WEBVTT

00:00:00.340 --> 00:00:05.720
Okay, just, I just wanted to show the view like
at 4, 5 p.m.

00:00:07.260 --> 00:00:14.900
I don't know if you can see the tower, like the
Bitexco Tower, but the sunlight is shining off.

00:00:11.820 --> 00:00:13.040
Wow, it's a lovely day.

00:00:15.260 --> 00:00:20.460
Yeah, well, this is the regular day in Vietnam,
but you can see how the 4 p.m. sun.

00:00:18.320 --> 00:00:22.980
Oh, can you angle the camera down a bit so I
can see the buildings?

00:00:23.000 --> 00:00:23.480
Oh, yeah.

00:00:23.820 --> 00:00:24.860
It looks like Singapore.

00:00:26.120 --> 00:00:27.520
Yeah, so this is, this is.

00:00:28.160 --> 00:00:30.460
the Phu My Hung area, which used to be swamps,

00:00:30.660 --> 00:00:32.340
but now it's full of construction.

00:00:32.640 --> 00:00:35.480
So all of those villas that were there for like
50 years

00:00:35.480 --> 00:00:37.540
now are being replaced by restaurants

00:00:37.540 --> 00:00:40.080
and some of them being removed.

00:00:40.340 --> 00:00:41.720
You see, they're all kind of the same, right?

00:00:41.780 --> 00:00:44.180
It was a huge development and people buy them

00:00:44.180 --> 00:00:46.280
and then they open a restaurant

00:00:46.280 --> 00:00:48.840
and then you have motorbikes everywhere on the
street.

00:00:49.160 --> 00:00:50.480
They open an office.

00:00:50.680 --> 00:00:54.080
So there's no zoning laws here, which is kind
of sad.

00:00:54.600 --> 00:00:56.140
In a way, it's kind of, you know,

00:00:56.140 --> 00:01:00.400
it's nice that everybody just does whatever
they want but in a way if you live in a nice
area

00:01:00.400 --> 00:01:08.160
then you suddenly have cars everywhere and and
then people get upset because they want um

00:01:08.160 --> 00:01:16.300
they don't want cars parked on the street so
they um they put all kinds of cones uh so
there's a

00:01:16.300 --> 00:01:21.620
restaurant but then the neighbors put cones and
then the street is like overloaded i don't know

00:01:21.620 --> 00:01:23.800
you don't have that problem in the UK I guess

00:01:23.800 --> 00:01:25.780
because there's strict laws, zoning

00:01:25.780 --> 00:01:27.740
laws of what you can do in each

00:01:27.740 --> 00:01:28.420
area, right?

00:01:28.480 --> 00:01:45.200
Yeah. Yeah. I'm not a huge fan of zoning laws
because I guess it's just a sliding scale or
something, right? Like you want to be free to
do things and you don't want the government to
basically weigh in all the time.

00:01:42.000 --> 00:01:42.440
Yeah.

00:01:47.040 --> 00:01:56.200
So you do need building control and zoning, but
sometimes it's a bit much, I feel. I think the
balance can be quite weird.

00:01:56.600 --> 00:02:00.100
So in Belgium, we were really bad.

00:02:00.220 --> 00:02:00.680
Well, they are.

00:02:00.740 --> 00:02:02.140
Oui.

00:02:00.800 --> 00:02:01.080
We are.

00:02:01.900 --> 00:02:02.800
I grew up.

00:02:03.560 --> 00:02:04.260
I don't know.

00:02:04.380 --> 00:02:08.180
I think at this point in my life, I've lived
more outside of Belgium than in Belgium.

00:02:09.840 --> 00:02:14.020
But we, Belgians, have a lot of problems.

00:02:14.200 --> 00:02:15.520
And we have a name for it.

00:02:15.640 --> 00:02:16.760
It's called lintbebouwing.

00:02:16.980 --> 00:02:19.140
It means that there's like one street.

00:02:19.140 --> 00:02:22.700
and because there's no regulation on how people
build,

00:02:22.920 --> 00:02:24.440
you literally have like a street

00:02:24.440 --> 00:02:27.440
and then the houses are all built along the
street.

00:02:27.880 --> 00:02:29.420
And then behind those houses is nothing.

00:02:29.720 --> 00:02:35.140
It's like a strip with like a lint really means
like a line.

00:02:35.380 --> 00:02:38.760
So if you drive from Belgium into the
Netherlands,

00:02:39.140 --> 00:02:40.840
you immediately see the difference

00:02:40.840 --> 00:02:43.220
between a nice well-organized country

00:02:43.220 --> 00:02:46.940
where everything is neatly zoned

00:02:46.940 --> 00:02:48.800
and then Belgium, which is

00:02:49.760 --> 00:02:50.700
the roads.

00:02:51.080 --> 00:02:52.880
Your car starts bumping on the roads

00:02:52.880 --> 00:02:54.920
at least back in 2006

00:02:54.920 --> 00:02:56.460
when I left.

00:02:58.100 --> 00:02:59.200
Crossing the border

00:02:59.200 --> 00:03:00.580
is like you hit

00:03:00.580 --> 00:03:02.260
the road starts bumping.

00:03:02.220 --> 00:03:02.860
dear

00:03:02.620 --> 00:03:04.920
The lights on the street are

00:03:04.920 --> 00:03:06.880
triple the lights that you have

00:03:06.880 --> 00:03:08.660
in the Netherlands because Belgium

00:03:08.660 --> 00:03:10.980
even you can see them from space

00:03:10.980 --> 00:03:12.280
because they put lights everywhere.

00:03:12.720 --> 00:03:14.460
I'm looking forward to go back in

00:03:14.460 --> 00:03:15.260
January.

00:03:15.080 --> 00:03:16.080
so yeah

00:03:16.080 --> 00:03:18.840
I'm looking forward to the Belgian trip

00:03:19.880 --> 00:03:23.240
how long has it been since you last were in
Belgium again?

00:03:26.000 --> 00:03:26.740
I think

00:03:28.660 --> 00:03:29.140
probably

00:03:30.720 --> 00:03:32.280
since at least 10 years

00:03:32.280 --> 00:03:34.200
I think the last time I went was in

00:03:34.200 --> 00:03:34.900
2016

00:03:34.600 --> 00:03:38.340
wow that's going to be a lot i suspect

00:03:36.760 --> 00:03:37.240
yeah

00:03:39.140 --> 00:03:40.240
I'm going to see all the

00:03:40.240 --> 00:03:41.240
cultural enrichment

00:03:44.340 --> 00:03:46.020
I mean in a good way

00:03:44.380 --> 00:03:45.760
hashtag controversial

00:03:49.100 --> 00:03:55.700
the the obviously the the thing that
disappoints me about about the belgium thing is
that

00:03:55.700 --> 00:03:58.880
is that my AI related proposals,

00:03:59.320 --> 00:04:01.080
my talks at Config Management Camp,

00:04:02.620 --> 00:04:04.200
I guess I didn't get accepted.

00:04:04.540 --> 00:04:07.040
I even followed up asking for feedback

00:04:07.040 --> 00:04:08.060
and I didn't get feedback.

00:04:08.280 --> 00:04:12.720
So I'm a little bit disappointed at the
organizers,

00:04:13.000 --> 00:04:14.480
but I'm still booked in to go.

00:04:14.760 --> 00:04:15.680
You tell them.

00:04:16.200 --> 00:04:17.700
You tell them in their face.

00:04:18.800 --> 00:04:20.660
You didn't recognize me?

00:04:20.880 --> 00:04:21.660
I go every year.

00:04:22.340 --> 00:04:23.920
Finally, the first year I propose

00:04:23.920 --> 00:04:25.440
and you guys don't even give me feedback

00:04:25.440 --> 00:04:26.820
on what's wrong with my proposal.

00:04:26.820 --> 00:04:27.860
That's not very encouraging.

00:04:28.020 --> 00:04:30.840
That's not a very nice community organized
event.

00:04:31.280 --> 00:04:32.660
I guess they're super busy.

00:04:32.820 --> 00:04:33.120
I don't know.

00:04:34.960 --> 00:04:39.880
Well, maybe my proposals genuinely sucked, but,
well, there's that.

00:04:39.500 --> 00:04:41.680
Maybe you don't know what they compare to.

00:04:42.080 --> 00:04:45.520
I mean, I got my talk about CDK Terrain.

00:04:45.520 --> 00:04:49.600
I guess the only reason is because there might
be an audience that actually like,

00:04:49.680 --> 00:04:56.520
I mean, there might be people that like to use
CDK for Terraform after the...

00:04:55.700 --> 00:05:04.400
I think, as I was saying, my colleagues
independently started using CDK Terrain.

00:05:04.580 --> 00:05:07.040
Like, what you've built is commendable,
Vincent.

00:05:07.360 --> 00:05:10.880
So, oh, yeah, of course.

00:05:07.580 --> 00:05:08.760
Not just me, right?

00:05:08.880 --> 00:05:09.680
I mean, it's a community.

00:05:10.080 --> 00:05:11.080
I couldn't do it alone.

00:05:11.080 --> 00:05:12.140
yeah

00:05:12.460 --> 00:05:13.940
Claude, what's his name?

00:05:14.320 --> 00:05:16.620
John Steinig, Eduardo

00:05:15.020 --> 00:05:15.580
Codex.

00:05:16.880 --> 00:05:17.440
Codex.

00:05:18.240 --> 00:05:19.020
Simon

00:05:19.020 --> 00:05:20.340
Heather

00:05:20.340 --> 00:05:21.360
Gabe

00:05:22.160 --> 00:05:24.680
there's quite a few people that are

00:05:24.680 --> 00:05:26.780
and Torsten of course

00:05:26.780 --> 00:05:29.020
the person who offered the Open Construct

00:05:29.020 --> 00:05:31.060
and basically manages

00:05:31.060 --> 00:05:32.300
the Open Constructs Foundation

00:05:32.300 --> 00:05:34.440
that originally was created with

00:05:34.440 --> 00:05:36.760
Elad Ben-Israel and Sebastian Korfmann

00:05:36.760 --> 00:05:38.700
you know the foundational people

00:05:37.260 --> 00:05:38.660
okay

00:05:38.700 --> 00:05:40.040
behind the framework

00:05:40.040 --> 00:05:46.880
So it's only possible and also because of,
yeah, there is a kind of a foundation there,
yeah.

00:05:45.480 --> 00:05:57.920
And so you were messaging me earlier about a
demo that you have with the new AWS emulator.

00:05:58.600 --> 00:06:06.260
Am I right to assume that you somehow got CDKTN
running on the emulator

00:06:06.260 --> 00:06:10.760
and you've got some value out of it.

00:06:11.740 --> 00:06:18.620
Yeah, actually, that would be an interesting
little demo.

00:06:19.340 --> 00:06:22.480
Where is that session gone?

00:06:23.740 --> 00:06:24.800
God darn it.

00:06:26.020 --> 00:06:28.160
Anyway, did I close it?

00:06:31.480 --> 00:06:33.060
I don't even know where it is.

00:06:33.260 --> 00:06:34.060
It doesn't matter.

00:06:34.200 --> 00:06:36.760
I can ask any new agent to spin it up.

00:06:38.500 --> 00:06:39.980
I'll spin up the agent

00:06:39.980 --> 00:06:42.440
so basically what I've built

00:06:43.900 --> 00:06:44.540
is

00:06:44.540 --> 00:06:46.900
I'm running a workshop in Singapore

00:06:46.900 --> 00:06:48.560
on the 28th

00:06:48.560 --> 00:06:50.300
on Saturday which is with

00:06:50.300 --> 00:06:52.200
you know, Co Dojo

00:06:52.200 --> 00:06:54.180
from Michael Cheng who

00:06:54.180 --> 00:06:56.040
does work for GovTech

00:06:56.040 --> 00:06:58.200
and talks a lot about, he's very

00:06:57.960 --> 00:07:00.860
Michael Cheng is awesome.

00:06:58.200 --> 00:06:59.940
yes, sorry

00:07:00.860 --> 00:07:02.600
and if Michael

00:07:01.600 --> 00:07:02.240
yeah

00:07:02.240 --> 00:07:02.500
he

00:07:02.600 --> 00:07:04.060
is listening

00:07:04.060 --> 00:07:05.480
howdy

00:07:05.480 --> 00:07:06.540
howdy from the UK

00:07:06.060 --> 00:07:06.700
he's

00:07:06.700 --> 00:07:12.360
he's the OG behind a lot of the meetups in
Singapore, right?

00:07:12.520 --> 00:07:15.200
And then he started going to meetups with a
camera

00:07:15.200 --> 00:07:16.900
and a sound setup and everything,

00:07:16.780 --> 00:07:17.620
yeah

00:07:17.080 --> 00:07:18.000
and then getting volunteers,

00:07:18.420 --> 00:07:21.800
trying to cover every meetup that's happening
across Singapore

00:07:21.100 --> 00:07:22.500
yeah

00:07:21.800 --> 00:07:27.200
and setting up cameras and having everything
recorded and online,

00:07:27.600 --> 00:07:29.420
creating speaker profiles,

00:07:30.700 --> 00:07:32.800
capturing the event and the attendees

00:07:32.800 --> 00:07:33.920
and what the meetup was about,

00:07:33.920 --> 00:07:38.560
creating a little website that gives you the
ability to review that even today i'm like oh

00:07:38.560 --> 00:07:45.200
yeah there's a there's a recording of of that
like event that's only thanks to michael and
his

00:07:40.720 --> 00:07:42.120
yeah

00:07:45.200 --> 00:07:46.060
volunteers yeah

00:07:45.680 --> 00:07:50.320
Yeah, recording this stuff and documenting this
stuff is so important.

00:07:51.360 --> 00:07:51.940
yeah

00:07:51.980 --> 00:07:59.200
I just, a little aside, with AI, it's so
important to record your meeting correctly.

00:07:59.420 --> 00:08:07.080
Like, it's so important to have that artifact
so that as AI improves,

00:08:07.080 --> 00:08:10.000
that you can sort of throw in those meeting
artifacts

00:08:10.000 --> 00:08:14.620
and summarize and learn from those artifacts.

00:08:14.980 --> 00:08:20.020
So I don't know if you feel the same way,

00:08:20.160 --> 00:08:24.900
but I'm pretty keen, even when we're meeting
face-to-face

00:08:24.900 --> 00:08:27.560
in Belgium, Vincent, that we record our
conversations.

00:08:28.340 --> 00:08:32.900
oh yeah that's not a bad idea to set up some
time for that i'm there for a whole week

00:08:32.900 --> 00:08:35.020
we can sit in a nice Belgian café

00:08:35.020 --> 00:08:36.520
in the Belgian winter

00:08:36.520 --> 00:08:38.080
have some french fries

00:08:38.080 --> 00:08:40.560
a beer

00:08:40.560 --> 00:08:41.740
a nice duvel

00:08:41.740 --> 00:08:43.600
and listen to the horses

00:08:43.600 --> 00:08:46.440
in the center of Gent

00:08:46.440 --> 00:08:47.580
if that's still the case

00:08:47.720 --> 00:08:50.500
I didn't see horses last time I was in Ghent

00:08:50.500 --> 00:08:51.580
but Jesus

00:08:51.060 --> 00:08:52.300
I'm gonna be sad

00:08:52.520 --> 00:08:54.180
it was when?

00:08:54.640 --> 00:08:56.220
1820 were you last in?

