WEBVTT

00:00:00.080 --> 00:00:01.040
How are you?

00:00:01.040 --> 00:00:03.120
>> Yeah, good. I've been on holiday this

00:00:03.120 --> 00:00:06.640
week, so I've but I've just been at

00:00:06.640 --> 00:00:09.920
home, so I in the end of the day, I just

00:00:09.920 --> 00:00:12.000
just been trying to spend time with my

00:00:12.000 --> 00:00:13.920
kids and

00:00:13.920 --> 00:00:16.720
not too sure I'm doing the greatest job.

00:00:16.720 --> 00:00:19.600
I just I just end up arranging

00:00:19.600 --> 00:00:21.359
activities and then watching them play

00:00:21.359 --> 00:00:22.720
from afar.

00:00:22.720 --> 00:00:24.400
>> How old are they?

00:00:24.400 --> 00:00:26.240
>> 10 and eight.

00:00:26.240 --> 00:00:27.840
>> Oh, that's nice. They're close to each

00:00:27.840 --> 00:00:29.279
other. Um

00:00:29.279 --> 00:00:30.640
>> yeah,

00:00:30.640 --> 00:00:32.160
>> and they can entertain themselves.

00:00:32.160 --> 00:00:34.399
They're at an age.

00:00:34.399 --> 00:00:36.559
>> Yeah, it's it's quite good. Yeah. And

00:00:36.559 --> 00:00:39.120
then and then um the big news is that my

00:00:39.120 --> 00:00:41.920
c my sister and her husband's moving in

00:00:41.920 --> 00:00:43.840
here and they're going to live next to

00:00:43.840 --> 00:00:46.000
us so that they got two cousins to play

00:00:46.000 --> 00:00:48.480
with all the time.

00:00:48.480 --> 00:00:50.480
>> So I guess I don't need to be around

00:00:50.480 --> 00:00:52.879
anymore. I can just fully devote myself

00:00:52.879 --> 00:00:56.879
to AI. Oh my god, that sounds so I mean

00:00:56.879 --> 00:00:59.120
not not at all what I'm doing though.

00:00:59.120 --> 00:01:01.840
>> No, no, no, no. You got the balance.

00:01:01.840 --> 00:01:04.879
You've got the balance just right.

00:01:04.879 --> 00:01:08.799
>> My son is 17 and my daughter is 12. No,

00:01:08.799 --> 00:01:11.680
10. Uh, she's 6 years younger, so she

00:01:11.680 --> 00:01:13.760
must be 11. Yeah, she's going to be 11.

00:01:13.760 --> 00:01:16.240
>> Yeah. And so it's a bigger gap and

00:01:16.240 --> 00:01:18.560
there's a bit more a different power

00:01:18.560 --> 00:01:22.640
dynamic. Like he acts like,

00:01:22.640 --> 00:01:25.280
you know, cuz he he feels like I spoil

00:01:25.280 --> 00:01:27.759
her because she's my little princess. So

00:01:27.759 --> 00:01:29.600
he's there to set things right, you

00:01:29.600 --> 00:01:31.759
know, to be a bit more disciplined when

00:01:31.759 --> 00:01:33.680
she's, you know, taking all this for a

00:01:33.680 --> 00:01:35.119
walk. I don't know if you can see this

00:01:35.119 --> 00:01:37.119
ring like in in the reflection in the

00:01:37.119 --> 00:01:37.520
glasses.

00:01:37.520 --> 00:01:39.200
>> I can actually. I can. Is this is this

00:01:39.200 --> 00:01:42.960
your your light or something? It's this

00:01:42.960 --> 00:01:46.159
feature on when I got this MacBook and I

00:01:46.159 --> 00:01:47.920
joined the meeting. I wanted I was in a

00:01:47.920 --> 00:01:50.720
dark room and it says said the the light

00:01:50.720 --> 00:01:52.560
is bad. Do you want to turn on the c the

00:01:52.560 --> 00:01:54.560
ring camera? And I thought it was a

00:01:54.560 --> 00:01:56.399
Microsoft Teams thing and I was like

00:01:56.399 --> 00:01:57.680
I'll figure out how to turn it off at

00:01:57.680 --> 00:02:00.399
some point. But it's everywhere and it's

00:02:00.399 --> 00:02:02.479
annoying cuz it's and I haven't figured

00:02:02.479 --> 00:02:04.079
out how to turn it off yet.

00:02:04.079 --> 00:02:06.399
>> I should ask Claw to turn it off for me.

00:02:06.399 --> 00:02:08.800
>> Yeah. So you're in a different room now.

00:02:08.800 --> 00:02:11.760
>> Yeah. So my I had to give my

00:02:11.760 --> 00:02:13.840
>> son's room back so he could spend some

00:02:13.840 --> 00:02:15.760
private time. He can get some some

00:02:15.760 --> 00:02:16.879
private time.

00:02:16.879 --> 00:02:19.680
>> Why you say it like that? [laughter]

00:02:19.680 --> 00:02:22.319
>> Well, yeah. It's It's been good to

00:02:22.319 --> 00:02:25.280
disconnect for a little bit, but I still

00:02:25.280 --> 00:02:26.879
like read hacker news in the morning

00:02:26.879 --> 00:02:28.959
like a lunatic.

00:02:28.959 --> 00:02:32.400
Funny enough, uh, Paul Graham

00:02:32.400 --> 00:02:35.280
was was talking about that one problem

00:02:35.280 --> 00:02:38.000
that he felt that was not solved with

00:02:38.000 --> 00:02:40.959
white carbonator is is how to balance

00:02:40.959 --> 00:02:44.400
family life with with startup culture.

00:02:44.400 --> 00:02:47.920
Family life, startup culture. Yeah,

00:02:47.920 --> 00:02:51.120
there doesn't seem to be any way of

00:02:51.120 --> 00:02:55.120
making those two mix very very well.

00:02:55.120 --> 00:02:58.160
I I mean I personally know at least a at

00:02:58.160 --> 00:03:00.560
least a couple of people that have have

00:03:00.560 --> 00:03:03.040
thrown themselves deep into tech and of

00:03:03.040 --> 00:03:05.680
course they don't have a family I would

00:03:05.680 --> 00:03:07.280
say

00:03:07.280 --> 00:03:08.879
>> which I think is a bit of a shame. I

00:03:08.879 --> 00:03:11.920
don't know if you have got any of those

00:03:11.920 --> 00:03:15.200
examples to hand but I know yeah I

00:03:15.200 --> 00:03:16.480
definitely know a couple of people who

00:03:16.480 --> 00:03:20.720
went deep. Yeah, I aside from myself, I

00:03:20.720 --> 00:03:22.640
have a family, but

00:03:22.640 --> 00:03:23.280
>> just

00:03:23.280 --> 00:03:26.239
>> cuz they're quite accommodating to to to

00:03:26.239 --> 00:03:28.640
me being so like honestly it was

00:03:28.640 --> 00:03:31.280
supposed to be temporary right after co

00:03:31.280 --> 00:03:33.840
you know it was supposed to be like you

00:03:33.840 --> 00:03:37.440
know just you know lock in for a while.

00:03:37.440 --> 00:03:40.720
It's been five years now um to to

00:03:40.720 --> 00:03:44.799
prepare for the future because you know

00:03:44.799 --> 00:03:48.640
co made remote work possible and

00:03:48.640 --> 00:03:49.599
>> exactly

00:03:49.599 --> 00:03:52.400
>> it allowed me to try and put away as

00:03:52.400 --> 00:03:54.720
much as I can and retire early except

00:03:54.720 --> 00:03:56.640
I've been doing that for 5 years and I'm

00:03:56.640 --> 00:03:58.400
not retiring early. So

00:03:58.400 --> 00:04:02.319
>> yeah, I yeah, like with AI and the

00:04:02.319 --> 00:04:05.360
investments I'm making, I feel like

00:04:05.360 --> 00:04:07.439
I'm going to be working to the day I

00:04:07.439 --> 00:04:10.319
well until the into retirement age at

00:04:10.319 --> 00:04:12.640
the very least. But yeah, I I feel the

00:04:12.640 --> 00:04:16.160
pressure with AI. I mean, I do enjoy AI.

