10.239 --> 01:08:27.920
And this one I love because when the internet first came out, we all believed, and this is the whole tech generation believed, now we will have truth because information will be free and therefore truth will float to the surface and lies will sink below.
01:08:28.079 --> 01:08:28.800
But guess what?
01:08:28.880 --> 01:08:29.760
It's the opposite.
01:08:30.000 --> 01:08:32.720
Fake news is everywhere, truth has buried deep.
01:08:32.960 --> 01:08:40.079
And so we have this tech utopian view every single time of how technology is gonna make life fundamentally better.
01:08:40.239 --> 01:08:45.600
And I love recommending it because we're now in an era where, don't worry, AI will change everything.
01:08:45.760 --> 01:08:53.199
Jobs will be so much easier to do, we'll all be sitting at home clicking, you know, yes to Claude while we sit back on our margaritas.
01:08:53.439 --> 01:08:54.960
Like, that's not how it's gonna play out.
01:08:55.199 --> 01:08:57.439
100% that is not gonna be how it plays out.
01:08:58.079 --> 01:09:02.079
Because every single technology, social media, like, oh now we'll all be friends.
01:09:02.319 --> 01:09:04.800
Nope, we're all enemies because of social media.
01:09:05.119 --> 01:09:10.880
So every single technology starts off with this utopian hype and ends with this kind of dystopian collapse.
01:09:11.279 --> 01:09:13.199
We're in the same utopian hype today.
01:09:13.760 --> 01:09:18.159
I've been through the cycle many times and I'm a cybersecurity professional, so I understand digital risks.
01:09:18.319 --> 01:09:19.600
We're gonna go through it again.
01:09:19.760 --> 01:09:29.439
But preparing for it and understanding how to manage that allows you to use social media wisely, use the internet wisely, and use AI wisely.
01:09:30.239 --> 01:09:30.560
Yeah.
01:09:30.800 --> 01:09:36.079
What's a question you think more people should be asking about digital risks?
01:09:37.840 --> 01:09:39.039
Just what are they?
01:09:39.199 --> 01:09:45.680
I mean, people are so enamored by the opportunities that they don't think, wait, what is this costing me?
01:09:46.079 --> 01:09:53.920
It's like when social media first came out, and then later we discovered, oh, it's it's taking all of our data and it's using that to change our worldviews.
01:09:54.399 --> 01:09:58.000
Ooh, I don't know if I'm
comfortable with my worldviews being shaped by an algorithm.
01:09:58.560 --> 01:09:58.800
Yeah.
01:09:59.039 --> 01:09:59.680
Too late.
01:10:00.079 --> 01:10:02.479
So we we we should be asking, you know, what does it cost us?
01:10:02.560 --> 01:10:02.800
Yeah.
01:10:02.960 --> 01:10:05.600
And and what are the long-term implications of this on society?
01:10:06.479 --> 01:10:10.479
What do my children think of a world in which we are talking to our toasters?
01:10:10.720 --> 01:10:13.600
Wow, this quickfire is getting way more interesting.
01:10:13.760 --> 01:10:18.479
Um what's a favorite show or film or content that you enjoy?
01:10:19.760 --> 01:10:22.800
Ooh, uh Guilty Pleasures, I love sci-fi.
01:10:23.520 --> 01:10:29.680
So Dune, um Harry Potter, anything that is sci-fi, I'm hard into it.
01:10:29.760 --> 01:10:34.479
I am I was a loyal Star Wars fan until they destroyed the universe.
01:10:34.800 --> 01:10:36.960
So my call sign as a pilot actually was Yoda.
01:10:37.279 --> 01:10:39.600
So very loyal original series fan.
01:10:39.680 --> 01:10:41.840
Uh and then and then they destroyed it for me.
01:10:41.920 --> 01:10:44.319
But sci-fi is my guilty pleasure because it's escapism.
01:10:44.479 --> 01:10:46.159
It's it's a different universe entirely.
01:10:46.399 --> 01:10:46.640
Great.
01:10:46.880 --> 01:10:48.800
A place that is special to you in Singapore?
01:10:49.920 --> 01:10:51.920
Like food, anything, yeah.
01:10:53.039 --> 01:11:00.000
I don't have a place that's special to me, but I have a specific thing that's special to me, and that's Kopi.
01:11:00.640 --> 01:11:04.319
So I thrive on Kopi in particular.