00:08:56.700 --> 00:08:58.600
no so even when I was

00:08:58.600 --> 00:09:01.040
I go there from high school

00:09:01.040 --> 00:09:02.160
all the way to uni right

00:09:02.160 --> 00:09:12.000
So from when I started high school, goddammit,
I finished in 2001, so 18 years, sorry, six
years before that.

00:09:12.480 --> 00:09:15.420
So 1997, 98?

00:09:17.460 --> 00:09:21.000
we must be very close in age i don't think i

00:09:20.720 --> 00:09:23.240
I'm from 82, you're like three years older, I
think.

00:09:24.100 --> 00:09:28.300
oh all right but i i i went to university in 98
myself

00:09:33.160 --> 00:09:35.500
maybe because yeah

00:09:33.580 --> 00:09:34.100
98.

00:09:35.000 --> 00:09:35.520
Sorry.

00:09:35.700 --> 00:09:39.040
Yeah, well, I started in 2001, Uni, in Ghent.

00:09:39.240 --> 00:09:40.840
So the thing what I was referring to is,

00:09:41.000 --> 00:09:43.880
at some point, they had a project around the
city of Ghent

00:09:43.880 --> 00:09:48.800
where they diverted all the cars into parking
areas around the city.

00:09:48.980 --> 00:09:51.600
And the center of the city is car-free.

00:09:50.680 --> 00:09:52.080
cool

00:09:51.840 --> 00:09:52.800
There's no cars allowed.

00:09:52.080 --> 00:09:53.240
yeah

00:09:53.060 --> 00:09:57.460
So you're only having bicycles, trolleybuses,
and horses.

00:09:57.460 --> 00:10:02.120
So, like, I have a story of a friend of mine
who had a hearing problem.

00:10:02.400 --> 00:10:11.640
And his parents were traveling at the time he
was receiving his in-ear, the first time they
built the in-ear, according to the ear channel.

00:10:11.880 --> 00:10:18.820
So, he stayed in Belgium and didn't travel with
his parents because he had behind-the-ear
hearing aids.

00:10:18.980 --> 00:10:24.980
And he wanted to get the new in-ear ones
because they are obviously better and less
visible.

00:10:24.240 --> 00:10:25.640
Yeah.

00:10:25.140 --> 00:10:26.520
And so, he was very excited.

00:10:26.520 --> 00:10:32.500
So he stayed with us for a couple of weeks and
we took him there to pick up his earring aids,
new ones.

00:10:33.420 --> 00:10:38.960
And right after, well, they asked him, you
know, he's a funny guy.

00:10:39.060 --> 00:10:41.060
So they asked him, like, what do you think
about your voice?

00:10:41.260 --> 00:10:43.240
And he says, good, I guess.

00:10:43.740 --> 00:10:48.340
It was like, what am I supposed to say about my
own voice that I, you know, never really
clearly heard?

00:10:48.580 --> 00:10:53.900
But anyway, well, once we go outside, he was
like, horses?

00:10:54.260 --> 00:10:55.300
Do I hear horses?

00:10:56.040 --> 00:10:56.560
so

00:10:56.120 --> 00:11:02.820
yes there are like chariots and horses uh at
least back then you know um even in 2000

00:10:56.560 --> 00:10:57.440
click click click click click

00:11:02.740 --> 00:11:05.140
so you never heard them before oh sweet

00:11:05.560 --> 00:11:11.100
so many like small little details um that he
shared with us about like you know differences

00:11:11.680 --> 00:11:16.880
he had a lot of techniques like whenever
somebody would say something he was actually
very good at

00:11:16.880 --> 00:11:22.080
lib reading so he even though he didn't hear or
understand you always he would very he would
pick

00:11:17.440 --> 00:11:18.340
Mm-mm.

00:11:22.080 --> 00:11:29.760
up very quickly uh and then the second thing is
he would always like say really is it real

00:11:29.760 --> 00:11:35.520
because he wants you to repeat it or he he's
trying to guess more or get more context like

00:11:35.520 --> 00:11:41.420
he didn't actually like fully hear what you
said but he's he's got pieces so he's just
trying to

00:11:41.420 --> 00:11:46.160
without revealing that he didn't hear it he had
a lot of like defensive mechanisms and when you

00:11:46.160 --> 00:11:52.020
don't know it you like some people did they
were like what's wrong with you like are you
not

00:11:51.940 --> 00:11:52.880
That's interesting.

00:11:52.020 --> 00:11:54.920
hearing me and yeah

00:11:53.060 --> 00:11:53.500
That's interesting.

00:11:53.740 --> 00:11:58.220
I just, I'm comparing that with my experience
with AI.

00:11:58.600 --> 00:12:02.500
And if anything, I really wish AI would ask me
more questions.

00:12:02.500 --> 00:12:09.320
I'm always having to add like a suffix, ask me
how, or ask me for clarifications.

00:12:10.100 --> 00:12:11.080
You know what I mean?

00:12:11.180 --> 00:12:23.300
Like AI doesn't have that sense sometimes, at
least the ones I'm using to ask me for
clarification sometimes.

00:12:24.880 --> 00:12:25.980
It's probably just me.

00:12:28.460 --> 00:12:30.720
So anyway, we got a bit distracted there.

00:12:30.720 --> 00:12:33.420
Did you want to show the...

00:12:33.420 --> 00:12:35.740
I guess you don't have the demo ready, right?

00:12:37.100 --> 00:12:40.420
Of the AWS emulator and CDK Terrain.

00:12:37.620 --> 00:12:38.660
I just...

00:12:43.220 --> 00:12:51.460
Yeah, I just asked, I thought Haiku would just
be able to find the directory on my laptop and
spin up the EC2 instance.

00:12:51.460 --> 00:12:53.820
and it's reading the Terraform state

00:12:53.820 --> 00:12:55.560
and it's getting really confused.

00:12:55.960 --> 00:12:56.820
It's simple.

00:12:56.960 --> 00:12:58.640
It just needs to flip a toggle

00:12:58.640 --> 00:12:59.940
and run Terraform apply

00:12:59.940 --> 00:13:01.520
and then the EC2 instance spins up.

00:13:00.720 --> 00:13:04.520
Okay, so you've used the AWS emulator in anger.

00:13:04.600 --> 00:13:05.960
How does it compare with LocalStack

00:13:04.940 --> 00:13:05.500
No.

00:13:08.080 --> 00:13:13.480
and floky, flotchy, floky?

00:13:14.260 --> 00:13:16.940
Yeah, I didn't try Floki in this case.

00:13:17.540 --> 00:13:20.000
What I really wanted was the...

00:13:20.000 --> 00:13:35.140
So basically, the CDK workshop is API gateway
invoking Lambda, IAM policies, a Lambda
invoking another Lambda, like a proxy Lambda,
and I guess S3 buckets, not too much.

00:13:35.420 --> 00:13:40.740
So I thought, I know that the AWS emulator
announced last week is very early stage and has
a lot of gaps.

00:13:41.000 --> 00:13:45.100
So they even say it's absolutely not complete.

00:13:45.160 --> 00:13:46.760
It's a very early release.

00:13:46.760 --> 00:14:02.480
And when I asked Opus, it says, yeah, looks
like most of it is there, but it's all
automated built from the AWS API services, like
from the Smithy API contracts, right?

00:14:03.040 --> 00:14:03.760
So...

00:14:03.520 --> 00:14:05.020
expected something like that.

00:14:07.580 --> 00:14:09.020
It's cool that they've

00:14:09.020 --> 00:14:10.380
sanctioned something because

00:14:10.380 --> 00:14:13.040
there was a few sort of AWS

00:14:13.040 --> 00:14:14.500
projects I've been following

00:14:14.500 --> 00:14:17.300
which were promising

00:14:17.300 --> 00:14:18.200
to be

00:14:18.200 --> 00:14:21.280
some sort of emulator but didn't really

00:14:22.860 --> 00:14:23.580
transpire

00:14:23.580 --> 00:14:24.080
to be one.

00:14:26.100 --> 00:14:31.280
yeah so what what the so the first thing that
happened is i wanted to say it's it's all built

00:14:31.280 --> 00:14:39.540
like rust a single binary um on rust that spins
up really fast uh but it's like 900 megabyte at

00:14:39.540 --> 00:14:46.180
least the one that i built so to build it it
also takes a lot like when i first evaluated it
on the

00:14:40.140 --> 00:14:41.540
relaxed

00:14:46.180 --> 00:14:50.720
macOS, Opus said like, yeah, it doesn't have a
Darwin release.

00:14:50.880 --> 00:14:53.720
Like it doesn't have an ARM Darwin release.

00:14:53.900 --> 00:14:56.680
So it spun it all up in Docker to avoid.

00:14:56.820 --> 00:14:58.780
Like it does have an ARM Linux release.

00:14:58.940 --> 00:15:01.880
So when you run it in Docker or Colima or
whatever,

00:15:02.140 --> 00:15:04.720
it runs a little Linux VM on your machine.

00:15:04.820 --> 00:15:06.960
And then you can run the AWS emulator in there

00:15:06.960 --> 00:15:08.220
without having to rebuild it.

00:15:08.460 --> 00:15:09.900
Because if you want to rebuild it,

00:15:10.780 --> 00:15:14.380
the estimates of Opus is you need like a lot of
CPU kernels,

00:15:14.380 --> 00:15:21.380
vCPU and a lot of memory a lot of disk space
and because i'm like running agent fleets my
disk

00:15:21.380 --> 00:15:27.160
is actually running out so i just told opus all
right spin up a build instance so it's it spun
up

00:15:27.160 --> 00:15:34.500
like this 16 core 128 gigabyte memory instance
to build the emulator and then estimated about
40

00:15:34.500 --> 00:15:41.440
minutes build time but it got stuck after 30
minutes um so it builds every module like every

00:15:41.440 --> 00:15:43.600
Rust create independently. That's all in
parallel.

00:15:43.900 --> 00:15:45.000
So you have a lot of cores

00:15:45.000 --> 00:15:47.220
in use at the same time. But then

00:15:47.220 --> 00:15:49.220
after that, it needs to link all of those

00:15:50.780 --> 00:15:51.740
library files

00:15:51.740 --> 00:15:53.420
with a dynamic

00:15:53.420 --> 00:15:54.760
linker, like LD,

00:15:55.220 --> 00:15:57.460
Linux linking. So that needs

00:15:57.460 --> 00:15:59.600
to like, that actually was a single

00:15:59.600 --> 00:16:01.380
CPU, like a single core at

00:16:01.380 --> 00:16:02.900
100% load that

00:16:02.900 --> 00:16:05.140
was not benefiting from the

00:16:05.140 --> 00:16:07.160
multi-core instance. So

00:16:07.160 --> 00:16:08.900
Opus, funny enough,

00:16:09.480 --> 00:16:11.280
I don't know, I asked it, isn't there

00:16:11.280 --> 00:16:18.940
like an instruction in the readme but there is
a ldd or like lld there's like a parallel
linker

00:16:18.940 --> 00:16:25.660
that can link up things in parallel and opus
suggested it so while that initial one cpu

00:16:25.660 --> 00:16:31.560
linker was running it started a second build
which just took like a few seconds on the
original

00:16:31.560 --> 00:16:38.080
crates because nothing changed and then the
linker itself also just took like a few minutes
so like

00:16:38.080 --> 00:16:42.880
The total build time is just, I don't know, 30
minutes or so if you build from scratch.

00:16:43.120 --> 00:16:45.760
So that's pretty beefy.

00:16:45.840 --> 00:16:51.340
But if you use the right multi-threaded linking
process, it can build really fast.

00:16:52.280 --> 00:16:58.440
and you're talking about building cdk tn with
awc so i didn't quite understand what you're
building

00:16:57.800 --> 00:17:02.900
No, here I'm talking about because the emulator
is very early stage and has a lot of gaps,

00:17:03.120 --> 00:17:04.820
I had to make some changes.

00:17:04.860 --> 00:17:06.220
oh i see

00:17:04.980 --> 00:17:06.320
They do not accept PRs.

00:17:06.220 --> 00:17:24.120
and then yeah i can see it's all rust i was
going to mention that the other sort of

00:17:06.320 --> 00:17:08.540
so I just created my own

00:17:08.540 --> 00:17:10.100
like get patch

00:17:11.240 --> 00:17:12.260
so basically

00:17:12.260 --> 00:17:14.340
Fable came up with the actual patch

00:17:14.340 --> 00:17:16.260
that I needed to support the workshop

00:17:16.260 --> 00:17:18.700
and then it rebuilt the whole thing

00:17:20.560 --> 00:17:21.200
yeah

00:17:26.340 --> 00:17:34.780
aws thing i noticed was this project which was
in local web services and the guy that was
working on

00:17:34.780 --> 00:17:43.620
it uh is an aws employee so i was i was at
least i think he was maybe he isn't now i the i
was

00:17:43.620 --> 00:17:50.540
thinking that this would be the the the aws
emulator but but this one here yeah you're
right

00:17:47.400 --> 00:17:48.800
Python

00:17:48.800 --> 00:17:52.740
I think

00:17:50.540 --> 00:17:51.660
It's not going to fly.

00:17:51.860 --> 00:17:53.960
This one seems...

00:17:52.740 --> 00:17:54.800
Floki is in Java

00:17:54.800 --> 00:17:55.360
I think

00:17:55.360 --> 00:17:56.600
yeah

00:17:55.920 --> 00:17:57.120
This one's looking promising.

00:17:57.360 --> 00:17:59.240
So once you have that Rust binary,

00:18:00.020 --> 00:18:02.620
it's, I'm assuming, super fast to work with.

00:18:03.220 --> 00:18:05.440
It's way faster than the cloud, right?

00:18:07.200 --> 00:18:14.440
yes the goal of this emulator is for local like
agents to quickly um you know spin up things

00:18:14.440 --> 00:18:19.920
against an emulated environment and verify
things and spin things down with a lot less
memory

00:18:19.920 --> 00:18:22.020
than the other solutions.

00:18:22.400 --> 00:18:26.740
But I heard it's still quite beefy.

00:18:26.960 --> 00:18:30.040
Like for my workshop, it estimates that each
candidate,

00:18:30.340 --> 00:18:32.440
if we all apply at the same time,

00:18:32.560 --> 00:18:34.360
because that's how workshops usually go, right?

00:18:34.460 --> 00:18:35.720
Everyone kind of follows along.

00:18:35.860 --> 00:18:38.180
So everyone runs Terraform apply at the same
time.

00:18:38.260 --> 00:18:41.080
So it estimates that the memory requirements

00:18:41.080 --> 00:18:44.100
for a single attendee is easily five gigabyte.

00:18:45.000 --> 00:18:47.460
And you know what actually took up a lot of
that memory?

00:18:48.480 --> 00:18:49.880
No.

00:18:49.120 --> 00:18:52.760
I'm running a VS Code like Monaco Editor

00:18:52.760 --> 00:18:54.560
and it runs the language server,

00:18:54.740 --> 00:18:56.040
the TypeScript language server.

00:18:57.320 --> 00:19:00.100
Yeah, and the TypeScript language server with
VS Code

00:19:00.100 --> 00:19:02.940
spikes up to like six gigabytes of memory

00:19:02.940 --> 00:19:04.740
just parsing node modules

00:19:04.740 --> 00:19:07.040
and providing type information.