00:04:16.160 --> 00:04:19.680
It's just that of course there is there

00:04:19.680 --> 00:04:22.160
is like an under an undertone an

00:04:22.160 --> 00:04:25.520
underlying pressure to to stay relevant

00:04:25.520 --> 00:04:27.040
and

00:04:27.040 --> 00:04:28.639
and and

00:04:28.639 --> 00:04:30.320
stay ahead of the pack and it's yeah it

00:04:30.320 --> 00:04:34.080
is a bit it is a bit much.

00:04:34.080 --> 00:04:37.040
It is a bit much but like got to keep on

00:04:37.040 --> 00:04:39.360
doing it and try and make it fun. So

00:04:39.360 --> 00:04:40.560
hopefully you have some fun things to

00:04:40.560 --> 00:04:43.199
tell me Vincent. Otherwise, I'm just I'm

00:04:43.199 --> 00:04:45.440
gonna sign off now and spend some time

00:04:45.440 --> 00:04:46.960
with my children.

00:04:46.960 --> 00:04:50.240
>> Stand still and contemplate or decide

00:04:50.240 --> 00:04:52.320
like make sense of what we're doing and

00:04:52.320 --> 00:04:54.400
let's just dive in and focus on what

00:04:54.400 --> 00:04:57.199
what what other fun stuff we read about.

00:04:57.199 --> 00:04:57.680
>> Yeah.

00:04:57.680 --> 00:05:00.400
>> Never stand still. Never look back.

00:05:00.400 --> 00:05:03.120
>> Got to keep it fun. Yeah. Don't ask uh

00:05:03.120 --> 00:05:05.520
you know, ethical questions.

00:05:05.520 --> 00:05:06.000
>> Yeah.

00:05:06.000 --> 00:05:08.160
>> Oh, I I got into I got into a silly

00:05:08.160 --> 00:05:09.759
debate yesterday with one of the people

00:05:09.759 --> 00:05:12.479
I played tennis with at I'm a proponent

00:05:12.479 --> 00:05:14.960
of having of building solar farms. I

00:05:14.960 --> 00:05:16.240
don't know what it's like in Vietnam,

00:05:16.240 --> 00:05:18.320
but like this there's been a bit of a

00:05:18.320 --> 00:05:22.240
buildout of of zero my god, what's it

00:05:22.240 --> 00:05:26.400
called? Zero whatever um zero footprint

00:05:26.400 --> 00:05:28.080
energy, whatever it's called. So

00:05:28.080 --> 00:05:30.800
basically solar farms in in England and

00:05:30.800 --> 00:05:32.240
there's like a real opposition to it

00:05:32.240 --> 00:05:33.919
because like oh what about our food

00:05:33.919 --> 00:05:35.759
security? But like I can't help but

00:05:35.759 --> 00:05:38.080
argue like isn't energy security

00:05:38.080 --> 00:05:40.000
slightly more important?

00:05:40.000 --> 00:05:41.440
>> Yeah.

00:05:41.440 --> 00:05:44.000
and saving the world. Anyway, okay,

00:05:44.000 --> 00:05:46.639
those distractions aside, Vincent, you

00:05:46.639 --> 00:05:48.800
look a bit drained and distracted. Come

00:05:48.800 --> 00:05:49.919
on.

00:05:49.919 --> 00:05:52.720
>> Yeah. No, let's let me share the the

00:05:52.720 --> 00:05:55.759
screen. Entire screen.

00:05:55.759 --> 00:05:57.440
I'm going to do it on the screen that

00:05:57.440 --> 00:06:00.240
has the Coo massive

00:06:00.240 --> 00:06:01.840
um inception right there.

00:06:01.840 --> 00:06:06.319
>> Okay. So, you're back on the hero stuff.

00:06:06.319 --> 00:06:10.880
>> Yeah. So I'm actually running like a

00:06:10.880 --> 00:06:14.080
local build um where I did a couple of

00:06:14.080 --> 00:06:16.479
really fun things. So my version might

00:06:16.479 --> 00:06:19.919
not actually work properly uh because I

00:06:19.919 --> 00:06:25.199
went quite invasive and I mean Kiro Crew

00:06:25.199 --> 00:06:27.759
is kind of AWS

00:06:27.759 --> 00:06:30.479
openclaw/harmus

00:06:30.479 --> 00:06:34.880
solution for enterprise right and since

00:06:34.880 --> 00:06:38.319
they made it public they now are also

00:06:38.319 --> 00:06:43.199
sharing about how quickly it evolves

00:06:43.199 --> 00:06:45.840
similar to openclaw or hermas you know

00:06:45.840 --> 00:06:48.479
having agents on their own contribute

00:06:48.479 --> 00:06:51.120
the features and then they have a whole

00:06:51.120 --> 00:06:54.880
testing harness and um review harness.

00:06:54.880 --> 00:06:57.120
So this is actually very interesting to

00:06:57.120 --> 00:06:59.199
to see and also to

00:06:59.199 --> 00:07:01.520
>> there's a link up with the the AWS bench

00:07:01.520 --> 00:07:04.319
that we were talking about last time.

00:07:04.319 --> 00:07:07.280
Uh

00:07:07.280 --> 00:07:10.240
maybe because I do believe they they

00:07:10.240 --> 00:07:14.240
they do benchmark hero's capabilities

00:07:14.240 --> 00:07:17.759
on specifically AWS type of um [snorts]

00:07:17.759 --> 00:07:20.000
you know problems. Do you want to focus

00:07:20.000 --> 00:07:23.199
on that first or like let's just look at

00:07:23.199 --> 00:07:25.759
>> So this is called Kiro Crew. I kind of

00:07:25.759 --> 00:07:28.000
>> Yes. So there's a big difference here.