01:11:04.960 --> 01:11:05.840
So here's the thing.
01:11:06.560 --> 01:11:08.239
And this is why this is why it gets irritating.
01:11:08.640 --> 01:11:11.279
Kopi, by definition, does not have creamer.
01:11:11.520 --> 01:11:14.640
It is the normal Kopi plus condensed milk.
01:11:14.880 --> 01:11:15.600
Correct.
01:11:15.920 --> 01:11:23.119
But if you go to the more Atas places in shopping centers, they will add the creamer as a default addition for free.
01:11:23.359 --> 01:11:26.079
So I have to order Kopi no C, which is not a thing.
01:11:26.239 --> 01:11:28.319
Kopi no see is not a thing.
01:11:28.479 --> 01:11:31.520
But if you go to like a Toast Box or a Yakun or whatever it is, or Mr.
01:11:31.600 --> 01:11:33.760
Wang's, you have to order specifically Kopi no see.
01:11:33.920 --> 01:11:35.199
Then they'll be like, oh, no creamer, is it?
01:11:35.279 --> 01:11:37.920
I was like, yes, because by right, it doesn't come with creamer, damn it.
01:11:38.399 --> 01:11:42.960
But I love my kopi so much that there was a year that I was studying overseas in Boston.
01:11:43.359 --> 01:11:46.479
I bought the sock and the coffee bean.
01:11:46.560 --> 01:11:48.960
Like I brought it all the way to UK, to the US.
01:11:49.119 --> 01:11:58.880
And every morning I'll sit there with my like metal thing and like the coffee powder and like straining my coffee beans and like making my kopi with condensed milk that I had to buy in the US to make my Kopi every day.
01:11:59.039 --> 01:12:01.279
And today, every morning, my day starts with Kopi.
01:12:01.439 --> 01:12:08.319
And my staff, both in my current organization as well as all of my previous organizations, know do not talk to Gorf before he's had his Koe.
01:12:09.600 --> 01:12:13.039
It's not that I'm mean or rude or anything, it's just that I'm not functional as a human being.
01:12:13.119 --> 01:12:16.159
Like I have no intelligent things to add to the conversation.
01:12:18.159 --> 01:12:20.800
What's the product that you recently got excited about?
01:12:21.119 --> 01:12:22.000
Man, AI.
01:12:22.399 --> 01:12:24.960
And I'll say that the thing that got me most excited is probably Suno.
01:12:25.760 --> 01:12:35.279
Because it's it's it allowed me to hear my own music just sung back at me in a at a quality and a level and in a way I never thought possible.
01:12:35.600 --> 01:12:37.600
I never thought I'd hear my own music that way.
01:12:37.840 --> 01:12:43.600
And to hear your guitar come back that clean, that crisp, and better, frankly.
01:12:43.920 --> 01:12:44.479
It's amazing.
01:12:44.640 --> 01:12:45.279
It's an amazing feeling.
01:12:45.600 --> 01:12:46.479
Maybe a last one.
01:12:46.560 --> 01:12:50.880
Uh, what's a line or a life philosophy you often go back to?
01:12:51.279 --> 01:12:52.000
I mean, there's two.
01:12:52.159 --> 01:12:54.720
So one is we've spoken about it a few times.
01:12:55.039 --> 01:12:57.199
Try to do well by doing good.
01:12:57.520 --> 01:12:59.840
There are many ways you can do well in life.
01:13:00.479 --> 01:13:02.720
Many of those don't involve you doing good.
01:13:03.279 --> 01:13:04.960
I try to do well by doing good.
01:13:05.039 --> 01:13:06.000
So that's the first one.
01:13:06.319 --> 01:13:20.479
Second one is I think uh, like going back to that Forrest Gump analogy, I think I've done well in life, not necessarily because I am awesome, because I've just been lucky to get certain opportunities that somebody else didn't get.
01:13:20.880 --> 01:13:23.039
Uh and there's nothing special.
01:13:23.279 --> 01:13:33.680
I mean, I worked hard, but I was also really, really lucky to get certain opportunities and certain exposure and certain mentors and certain, you know, placements.
01:13:34.000 --> 01:13:41.279
And I need to find a way to give that forward, to pay that forward, to make sure that somebody else benefits from all of this beyond just me.
01:13:41.600 --> 01:13:45.039
So I mean, I spend a lot of time mentoring very aggressively.