00:19:07.460 --> 00:19:10.740
So it was really fun to spin up a workshop

00:19:10.740 --> 00:19:12.460
because Opus was like,

00:19:12.520 --> 00:19:13.000
hey, you know what?

00:19:13.040 --> 00:19:14.460
We can modify the configuration

00:19:14.460 --> 00:19:16.300
of the TypeScript language server

00:19:16.300 --> 00:19:19.500
to eliminate the TypeScript type definition

00:19:20.140 --> 00:19:21.560
for all these node modules.

00:19:21.760 --> 00:19:24.820
And then I said, why don't we use TypeScript
Go?

00:19:24.960 --> 00:19:26.340
Why don't we use TypeScript 7?

00:19:27.420 --> 00:19:31.800
And then it found the TypeScript 7 Visual
Studio Code extension,

00:19:32.020 --> 00:19:34.340
which is not allowed to be used.

00:19:34.560 --> 00:19:36.540
The licensing does not allow you to use it

00:19:35.420 --> 00:19:39.740
So why don't you run your demo in the cloud?

00:19:36.540 --> 00:19:38.480
with non-Visual Studio Code, apparently.

00:19:40.020 --> 00:19:44.860
Because like, oh, you are.

00:19:41.420 --> 00:19:43.660
That's exactly what I am telling you that I'm
doing.

00:19:43.940 --> 00:19:45.300
I'm running, I'm holding.

00:19:45.300 --> 00:19:48.960
I need 5 gigabytes of memory per attendee.

00:19:49.800 --> 00:19:53.800
but so for each participant you're going to
have a cloud instance for them

00:19:54.340 --> 00:19:55.780
So I'm not doing that.

00:19:56.000 --> 00:20:00.560
Yes, we could spin up ECS task definitions per
attendee,

00:20:00.680 --> 00:20:01.400
like one container,

00:20:01.600 --> 00:20:04.040
but I'm actually using a single EC2 instance

00:20:04.040 --> 00:20:06.160
and it can hold,

00:20:07.640 --> 00:20:10.960
I think it's R7 Graviton,

00:20:11.160 --> 00:20:12.520
holds about seven.

00:20:12.520 --> 00:20:15.320
I think we estimated seven attendees per
instance.

00:20:15.560 --> 00:20:18.960
So I'm going to open source the whole
infrastructure for it.

00:20:19.060 --> 00:20:22.040
But it spins up an instance and shards the
attendees.

00:20:22.420 --> 00:20:26.300
So if you say I have 14 attendees and you say
the instance type, the size,

00:20:26.460 --> 00:20:29.080
it's going to shard the attendees across the
instances.

00:20:29.080 --> 00:20:35.420
So if one instance dies, then not everyone in
the workshop stops having access to it.

00:20:35.500 --> 00:20:37.560
But yes, there is a blast impact.

00:20:39.940 --> 00:20:41.140
basically I'm

00:20:41.140 --> 00:20:43.260
saving money by not using

00:20:43.260 --> 00:20:43.840
ECS

00:20:43.840 --> 00:20:45.940
I don't have to do

00:20:45.940 --> 00:20:48.980
like it's just simply having a single

00:20:48.980 --> 00:20:50.440
EC2 instance with a public endpoint

00:20:50.440 --> 00:20:53.020
and I put Caddy in front of it

00:20:53.020 --> 00:20:54.960
it serves the workshop like

00:20:54.960 --> 00:20:56.940
a dedicated workshop per

00:20:56.940 --> 00:20:57.480
attendee

00:20:58.860 --> 00:21:00.220
in Docker containers

00:21:00.220 --> 00:21:02.140
everything, it's running now

00:21:02.140 --> 00:21:04.580
let me open the link

00:21:02.720 --> 00:21:08.360
okay that's how did you get in roped into doing
are you getting obviously you're getting paid
to

00:21:08.360 --> 00:21:10.340
run this workshop, right?

00:21:09.280 --> 00:21:15.500
no no this is no um i don't know where which
screen i should share

00:21:13.260 --> 00:21:13.940
What?

00:21:14.960 --> 00:21:16.320
Who's paying

00:21:16.320 --> 00:21:17.980
for the compute

00:21:17.980 --> 00:21:19.880
to get the stacks running?

00:21:18.620 --> 00:21:23.520
i'm using one ec2 instance for five hours it's
like five dollars or less like three dollars

00:21:23.520 --> 00:21:30.100
to host seven attendees yeah it's on my account
plus i have a community builder um uh

00:21:23.820 --> 00:21:24.500
Okay.

00:21:31.840 --> 00:21:34.340
So what is your goal to run

00:21:32.060 --> 00:21:32.640
budget.

00:21:34.340 --> 00:21:36.100
this demo just so that people

00:21:37.200 --> 00:21:38.440
so you teach

00:21:37.640 --> 00:21:39.140
Because I wanted to do it.

00:21:40.960 --> 00:21:42.380
who pays for your flight

00:21:42.380 --> 00:21:43.760
and everything for Christ's sake

00:21:44.200 --> 00:21:45.920
I wanted to go to Singapore.

00:21:47.400 --> 00:21:48.460
so you're doing this

00:21:48.080 --> 00:21:49.560
I pay for my flight.

00:21:48.460 --> 00:21:51.040
so you're upscaling people

00:21:50.280 --> 00:21:50.400
Yeah.

00:21:51.040 --> 00:21:52.360
through the goodness of your heart

00:21:52.360 --> 00:21:54.200
but it's going to be recorded by Michael Cheng

00:21:54.200 --> 00:21:56.580
so it should have some distribution

00:21:56.580 --> 00:21:56.880
right

00:21:57.680 --> 00:22:06.880
okay so this is the the every attendee will get
a key a session name and a key and um

00:22:07.660 --> 00:22:16.080
and basically when they go to that dns it goes
into a reverse proxy that is caddy ZeroSSL

00:22:16.080 --> 00:22:17.300
with TLS via

00:22:18.320 --> 00:22:19.600
I don't know if you know Kadi

00:22:19.600 --> 00:22:20.720
do you know Kadi

00:22:21.800 --> 00:22:23.740
something's running I'll just give it access

00:22:24.580 --> 00:22:25.200
I don't know.

00:22:24.960 --> 00:22:25.800
so Caddy

00:22:25.420 --> 00:22:25.840
Oh, Caddy.

00:22:26.120 --> 00:22:26.480
Yeah, yeah.

00:22:26.600 --> 00:22:28.940
I was one of the first users of Caddy.

00:22:28.520 --> 00:22:29.280
Kadi

00:22:29.280 --> 00:22:30.500
oh yeah sure you were

00:22:31.200 --> 00:22:31.580
Okay.

00:22:31.760 --> 00:22:32.560
weren't we all

00:22:33.120 --> 00:22:33.880
aren't we all

00:22:36.540 --> 00:22:37.660
at least you didn't say

00:22:37.660 --> 00:22:39.560
you were one of the first users of Traefik

00:22:41.540 --> 00:22:41.940
Victor

00:22:43.340 --> 00:22:44.040
how is

00:22:44.040 --> 00:22:45.500
vector is not

00:22:45.500 --> 00:22:46.240
traffic

00:22:47.640 --> 00:22:48.820
no I wasn't one of the

00:22:50.480 --> 00:22:51.460
yeah I know Traefik

00:22:51.460 --> 00:22:53.840
I mean I'm more of a Kong kind of guy

00:22:54.020 --> 00:22:55.340
oh my god

00:22:55.340 --> 00:22:57.160
that's just Lua and Nginx

00:22:57.160 --> 00:22:58.240
that's OpenResty

00:22:58.240 --> 00:23:00.880
with a brand on it

00:22:58.580 --> 00:22:59.360
yeah I know

00:22:59.360 --> 00:22:59.940
I know

00:23:00.880 --> 00:23:02.280
anyway

00:23:01.540 --> 00:23:03.260
I know but it works really well though

00:23:02.280 --> 00:23:04.200
Caddy is amazing

00:23:04.200 --> 00:23:05.880
it's the best

00:23:04.600 --> 00:23:05.320
I love it

00:23:07.560 --> 00:23:16.400
the https integration with uh Let's Encrypt was
chef kiss chef kiss

00:23:12.160 --> 00:23:13.100
ZeroSSL

00:23:16.060 --> 00:23:20.940
Yeah, it's my favorite reverse proxy at the
moment for simple projects.

00:23:18.900 --> 00:23:24.800
and i'm trying to remember the guy the the guy
behind it he's from the church of the

00:23:24.800 --> 00:23:25.200
Saints.

00:23:27.200 --> 00:23:28.980
Matt Holt or something like that.

00:23:28.100 --> 00:23:28.660
Okay.

00:23:29.420 --> 00:23:31.780
All right, so this is what the attendees get.

00:23:29.560 --> 00:23:32.960
M. Holt.

00:23:32.080 --> 00:23:38.460
They get a key, and when they join, it also
runs a workshop.

00:23:38.460 --> 00:23:45.220
it actually runs a statically like built off
that of the workshop uh which has a couple of

00:23:45.220 --> 00:23:51.800
modifications because there's no AWS required
here um so if you compare that to the the one

00:23:51.800 --> 00:23:58.040
that's live that is the actual you know uh it
has a disclaimer about cost and blah blah blah

00:23:56.700 --> 00:23:58.100
Mmm.

00:23:58.040 --> 00:24:03.480
so this one is meant you use your own AWS
account um I don't provide you anything this
one is

00:24:03.480 --> 00:24:09.540
actually, it is a static compile of the box
mode because it's inside the box.

00:24:09.720 --> 00:24:12.120
So then it has a couple of different steps.

00:24:12.300 --> 00:24:14.340
For example, I don't have AWS UI.

00:24:14.640 --> 00:24:23.680
So if I go to the writing constructs, here
there's a whole section about finding the

00:24:23.680 --> 00:24:28.140
function and looking at the, like going to the
monitor in the UI.

00:24:28.140 --> 00:24:34.160
If you go to the same one here, CloudWatch
Logs, it's all CLI commands

00:24:34.160 --> 00:24:36.220
because we don't have a UI in the emulator.

00:24:36.480 --> 00:24:37.200
And that's fine.

00:24:37.460 --> 00:24:38.280
That's fine.

00:24:39.680 --> 00:24:41.260
So that's the box.

00:24:41.800 --> 00:24:42.920
Sorry, did you just repeat that?

00:24:43.000 --> 00:24:44.380
You don't have a what in the emulator?

00:24:45.380 --> 00:24:47.940
AWS emulator doesn't come with the console.

00:24:48.340 --> 00:24:49.720
It doesn't come with the UI.

00:24:50.120 --> 00:24:52.560
There's no AWS UI there for you.

00:24:52.960 --> 00:24:53.580
Oh, okay.

00:24:53.580 --> 00:24:54.880
I understand.

00:24:54.480 --> 00:24:55.040
Yeah.

00:24:56.600 --> 00:24:58.400
That's interesting, actually.

00:25:00.280 --> 00:25:00.820
Yeah.

00:25:01.900 --> 00:25:02.440
Okay.

00:25:02.760 --> 00:25:04.980
You can combine these two.

00:25:05.620 --> 00:25:07.420
How do I add tap?

00:25:07.780 --> 00:25:08.820
How do I split?

00:25:09.060 --> 00:25:09.940
Split view.

00:25:10.080 --> 00:25:10.620
Yeah.

00:25:11.180 --> 00:25:12.320
Oh, man, I'm loving split blue.

00:25:12.460 --> 00:25:16.460
Do you know you can hold down command, option,
and then click,

00:25:16.660 --> 00:25:18.400
and then it opens on the right all the time?

00:25:18.900 --> 00:25:19.660
Okay.

00:25:20.180 --> 00:25:21.320
So here it is.

00:25:20.700 --> 00:25:21.240
Okay.

00:25:22.240 --> 00:25:25.360
You can have the workshop on...

00:25:26.680 --> 00:25:30.360
when i do it in the i i recommend that the
workshop will actually be on the

00:25:30.920 --> 00:25:35.420
on the display and i will walk through it and
anyone who wants to do it at their own pace

00:25:35.420 --> 00:25:38.900
and then ask me questions they can do it at
their own pace um

00:25:38.240 --> 00:25:39.820
Have you done this before?

00:25:40.220 --> 00:25:45.740
It sounds like you're going to be chucking down
a lot of knowledge down, people.

00:25:46.060 --> 00:25:47.520
I don't know if people are for that.

00:25:48.200 --> 00:25:49.560
This is going to be a lot of stuff.

00:25:48.700 --> 00:25:54.380
ready for it yeah no this workshop is really
good have you never done the cdk workshop like

00:25:49.980 --> 00:25:50.140
Yeah.

00:25:54.380 --> 00:25:58.740
AWS actually runs it for companies, and we ran
it in my previous company.

00:25:56.420 --> 00:25:57.120
oh

00:25:58.560 --> 00:25:59.980
no I haven't actually

00:25:59.600 --> 00:26:03.740
Yeah, it's the official...

00:25:59.980 --> 00:26:01.180
I didn't know about this

00:26:02.000 --> 00:26:02.700
so

00:26:02.700 --> 00:26:06.160
so you're basically building on the CDK

00:26:06.160 --> 00:26:06.940
workshop

00:26:07.660 --> 00:26:08.980
well that's interesting

00:26:08.060 --> 00:26:13.940
Yeah, so this CDK workshop is a repository that
uses Hugo in the past.

00:26:13.940 --> 00:26:16.640
It's like seven years old by Elad Ben-Israel
and his team.

00:26:14.820 --> 00:26:15.520
hmm

00:26:16.820 --> 00:26:20.660
And then they archived that repo, and they made
it in the catalog,

00:26:20.880 --> 00:26:21.900
so it's no longer public.

00:26:21.900 --> 00:26:24.720
but I could still access the original markdown

00:26:24.720 --> 00:26:26.360
so I modified it to

00:26:26.360 --> 00:26:28.460
basically run

00:26:27.620 --> 00:26:29.720
oh that's really cool man

00:26:30.140 --> 00:26:30.740
completely

00:26:30.740 --> 00:26:32.720
on Terraform. So if we look

00:26:32.720 --> 00:26:34.100
here, hello

00:26:34.100 --> 00:26:36.660
cdktn, it's exactly

00:26:36.660 --> 00:26:38.660
the same thing. I just didn't

00:26:38.660 --> 00:26:40.640
do the Python one, so it's

00:26:40.640 --> 00:26:42.800
the same except I want it nicer.

00:26:41.120 --> 00:26:46.680
yeah well it's great that you're doing this i i
there's been a lot in the news lately about

00:26:46.680 --> 00:26:52.020
unemployed youth and upskilling people who
don't know about stuff.

00:26:52.340 --> 00:26:56.160
And I sometimes think to myself, like the
videos and the podcasts that I make,

00:26:56.280 --> 00:27:00.720
that's my attempt to basically share knowledge.

00:27:00.940 --> 00:27:06.500
But I really like how you're going out on a
limb and creating this material,

00:27:07.060 --> 00:27:10.880
this structured material, and doing a workshop.

00:27:10.880 --> 00:27:11.780
This is good.

00:27:11.060 --> 00:27:13.860
do you see this little progress tracker at the
top

00:27:13.960 --> 00:27:14.600
Beautiful.