00:07:28.000 --> 00:07:29.599
Maybe you see it at the top. I am

00:07:29.599 --> 00:07:32.240
currently running the Mac OS app and

00:07:32.240 --> 00:07:34.240
that comes bundled with a gateway that

00:07:34.240 --> 00:07:38.319
then hooks into your local MacBook um

00:07:38.319 --> 00:07:40.639
where you will have Kiro CLI

00:07:40.639 --> 00:07:44.080
authenticated and the Kiro CLI and the

00:07:44.080 --> 00:07:46.560
Kiro IDE which is the original IDE they

00:07:46.560 --> 00:07:47.919
launched with specdriven development

00:07:47.919 --> 00:07:51.199
last year has all been rewritten to work

00:07:51.199 --> 00:07:54.160
on a single harness a single bankant and

00:07:54.160 --> 00:07:56.319
it uses the agent client protocol to

00:07:56.319 --> 00:07:57.759
connect everything. So you get you get

00:07:57.759 --> 00:08:00.319
all of these different experiences um

00:08:00.319 --> 00:08:04.080
and with with like one like if I go into

00:08:04.080 --> 00:08:06.560
the builder article I wrote

00:08:06.560 --> 00:08:08.960
>> I noticed that AWS was contributing to

00:08:08.960 --> 00:08:13.599
the whole MCP thing and making it easier

00:08:13.599 --> 00:08:14.479
to

00:08:14.479 --> 00:08:18.479
>> so this is the agent client protocol not

00:08:18.479 --> 00:08:19.759
not MCP

00:08:19.759 --> 00:08:21.360
>> agent client protocol okay

00:08:21.360 --> 00:08:22.720
>> yeah it's not the model

00:08:22.720 --> 00:08:25.440
>> it's not the model context protocol so

00:08:25.440 --> 00:08:28.319
what what Kira crew is here. Um it's a

00:08:28.319 --> 00:08:31.360
it's a gateway and um

00:08:31.360 --> 00:08:34.080
>> very much like how how um you know

00:08:34.080 --> 00:08:36.080
OpenCloud originally was a gateway to

00:08:36.080 --> 00:08:38.719
Telegram or WhatsApp uh for you to talk

00:08:38.719 --> 00:08:42.479
to your um agent um wherever it ran like

00:08:42.479 --> 00:08:45.040
maybe it's your um you know codec cli or

00:08:45.040 --> 00:08:46.640
your Claude Code you know originally it

00:08:46.640 --> 00:08:50.160
was open claw claw like cloud cloudbot

00:08:50.160 --> 00:08:52.480
it was related to Claude Code right until

00:08:52.480 --> 00:08:55.120
anthropic shut it down and then codex u

00:08:55.120 --> 00:08:57.600
you know openai hired hurts Peter and

00:08:57.600 --> 00:09:00.000
anyway this is the same thing right it

00:09:00.000 --> 00:09:03.120
starts as a gateway and the gateway

00:09:03.120 --> 00:09:06.160
connects to MCP tools and this is uh the

00:09:06.160 --> 00:09:08.399
dashboard app um this is the actual

00:09:08.399 --> 00:09:09.600
dashboard app

00:09:09.600 --> 00:09:11.920
>> you can also have it in your browser so

00:09:11.920 --> 00:09:14.240
this is the dashboard app running in the

00:09:14.240 --> 00:09:16.959
>> I was going to ask is there a mobile app

00:09:16.959 --> 00:09:19.040
>> um potentially I'm not I'm not sure I

00:09:19.040 --> 00:09:20.959
haven't looked into that uh but anyway

00:09:20.959 --> 00:09:23.279
so so this is the whole component layout

00:09:23.279 --> 00:09:25.440
which I think is a little bit murky when

00:09:25.440 --> 00:09:26.959
you first inst install it and you don't

00:09:26.959 --> 00:09:28.880
know like when you first download or

00:09:28.880 --> 00:09:30.560
built the the dashboard app on your

00:09:30.560 --> 00:09:32.480
machine, it's actually running on your

00:09:32.480 --> 00:09:35.600
on your laptop and this this Mac OS app

00:09:35.600 --> 00:09:36.640
right here

00:09:36.640 --> 00:09:41.120
>> comes with sessions to for chat schedule

00:09:41.120 --> 00:09:43.200
um everything is is is there and it runs

00:09:43.200 --> 00:09:45.519
on your laptop which I did not like I

00:09:45.519 --> 00:09:48.000
don't want this on my laptop I wanted to

00:09:48.000 --> 00:09:49.920
>> feel a little bit like an antiattern you

00:09:49.920 --> 00:09:51.680
know when people install openclaw on

00:09:51.680 --> 00:09:54.959
their their Mac minis and like yeah take

00:09:54.959 --> 00:09:56.480
over everything.

00:09:56.480 --> 00:09:58.640
>> Yeah. So, so that's where I got confused

00:09:58.640 --> 00:09:59.920
and that's where I really like this

00:09:59.920 --> 00:10:01.360
diagram because I didn't know what the

00:10:01.360 --> 00:10:03.839
components were within and uh but the

00:10:03.839 --> 00:10:04.959
most important part is that this

00:10:04.959 --> 00:10:07.279
dashboard app comes with like a bundled

00:10:07.279 --> 00:10:09.760
gateway uh and then it connects to your

00:10:09.760 --> 00:10:12.080
Kiro CLI on your machine and then if you

00:10:12.080 --> 00:10:14.320
authenticate Kiro CLI or you use Kiro

00:10:14.320 --> 00:10:16.240
before it will inherit everything you

00:10:16.240 --> 00:10:18.480
already have there. So, um then it it's

00:10:18.480 --> 00:10:20.160
a little bit like herder what you

00:10:20.160 --> 00:10:22.079
mentioned you know it gives you an

00:10:22.079 --> 00:10:23.279
overview of all this of all the

00:10:23.279 --> 00:10:23.680
sessions.

00:10:23.680 --> 00:10:25.279
>> Where does this gateway run? Is this an

00:10:25.279 --> 00:10:26.800
AWS thing

00:10:26.800 --> 00:10:30.720
>> here? Um, let me just

00:10:30.720 --> 00:10:33.279
>> This is you get like all your chats of

00:10:33.279 --> 00:10:35.680
your your local session. So,

00:10:35.680 --> 00:10:38.320
>> when you do the gateway in the Mac OS

00:10:38.320 --> 00:10:40.399
app, it's it's basically just it talks

00:10:40.399 --> 00:10:43.040
to your uh CLI on your machine using

00:10:43.040 --> 00:10:45.920
ACP, right? Agent client protocol. Okay,

00:10:45.920 --> 00:10:52.160
cool. So, um Oops. Yeah. So, so that's

00:10:52.160 --> 00:10:53.600
where does the gateway run on your

00:10:53.600 --> 00:10:54.880
machine?

00:10:54.880 --> 00:10:56.720
Yes, I'm going to say it the third time

00:10:56.720 --> 00:11:00.079
now for you. Okay, the local bundle.

00:11:00.079 --> 00:11:02.640
This thing runs the gateway inside the

00:11:02.640 --> 00:11:05.600
Mac OS DMG. If you build it, it it

00:11:05.600 --> 00:11:08.880
bundles the single page app like a React

00:11:08.880 --> 00:11:12.640
vit and the gateway which is a Python uh

00:11:12.640 --> 00:11:14.480
gateway and that connect with the

00:11:14.480 --> 00:11:17.920
protocol to your KRO on your machine.

00:11:17.920 --> 00:11:20.560
Okay. So, this this Mac OS app

00:11:20.560 --> 00:11:22.560
>> runs the gateway.

00:11:22.560 --> 00:11:24.000
>> Got it. But then

00:11:24.000 --> 00:11:26.399
>> and now let me let me tell you one more

00:11:26.399 --> 00:11:29.279
thing. One more thing here is that this

00:11:29.279 --> 00:11:31.440
particular gateway has a remote crew

00:11:31.440 --> 00:11:33.360
capability here.

00:11:33.360 --> 00:11:34.640
>> Ah

00:11:34.640 --> 00:11:39.279
>> yeah so with this remote crew you can um

00:11:39.279 --> 00:11:42.320
if you use the Kiro crew CLI you can you

00:11:42.320 --> 00:11:44.079
have a cloud subcomand. So actually

00:11:44.079 --> 00:11:46.079
that's um the next part of this of this

00:11:46.079 --> 00:11:48.079
article here uh is that you can set up

00:11:48.079 --> 00:11:51.120
remote crew and you can run Kiro Crew

00:11:51.120 --> 00:11:56.079
cloud um to manage remote instances and

00:11:56.079 --> 00:11:58.160
and that basically bootstrap a cloud for

00:11:58.160 --> 00:12:00.560
template into your AWS account and spins

00:12:00.560 --> 00:12:03.040
up an instance with the gateway as well.

00:12:03.040 --> 00:12:05.600
So that's a separate remote gateway that

00:12:05.600 --> 00:12:08.480
you then that you then connect to. Okay.

00:12:08.480 --> 00:12:09.440
So that's

00:12:09.440 --> 00:12:11.920
>> that's a separate machine. You may run

00:12:11.920 --> 00:12:13.839
the Kiro CLI on it. Then you need to

00:12:13.839 --> 00:12:16.880
authenticate your Kiro CLI and um with

00:12:16.880 --> 00:12:18.880
the device mode authentication on your

00:12:18.880 --> 00:12:22.079
subscription and then you you get um you

00:12:22.079 --> 00:12:24.720
know [snorts] Kira access and all the

00:12:24.720 --> 00:12:28.240
models that Kiro support. Um and you you

00:12:28.240 --> 00:12:32.079
you get the same thing here. Um sorry

00:12:32.079 --> 00:12:36.320
you get the where was I?

00:12:36.320 --> 00:12:38.800
Uh where is the app? Oh, here it is. So,

00:12:38.800 --> 00:12:41.600
so you get you get you can connect. So,

00:12:41.600 --> 00:12:43.600
let let's let's say you spin up one and

00:12:43.600 --> 00:12:46.399
then you can connect here and then it

00:12:46.399 --> 00:12:49.279
will pop up at the top and now you have

00:12:49.279 --> 00:12:51.120
uh all of the chat sessions you have

00:12:51.120 --> 00:12:53.200
with that instance over there

00:12:53.200 --> 00:12:55.920
>> and it can run um you know you can set

00:12:55.920 --> 00:12:58.639
up tasks to run on on a on a schedule.