01:13:45.279 --> 01:13:47.840
Now I don't have the capacity to do one-on-one mentoring anymore.
01:13:47.920 --> 01:13:49.680
So that's why I share so openly.
01:13:50.000 --> 01:13:52.720
You know, the minute you guys ask me, I was like, yeah, let's do this.
01:13:52.880 --> 01:13:56.960
The minute, you know, the our chat group
is like, hey, do you want to share how you're building things?
01:13:57.199 --> 01:13:58.079
Like, yeah, let's do it.
01:13:58.399 --> 01:14:06.239
Uh and and I think that by sharing, I believe in karma, actually a lot of good stuff has come back to me.
01:14:06.399 --> 01:14:13.520
And people have been like, hey, you gave a speech in my secondary school when I was like, you know, 14 years old, and I loved it.
01:14:13.680 --> 01:14:15.359
And now I'm running a company that does this.
01:14:15.520 --> 01:14:16.960
Can I help you in any way?
01:14:17.359 --> 01:14:18.000
Like, huh?
01:14:18.159 --> 01:14:21.119
You remember a speech I gave, like, and I have this regularly.
01:14:21.279 --> 01:14:24.800
It's like, hey, you coached me in my like primary school debate.
01:14:24.960 --> 01:14:27.520
Like, now I am you know the CISO of this company.
01:14:27.680 --> 01:14:28.560
Can I help you?
01:14:29.119 --> 01:14:31.359
So I didn't do it back then to get this favor.
01:14:31.520 --> 01:14:35.359
Like, I didn't like, well, 15 years from now, this guy confirm will be CISO one.
01:14:35.520 --> 01:14:36.319
I will invest in him.
01:14:36.640 --> 01:14:37.359
Like, you just do it.
01:14:37.600 --> 01:14:39.279
And then who knows where life takes you.
01:14:39.439 --> 01:14:44.000
But no matter what, having people who think positively of you is always a net good.
01:14:44.319 --> 01:14:48.479
And always ends up with certain, like again, doing well through doing good.
01:14:48.960 --> 01:14:50.319
Always ends up in a great way.
01:14:50.399 --> 01:14:52.880
So I I always believe in that philosophy.
01:14:53.199 --> 01:14:53.920
Awesome.
01:14:54.159 --> 01:14:58.880
And where can listeners find you and how can they be useful to you?
01:14:59.760 --> 01:15:01.279
Well, how can I be useful to them?
01:15:01.439 --> 01:15:05.520
Um, so they can find me on our website, it's strongkeep.com.
01:15:05.680 --> 01:15:08.479
So strong muscles, keep, you know, keep something.
01:15:08.560 --> 01:15:14.159
It's actually uh the str the safest part of a castle in medieval architecture is called a keep.
01:15:14.239 --> 01:15:18.800
And so the strong keep, the stronghold, both the synonyms, you know, the safest part of a castle.
01:15:18.880 --> 01:15:20.000
So that's where the team came up with the name.
01:15:20.079 --> 01:15:21.439
And the team came up with the name, not me.
01:15:21.520 --> 01:15:23.119
Uh but so strongkeep.com.
01:15:23.199 --> 01:15:27.199
Uh so they can email me, gorov at strongkeep.com, or they can find me at LinkedIn.
01:15:27.279 --> 01:15:35.439
Uh luckily, I'm the only Gorov Kiti in the world, so it is relatively easy to find me, which is also hard because we want to hide from people, really tricky.
01:15:35.520 --> 01:15:36.319
But that's that's me.
01:15:37.119 --> 01:15:37.680
Awesome.
01:15:37.760 --> 01:15:39.279
Thank you so much for your time, Gorov.
01:15:39.439 --> 01:15:41.039
It was really nice chatting with you.
01:15:41.359 --> 01:15:44.880
Yeah, likewise, thanks uh thanks both of you for all the great questions.
01:15:45.199 --> 01:15:46.399
Yeah, thank you for your time.
01:15:46.560 --> 01:15:51.119
Yeah, thank you so much for listening to our episode with Garof Kirti.
01:15:51.359 --> 01:15:54.239
We hope you enjoyed it and learned as much as we did.
01:15:54.399 --> 01:15:56.319
You can check out our sub stack.
01:15:56.479 --> 01:16:07.840
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.
01:16:08.079 --> 01:16:10.640
And we hope to see you in the next episode.