00:27:15.520 --> 00:27:16.460
what this does

00:27:17.100 --> 00:27:18.220
I need to white

00:27:18.220 --> 00:27:21.080
so this is using Cloudflare durable objects

00:27:21.080 --> 00:27:23.400
so when this

00:27:23.400 --> 00:27:25.460
do I have the Cloudflare here

00:27:27.880 --> 00:27:28.600
dashboard

00:27:29.700 --> 00:27:30.920
I guess here

00:27:30.920 --> 00:27:33.240
so this is all Opus

00:27:33.240 --> 00:27:35.600
built for your

00:27:35.600 --> 00:27:37.480
question that you asked me about

00:27:37.480 --> 00:27:39.580
like the Cloudflare MCP and the

00:27:39.580 --> 00:27:41.600
code mode so I build a little

00:27:41.600 --> 00:27:42.120
dashboard

00:27:44.020 --> 00:27:45.660
for data so I

00:27:45.660 --> 00:27:47.700
have data these are my

00:27:47.700 --> 00:27:49.800
deployments so this

00:27:49.800 --> 00:27:57.540
tracker here at the top when you scroll and you
complete you can put yourself on the on the
board

00:27:57.540 --> 00:28:03.560
if you want to right so you can put in your
name your email so this is the sensitive data
that the

00:28:03.560 --> 00:28:08.780
agents should never see that you want code mode
to protect right so that comes to the next part
of

00:28:08.780 --> 00:28:17.160
this this this talk um for you okay your
question um you have sensitive data in
cloudflare and you

00:28:17.160 --> 00:28:24.640
don't want the agents to show those emails in
the transaction log that's my yeah so so people
can

00:28:20.640 --> 00:28:21.520
Oh, yeah.

00:28:24.640 --> 00:28:35.040
if they want enter their name and then they
will be on poc this dashboard oh shit i don't
know the

00:28:35.040 --> 00:28:52.380
secret what's the secret for so no no no i'm
just gonna put up a dashboard of uh the
progress of

00:28:40.680 --> 00:28:45.280
Oh, so not only are you just introducing people
to CDK Terrain,

00:28:52.380 --> 00:28:57.380
everyone during the workshop they don't have to
put their name they don't have to put their
name

00:28:52.940 --> 00:28:55.480
oh that's cool

00:28:57.380 --> 00:29:03.380
but that's gonna give me data for my next
presentation at in singapore which is about

00:29:03.380 --> 00:29:10.620
policies and preventing agents from raising
private data such as email addresses and if

00:29:08.060 --> 00:29:10.860
cool man

00:29:10.620 --> 00:29:15.540
nobody signs up with their email then i will
just generate a bunch of fake data which you
will see now

00:29:15.480 --> 00:29:18.400
Oh my God, you're prolific compared to me.

00:29:18.600 --> 00:29:19.320
This is really good stuff.

00:29:19.440 --> 00:29:22.040
I think you're getting your hopes up a little
bit much.

00:29:20.180 --> 00:29:20.780
why

00:29:22.180 --> 00:29:24.900
I mean, I'm going to be amazed

00:29:23.680 --> 00:29:25.320
i cannot find a secret

00:29:24.900 --> 00:29:28.920
if one person in your workshop gets through all
this material.

00:29:29.340 --> 00:29:32.980
so that's what i'm gonna know because i'm gonna
see

00:29:32.980 --> 00:29:34.940
if people complete it.

00:29:35.580 --> 00:29:36.460
Ha ha ha.

00:29:35.780 --> 00:29:37.560
There's even a validation step.

00:29:37.980 --> 00:29:39.580
So when you say use the hit counter,

00:29:39.780 --> 00:29:40.560
like for example here,

00:29:40.780 --> 00:29:42.440
when you deploy the API gateway,

00:29:43.140 --> 00:29:45.920
when you run the Terraform apply,

00:29:46.340 --> 00:29:51.120
it's going to tell you to put that endpoint in
here.

00:29:51.260 --> 00:29:52.660
And when you click next,

00:29:53.180 --> 00:29:59.620
oh man you're completely gamified can i run
through this as a can i offer to run through

00:29:53.740 --> 00:29:55.880
it's going to call a verification.

00:29:59.620 --> 00:30:06.440
this as a guinea pig yeah

00:30:02.660 --> 00:30:08.940
like i can spin up an instance uh and give you
access yes yes it would be good i suggested

00:30:08.940 --> 00:30:13.160
michael as well to go through it first because
i think the the biggest issue with the workshop

00:30:13.160 --> 00:30:18.120
like that in limited amount of time is usually
it'll be a second set up laptop uh you know is

00:30:18.120 --> 00:30:22.320
the laptop have enough memory and capability to
you know have the node cs version and
everything

00:30:22.320 --> 00:30:28.240
setup and and this kind of like takes that away
right people just get a box they they sign in

00:30:25.880 --> 00:30:27.280
hmm

00:30:28.240 --> 00:30:37.940
they can run the um the cdk in it so i'm i can
copy this i added oh what the hell it's offline

00:30:35.120 --> 00:30:36.340
Okay, let's

00:30:36.340 --> 00:30:40.220
can I share my screen? I wanted to have a

00:30:40.220 --> 00:30:41.680
closer look at

00:30:42.380 --> 00:30:45.180
wait one one one thing where the hell is the

00:30:42.720 --> 00:30:43.740
the emulator.

00:30:44.200 --> 00:30:45.860
One last thing.

00:30:47.140 --> 00:30:54.520
why haiku is so bad oh it found it okay hold on
a minute i need to run one comment

00:30:53.140 --> 00:30:54.100
You're using

00:30:54.100 --> 00:30:55.880
what's haiku again? You're using

00:30:55.880 --> 00:30:57.280
the very basic

00:30:57.500 --> 00:30:58.380
Sonnet, Claude

00:30:59.060 --> 00:31:01.100
you're using haiku

00:31:01.100 --> 00:31:02.300
just because it's cheap

00:31:03.740 --> 00:31:08.560
yeah there it is yeah i just want to see
sometimes you just want to know if like the
cheapest model

00:31:08.560 --> 00:31:12.280
can do what it needs to do like does your
repository have enough so you see obviously

00:31:12.280 --> 00:31:16.760
these are fake email addresses that Opus was
using when it was trying to verify that
everything

00:31:16.760 --> 00:31:25.100
works as expected. So here's my trial. So you
can see wherever it actually verified, it
recorded

00:31:25.100 --> 00:31:31.000
that endpoint. Again, this is, I don't know why
it needs to keep it, but maybe that's part of
the

00:31:31.000 --> 00:31:35.040
sensitive data demo, right? That the agents
would not see these live endpoints or whatever.

00:31:35.740 --> 00:31:41.740
That's just a little scenario that I told Opus
to come up with. Yeah. And another cool thing,

00:31:37.500 --> 00:31:40.040
okay well

00:31:41.740 --> 00:31:49.340
hold on when you click define here when then it
goes into like verify it like hit it every curl

00:31:49.340 --> 00:31:53.420
command that you get here in the workshop is
then showing that endpoint that you created

00:31:54.020 --> 00:31:59.700
back then you see that endpoint is the one that
you so you can just copy it and run it

00:31:57.460 --> 00:32:02.920
oh so the the docs live update and

00:32:02.920 --> 00:32:04.780
yeah

00:32:03.340 --> 00:32:07.700
yeah i mean i had a little bit of fun anyway
that's the last thing so that's all i did for

00:32:07.700 --> 00:32:13.020
the workshop that's going to happen in two
weeks time and it literally just took like two
days to

00:32:10.360 --> 00:32:30.920
Cool. So if anyone's listening to this who is
around Singapore, you should go and meet
Vincent and do his demo and learn a lot about
how to stand up infrastructure using the
cutting edge tools, AI and CDK and all the rest
of it.

00:32:13.020 --> 00:32:14.860
build all of this it's insane

00:32:32.980 --> 00:32:36.120
And speaking of, can I just replace the share
here?

00:32:34.800 --> 00:32:36.980
now let me stop yeah yes

00:32:36.740 --> 00:32:40.660
I just wanted to have a closer look at the AWS
emulator.

00:32:40.900 --> 00:32:43.200
So can you see my screen?

00:32:43.620 --> 00:32:46.720
It's like 64,000 files of Rust.

00:32:49.000 --> 00:32:50.520
Yeah, I can see the screen.

00:32:49.340 --> 00:32:51.880
Quite a lot of markdown.

00:32:52.060 --> 00:32:53.720
I don't know what the hell that markdown is.

00:32:54.840 --> 00:32:55.260
Steering.

00:32:59.180 --> 00:32:59.740
Operations.

00:33:02.720 --> 00:33:03.860
what I'm guessing

00:33:03.860 --> 00:33:06.300
is that that's context for the

00:33:06.300 --> 00:33:07.660
agents to build

00:33:08.480 --> 00:33:10.340
maybe there's the Smithy like

00:33:11.320 --> 00:33:13.080
let's have a look at

00:33:12.260 --> 00:33:12.900
docs

00:33:12.900 --> 00:33:14.280
I opened one of them

00:33:13.080 --> 00:33:14.140
Lambda or something

00:33:15.180 --> 00:33:17.500
but that's the coolest thing about my
experiment

00:33:17.500 --> 00:33:18.460
you know building this

00:33:18.460 --> 00:33:20.540
you know this

00:33:18.860 --> 00:33:20.760
oh it's like issues and

00:33:21.280 --> 00:33:22.460
building this web

00:33:22.460 --> 00:33:25.200
this web box

00:33:25.200 --> 00:33:26.380
for people to attend.

00:33:26.640 --> 00:33:28.420
I never even looked at the code.

00:33:29.080 --> 00:33:31.580
And I just, even Opus just went,

00:33:31.760 --> 00:33:33.360
okay, step one of the workshop

00:33:33.360 --> 00:33:35.720
tells me to run Terraform Apply.

00:33:35.980 --> 00:33:37.780
So it stands up the emulator,

00:33:38.160 --> 00:33:39.120
runs Terraform Apply,

00:33:39.320 --> 00:33:41.200
and then does the next step.

00:33:41.340 --> 00:33:43.280
So it even built like a whole bash script

00:33:43.280 --> 00:33:46.060
of part one, part two, part three.

00:33:46.240 --> 00:33:47.180
It runs them in line

00:33:47.180 --> 00:33:50.160
and then validates if the emulator can do it.

00:33:50.020 --> 00:33:55.540
um yeah so you have a whole end-to-end
verification thing

00:33:50.440 --> 00:33:51.280
It's insane.

00:33:56.100 --> 00:34:02.040
Yeah, and then it told me, like, okay, so the
emulator does not, like, it runs QuickJS.

00:34:02.200 --> 00:34:06.880
So when you build a Lambda, a Node.js Lambda,
it actually runs a QuickJS engine.

00:34:06.880 --> 00:34:09.360
It's not the real Node runtime that you run in
AWS.

00:34:09.720 --> 00:34:19.460
So it doesn't support some of the, like, the
Lambda handler, for example, that is currently
the official one on my website, doesn't work
with QuickJS.

00:34:19.460 --> 00:34:27.260
So the workshop version of the website, the
workshop, is using the JavaScript that works
with QuickJS.

00:34:28.120 --> 00:34:35.240
And then the second thing is the QuickJS can
invoke, I don't know, it can invoke like many
services.

00:34:35.240 --> 00:34:39.720
So you can invoke from your Lambda, you can
invoke DynamoDB to write data into it.

00:34:39.840 --> 00:34:41.600
You can do this and do that.

00:34:41.700 --> 00:34:43.760
But you cannot invoke another Lambda.

00:34:43.760 --> 00:34:50.540
Like that's blocked in the emulator on the
version that I had last week on Thursday.

00:34:50.940 --> 00:35:01.380
But Fable just highlighted part three, step
four fails because the emulator doesn't support
one Lambda invoking another Lambda.

00:35:01.740 --> 00:35:03.180
And then it built a little patch.

00:35:03.340 --> 00:35:09.660
It just patched that one emulator feature and
just made it so that it can invoke a Lambda.

00:35:10.360 --> 00:35:13.260
Another thing that the emulator didn't do is
IAM policies.

00:35:13.760 --> 00:35:20.440
So part of the workshop is to actually create a
DynamoDB table to record data.

00:35:20.640 --> 00:35:23.820
So Lambda has to get access to write into
DynamoDB.

00:35:24.060 --> 00:35:26.580
And this is a very important step of the
workshop

00:35:26.580 --> 00:35:31.040
because it teaches you how to grant permissions
across different services in AWS

00:35:31.040 --> 00:35:34.360
and how easy it is with AWS CDK because you
just take your table

00:35:34.360 --> 00:35:38.420
and then you grant your Lambda to have access
to that table to write or read only.

00:35:38.600 --> 00:35:40.660
And this is a very important part of the
workshop.

00:35:40.660 --> 00:35:43.380
But when the emulator doesn't support IAM
policies,

00:35:43.760 --> 00:35:46.220
it just gets 200 OK on the first step,

00:35:46.360 --> 00:35:48.800
even when you didn't grant any permissions.

00:35:49.160 --> 00:35:53.320
So again, Fable built the IAM policy component

00:35:53.320 --> 00:35:58.040
in such a way that if you don't grant the table
permissions

00:35:58.040 --> 00:36:00.400
to be written to from the Lambda,

00:36:00.660 --> 00:36:02.220
it will actually fail the workshops

00:36:02.220 --> 00:36:03.520
and you have to go through that step.

00:36:03.920 --> 00:36:06.760
Again, small little patches that Fable did,

00:36:06.960 --> 00:36:09.940
then build a new binary of the emulator,

00:36:10.340 --> 00:36:11.800
put it into the container,

00:36:12.040 --> 00:36:12.960
build a container image,

00:36:12.960 --> 00:36:15.000
spun up and do the whole test.

00:36:15.760 --> 00:36:18.340
Fable did all of that in just two, three hours

00:36:18.340 --> 00:36:19.680
to build the whole workshop

00:36:19.680 --> 00:36:21.640
with the custom emulator and everything.

00:36:21.980 --> 00:36:22.720
It's insane.

00:36:22.920 --> 00:36:26.160
That is insane, sorry that's very hard for me
to

00:36:25.780 --> 00:36:29.040
I mean, it's a very, very well-defined problem.

00:36:29.320 --> 00:36:31.380
It's very easy for Fable to define

00:36:32.000 --> 00:36:33.800
what is the steps that need to pass.

00:36:33.960 --> 00:36:37.600
It's very easy for Fable to hand that off to
Opus to build

00:36:37.600 --> 00:36:39.460
and then verify that the result works.

00:36:39.640 --> 00:36:42.440
So when you have a very well-defined task,

00:36:42.960 --> 00:36:44.900
You can have it completely autonomously

00:36:44.900 --> 00:36:48.540
be done by the agents, depending on how many
tokens you have.

00:36:47.640 --> 00:36:49.960
Yeah, I've got to try this emulator.

00:36:50.240 --> 00:36:50.680
I've got to try.

00:36:51.500 --> 00:36:54.820
I was just noticed that there is an IAM
directory here, but.

00:36:56.420 --> 00:36:58.760
Well, maybe the version that I tried on Terzid

00:36:59.720 --> 00:37:01.380
is already outdated, right?

00:37:01.520 --> 00:37:03.620
Maybe if I try a new version of the emulator,

00:37:04.320 --> 00:37:06.380
it will already do some of this stuff.