00:12:58.639 --> 00:13:00.639
So, the thing runs on its own

00:13:00.639 --> 00:13:03.200
>> and um I mean I guess this is like what

00:13:03.200 --> 00:13:05.440
you would get if you runs the Hermes app

00:13:05.440 --> 00:13:08.639
which I never ran. So um I have a Hermes

00:13:08.639 --> 00:13:10.800
VM and right now the only way I interact

00:13:10.800 --> 00:13:13.519
with it is through the Discord channel

00:13:13.519 --> 00:13:15.440
and it's not a very nice experience,

00:13:15.440 --> 00:13:17.040
right? Every interaction is either

00:13:17.040 --> 00:13:20.160
through a Discord uh thread um and and

00:13:20.160 --> 00:13:22.880
in this case actually I get a nice UI

00:13:22.880 --> 00:13:25.519
and I do think Hermes has this as well

00:13:25.519 --> 00:13:28.399
but this is the AWS version. I I've

00:13:28.399 --> 00:13:30.079
shown you this before, but I I use this

00:13:30.079 --> 00:13:32.560
thing called command and control,

00:13:32.560 --> 00:13:34.880
which has a mobile app and a web page

00:13:34.880 --> 00:13:38.399
where I I I I connect to all my my

00:13:38.399 --> 00:13:40.560
different uh

00:13:40.560 --> 00:13:44.160
Okay. So, so this this this is confusing

00:13:44.160 --> 00:13:46.079
because I have another sessions here

00:13:46.079 --> 00:13:48.000
settings here and then here I don't have

00:13:48.000 --> 00:13:50.639
the remote crew uh option because this

00:13:50.639 --> 00:13:53.200
is a remote gateway already. Um, and

00:13:53.200 --> 00:13:54.560
what's very interesting is when you're

00:13:54.560 --> 00:13:56.000
developing and you're making changes to

00:13:56.000 --> 00:13:57.920
the gateway and then it goes and deploys

00:13:57.920 --> 00:14:00.079
the gateway into that remote instance,

00:14:00.079 --> 00:14:03.199
you have to like reload this whole pane

00:14:03.199 --> 00:14:06.000
because this whole pane is served over

00:14:06.000 --> 00:14:09.360
SSH or over SSM via private tunnel. So,

00:14:09.360 --> 00:14:10.880
so this actually is a separate instance

00:14:10.880 --> 00:14:12.959
of a gateway running separately that you

00:14:12.959 --> 00:14:15.680
that you get over um over a tunnel like

00:14:15.680 --> 00:14:17.040
an SSH tunnel.

00:14:17.040 --> 00:14:19.360
>> It's very interesting. Yeah. So, okay,

00:14:19.360 --> 00:14:20.720
>> that that was interesting because I was

00:14:20.720 --> 00:14:22.560
building I was re uh changing things and

00:14:22.560 --> 00:14:24.720
and and reloading the app is actually a

00:14:24.720 --> 00:14:26.240
little bit unintuitive because I was

00:14:26.240 --> 00:14:29.040
like do I need to rebuild my my local uh

00:14:29.040 --> 00:14:32.079
Mac OS app and and uh cloud was like no

00:14:32.079 --> 00:14:34.399
uh the whole app loads over SSH

00:14:34.399 --> 00:14:37.920
remotely. So once the the whole um asset

00:14:37.920 --> 00:14:39.680
all of the assets have been downloaded

00:14:39.680 --> 00:14:42.320
over SSH then it runs very quickly. Um,

00:14:42.320 --> 00:14:43.920
but the first time it can take a while

00:14:43.920 --> 00:14:46.000
if you run a a crew somewhere in in

00:14:46.000 --> 00:14:47.680
Singapore and you're in Vietnam and your

00:14:47.680 --> 00:14:50.079
connection is not great. Um, then then

00:14:50.079 --> 00:14:51.600
the first time loading then is a bit

00:14:51.600 --> 00:14:53.600
slow. Uh, but once it's loaded it's it's

00:14:53.600 --> 00:14:54.160
pretty fast. I

00:14:54.160 --> 00:14:56.160
>> I like the look of this whole ecosystem.

00:14:56.160 --> 00:14:58.639
I guess I guess this is like remote

00:14:58.639 --> 00:15:01.600
control claw remote control on steroids.

00:15:01.600 --> 00:15:03.040
I guess we're going to see more of these

00:15:03.040 --> 00:15:05.600
sort of buildouts, right, where you you

00:15:05.600 --> 00:15:09.839
you able to see your a number of agents

00:15:09.839 --> 00:15:13.519
in essentially one place, right? Yeah.

00:15:13.519 --> 00:15:17.199
And there's also the ability here. Okay.

00:15:17.199 --> 00:15:18.880
I I want to really go into the features

00:15:18.880 --> 00:15:20.720
that this thing has because it has a lot

00:15:20.720 --> 00:15:22.320
more like my experience setting up

00:15:22.320 --> 00:15:25.360
Hermes VM and then coming into this um

00:15:25.360 --> 00:15:29.040
is like holy, you know, sword. This is

00:15:29.040 --> 00:15:33.360
them um packed with features. Um you I

00:15:33.360 --> 00:15:35.120
haven't tried this yet, but right now

00:15:35.120 --> 00:15:37.120
this remote uh instance is actually a

00:15:37.120 --> 00:15:38.880
virtual machine on my desktop at home

00:15:38.880 --> 00:15:41.040
only accessible over LAN. Uh I'm very

00:15:41.040 --> 00:15:43.760
excited to enable like tail net uh or or

00:15:43.760 --> 00:15:45.680
like tail scale and then being able to

00:15:45.680 --> 00:15:48.320
connect to it remotely. Um

00:15:48.320 --> 00:15:50.720
>> ideally over a control and open claw

00:15:50.720 --> 00:15:53.759
have that kind of up the box.

00:15:53.759 --> 00:15:56.160
>> Uh yeah sure. Now let's go and talk

00:15:56.160 --> 00:15:57.680
about some of these cool things here

00:15:57.680 --> 00:16:00.480
like knowledge. So I don't know if you

00:16:00.480 --> 00:16:02.320
set up like I don't know about open claw

00:16:02.320 --> 00:16:04.880
but Hermes when you set it up one of

00:16:04.880 --> 00:16:06.560
the first thing you have to do is like

00:16:06.560 --> 00:16:09.040
what is your memory system it's flexible

00:16:09.040 --> 00:16:10.399
right you can choose which one you want

00:16:10.399 --> 00:16:12.959
you can use hindsight.ai AI uh you can

00:16:12.959 --> 00:16:15.920
use another like you ideally nose knows

00:16:15.920 --> 00:16:18.560
no new uh the the company behind Hermes

00:16:18.560 --> 00:16:20.079
wants you to use their control their

00:16:20.079 --> 00:16:22.160
cloud to use their services right but

00:16:22.160 --> 00:16:24.240
you are freeing one I don't remember

00:16:24.240 --> 00:16:25.839
choosing I don't remember choosing one

00:16:25.839 --> 00:16:27.680
but was

00:16:27.680 --> 00:16:30.480
>> yeah soh yeah it also can use markdown

00:16:30.480 --> 00:16:33.600
files so on disk it will have a a soulm

00:16:33.600 --> 00:16:36.560
and maybe some other uh markdown files

00:16:36.560 --> 00:16:38.240
but what I found was that it very

00:16:38.240 --> 00:16:40.320
quickly hit a ceiling on like there are

00:16:40.320 --> 00:16:41.680
restrictions on the markdown on file

00:16:41.680 --> 00:16:43.519
size and also when it tries to save

00:16:43.519 --> 00:16:45.600
memories it writes to down to markdown

00:16:45.600 --> 00:16:47.519
and then it hit that size limit. I

00:16:47.519 --> 00:16:49.040
basically used Hermes without really

00:16:49.040 --> 00:16:50.399
knowing what what it was doing for a

00:16:50.399 --> 00:16:51.920
week and then I asked Sonnet what are the

00:16:51.920 --> 00:16:53.279
common failures what are the errors that

00:16:53.279 --> 00:16:54.959
you see in the logs and one of the first

00:16:54.959 --> 00:16:57.040
things Sonnet uh highlighted or cloud

00:16:57.040 --> 00:16:59.040
code highlighted was that well it's

00:16:59.040 --> 00:17:00.720
having a lot of issues writing memories

00:17:00.720 --> 00:17:03.440
like it's hitting markdown file CS and

00:17:03.440 --> 00:17:04.799
things like that so the memory system

00:17:04.799 --> 00:17:06.559
was the first thing that I replaced uh

00:17:06.559 --> 00:17:07.760
within Hermes.