00:37:07.840 --> 00:37:08.720
I'm also

00:37:08.280 --> 00:37:09.040
What happened?

00:37:08.720 --> 00:37:10.280
I wonder how

00:37:09.920 --> 00:37:10.780
Oh, shit.

00:37:10.280 --> 00:37:12.060
I wonder how it

00:37:15.080 --> 00:37:17.420
it just like smithy changes

00:37:15.640 --> 00:37:16.720
So guess what?

00:37:17.420 --> 00:37:18.360
and stuff like that

00:37:18.400 --> 00:37:19.660
You know what happened?

00:37:20.540 --> 00:37:21.220
what

00:37:20.960 --> 00:37:23.260
So my whole workshop is on spot instances,

00:37:23.700 --> 00:37:25.400
which is like 60% discount.

00:37:25.700 --> 00:37:27.200
The instance got terminated.

00:37:28.520 --> 00:37:32.480
I guess I'll pay $10 instead of $3.

00:37:34.320 --> 00:37:35.600
issue ID

00:37:38.020 --> 00:37:40.100
just seeing the engineering

00:37:40.100 --> 00:37:41.560
behind this is quite interesting

00:37:41.560 --> 00:37:42.140
it's like

00:37:42.140 --> 00:37:43.400
it's all

00:37:43.400 --> 00:37:45.820
no one's name right

00:37:45.820 --> 00:37:47.160
it's this particular

00:37:47.160 --> 00:37:49.060
issue fixer agent

00:37:50.100 --> 00:37:55.660
So when AWS announced this emulator, they said
it's been built by agents running for I don't
know how long.

00:37:56.220 --> 00:38:02.660
They said like thousands of agents have been
building it over a certain period of time, and
they don't accept any PRs.

00:38:02.800 --> 00:38:04.900
I think it makes total sense.

00:38:04.900 --> 00:38:08.240
They have just an agentic process running
along,

00:38:08.540 --> 00:38:12.940
just slowly ticking off and verifying and
building everything.

00:38:14.540 --> 00:38:18.220
Unfortunately, for one of the big projects that
I'm working on,

00:38:15.420 --> 00:38:17.380
but I

00:38:18.620 --> 00:38:20.060
the gaps are too big.

00:38:20.160 --> 00:38:22.280
I can't use it for a lot of stuff.

00:38:21.020 --> 00:38:23.960
Do you see the spaces between the commits?

00:38:24.260 --> 00:38:25.440
It's kind of insane, right?

00:38:25.580 --> 00:38:28.880
It's like two minutes between each commit here.

00:38:31.120 --> 00:38:36.720
i wonder if they if they're still using git or
they are using like if they used already

00:38:36.720 --> 00:38:40.120
another cicd mechanisms and version control

00:38:38.340 --> 00:38:41.340
It's going 24 hours a day.

00:38:42.380 --> 00:38:43.880
It almost doesn't make sense.

00:38:44.020 --> 00:38:44.400
Look at this.

00:38:44.520 --> 00:38:44.920
It's like two.

00:38:45.300 --> 00:38:46.740
There's a two-minute gap.

00:38:46.740 --> 00:38:49.720
I mean, I'm exaggerating, but it's about that.

00:38:50.880 --> 00:38:52.740
Between every commit here, it's...

00:38:52.400 --> 00:38:55.880
I can show you some of my repositories

00:38:55.880 --> 00:38:57.900
with one or two minute commits

00:38:57.900 --> 00:39:00.440
and 100 commits built by agents.

00:38:59.480 --> 00:39:00.800
Oh my God, I can't...

00:39:00.800 --> 00:39:04.500
I've got one of those sort of like feelings of
dread almost.

00:39:04.680 --> 00:39:06.780
Like we're just living in a new fucking era.

00:39:07.300 --> 00:39:08.740
When shit like this is...

00:39:08.740 --> 00:39:10.620
When you look at a commit history...

00:39:10.620 --> 00:39:11.740
Like I remember...

00:39:12.300 --> 00:39:15.520
You keep telling about all of these tools

00:39:15.520 --> 00:39:16.600
to control commits,

00:39:16.620 --> 00:39:19.680
and I keep telling you your metrics are all
wrong

00:39:19.680 --> 00:39:21.500
for this new, like,

00:39:21.600 --> 00:39:23.780
agentic era. Everyone is

00:39:23.780 --> 00:39:25.820
moving to the higher layer, the intent

00:39:25.820 --> 00:39:27.720
layer. You cannot verify or

00:39:27.720 --> 00:39:29.660
review code. You have to

00:39:29.660 --> 00:39:31.680
define, like I said, you have to have

00:39:29.760 --> 00:39:31.700
Yeah, I know that.

00:39:31.680 --> 00:39:33.320
a very well-defined problem.

00:39:32.560 --> 00:39:33.480
I know that, Vincent.

00:39:33.660 --> 00:39:36.880
It's just that I just can't help but think that
it wasn't that long ago

00:39:36.880 --> 00:39:39.700
when I was sitting with my team,

00:39:40.400 --> 00:39:46.840
going through the commits that landed on Trunk
in the last day

00:39:46.840 --> 00:39:52.140
and making sure that the team was aware of what
was happening

00:39:52.140 --> 00:39:57.820
with the code base, including myself, as a
lead.

00:39:56.140 --> 00:39:57.840
but that you don't relinquish that

00:39:58.360 --> 00:40:01.800
this is a big problem with the way that a lot
of people are adopting AI

00:40:01.800 --> 00:40:05.540
and vibe coding, they think that you completely
do not control anything

00:40:05.540 --> 00:40:09.840
but you control very much the verification
process

00:40:09.840 --> 00:40:10.620
and everything, right?

00:40:11.800 --> 00:40:18.200
yeah yeah yeah i mean but it's still come on
man i just wanted to acknowledge that like

00:40:18.200 --> 00:40:24.000
the world has changed like freaking big time
when when you see things like this

00:40:26.200 --> 00:40:35.280
it's insane how what is the extent of the good
history here so it's it all kicked off in
october

00:40:35.860 --> 00:40:37.200
It's not even that long.

00:40:39.040 --> 00:40:40.060
It's a folder.

00:40:44.060 --> 00:40:45.700
Yeah, I've got to study this.

00:40:45.900 --> 00:40:47.100
I mean, this is fascinating.

00:40:47.560 --> 00:40:48.620
I mean, this is a game changer.

00:40:50.320 --> 00:40:54.440
And is it fair to say that AWS is doing pretty
well?

00:40:54.640 --> 00:40:59.980
I wonder how the other clouds are doing in the
face of AI modernization

00:40:59.980 --> 00:41:01.140
or what do we call it?

00:41:01.480 --> 00:41:06.800
I think this is just basic everyone.

00:41:02.380 --> 00:41:03.260
AI-ification?

00:41:07.100 --> 00:41:10.140
I mean, in terms of when you say cloud
providers,

00:41:10.460 --> 00:41:12.700
I think everyone has to adopt this, right?

00:41:13.000 --> 00:41:15.280
You see, Microsoft is adopting it very well.

00:41:13.780 --> 00:41:18.960
sure but like i'm you know since we're a
podcast here we're quick

00:41:17.660 --> 00:41:19.920
Microsoft did a really great job

00:41:19.920 --> 00:41:20.680
adopting it

00:41:21.600 --> 00:41:23.400
that's sarcasm i'm sure

00:41:24.340 --> 00:41:26.040
yeah because Microsoft

00:41:26.040 --> 00:41:27.520
didn't really go down for like

00:41:27.520 --> 00:41:29.640
24 hours at all

00:41:29.880 --> 00:41:35.680
what is the google gcp story because everyone
seems to to bash

00:41:38.280 --> 00:41:40.460
Google for having crappy models.

00:41:40.200 --> 00:41:42.000
have you forgotten

00:41:42.000 --> 00:41:43.760
that Microsoft 365

00:41:43.760 --> 00:41:45.920
65 was down for 24 hours.

00:41:45.700 --> 00:41:47.040
I'm talking about Google now.

00:41:47.580 --> 00:41:48.520
Everyone's bashing Google.

00:41:47.640 --> 00:41:48.200
Google.

00:41:48.760 --> 00:41:49.000
Yes.

00:41:50.640 --> 00:41:51.760
Why we...

00:41:51.500 --> 00:41:52.700
Googly gun drops.

00:41:51.760 --> 00:41:54.120
So everyone likes to hate on Google

00:41:54.120 --> 00:41:55.520
because they have these...

00:41:55.520 --> 00:41:56.740
They were TensorFlow.

00:41:57.140 --> 00:42:00.960
They had these custom hardware built for LLMs.

00:42:01.120 --> 00:42:03.400
And then Gemini was like blowing everyone out
of the water

00:42:03.400 --> 00:42:04.460
with the 1 million context.

00:42:04.880 --> 00:42:07.580
But they didn't really recover from that,
right?

00:42:07.680 --> 00:42:11.080
They're still kind of struggling, it seems.

00:42:09.300 --> 00:42:12.200
Well, yeah, that's the story.

00:42:13.760 --> 00:42:19.600
They open sourced their Goldfish versus
Elephant practice.

00:42:19.600 --> 00:42:23.520
That was a blog post that became, they made it
a framework.

00:42:24.340 --> 00:42:25.280
You know this one?

00:42:24.860 --> 00:42:25.820
Gold? No.

00:42:26.400 --> 00:42:27.200
You don't want to share that?

00:42:28.540 --> 00:42:33.040
Yeah, Google Goldfish Elephant.

00:42:33.700 --> 00:42:40.980
So there was this Google engineer that shared
months ago.

00:42:41.440 --> 00:42:43.060
Where is the share?

00:42:43.900 --> 00:42:47.380
We talked about it on the podcast as well.

00:42:48.840 --> 00:42:53.040
Because to be honest, he's not introducing
anything revolutionary.

00:42:53.280 --> 00:42:58.160
To me, this is just exactly a repeat of what
Steven Yege wrote,

00:42:58.320 --> 00:43:01.380
beats are for Blobfish.

00:43:03.280 --> 00:43:06.860
There's a couple of things that I really liked
about this.

00:43:07.080 --> 00:43:07.980
It's a good summary.

00:43:08.720 --> 00:43:10.640
So Beads are for Blobfish is very clear.

00:43:10.640 --> 00:43:17.780
I like the comparison that an agent is born
with zero context.

00:43:17.780 --> 00:43:22.140
You fill up the context, and very quickly the
agent is useless

00:43:22.140 --> 00:43:24.440
because the context has filled.

00:43:24.720 --> 00:43:29.960
So beads is a way to keep agent context
tracked.

00:43:31.760 --> 00:43:34.860
Do you still use beads, by the way, or you
don't feel the need?

00:43:37.040 --> 00:43:40.480
Honestly, I need something like that.

00:43:40.640 --> 00:43:43.960
With Atlassian MCP, the V2 one, it's better.

00:43:44.220 --> 00:43:45.620
But anyway, let's get back on points.

00:43:46.180 --> 00:43:50.680
So then you have this one from Elephant and
Goldfish model,

00:43:51.900 --> 00:43:54.440
which there was an original Medium article.

00:43:55.060 --> 00:43:58.760
Dave Ranson did that article, I think, in
March, or was it?

00:44:00.180 --> 00:44:05.400
2026, where he talks about April 2026.

00:44:05.400 --> 00:44:07.960
and he shared about some of the practices

00:44:07.960 --> 00:44:12.620
that within the engineering teams within Google
were following.

00:44:12.840 --> 00:44:16.060
And they're basically grill me type of
practices, right?

00:44:16.160 --> 00:44:23.340
You ask the agent to ask questions for you to
fine tune onto your...

00:44:21.860 --> 00:44:23.260
Ask

00:44:23.340 --> 00:44:26.920
Find the alignment between the user intention

00:44:26.920 --> 00:44:28.880
and the agent's understanding, right?

00:44:28.960 --> 00:44:30.320
Because you can say one word

00:44:30.320 --> 00:44:32.940
and it can be interpreted into many, many
different ways.

00:44:35.020 --> 00:44:36.360
so there was

00:44:35.060 --> 00:44:41.020
I mean, I like to use a prompt like, ask me how
I want it.

00:44:41.840 --> 00:44:43.080
Ask me how I want it.

00:44:45.000 --> 00:44:49.400
yeah so he has some really good examples and
prompts in here which all honestly like

00:44:49.400 --> 00:44:55.500
Matt Pocock's grill-me skill with domain driven
design you know is the same except it's a
little

00:44:55.500 --> 00:45:02.540
bit more useful because people you know kind of
really like the ideas of domain driven design

00:45:02.540 --> 00:45:05.500
and bounded context and Matt Pocock's skills
really help there.

00:45:05.680 --> 00:45:07.360
So it's very similar.

00:45:07.540 --> 00:45:10.900
You ask questions, you find alignment, you find
a shared language.

00:45:11.480 --> 00:45:15.380
And that's your, if I understand correctly,
this is like your elephant, right?

00:45:15.440 --> 00:45:18.900
You build your whole, your overall view, no
code, right?

00:45:18.960 --> 00:45:19.680
There's no code.

00:45:21.020 --> 00:45:22.840
You're defining a shared language.

00:45:23.060 --> 00:45:26.480
And then you have your goldfish, right?

00:45:23.260 --> 00:45:23.840
Hmm.

00:45:26.860 --> 00:45:30.900
They come along and they pick up pieces of work
in that shared language

00:45:29.180 --> 00:45:29.860
Yeah.

00:45:30.900 --> 00:45:31.800
and they deliver.

00:45:31.380 --> 00:45:32.560
Okay, thanks.

00:45:32.860 --> 00:45:34.580
I'm beginning to understand now.

00:45:34.760 --> 00:45:35.620
Thanks for the explanation.

00:45:36.420 --> 00:45:41.240
yeah and then they made it they made it in
october they made it like in october 22

00:45:41.240 --> 00:45:46.700
they said here's the the thing as a as a
framework for you to use so

00:45:45.340 --> 00:45:47.660
So what does the elephant look like?

00:45:47.660 --> 00:45:50.180
A bunch of intentions or...

00:45:50.180 --> 00:45:52.140
What is...

00:45:50.600 --> 00:45:55.240
yeah i'm actually curious because we we
actually we use Google Conductor as well right

00:45:57.180 --> 00:46:02.080
so i think it's like an evolution of of uh of
conductor or a different approach

00:46:02.080 --> 00:46:05.380
which it's like there's these different schools
of thoughts, right?

00:46:05.400 --> 00:46:08.520
You have the spec-driven development, which
people say is like waterfall.

00:46:08.760 --> 00:46:12.320
You have the, you know, Conductor, Spec Kit.

00:46:12.460 --> 00:46:15.020
All of those are spec-driven development.

00:46:15.360 --> 00:46:17.300
Then you have the get shit done.

00:46:17.720 --> 00:46:20.960
You know, I want to work closely with the agent
on small little problems

00:46:20.960 --> 00:46:27.760
and, you know, iteratively develop them instead
of like more waterfall type of processes.

00:46:27.760 --> 00:46:32.440
as conductor was their attempt in this spec-
driven development

00:46:32.440 --> 00:46:32.920
world.

00:46:33.240 --> 00:46:36.140
And I feel like I never used Get Shit Done.

00:46:36.280 --> 00:46:37.780
I have friends who really like it.

00:46:38.000 --> 00:46:41.700
I feel like maybe the elephant goldfish model

00:46:41.700 --> 00:46:42.540
is similar to that.

00:46:42.700 --> 00:46:44.800
Again, I'm out of that here.

00:46:44.800 --> 00:46:45.580
I actually don't know.

00:46:47.940 --> 00:46:50.300
I haven't tried these because I'm still

00:46:50.300 --> 00:46:53.880
using a Spec Kit-like variation.

00:46:53.260 --> 00:47:05.580
So speaking of skills and approaches and
techniques, I recently tried P-Stack from
Potato on one of my apps I can maybe share.

00:47:06.300 --> 00:47:10.040
yeah i i use a lot of her skills but like i
said i have my own

00:47:10.040 --> 00:47:14.000
so i i don't i just take what i like from other
people

00:47:12.600 --> 00:47:16.380
Oh, yeah, I used her verification skill.

00:47:16.840 --> 00:47:17.860
yeah i use the same

00:47:18.660 --> 00:47:21.720
it was a bit weird because I had my end-to-end
tests already

00:47:23.200 --> 00:47:27.340
but it just seems like I guess now I have a
skill

00:47:26.180 --> 00:47:28.860
I don't know.

00:47:27.340 --> 00:47:29.820
so it's a little bit easier, it's a little bit
more Asian friendly

00:47:31.720 --> 00:47:34.680
and then the other thing I thought that was
really interesting

00:47:35.400 --> 00:47:36.920
am I sharing my screen?

00:47:37.410 --> 00:47:41.620
Yes, I can see the verify LTA bus.

00:47:40.100 --> 00:47:43.000
so what was interesting I thought was

00:47:43.000 --> 00:47:49.440
was the fact that it managed to discover the
features

00:47:49.440 --> 00:47:55.400
of my Singapore bus app, like the fact that it
auto-refreshes,

00:47:55.500 --> 00:47:57.420
the fact that it shows you the closest stop.

00:47:58.140 --> 00:48:02.160
yeah what is this opus did this like custom

00:48:01.240 --> 00:48:05.840
Opus 5.5 and the verification, create
verification skill.

00:48:04.940 --> 00:48:06.620
yeah nice

00:48:06.060 --> 00:48:07.180
And I thought that was really interesting.

00:48:07.180 --> 00:48:13.140
And then you can also ask it to create a
feature map.

00:48:09.920 --> 00:48:12.420
are those specs Playwright

00:48:13.520 --> 00:48:14.040
Sorry?

00:48:14.560 --> 00:48:17.380
are those spec.js Playwright specs

00:48:17.380 --> 00:48:19.020
or what are those

00:48:19.020 --> 00:48:21.360
they are hand to hand drive

00:48:19.520 --> 00:48:21.140
Yeah, yeah, that Playwright.

00:48:21.360 --> 00:48:23.460
yeah it's what I told you

00:48:21.460 --> 00:48:22.060
It's Playwright.

00:48:23.460 --> 00:48:25.100
how people should use Playwright right

00:48:25.100 --> 00:48:27.280
no stupid MCP you write proper

00:48:27.280 --> 00:48:29.120
specs and you run them and the models get

00:48:29.120 --> 00:48:30.220
you know nice feedback

00:48:29.260 --> 00:48:29.940
Yeah, yeah.

00:48:30.060 --> 00:48:34.780
I mean, I was doing that before.

00:48:30.940 --> 00:48:32.560
like code mode for tests

00:48:34.780 --> 00:48:42.160
like in fact it didn't seem worryingly it
didn't seem to pick up on some of my old specs
so whatever

00:48:41.000 --> 00:48:41.720
uh huh

00:48:43.640 --> 00:48:48.820
so but but maybe we should talk about the ideas
behind these things rather than like

00:48:43.840 --> 00:48:44.380
but

00:48:48.820 --> 00:48:56.340
i mean it's it's because honestly Lauren Tan
Matt Pocock they both have a lot of overlapping
ideas

00:48:56.340 --> 00:49:01.600
they have some really good presentations um
Matt Pocock has a course that my colleague did

00:48:56.480 --> 00:48:57.020
yeah

00:48:57.020 --> 00:48:57.460
yeah

00:49:01.600 --> 00:49:10.580
and um it's interesting but but i really like
uh Lauren Tan's presentation where she
highlights

00:49:11.560 --> 00:49:13.320
if you follow her on Twitter

00:49:13.320 --> 00:49:15.500
you basically saw everything

00:49:15.500 --> 00:49:17.660
that she presented but in the presentation

00:49:17.660 --> 00:49:19.600
she really like organizes

00:49:19.600 --> 00:49:20.680
it in a very nice

00:49:19.800 --> 00:49:20.800
Yeah, yeah.

00:49:20.680 --> 00:49:23.680
when I see it I'm like shit I wish I was

00:49:23.680 --> 00:49:25.720
presenting like that because it ties

00:49:25.720 --> 00:49:27.560
everything neatly together and how everything
is

00:49:25.900 --> 00:49:26.600
Yeah.

00:49:26.600 --> 00:49:33.840
Yeah, her AI engineer video was great.

00:49:27.560 --> 00:49:28.000
connected

00:49:32.260 --> 00:49:33.360
talk is really good

00:49:33.840 --> 00:49:41.440
and and yeah like i um yeah i just wanted to
quickly show the the artifact where it sort of

00:49:34.600 --> 00:49:35.120
yeah

00:49:41.440 --> 00:49:47.100
like you know shows shows the the feature map a
little bit better hopefully i'm showing the
right

00:49:42.680 --> 00:49:43.520
the UI.

00:49:47.100 --> 00:49:58.740
screen okay uh yeah Potato, Pocock they're
they're great and it and the fact that they're
so popular

00:49:47.720 --> 00:49:48.860
Yes, it's the

00:49:48.860 --> 00:49:49.700
Itobus feature map.

00:49:58.740 --> 00:50:04.240
that means that they're actually filling a huge
demand, right?

00:50:04.060 --> 00:50:12.340
but they have different angles like Matt Pocock
is a great explainer he has beautiful courses

00:50:04.760 --> 00:50:06.980
I mean, yeah.

00:50:12.340 --> 00:50:18.200
he visualizes things and explains them really
well but i feel and this is my personal opinion

00:50:18.680 --> 00:50:22.060
that he doesn't have the hands-on large project
experience

00:50:22.060 --> 00:50:23.680
that then you get from Lauren Tan.

00:50:22.940 --> 00:50:24.340
Mmm

00:50:23.840 --> 00:50:26.340
And Lauren Tan is great at explaining as well,

00:50:24.540 --> 00:50:25.240
Mmm

00:50:26.380 --> 00:50:28.720
but she has all of these hands-on skills.

00:50:28.620 --> 00:50:29.380
yeah

00:50:28.920 --> 00:50:32.320
She's like, yeah, but it shows that she
actually runs

00:50:32.900 --> 00:50:35.600
3,000 PRs merged in a month,

00:50:34.300 --> 00:50:34.880
yeah

00:50:35.600 --> 00:50:36.700
even if it's a stupid metric,

00:50:36.720 --> 00:50:37.520
and she admits that.

00:50:37.580 --> 00:50:39.560
She says, it just visualizes a little bit

00:50:38.880 --> 00:50:42.440
Vincent your explanation of the differences
there

00:50:39.560 --> 00:50:40.240
what we are doing.

00:50:42.440 --> 00:50:44.040
was 100%

00:50:43.960 --> 00:50:45.200
Exactly how you feel, right?

00:50:45.000 --> 00:50:45.760
exactly

00:50:45.760 --> 00:50:47.940
so yeah Lauren is

00:50:46.660 --> 00:50:47.080
Yeah.

00:50:47.940 --> 00:50:49.240
doing great

00:50:49.240 --> 00:50:52.680
and you know how it is

00:50:52.680 --> 00:50:54.780
it's like you and I are bashing

00:50:54.780 --> 00:51:02.540
away at AI. And when I use the verification
skill, I go, yeah, but it has value. But I'm at
the same

00:51:02.540 --> 00:51:13.120
time like, yeah, I knew that. But yeah, I mean,
yeah, exactly. Yeah, that said, my depreciating

00:51:06.180 --> 00:51:07.400
they make it very applicable

00:51:07.800 --> 00:51:09.000
they make it very applicable

00:51:13.120 --> 00:51:22.000
comments aside, it's very good. And like the
other things that she's doing around

00:51:22.000 --> 00:51:24.580
formal verification.

00:51:24.920 --> 00:51:27.720
I think we talked about that last pod.

00:51:27.720 --> 00:51:32.060
I use her state machine graph and state machine

00:51:32.060 --> 00:51:41.840
like um Quint modelling and uh what's the other
one TLA+ yes i run them i literally had an
agent

00:51:36.040 --> 00:51:37.100
TLA+.

00:51:41.840 --> 00:51:46.660
last night doing a full because i had the race
condition um and then i want to make sure that

00:51:46.440 --> 00:51:47.020
Ooh.

00:51:46.660 --> 00:51:53.380
the fix meets like the you know doesn't
actually solve the race condition in a in a
proper way

00:51:53.380 --> 00:52:00.100
so i usually the way it ends up with me i don't
take on her stack but i ask opus uh here's a
tweet

00:52:00.100 --> 00:52:01.840
where she talks about state graph

00:52:01.840 --> 00:52:04.620
and showing all of the ways that the state
machine

00:52:04.620 --> 00:52:08.540
can end up in an invalid state or a faulty
state.

00:52:08.900 --> 00:52:13.200
So I just point Opus at that tweet

00:52:13.200 --> 00:52:15.060
and also the repo where it lives,

00:52:15.240 --> 00:52:17.440
and I tell it, clone it locally, find the
skill,

00:52:17.580 --> 00:52:18.720
and see how it applies.

00:52:19.540 --> 00:52:20.700
It's all on the cursor.

00:52:21.980 --> 00:52:22.420
the tweet

00:52:23.520 --> 00:52:24.280
Usually, yeah.

00:52:24.460 --> 00:52:26.960
Usually I ask Opus, like,

00:52:27.800 --> 00:52:30.100
find how I can apply it into this particular

00:52:30.100 --> 00:52:31.360
problem that I have

00:52:31.980 --> 00:52:34.000
and then it picks up the skill and then I

00:52:34.000 --> 00:52:34.980
inline it or

00:52:35.820 --> 00:52:37.620
it's a separate agent usually

00:52:37.780 --> 00:52:39.780
there's so much good stuff here actually

00:52:39.780 --> 00:52:41.740
I like the no comments one

00:52:41.740 --> 00:52:42.480
because I do

00:52:42.080 --> 00:52:43.440
I don't know about that one

00:52:44.640 --> 00:52:45.200
she

00:52:45.200 --> 00:52:47.900
could explain it better than I can though

00:52:47.900 --> 00:52:48.740
you know sometimes

00:52:49.340 --> 00:52:50.480
I can tell you

00:52:49.540 --> 00:52:50.380
AI just

00:52:50.480 --> 00:52:51.940
where I don't agree with it

00:52:51.660 --> 00:52:54.600
AI adds a lot of comments which can

00:52:55.540 --> 00:52:58.560
cultify the project. So this one is about

00:52:59.440 --> 00:53:01.420
removing those comments that sort of

00:53:02.500 --> 00:53:05.600
make no sense anymore, which is

00:53:05.380 --> 00:53:12.960
well i can tell you i agree a lot with what she
says but and i and one of the biggest issues
that

00:53:05.600 --> 00:53:06.260
...

00:53:12.960 --> 00:53:21.220
biggest concerns that i have is you know how
agents are working on independent pieces of
work

00:53:21.220 --> 00:53:27.280
right so sometimes and this is a very common
pattern that they do is they have to put a mock

00:53:27.280 --> 00:53:33.460
somewhere or they will put like a little proxy
or like a function um like a placeholder that

00:53:33.460 --> 00:53:37.480
does like actually mocks the function doesn't
do anything like not implemented error which

00:53:35.720 --> 00:53:37.700
Yeah, I work around.

00:53:38.480 --> 00:53:44.080
not a workaround um of a seam that you don't
have the other side of the implementation or

00:53:44.080 --> 00:53:48.860
somebody else is working on it you have a clear
contract and you may have an api mock between
them

00:53:44.380 --> 00:53:45.760
Okay, okay, yeah.

00:53:49.040 --> 00:53:50.100
Right, right, right.

00:53:50.200 --> 00:53:54.960
right right so it's not a workaround it's it's
it's a placeholder it's a temporary

00:53:54.960 --> 00:54:02.680
like function and i'm always afraid that they
they will be lost and so one of the biggest
things that

00:54:02.680 --> 00:54:08.460
she says is if you give the agents permission
to put to do's then another agent comes along

00:54:08.460 --> 00:54:14.460
sees the to do and follows that pattern which
obviously absolutely you know to do's get out

00:54:14.460 --> 00:54:23.580
of hand. But I want to have a way for agents to
really know that, like, okay, I can put a to-do

00:54:15.020 --> 00:54:16.100
Mm-hmm.

00:54:23.580 --> 00:54:29.280
here. And it's very tricky, right? The to-do
should be... So this is where I think comments
still need

00:54:29.280 --> 00:54:43.920
to be there. Yeah, that's quite a lot. I mean,
again, a lot of these skills... You know,
another

00:54:29.580 --> 00:54:35.040
yeah i love this i love this skill here i gotta
use this one bro

00:54:37.180 --> 00:54:37.640
bro

00:54:43.920 --> 00:54:48.540
the presentation that I don't know if I shared
it with you, but there was this five days ago,

00:54:48.620 --> 00:54:56.920
AI engineer released a video from Anthropic. I
forgot her name, but she shares, she talks
about

00:54:53.240 --> 00:54:54.640
hmm

00:54:56.920 --> 00:55:05.700
how you used to use Opus 4.6, how that evolved
towards 5.5 and 5.2 Fable. So the video came
out

00:55:05.700 --> 00:55:11.600
after they released 5.2 Fable. And she talks
about, this is like how you change, because
this

00:55:11.600 --> 00:55:15.300
bro skill, for example, I think it was very
relevant when you have Opus 5.

00:55:15.500 --> 00:55:16.900
mm

00:55:15.520 --> 00:55:19.020
But then they really trained Opus 5.5 to make
it

00:55:20.320 --> 00:55:23.660
better. I will admit I'm still having

00:55:24.460 --> 00:55:27.500
pages and pages of agent text that I have to go
through where I'm like, also

00:55:27.500 --> 00:55:31.600
bro, I want you to shorten it. For this

00:55:31.600 --> 00:55:34.260
particular question that I have, I want you to
keep it short.

00:55:35.460 --> 00:55:35.900
Um...

00:55:35.700 --> 00:55:44.820
yeah the on the topic of skills uh in in my in
my client they have a skill repository but

00:55:44.820 --> 00:55:48.320
it's gone stale like there's no one really
actively updating it

00:55:49.500 --> 00:55:57.500
and uh and hence i'm always careful to disable
the company's skills and

00:56:00.220 --> 00:56:01.540
what's my point here

00:56:02.580 --> 00:56:03.620
I don't know, Clay.

00:56:05.160 --> 00:56:05.640
Sorry.