00:17:07.760 --> 00:17:08.799
>> Okay.

00:17:08.799 --> 00:17:10.959
>> Yeah. I feel like I'm being lazy because

00:17:10.959 --> 00:17:12.880
I've not looked through the Hermes logs

00:17:12.880 --> 00:17:14.240
to be honest.

00:17:14.240 --> 00:17:16.880
>> Yeah. So, so, so what that meant for me

00:17:16.880 --> 00:17:19.039
was that first off I wanted to figure

00:17:19.039 --> 00:17:20.880
out what does a memory system provide

00:17:20.880 --> 00:17:24.720
and like how can I use it and I started

00:17:24.720 --> 00:17:27.760
to learn a little bit more about

00:17:27.760 --> 00:17:30.160
um like on the memory side of things

00:17:30.160 --> 00:17:32.880
there's information about how do these

00:17:32.880 --> 00:17:35.679
memories uh link together

00:17:35.679 --> 00:17:38.400
um you know semantically you know recall

00:17:38.400 --> 00:17:40.480
rate like when were the last recall are

00:17:40.480 --> 00:17:42.720
they still fresh there's a lot of uh

00:17:42.720 --> 00:17:45.120
aspects about a memory system that you

00:17:45.120 --> 00:17:46.480
don't really want to bother with, right?

00:17:46.480 --> 00:17:48.000
But but I had to learn it because I had

00:17:48.000 --> 00:17:49.600
to like understand a little bit better

00:17:49.600 --> 00:17:51.520
about Hermes and setting up the memory

00:17:51.520 --> 00:17:53.520
system. So I I'm running on the desktop

00:17:53.520 --> 00:17:56.799
a complete full stack of uh containers

00:17:56.799 --> 00:17:58.880
which uh uses Postgres vector

00:17:58.880 --> 00:18:00.640
embedding. I have a embedding model that

00:18:00.640 --> 00:18:03.280
I had set up with my GPU and then you

00:18:03.280 --> 00:18:06.080
come to Kiro you install it and

00:18:06.080 --> 00:18:08.240
everything is there right I'm trying to

00:18:08.240 --> 00:18:10.559
find now but in terms of like if I look

00:18:10.559 --> 00:18:13.200
at the memory configuration it comes

00:18:13.200 --> 00:18:14.960
with a local embedding model where is

00:18:14.960 --> 00:18:16.080
the memory

00:18:16.080 --> 00:18:19.679
>> what does what does embedding mean here

00:18:19.679 --> 00:18:21.600
>> yeah so embedding models are basically

00:18:21.600 --> 00:18:23.840
converting text into token into into uh

00:18:23.840 --> 00:18:28.160
token arrays right so um a vector so so

00:18:28.160 --> 00:18:30.000
text becomes tokens tokens becomes a

00:18:30.000 --> 00:18:31.760
vector that gets stores into a vector

00:18:31.760 --> 00:18:34.880
database for semantic lookups.

00:18:34.880 --> 00:18:37.120
>> What what's the benefit of that?

00:18:37.120 --> 00:18:41.200
>> If I say cat and um it can it can relate

00:18:41.200 --> 00:18:42.960
to house cat even though they're not

00:18:42.960 --> 00:18:46.000
lexically close the concept of cat dog

00:18:46.000 --> 00:18:48.000
is is is semantically close to house

00:18:48.000 --> 00:18:50.320
pet. So converting a word like cat or

00:18:50.320 --> 00:18:53.120
dog to tokens and then calculating uh

00:18:53.120 --> 00:18:55.200
using an embedding model. So you have to

00:18:55.200 --> 00:18:56.799
use the same embedding model that you

00:18:56.799 --> 00:18:59.679
that you um store the data with the

00:18:59.679 --> 00:19:01.679
vectors with because it consistently

00:19:01.679 --> 00:19:03.919
converts all of this. It's it's a small

00:19:03.919 --> 00:19:08.320
open uh weights model that um gets text

00:19:08.320 --> 00:19:10.640
and then converts them into vectors and

00:19:10.640 --> 00:19:12.160
then those vectors they have like

00:19:12.160 --> 00:19:15.360
similarity. So cat, dog and house pet

00:19:15.360 --> 00:19:18.000
all have very close semantic meanings.

00:19:18.000 --> 00:19:19.679
So their vectors will be very close to

00:19:19.679 --> 00:19:21.840
each other. So if you do you can imagine

00:19:21.840 --> 00:19:24.080
like if you if you imagine two

00:19:24.080 --> 00:19:25.600
dimensional vectors you can calculate

00:19:25.600 --> 00:19:27.120
the distance between them easily right

00:19:27.120 --> 00:19:30.080
but these are vectors with like 1,024

00:19:30.080 --> 00:19:33.360
dimensions it's very hard to imagine how

00:19:33.360 --> 00:19:35.200
uh you know but you can calculate the

00:19:35.200 --> 00:19:37.039
similarity between them. So, so that's

00:19:37.039 --> 00:19:39.039
what an embedding model does and it

00:19:39.039 --> 00:19:40.559
means that if you have some type of

00:19:40.559 --> 00:19:44.640
semantic um you know lookups you need to

00:19:44.640 --> 00:19:47.039
run an embedding model uh when you get

00:19:47.039 --> 00:19:49.760
text store it into a vector database and

00:19:49.760 --> 00:19:51.440
then when you want to query the database

00:19:51.440 --> 00:19:55.520
you need to um embed that query send it

00:19:55.520 --> 00:19:56.880
um to the vector store to do a

00:19:56.880 --> 00:19:58.400
similarity lookup and then it's going to

00:19:58.400 --> 00:20:01.200
return facts about that are related to

00:20:01.200 --> 00:20:02.720
that query. So embedding is very

00:20:02.720 --> 00:20:04.799
important and running embedded locally

00:20:04.799 --> 00:20:06.960
on your machine is kind of it's

00:20:06.960 --> 00:20:08.559
happening. You have a lot of small

00:20:08.559 --> 00:20:11.039
little um models that are running on

00:20:11.039 --> 00:20:12.640
your machines already. I Claude Code

00:20:12.640 --> 00:20:15.039
ships with the local um model. It it it

00:20:15.039 --> 00:20:16.799
it runs a local model to do the auto

00:20:16.799 --> 00:20:18.799
mode classifier. I'm pretty sure. Oh

00:20:18.799 --> 00:20:20.080
actually no

00:20:20.080 --> 00:20:21.679
>> actually no because because when you

00:20:21.679 --> 00:20:23.039
sometimes you get an error saying the

00:20:23.039 --> 00:20:24.720
classifier hits an API error or

00:20:24.720 --> 00:20:26.160
something. So it is it is remote.

00:20:26.160 --> 00:20:29.679
>> Yeah it is remote. But the on the topic

00:20:29.679 --> 00:20:32.159
of memories. So

00:20:32.159 --> 00:20:33.600
how do you maintain your memories? Cuz

00:20:33.600 --> 00:20:35.280
like markdown is cool because you can

00:20:35.280 --> 00:20:38.880
just like SCP your your your markdown

00:20:38.880 --> 00:20:41.919
into a new machine. How do you maintain

00:20:41.919 --> 00:20:45.919
your your embeddings database here? It

00:20:45.919 --> 00:20:48.480
doesn't. So ideally you don't right like

00:20:48.480 --> 00:20:50.799
>> so if ideally that's where where these

00:20:50.799 --> 00:20:52.720
these memory systems have all kinds of

00:20:52.720 --> 00:20:54.640
like capabilities and cleanup jobs and

00:20:54.640 --> 00:20:57.039
and and and background functionality to

00:20:57.039 --> 00:20:59.440
maintain the memory right so I don't

00:20:59.440 --> 00:21:02.080
like with hindsight I don't even know if

00:21:02.080 --> 00:21:04.480
there's like a a chrome running to like

00:21:04.480 --> 00:21:06.960
identify any stale memories and and re

00:21:06.960 --> 00:21:09.200
reorganize the memories to see what's

00:21:09.200 --> 00:21:11.600
what's relevant and I al honestly don't

00:21:11.600 --> 00:21:13.200
really want to worry too much about it

00:21:13.200 --> 00:21:15.679
right um but you see how like that you

00:21:15.679 --> 00:21:17.039
miss a lot of that when you have simple

00:21:17.039 --> 00:21:18.720
markdown on disk, right?