00:56:05.760 --> 00:56:15.700
these these skills from a third party i.e
potato are great and and i would i would use
them judiciously

00:56:17.840 --> 00:56:19.760
and I have my own skills of course

00:56:21.360 --> 00:56:22.220
but the

00:56:22.220 --> 00:56:23.060
I really feel

00:56:23.060 --> 00:56:24.280
I wanted to basically

00:56:24.280 --> 00:56:25.840
make a point that

00:56:26.800 --> 00:56:29.080
the skills that are produced by

00:56:29.080 --> 00:56:31.240
by my client

00:56:31.240 --> 00:56:33.860
really need to be updated

00:56:33.860 --> 00:57:02.020
Same edition as

00:56:35.720 --> 00:56:38.120
So this is the thing that I wanted to go over,

00:56:39.400 --> 00:56:42.700
which I think highlights the problem that
you're mentioning.

00:56:43.700 --> 00:56:45.820
Skills need to be...

00:56:46.340 --> 00:56:48.660
I think that's what PeaceTech also...

00:56:48.660 --> 00:56:51.240
Sorry, Potato, Laurentan really highlights

00:56:51.240 --> 00:56:54.020
and Matt Pocock, I really like the word that
you use.

00:56:55.740 --> 00:56:57.160
Their interviews are interesting.

00:56:57.360 --> 00:56:59.140
But it's called gardening, right?

00:56:59.460 --> 00:57:03.180
We're basically making sure that the agents and
the plants can grow

00:57:03.180 --> 00:57:04.840
and are flourishing.

00:57:04.840 --> 00:57:07.960
So we are responsible for maintaining those
skills.

00:57:05.040 --> 00:57:05.700
Hmm.

00:57:08.120 --> 00:57:11.500
identifying where the agents are repeating
mistakes

00:57:11.340 --> 00:57:12.000
Hmm.

00:57:11.500 --> 00:57:17.900
and curating the environment for the plants to
grow nicely.

00:57:16.520 --> 00:57:17.180
Yeah.

00:57:17.880 --> 00:57:22.220
Matt, one thing that struck me about what Matt
said is the vocabulary,

00:57:22.700 --> 00:57:26.820
the power of certain words is so critical.

00:57:27.380 --> 00:57:32.720
And I often find that every company has their
own glossary, right?

00:57:32.720 --> 00:57:36.280
But sometimes it's just not really well stated
anywhere.

00:57:36.660 --> 00:57:38.480
And it needs to be, right?

00:57:38.860 --> 00:57:43.440
I went very hard in on the glossary concept and
the grill me and the domain driven.

00:57:45.480 --> 00:57:50.200
The danger is you need several iterations on
getting it right.

00:57:50.420 --> 00:57:51.300
And it's very annoying.

00:57:51.940 --> 00:57:56.220
You know, you can do the glossary definition
with the agents of terminology.

00:57:57.260 --> 00:58:03.400
But you have to be careful because what
happened with me is some words, I forgot what
they meant.

00:58:03.400 --> 00:58:09.960
So when we talk and then we use this glossary,
I was like, can you repeat to me again what
exactly that word means?

00:58:09.060 --> 00:58:10.460
mmm

00:58:09.980 --> 00:58:13.360
Or is that including that particular concept in
that word?

00:58:13.060 --> 00:58:14.460
Haha

00:58:13.400 --> 00:58:14.200
Because I'm not sure.

00:58:14.460 --> 00:58:16.060
So this glossary, yes.

00:58:16.480 --> 00:58:20.240
That's what Matt talks about.

00:58:20.340 --> 00:58:25.620
He says what people always really like about my
course is that they actually really like
domain-driven design.

00:58:25.740 --> 00:58:27.760
They write the shared language.

00:58:28.540 --> 00:58:29.800
And it's a rediscovery.

00:58:29.800 --> 00:58:35.780
and i think it's a really like a lot of these
original um practices uh of of like software

00:58:35.780 --> 00:58:42.380
development get a new breath of fresh air like
life uh in this new new age so they become a
bit

00:58:40.960 --> 00:58:47.020
Yeah, but just a quick aside about glossary.

00:58:42.380 --> 00:58:45.240
more practical when you work with agents but
but

00:58:47.320 --> 00:58:51.640
In the real world, in enterprises, especially
when they operate in different markets,

00:58:52.520 --> 00:59:00.600
the meaning of different things becomes pretty
blurred depending on a lot of environments or a
lot of context.

00:59:00.880 --> 00:59:05.200
So it's a hard task for any enterprise.

00:59:06.120 --> 00:59:12.700
yeah that's kind of what my my team members
also were like i have um he's kind of a bit of
a thought

00:59:12.700 --> 00:59:18.220
leader for me at least in terms of like domain
driven language he highlights how how it really

00:59:18.220 --> 00:59:24.020
cuts at the edges like you just highlighted he
said the same thing um but but right now it was

00:59:19.240 --> 00:59:20.640
Hmm.

00:59:24.020 --> 00:59:29.440
more a language that me and the agents were
sharing for a very like i'm working on one one

00:59:29.440 --> 00:59:36.880
platform and there's different systems within
and they each have their responsibilities so we
don't

00:59:36.880 --> 00:59:41.360
have too many other people it's just me and the
agents and i'm trying to get more people
working

00:59:41.360 --> 00:59:48.440
on this but like it's not that like when we
defer define one term um if i if it's a if it's
an

00:59:48.440 --> 00:59:52.760
overlap between the bounded context and the
agents will very quickly highlight and i don't
have a lot

00:59:52.760 --> 00:59:57.800
of problems like disembogating those but this
is the talk the one that i wanted to to mention

00:59:57.800 --> 01:00:04.040
this is the Rethinking Claude Code best
practices um by lydia from anthropic

01:00:02.500 --> 01:00:03.280
All right.

01:00:03.620 --> 01:00:08.160
So, oh, so you've taken screenshots and
you've...

01:00:10.480 --> 01:00:11.480
what I wanted to

01:00:11.280 --> 01:00:13.200
You're sharing that with your team or
something.

01:00:13.580 --> 01:00:15.420
yes I was sharing it with my team

01:00:15.420 --> 01:00:18.180
but I didn't take the screenshot where she
talks about

01:00:18.180 --> 01:00:21.200
the curating part because cloud code

01:00:21.200 --> 01:00:22.080
is

01:00:24.280 --> 01:00:25.180
very well

01:00:27.000 --> 01:00:28.240
designed based on

01:00:28.240 --> 01:00:31.220
many many users within Anthropic and how they
use it

01:00:31.220 --> 01:00:33.500
so she goes over some of the practices that
changed

01:00:33.500 --> 01:00:38.340
this is how we used to do things this is how we
do things this is how you migrate from your

01:00:38.340 --> 01:00:43.720
existing Claude Code setup towards the new way
of how you should steer the model and so she

01:00:43.720 --> 01:00:50.940
highlighted one of the skills was like slash
doctor um is it prompt improve or or it's
basically

01:00:50.940 --> 01:00:57.820
a skill that helps Claude highlight old ways of
doing things that opus 4.8 needed or cleaning

01:00:57.820 --> 01:01:02.240
up your CLAUDE.md cleaning up your skills and
the words within your skill if you have like
big

01:01:02.240 --> 01:01:03.540
important, important.

01:01:03.440 --> 01:01:05.420
Yeah, that's...

01:01:04.840 --> 01:01:06.040
She says today,

01:01:06.260 --> 01:01:08.000
that was necessary back then.

01:01:07.320 --> 01:01:08.520
Yeah.

01:01:08.160 --> 01:01:10.200
But today, it actually causes the model

01:01:10.200 --> 01:01:11.000
to get distracted

01:01:11.000 --> 01:01:12.420
and to maybe do steps

01:01:12.420 --> 01:01:13.680
that it can actually skip

01:01:13.680 --> 01:01:15.300
because it knows more.

01:01:13.800 --> 01:01:14.300
Yeah.

01:01:15.460 --> 01:01:16.780
Like Fable 5.2 is really good.

01:01:16.980 --> 01:01:18.660
And also, this is really

01:01:18.660 --> 01:01:20.700
how I've been using Fable.

01:01:21.200 --> 01:01:22.620
And she's highlighting

01:01:22.620 --> 01:01:23.340
and that's a pattern

01:01:23.340 --> 01:01:26.000
that they are making available

01:01:26.000 --> 01:01:28.180
as a new feature called projects in beta.

01:01:28.400 --> 01:01:30.260
But you can do this already without,

01:01:30.260 --> 01:01:35.300
without and we've talked about this with with
herder and and rex maybe but you have your

01:01:35.300 --> 01:01:42.420
fable as the orchestrator that keeps the
overall goal and then you have it can spin up a
separate

01:01:37.280 --> 01:01:38.600
Yeah, that's been up.

01:01:42.420 --> 01:01:48.320
session like you highlight i actually have
dedicated agents that um like i have an opus
5.5

01:01:48.320 --> 01:01:53.840
agent that needs to deliver the work and it
does that by actually like spinning up sub
agents but

01:01:53.840 --> 01:01:58.260
even then they say you shouldn't ask opus to
spin up too many sub agents anymore apparently
it's

01:01:58.260 --> 01:02:02.780
better when it does itself these days, himself,
themselves, I

01:02:00.680 --> 01:02:01.540
All right.

01:02:02.300 --> 01:02:07.720
So, do you manage to use Rex or you just
orchestrate inside Claude yourself?

01:02:02.780 --> 01:02:14.480
don't know. So, so we have this initiative
about always updating

01:02:08.200 --> 01:02:08.620
Out of curiosity.

01:02:14.480 --> 01:02:19.360
your tools and finding time to breathe. So I
haven't taken a

01:02:19.360 --> 01:02:23.320
pause. I think this is another thing you need
to make time for.

01:02:23.500 --> 01:02:26.900
I spending a lot of time to read and I try out
skills to fit

01:02:26.900 --> 01:02:29.560
them into my existing workflow. For example,
you asked me to use

01:02:29.560 --> 01:02:33.780
beats, I really like a proper task tracking
that I can hand

01:02:33.780 --> 01:02:37.720
off to the agents. Right now, I feel like I
regressed because

01:02:37.680 --> 01:02:39.060
Mm-hmm.

01:02:37.720 --> 01:02:43.080
of because of pressure. So I really need to
take some time to

01:02:43.080 --> 01:02:45.860
to re configure. Yeah.

01:02:43.720 --> 01:02:50.820
to consolidate yeah yeah there's there's a lot
of i mean i love this actually i love uh

01:02:52.040 --> 01:02:55.620
rethinking or looking at the way i do things
and improving on it

01:02:55.620 --> 01:02:57.780
and you have to do a lot of that sort of
activity

01:02:57.020 --> 01:02:59.880
It's a luxury you have if you don't have like,

01:03:00.140 --> 01:03:02.560
maybe if you're in an enterprise environment

01:03:02.560 --> 01:03:04.880
where there's a lot of approvals and paper
pushing,

01:03:05.060 --> 01:03:06.780
but if you're in a very like,

01:03:07.980 --> 01:03:10.360
delivered this yesterday environment,

01:03:11.100 --> 01:03:14.220
it's very hard to stand still and improve your
tools.

01:03:12.940 --> 01:03:13.460
maybe

01:03:15.540 --> 01:03:16.060
yeah

01:03:17.180 --> 01:03:17.700
like

01:03:17.520 --> 01:03:20.780
It's a luxury to stand still and change your
workflow.

01:03:21.440 --> 01:03:21.960
yeah

01:03:23.280 --> 01:03:25.600
here's something I could do with your advice on

01:03:25.600 --> 01:03:35.640
I was struggling to explain myself. A colleague
had a PR for driving an agent in AgentCore,
like one of those strands based ones.

01:03:35.980 --> 01:03:45.800
And his prompt was like a page long, a page of
text, like do this, do that, do this, do this,
do that, do that.

01:03:46.420 --> 01:03:55.240
Ultimately, all I was doing was working out if
a security finding was a false positive or not.

01:03:58.160 --> 01:04:02.420
And instinctively, how do I explain this?

01:04:02.820 --> 01:04:07.180
But if you're using Opus 5.5, you don't have to
give it much to go on.

01:04:07.560 --> 01:04:09.740
You know, this is your context.

01:04:09.740 --> 01:04:11.120
This is the source code.

01:04:11.280 --> 01:04:12.300
This is the finding.

01:04:12.520 --> 01:04:13.820
Is it a false positive or not?

01:04:14.580 --> 01:04:18.100
You know, in my opinion, I could use Jev
almost.

01:04:20.240 --> 01:04:27.440
um but but though my colleague was like like a
full page like thing oh i need to uh

01:04:28.100 --> 01:04:34.380
what do you call it reach consensus maybe run
this with other agents to find out if it's a
finding

01:04:34.380 --> 01:04:40.380
and i was like no no you've made it too
complicated dude like models are good nowadays

01:04:40.420 --> 01:04:41.860
how old is the code

01:04:43.080 --> 01:04:51.420
well this this pr was was one day old but you
know what i mean it's like like yeah i'm trying

01:04:47.980 --> 01:04:48.840
like today

01:04:51.000 --> 01:04:52.140
I mean like this week

01:04:52.140 --> 01:04:52.880
like this month

01:04:52.340 --> 01:04:56.880
yeah this week i'm trying to tell my my
colleague here that like you don't need

01:04:56.880 --> 01:05:04.520
to engineer something like you would do maybe
three months ago you can quite easily get by

01:05:06.780 --> 01:05:08.520
Did he write everything himself

01:05:08.520 --> 01:05:10.260
or did he ask Opus to write it?

01:05:09.840 --> 01:05:12.740
I think he must have got Claude to write it.

01:05:13.340 --> 01:05:18.660
So the next thing I need to figure out is that
this is my task today,

01:05:13.620 --> 01:05:14.180
Yeah.

01:05:18.980 --> 01:05:22.160
my stretch goal, is that given his extensive
prompt,

01:05:22.180 --> 01:05:25.600
it's a bit like using Claude's skill creator.

01:05:25.600 --> 01:05:39.740
I need to basically prove a point that his
longer prompt is not any better than just a
simple prompt.

01:05:27.880 --> 01:05:29.600
Yes, evals.

01:05:31.130 --> 01:05:32.880
And obliration, what's it called?

01:05:33.500 --> 01:05:34.280
Obliration?

01:05:39.740 --> 01:05:43.940
uh how to operate

01:05:41.640 --> 01:05:43.240
It's called obliration.

01:05:44.540 --> 01:05:45.900
Prompt obliration.

01:05:46.920 --> 01:05:49.120
It's chaos engineering on prompts.

01:05:49.400 --> 01:05:53.220
You basically delete paragraphs and then run an
eval

01:05:53.220 --> 01:05:56.260
and see if the agent is still working or not,
randomly.

01:05:54.540 --> 01:06:00.300
okay i don't know if do Strands or AgentCore
give any frameworks for that because i need

01:06:00.300 --> 01:06:05.140
i need to do that now that's that's my one of
my things i need to figure out

01:06:05.140 --> 01:06:12.020
But I was struggling to argue with my
colleague, like, please, it doesn't need to be
so complicated.

01:06:12.020 --> 01:06:13.600
Just have it simpler.

01:06:13.820 --> 01:06:20.080
i mean instead of telling him did you ask him
how he evaluated and came to that prompt

01:06:21.300 --> 01:06:24.560
Well, I think he's gone through a whole bunch
of iterations.

01:06:24.560 --> 01:06:27.980
It's just built on one, just a cue.