00:21:18.720 --> 00:21:21.440
>> Yeah. Yeah. This is this is why I kind

00:21:21.440 --> 00:21:23.520
of I I always kind of gravitate back

00:21:23.520 --> 00:21:25.679
down to markdown because I can

00:21:25.679 --> 00:21:28.400
understand it in my human monkey brain.

00:21:28.400 --> 00:21:30.720
>> So, here it is. Embedding is using llama

00:21:30.720 --> 00:21:33.600
cpp. I want to find because it actually

00:21:33.600 --> 00:21:36.400
lists out the local models that that KO

00:21:36.400 --> 00:21:38.799
ships with and it runs. So, I can't find

00:21:38.799 --> 00:21:40.320
the model. I think it's in security. If

00:21:40.320 --> 00:21:42.480
I go to settings and security, I think

00:21:42.480 --> 00:21:44.960
the auto mode classifier

00:21:44.960 --> 00:21:47.600
um is somewhere in the settings. So,

00:21:47.600 --> 00:21:49.039
it's it's very interesting to go through

00:21:49.039 --> 00:21:50.960
and and and figure that out. Uh but I

00:21:50.960 --> 00:21:53.440
think it runs at least two or three.

00:21:53.440 --> 00:21:55.280
>> But the next burning question in my mind

00:21:55.280 --> 00:21:57.679
is what's the the multiplayer aspect? Is

00:21:57.679 --> 00:22:01.200
it just through just are you supposed to

00:22:01.200 --> 00:22:03.760
is one person supposed to admin this and

00:22:03.760 --> 00:22:05.840
then your team is on Slack or something

00:22:05.840 --> 00:22:09.520
or or was there some other usage

00:22:09.520 --> 00:22:14.080
paradigm that's supposed to happen here?

00:22:14.080 --> 00:22:17.919
>> I do know that Kiro um when you use it,

00:22:17.919 --> 00:22:20.880
it has this artifacts capability and I

00:22:20.880 --> 00:22:23.280
know if you look at the artifact, you

00:22:23.280 --> 00:22:27.360
you can leave comments on things. Um,

00:22:27.360 --> 00:22:30.880
and I do think that this comments thing

00:22:30.880 --> 00:22:32.240
I feel like that's really something that

00:22:32.240 --> 00:22:35.360
is a multiplayer capability. Um, but

00:22:35.360 --> 00:22:37.600
>> so that means every everyone in your

00:22:37.600 --> 00:22:38.720
team runs

00:22:38.720 --> 00:22:40.960
>> I don't know if you can like I don't I

00:22:40.960 --> 00:22:43.600
guess as far as I can see right now the

00:22:43.600 --> 00:22:45.520
gateway doesn't have like a multi-tenant

00:22:45.520 --> 00:22:49.200
or multi-user um capability. When I look

00:22:49.200 --> 00:22:51.200
at these and I want to click publish to

00:22:51.200 --> 00:22:54.159
share it with the team, it's going to

00:22:54.159 --> 00:22:58.240
use a skill to if this is a box running

00:22:58.240 --> 00:23:00.159
in my AWS, it's going to try and publish

00:23:00.159 --> 00:23:02.480
it through like a cloud front or S3. So

00:23:02.480 --> 00:23:04.240
the cool thing is that everything is

00:23:04.240 --> 00:23:05.919
yours, right? If you use Claude Code and

00:23:05.919 --> 00:23:07.840
you create an artifact and then you it

00:23:07.840 --> 00:23:10.720
gets like a cloud hosted URL publishing

00:23:10.720 --> 00:23:12.720
it is all controlled by entropic. This

00:23:12.720 --> 00:23:14.799
is all your box, your memories, your

00:23:14.799 --> 00:23:16.159
your S3, your account.

00:23:16.159 --> 00:23:18.159
>> The power of AWS, they can sort of

00:23:18.159 --> 00:23:20.000
connect the dots, can't they?

00:23:20.000 --> 00:23:21.600
>> Yeah. And then

00:23:21.600 --> 00:23:23.039
>> with your

00:23:23.039 --> 00:23:25.280
>> with this with this capability to just

00:23:25.280 --> 00:23:29.120
run one command Kiro Crew cloud and then

00:23:29.120 --> 00:23:31.280
you can just launch and then you you

00:23:31.280 --> 00:23:33.440
consider do you want light balance but

00:23:33.440 --> 00:23:34.880
you do realize that you need a

00:23:34.880 --> 00:23:37.200
significant amount of um CPU and memory

00:23:37.200 --> 00:23:38.320
because you're running all these local

00:23:38.320 --> 00:23:40.400
embedding models, right? you know, and

00:23:40.400 --> 00:23:42.159
you're running a whole memory system and

00:23:42.159 --> 00:23:45.200
and and more. So, so the light one is is

00:23:45.200 --> 00:23:46.159
not cheap.

00:23:46.159 --> 00:23:48.400
>> And I guess it's not serverless. The

00:23:48.400 --> 00:23:50.080
thing is running all the time, even if

00:23:50.080 --> 00:23:52.320
you don't use it, right?

00:23:52.320 --> 00:23:55.440
>> Even worse, I I went for the full p

00:23:55.440 --> 00:23:57.760
private uh instance. That means it was

00:23:57.760 --> 00:24:00.159
running on a private subnet, not public.

00:24:00.159 --> 00:24:01.520
And that means I had to have a net

00:24:01.520 --> 00:24:03.760
gateway. That means uh I get a whole

00:24:03.760 --> 00:24:05.200
bunch of extra costs, which is already

00:24:05.200 --> 00:24:07.520
$50 a month for like a hobby or a single

00:24:07.520 --> 00:24:09.840
user. That doesn't make sense, right?

00:24:09.840 --> 00:24:10.240
So,

00:24:10.240 --> 00:24:11.760
>> yeah. Yeah.

00:24:11.760 --> 00:24:13.840
>> Yeah. So, so I did set up scheduled

00:24:13.840 --> 00:24:16.000
scaling um which basically with

00:24:16.000 --> 00:24:18.960
EventBridge uh stops it at night in the

00:24:18.960 --> 00:24:20.400
evening and then scales it up again in

00:24:20.400 --> 00:24:21.440
the morning.

00:24:21.440 --> 00:24:24.400
>> But um but still it says the net gateway

00:24:24.400 --> 00:24:26.320
is really the the cause of your cost.

00:24:26.320 --> 00:24:28.000
>> Oh, this your own blog is it?

00:24:28.000 --> 00:24:29.760
>> Yeah, this was the blog post I posted. I

00:24:29.760 --> 00:24:32.559
have a I I want to post four parts. The

00:24:32.559 --> 00:24:34.720
first one was this initial cloud-based

00:24:34.720 --> 00:24:36.559
exploration. I did the same with Hermes,

00:24:36.559 --> 00:24:38.799
right? I just um you know spun up a

00:24:38.799 --> 00:24:41.520
Hermes instance, looked at how it felt,

00:24:41.520 --> 00:24:42.880
looked at what I could do with it,

00:24:42.880 --> 00:24:44.559
realized it's going to cost me $50 or

00:24:44.559 --> 00:24:46.559
more dollars a month. Definitely if I I

00:24:46.559 --> 00:24:48.240
wanted to have a memory system, you

00:24:48.240 --> 00:24:49.279
know, I wanted to have like the

00:24:49.279 --> 00:24:51.760
hindsight and all that uh fire crawl for

00:24:51.760 --> 00:24:52.799
for web search.

00:24:52.799 --> 00:24:54.799
>> Wait, why is it costing you $50? You

00:24:54.799 --> 00:24:56.799
don't have a you don't have a Pi or

00:24:56.799 --> 00:24:58.880
something? You don't have a VPS? What do

00:24:58.880 --> 00:24:59.760
you mean?

00:24:59.760 --> 00:25:02.080
>> Um yeah, this a different question. I'm

00:25:02.080 --> 00:25:04.000
talking about exploring and and getting

00:25:04.000 --> 00:25:06.720
started quickly. Um, I didn't have like

00:25:06.720 --> 00:25:09.600
I I I didn't have like

00:25:09.600 --> 00:25:11.600
pie lying around or anything like that.

00:25:11.600 --> 00:25:12.080
>> I see.

00:25:12.080 --> 00:25:14.799
>> Um, so I wanted to quickly, you know,

00:25:14.799 --> 00:25:18.240
try try this out. And and my quick start

00:25:18.240 --> 00:25:20.640
to try anything out is, you know, spin

00:25:20.640 --> 00:25:22.880
up an EC2 instance, do a bunch of

00:25:22.880 --> 00:25:23.520
things.

00:25:23.520 --> 00:25:24.960
>> If I don't like it, tear everything

00:25:24.960 --> 00:25:27.120
down. Uh, if I keep if I want to keep it

00:25:27.120 --> 00:25:28.240
running, it's going to cost me at least

00:25:28.240 --> 00:25:30.880
$50 a month. Um, but of course if I run

00:25:30.880 --> 00:25:32.159
it for an hour or two, it's going to

00:25:32.159 --> 00:25:35.279
cost me maybe $2, $3 for an experiment.

00:25:35.279 --> 00:25:37.360
I don't have to go and buy a a pie for

00:25:37.360 --> 00:25:39.840
that.

00:25:39.840 --> 00:25:43.760
>> Okay. So, um,

00:25:43.760 --> 00:25:46.080
>> N gateway and also funny um, because

00:25:46.080 --> 00:25:48.400
again the the the information for this

00:25:48.400 --> 00:25:51.120
blog post came mostly from cloud, but I

00:25:51.120 --> 00:25:52.559
actually rewrote it and wrote it all

00:25:52.559 --> 00:25:55.120
manually. I mean, large parts of it,

00:25:55.120 --> 00:25:56.960
most of it. [laughter]

00:25:56.960 --> 00:25:59.039
Why? Why did you have to rewrite it

00:25:59.039 --> 00:26:00.159
manually? Just because it

00:26:00.159 --> 00:26:01.679
>> cuz I wanted to make sure it wasn't to

00:26:01.679 --> 00:26:04.480
AI slop and it was actually um you know

00:26:04.480 --> 00:26:05.679
>> you tightened it up.

00:26:05.679 --> 00:26:07.760
>> Yeah, I tightened it up. Uh but the

00:26:07.760 --> 00:26:09.200
funny thing is that cloud is like Yeah,

00:26:09.200 --> 00:26:10.559
but you're running it in in in

00:26:10.559 --> 00:26:12.480
Singapore. In Singapore, the net gateway

00:26:12.480 --> 00:26:15.360
is costing you $40. The savings that you

00:26:15.360 --> 00:26:17.840
have on a nightly um if you scale it

00:26:17.840 --> 00:26:20.960
down night nightly is only three like

00:26:20.960 --> 00:26:23.039
$30. But if you would switch everything

00:26:23.039 --> 00:26:25.600
to the US East, you would uh already cut

00:26:25.600 --> 00:26:28.480
the bill by $10 without even the up down

00:26:28.480 --> 00:26:30.640
schedule, which is a significant

00:26:30.640 --> 00:26:32.400
difference, right? If you if you put it

00:26:32.400 --> 00:26:33.679
in proportion that way. I thought that

00:26:33.679 --> 00:26:36.559
was pretty interesting finding.

00:26:36.559 --> 00:26:36.960
>> Yeah.

00:26:36.960 --> 00:26:39.279
>> Yeah. I I'm I'm almost surprised how

00:26:39.279 --> 00:26:41.120
expensive the net gateway is. And it's

00:26:41.120 --> 00:26:43.200
it's just been a meme for a decade now,

00:26:43.200 --> 00:26:44.159
hasn't it?

00:26:44.159 --> 00:26:45.760
>> Yeah. So, there's two things I want to

00:26:45.760 --> 00:26:47.760
follow up on this article. It's to use

00:26:47.760 --> 00:26:50.799
fck-nat, right, which is the um

00:26:50.799 --> 00:26:52.640
>> Oh, yeah. You told me about that. You

00:26:52.640 --> 00:26:53.760
told me about that.

00:26:53.760 --> 00:26:55.840
>> Yeah, I use it in in dev environments at

00:26:55.840 --> 00:26:59.919
work. It's really cheap. Um, it's a T4G

00:26:59.919 --> 00:27:02.480
nano because ultimately if you don't

00:27:02.480 --> 00:27:05.360
need a high available multi- instance

00:27:05.360 --> 00:27:08.159
autoreovering high high throughput net

00:27:08.159 --> 00:27:10.799
gateway uh you can run on a single NAT

00:27:10.799 --> 00:27:14.480
and set up um what's this called? Net is

00:27:14.480 --> 00:27:16.880
really just a couple of uh

00:27:16.880 --> 00:27:18.640
>> IP tables is nothing.

00:27:18.640 --> 00:27:20.159
>> Yeah, exactly. You can just do it with

00:27:20.159 --> 00:27:20.960
IP tables.

00:27:20.960 --> 00:27:22.640
>> If AWS announced,

00:27:22.640 --> 00:27:23.600
>> that's great.

00:27:23.600 --> 00:27:26.640
>> That Net Gateway was being folded into

00:27:26.640 --> 00:27:29.039
something and you wouldn't be in charge

00:27:29.039 --> 00:27:30.880
for it. It would be so great.

00:27:30.880 --> 00:27:32.720
>> But anyway,

00:27:32.720 --> 00:27:35.760
>> yeah. So, great thing about Riverside is

00:27:35.760 --> 00:27:37.360
it connects. It records everything

00:27:37.360 --> 00:27:38.960
locally. It's just going to be weird in

00:27:38.960 --> 00:27:41.600
the edit. Anyway, [laughter]

00:27:41.600 --> 00:27:43.760
back back to Kiro Crew. So, already like

00:27:43.760 --> 00:27:45.279
this is great. Like I told you about all

00:27:45.279 --> 00:27:46.960
of the capabilities comes with out of

00:27:46.960 --> 00:27:48.559
the box, but how does it feel to

00:27:48.559 --> 00:27:51.440
actually use it? So

00:27:51.440 --> 00:27:53.600
I I tried some things out. For example,

00:27:53.600 --> 00:27:57.840
I was uh I was on a call. Let's uh close

00:27:57.840 --> 00:28:01.120
the sessions. Um and I wanted to ask a

00:28:01.120 --> 00:28:04.399
question about um agent core some

00:28:04.399 --> 00:28:08.159
capabilities there. So I was asking here

00:28:08.159 --> 00:28:10.720
it gave me some diagrams but then I

00:28:10.720 --> 00:28:13.679
asked it to to create

00:28:13.679 --> 00:28:15.200
I hope that it would automatically

00:28:15.200 --> 00:28:16.640
create like I wanted the sequence

00:28:16.640 --> 00:28:18.159
diagram because it shows these things

00:28:18.159 --> 00:28:20.240
and I want it as a sequence diagram. I

00:28:20.240 --> 00:28:22.000
was hoping it would create an artifact

00:28:22.000 --> 00:28:24.159
but it split out a bunch of mermaid. I

00:28:24.159 --> 00:28:26.000
said now please just create a proper

00:28:26.000 --> 00:28:28.320
artifact and um

00:28:28.320 --> 00:28:31.039
>> what's a proper artifact? HTML basically

00:28:31.039 --> 00:28:33.120
I suppose. Yeah, the same like if you

00:28:33.120 --> 00:28:34.640
talk in Claude Code and you say create an

00:28:34.640 --> 00:28:36.080
artifact and then it opens up your

00:28:36.080 --> 00:28:38.559
browser with that HTML. Uh

00:28:38.559 --> 00:28:40.559
>> I don't really like mermaid diagrams but

00:28:40.559 --> 00:28:41.919
oh well they could be just

00:28:41.919 --> 00:28:43.600
>> yeah well even Claude Code is doing most

00:28:43.600 --> 00:28:45.360
of it uh when it creates the artifact

00:28:45.360 --> 00:28:47.600
it's using u mermaid or other diagram

00:28:47.600 --> 00:28:51.120
tools under the hood um but anyway so

00:28:51.120 --> 00:28:52.960
while I was talking it created these

00:28:52.960 --> 00:28:57.840
diagrams um and then you can ideally

00:28:57.840 --> 00:29:00.559
easily share them uh with your team

00:29:00.559 --> 00:29:02.640
right so this was pretty cool like I

00:29:02.640 --> 00:29:04.320
found that a very nice experience uh

00:29:04.320 --> 00:29:07.600
it's all locally um on my Yeah, those

00:29:07.600 --> 00:29:08.880
look good. Those look good.