01:06:27.000 --> 01:06:44.340
Yeah, because I had the same with Opus when I
was working on an AI reviewer for Terraform
plans, where it was really constrained. And we
wanted to get like kind of a fixed response as
the end result.

01:06:45.580 --> 01:06:48.280
So part of it, oh, no, I remember where it was.

01:06:48.340 --> 01:06:50.940
It was in elicitation.

01:06:51.160 --> 01:06:54.120
So when an agent needs to stop and elicit more
information,

01:06:54.760 --> 01:06:58.760
I've had a deterministic, like the result from
the agent

01:06:58.760 --> 01:07:02.480
would go into another, you know, into a
deterministic function.

01:07:02.560 --> 01:07:04.540
So it needs to be able to parse that, right?

01:07:04.820 --> 01:07:07.900
So actually, I think it was using a tool to
verify

01:07:07.900 --> 01:07:09.880
that the response was matching the rules

01:07:09.880 --> 01:07:11.940
before it then resends it out back

01:07:11.940 --> 01:07:13.940
because then it goes into Slack that gets
formatted.

01:07:14.340 --> 01:07:20.660
into an um an interactive card that gets
presented to the user to say a b or c or
another response

01:07:20.660 --> 01:07:27.720
it's a bit like ask user question but from
AgentCore to Slack right so um and i also ended
up with

01:07:21.100 --> 01:07:22.500
Yeah.

01:07:27.720 --> 01:07:32.920
with with an iteration an iterative loop where
he kept saying okay i'm running the agent uh
and i'm

01:07:32.920 --> 01:07:38.680
i'm verifying um if it's responding according
to the format that we expect and then you know
it

01:07:38.680 --> 01:07:44.480
keeps growing, keeps, keeps, you know,
modifying and it says all the instructions
aren't good

01:07:44.480 --> 01:07:47.280
enough because the agent came back with
something that wasn't right or it took.

01:07:47.920 --> 01:07:51.880
yeah another thing that caught me off guard i
didn't realize Strands had memory

01:07:51.880 --> 01:07:58.900
so i was just like i was a little bit yeah i
was i was i noticed that it was building up on
its

01:07:54.120 --> 01:07:56.160
It's AgentCore memory, I think.

01:07:58.900 --> 01:08:04.840
memory and i'm like i don't actually want that
i want it to be lean and mean because i'm like

01:08:04.840 --> 01:08:07.860
when I switch up a model, you know, I'm
developing on Sonnet

01:08:07.860 --> 01:08:09.640
and then I might switch to Opus.

01:08:10.000 --> 01:08:15.560
I feel like all this memory that you've
gathered here is just dust, right?

01:08:15.740 --> 01:08:19.000
I mean, it's not going to help us in the
future, really.

01:08:19.860 --> 01:08:23.240
memory systems are a huge market as well

01:08:20.120 --> 01:08:25.380
It might be useful in the sense that, like,

01:08:25.400 --> 01:08:31.040
it's maybe seen the finding before and it
doesn't need to re-evaluate it,

01:08:31.080 --> 01:08:34.760
but then still it's kind of garbage in that
sense.

01:08:33.580 --> 01:08:37.700
oh it's obliteration prompt obliteration

01:08:37.360 --> 01:08:38.220
yeah so

01:08:39.860 --> 01:08:42.160
anyway that's my task today and I should
actually get

01:08:42.160 --> 01:08:43.700
back to work Vincent

01:08:43.700 --> 01:08:46.280
I need to do this eval on

01:08:46.280 --> 01:08:47.680
agent call and this

01:08:47.980 --> 01:08:48.480
Obliteration.

01:08:48.800 --> 01:08:50.400
It's obliteration.

01:08:51.040 --> 01:08:52.940
I need to do some prompt evals

01:08:51.520 --> 01:08:52.700
No, yeah, but

01:08:52.700 --> 01:08:54.920
but strengths

01:08:54.920 --> 01:08:56.780
has a lot of eval libraries

01:08:56.780 --> 01:08:58.740
and capabilities for you to run. I wouldn't be

01:08:58.740 --> 01:09:00.660
surprised if they even provide this as a

01:09:00.660 --> 01:09:02.920
capability. I know that Omni from AWS

01:09:02.920 --> 01:09:04.780
is an offering that allows

01:09:04.780 --> 01:09:05.480
you to take

01:09:06.340 --> 01:09:13.320
you can even get all of the sessions from like
actual live sessions and then feed that into

01:09:13.320 --> 01:09:20.480
your agent setup and then rerun evals over and
over to to see if you hit the mark on your
quality

01:09:14.520 --> 01:09:15.920
ok

01:09:20.480 --> 01:09:30.940
control um and then yeah the the Claude skill-
creator skill does that like locally very
quickly

01:09:20.880 --> 01:09:26.600
yeah yeah okay yeah i'm gonna i'm gonna do that
right now i'm excited uh

01:09:30.940 --> 01:09:33.140
but you have whole solutions

01:09:31.000 --> 01:09:31.520
yeah

01:09:33.140 --> 01:09:35.160
you know like judge evals

01:09:35.160 --> 01:09:37.200
even master like back more than

01:09:37.200 --> 01:09:39.160
a year ago March 2025 was doing

01:09:37.220 --> 01:09:43.540
yeah i guess the one issue with invoking the
AgentCore is that it is a little bit slow

01:09:39.160 --> 01:09:39.380
that

01:09:44.940 --> 01:09:45.540
it's

01:09:46.400 --> 01:09:47.580
it's not definitely

01:09:46.840 --> 01:09:47.640
which one

01:09:48.580 --> 01:09:50.120
the Bedrock AgentCore

01:09:50.120 --> 01:09:51.940
when you invoke it on the cloud

01:09:51.940 --> 01:09:53.720
it's a bit sluggish

01:09:52.920 --> 01:09:54.380
what is slow

01:09:55.520 --> 01:09:57.240
just getting up and running

01:09:59.720 --> 01:10:01.980
I think I have some ideas

01:10:01.980 --> 01:10:02.700
to make it fast

01:10:02.700 --> 01:10:05.900
anything else you want to say

01:10:05.900 --> 01:10:08.160
in the podcast besides like the video

01:10:08.160 --> 01:10:09.760
comment on the pod

01:10:10.680 --> 01:10:19.500
I promised you a demo on policy, like on
basically, yeah, the PII one.

01:10:16.200 --> 01:10:18.980
PII logging stuff.

01:10:19.860 --> 01:10:23.580
I have it all built, but it's the second part
of the workshop.

01:10:23.740 --> 01:10:27.980
So the first part is running the workshop,
getting information, like the dashboard,

01:10:27.980 --> 01:10:30.540
the trackers of the humans, the people,

01:10:28.680 --> 01:10:29.940
Mm-hmm.

01:10:30.740 --> 01:10:35.020
and then doing agentic work without actually
exposing PII data

01:10:35.020 --> 01:10:39.560
by enforcing strict policies that block these
calls.

01:10:40.140 --> 01:10:44.280
I've got the demo working, and thanks to your
questions,

01:10:44.280 --> 01:10:46.940
I created it as an example of my talk.

01:10:45.680 --> 01:10:52.400
okay we can look into that yeah this this topic
of of logging

01:10:48.620 --> 01:10:49.600
Next time.

01:10:53.680 --> 01:10:59.580
if you're a mature enterprise right you you
have your production uh boundary in place right
so that

01:10:59.580 --> 01:11:07.380
people don't view the logs of production
unfortunately um it's hard to get there
sometimes

01:11:08.160 --> 01:11:17.220
well it looks like like for example state graph
their positioning is you can't trust an agent

01:11:17.220 --> 01:11:23.600
so you scope the token so they generate short-
lived tokens scoped for certain tasks

01:11:23.600 --> 01:11:26.640
so that the agent can't do simply that thing,

01:11:26.780 --> 01:11:28.860
which is similar to having a policy

01:11:28.860 --> 01:11:30.520
that blocks certain tool calls, right?

01:11:30.640 --> 01:11:31.580
An MCP tool call.

01:11:31.680 --> 01:11:35.260
yeah but but there's a lot more to it because
like because people

01:11:34.500 --> 01:11:36.880
But it would also solve your code mode problem

01:11:36.880 --> 01:11:38.500
because if you execute script

01:11:38.500 --> 01:11:40.440
and the token isn't allowed to do things,

01:11:40.700 --> 01:11:41.840
then even if the script runs,

01:11:41.980 --> 01:11:42.660
the token will block it

01:11:42.660 --> 01:11:44.620
because the token doesn't have the scope for
it.

01:11:44.880 --> 01:11:46.900
So it could be one way of doing it.

01:11:47.000 --> 01:11:49.760
Your agent lives in a sandbox.

01:11:50.100 --> 01:11:51.340
It doesn't have any credentials.

01:11:51.340 --> 01:11:52.680
The requests that go out,

01:11:52.680 --> 01:11:54.360
the tool calls that go out

01:11:53.380 --> 01:11:54.780
Ok

01:11:54.360 --> 01:11:56.840
are injected with tokens that are scoped

01:11:56.840 --> 01:11:57.900
for that particular task.

01:11:58.060 --> 01:11:59.900
So if that token cannot read the logs

01:11:59.900 --> 01:12:02.140
then even if the agent uses a script

01:12:02.140 --> 01:12:03.800
to do A, B, C, D

01:12:03.800 --> 01:12:05.860
and one of those steps is read the logs

01:12:05.860 --> 01:12:07.580
the token will simply block it.

01:12:07.700 --> 01:12:10.420
So you don't need to do your complicated setup.

01:12:08.560 --> 01:12:15.920
yeah i i i get that but there's more to it like
for example uh it's a human problem here really

01:12:15.920 --> 01:12:20.400
because like you have a human who who owns a
particular service

01:12:21.840 --> 01:12:24.500
in a production or enterprise environment,

01:12:24.560 --> 01:12:28.000
you don't want that engineer to have production
access to the logs.

01:12:29.160 --> 01:12:32.340
But, of course, depending on where they are and
the maturity,

01:12:32.720 --> 01:12:35.060
they might need access to the production logs

01:12:35.060 --> 01:12:36.640
because they're debugging a problem, right?

01:12:37.080 --> 01:12:39.840
Because the software...

01:12:38.020 --> 01:12:43.020
and then a very common setup in an enterprise
is to have zero trust and to have break rest

01:12:43.780 --> 01:12:48.740
access so the people don't have standing access
to production except if there's an incident and

01:12:44.520 --> 01:12:45.640
Yeah, yeah.

01:12:48.740 --> 01:12:50.180
they raise the ticket.

01:12:48.760 --> 01:12:55.420
yeah but that but but but it gets a bit so so
people maybe break that glass and then they
copy

01:12:55.420 --> 01:13:01.120
in the logs into into claude and then you kind
of have the same problem you know what i mean
it's

01:13:01.120 --> 01:13:07.040
like you could you could have all these
controls around your agent not doing it but
then many in

01:13:07.040 --> 01:13:13.280
many cases uh engineers nowadays will use the
break glass and just copy the freaking logs
into

01:13:13.280 --> 01:13:13.800
Claude.

01:13:14.280 --> 01:13:15.980
but then your

01:13:15.980 --> 01:13:17.680
enterprise

01:13:17.680 --> 01:13:19.140
device management

01:13:19.140 --> 01:13:19.920
system has

01:13:19.920 --> 01:13:23.500
set up all these rules inside Claude

01:13:23.500 --> 01:13:26.000
and these monitors and metrics.

01:13:26.540 --> 01:13:30.480
And even if they use a not allowed cloud
enterprise,

01:13:30.900 --> 01:13:34.740
sorry, cloud session that gets private logs,

01:13:34.980 --> 01:13:36.920
then they get flagged.

01:13:38.520 --> 01:13:43.140
Which you said people are not...

01:13:40.760 --> 01:13:42.500
Yeah, like a detective control.

01:13:42.500 --> 01:13:44.080
Well, let's not talk about it now.

01:13:44.240 --> 01:13:47.000
It's actually pretty fresh stuff at my
workplace,

01:13:47.020 --> 01:13:50.240
so I don't want to talk about it right now.

01:13:49.400 --> 01:13:50.800
it's a common problem

01:13:50.800 --> 01:13:53.000
it's a problem I hear a lot

01:13:54.040 --> 01:13:54.780
Exactly, exactly.

01:13:54.360 --> 01:13:56.880
do you think if you move to background

01:13:56.880 --> 01:13:58.260
agents you remove that problem

01:13:58.700 --> 01:14:01.200
No, I don't think it's a technical problem.

01:14:01.060 --> 01:14:03.060
do you understand background agents

01:14:01.360 --> 01:14:03.140
It's more of a...

01:14:04.240 --> 01:14:05.040
No, I don't.

01:14:06.180 --> 01:14:09.000
it's an agent that doesn't run on the engineer
laptop

01:14:09.000 --> 01:14:09.780
it doesn't

01:14:11.280 --> 01:14:13.700
I think it's one of the talks

01:14:13.700 --> 01:14:15.700
of the evolution that she talks about

01:14:15.700 --> 01:14:16.820
how people work with Claude

01:14:16.820 --> 01:14:18.020
like you used to steer it

01:14:18.020 --> 01:14:19.300
you used to need to see the session

01:14:19.300 --> 01:14:20.640
you used to need to read the messages

01:14:20.640 --> 01:14:21.720
to understand the concept

01:14:21.720 --> 01:14:23.280
we're moving away from steering

01:14:23.280 --> 01:14:25.560
the agent is no longer running on your laptop

01:14:25.560 --> 01:14:27.840
the agent is no longer running with your
permissions

01:14:27.840 --> 01:14:30.000
there's no way for you to copy logs into it

01:14:33.820 --> 01:14:35.940
interesting that sounds very mature though

01:14:36.780 --> 01:14:38.320
okay i need the toilet

01:14:38.020 --> 01:14:44.480
it's my talk at Red Hat Singapore on the 27th
it's about background agents and the different

01:14:44.480 --> 01:14:46.780
enterprise agentic adopting patterns

01:14:44.840 --> 01:14:45.660
Okay, cool.

01:14:46.060 --> 01:14:47.760
Okay, well, I'm glad to talk to you

01:14:47.760 --> 01:14:49.820
because you can help me solve these problems.

01:14:50.080 --> 01:14:52.320
And if anyone's listening and they know better,

01:14:52.620 --> 01:14:53.460
please comment below.

01:14:54.140 --> 01:14:55.500
And if you learned something...

01:14:54.600 --> 01:14:56.200
and if you're in Singapore

01:14:56.200 --> 01:14:57.800
come to the Red Hat meetup

01:14:57.800 --> 01:14:59.800
about the

01:14:59.800 --> 01:15:02.140
enterprise agentic adoption

01:15:02.140 --> 01:15:03.140
patterns

01:15:02.540 --> 01:15:03.020
Awesome.

01:15:03.140 --> 01:15:05.000
with background agents

01:15:05.000 --> 01:15:07.040
to solve these problems

01:15:05.240 --> 01:15:06.300
Okay, cool.

01:15:06.440 --> 01:15:06.820
Thank you.

01:15:08.200 --> 01:15:08.560
Okay.

01:15:09.320 --> 01:15:10.120
Cheers, Vincent.

01:15:10.300 --> 01:15:11.100
yeah

01:15:10.360 --> 01:15:10.580
Thanks.

01:15:11.420 --> 01:15:12.400
Thanks ever so much.

01:15:12.500 --> 01:15:12.960
See you.