00:29:08.880 --> 00:29:10.640
>> I click on artifacts. I can see all of

00:29:10.640 --> 00:29:12.960
the previous artifacts it has created.

00:29:12.960 --> 00:29:14.159
Um, for example, here

00:29:14.159 --> 00:29:16.720
>> we can do better in this regard.

00:29:16.720 --> 00:29:19.679
>> Well, yeah, but remember that here,

00:29:19.679 --> 00:29:21.120
okay, this is one thing I haven't told

00:29:21.120 --> 00:29:24.720
you yet. I'm I'm running this all off um

00:29:24.720 --> 00:29:28.080
my local fork which uses my ChatGPT $20

00:29:28.080 --> 00:29:30.240
subscription. It doesn't like it's not

00:29:30.240 --> 00:29:32.799
Claude Code. I'm not paying uh a lot of

00:29:32.799 --> 00:29:35.520
money here. Uh basically my my family is

00:29:35.520 --> 00:29:37.600
using the the ChatGPT app on their phone

00:29:37.600 --> 00:29:39.919
and they're very happy with it. Uh and

00:29:39.919 --> 00:29:42.799
I'm using it to generate you know code

00:29:42.799 --> 00:29:44.080
on my my machine.

00:29:44.080 --> 00:29:45.679
>> So that's what you did. You you

00:29:45.679 --> 00:29:48.480
basically forked Kro so that I can work

00:29:48.480 --> 00:29:52.480
with you with your own personal

00:29:52.480 --> 00:29:54.399
>> GPT subscription. So, if I go here in

00:29:54.399 --> 00:29:56.720
settings and I click on uh usage

00:29:56.720 --> 00:29:59.840
overview here, I can see here my uh it's

00:29:59.840 --> 00:30:03.039
my GPT plus and weekly reset is in six

00:30:03.039 --> 00:30:06.000
days and I've only used 8% of my limit

00:30:06.000 --> 00:30:07.919
>> um how many chats I have. So, all of

00:30:07.919 --> 00:30:10.480
these statistics are are my ChatGPT

00:30:10.480 --> 00:30:13.039
subscription and and I think this is the

00:30:13.039 --> 00:30:15.919
greatest thing about Kiro Crew. Um if

00:30:15.919 --> 00:30:18.799
you go back to that diagram here, where

00:30:18.799 --> 00:30:20.080
is it? It's

00:30:20.080 --> 00:30:22.159
>> so the whole thing's open source is it?

00:30:22.159 --> 00:30:25.440
It's fully open source Apache 2 and it's

00:30:25.440 --> 00:30:28.640
using this ACP protocol or agent client

00:30:28.640 --> 00:30:30.320
protocol which means

00:30:30.320 --> 00:30:34.240
>> they ship it with Kiro but you um can

00:30:34.240 --> 00:30:36.480
contribute your own. So what I did I

00:30:36.480 --> 00:30:42.240
replaced the Kiro CLI with Codex and

00:30:42.240 --> 00:30:45.760
as a result my whole Kir crew is using

00:30:45.760 --> 00:30:49.200
CEX under the hood instead of KO. Ah,

00:30:49.200 --> 00:30:50.880
now I understand why you've been excited

00:30:50.880 --> 00:30:52.799
to tell me about this. Yeah, that does

00:30:52.799 --> 00:30:54.320
sound quite cool.

00:30:54.320 --> 00:30:56.960
>> Yeah. So, I've I've I've showed it to

00:30:56.960 --> 00:30:58.559
the team because when they announced it,

00:30:58.559 --> 00:31:01.200
they said that um this is ACP, so you

00:31:01.200 --> 00:31:03.919
can use any uh compliant and and

00:31:03.919 --> 00:31:05.600
somebody asked I think like can we

00:31:05.600 --> 00:31:08.159
contribute our own other than Kira? And

00:31:08.159 --> 00:31:11.200
they said yeah. [laughter] And I was

00:31:11.200 --> 00:31:14.000
like okay you you said we can. It wasn't

00:31:14.000 --> 00:31:16.480
me. Somebody else said, so I said uh to

00:31:16.480 --> 00:31:19.279
Fable, what would it take to rewrite

00:31:19.279 --> 00:31:21.520
this thing and not use Kiro uh but

00:31:21.520 --> 00:31:24.320
instead use use Codex and use this ACP?

00:31:24.320 --> 00:31:26.399
So I wanted to know and it's actually

00:31:26.399 --> 00:31:28.000
very interesting because this agent

00:31:28.000 --> 00:31:33.200
client protocol um is a way of like the

00:31:33.200 --> 00:31:35.279
client here is the gateway, right?

00:31:35.279 --> 00:31:38.559
Gateway is the client to the agent. So

00:31:38.559 --> 00:31:41.120
so the gateway advertises capabilities

00:31:41.120 --> 00:31:43.600
of what it can show to the user. Um, for

00:31:43.600 --> 00:31:46.159
example, elicitation. When I ask a

00:31:46.159 --> 00:31:49.840
question, um, it may ask like when the

00:31:49.840 --> 00:31:51.039
user asks a question through the

00:31:51.039 --> 00:31:53.360
gateway, the model might decide that

00:31:53.360 --> 00:31:55.360
it's better to get like, um, you know,

00:31:55.360 --> 00:31:56.960
feedback before it answers a question.

00:31:56.960 --> 00:31:58.320
So, it gives you a couple of options,

00:31:58.320 --> 00:32:00.159
right? Um, that's that's a ask user

00:32:00.159 --> 00:32:03.039
question tool in in in in um in cloud

00:32:03.039 --> 00:32:07.760
code. Um, so this elicitation is a

00:32:07.760 --> 00:32:10.240
defined capability in the agent client

00:32:10.240 --> 00:32:13.519
protocol. So there's a difference

00:32:13.519 --> 00:32:18.399
between the capabilities of Kira and

00:32:18.399 --> 00:32:22.399
Codex. So Codex implements elicitation.

00:32:22.399 --> 00:32:27.679
KO does not yet implement it. Okay. So

00:32:27.679 --> 00:32:30.240
what does that mean is that the gateway

00:32:30.240 --> 00:32:32.799
uh currently advertises that it can

00:32:32.799 --> 00:32:35.679
display. It doesn't actually eliminate

00:32:35.679 --> 00:32:38.320
elicitation as a capability. So, Codex

00:32:38.320 --> 00:32:39.840
assumes that the gateway can can

00:32:39.840 --> 00:32:41.919
display. That means that Codex will come

00:32:41.919 --> 00:32:44.000
back to the gateway and say, "Show the

00:32:44.000 --> 00:32:46.480
user a cart with three options and the

00:32:46.480 --> 00:32:48.640
gateway um didn't say that it doesn't

00:32:48.640 --> 00:32:50.880
support it." So, um it actually right

00:32:50.880 --> 00:32:53.120
now just responds with a 500 like

00:32:53.120 --> 00:32:55.840
internal server error. Um it actually

00:32:55.840 --> 00:32:58.320
responds with a reject code which keep

00:32:58.320 --> 00:33:01.519
which Codex inter interprets as it was

00:33:01.519 --> 00:33:03.919
a security rejection and it cancels the

00:33:03.919 --> 00:33:06.000
request. So that was kind of like the

00:33:06.000 --> 00:33:09.519
unexpected um you know kind of problems

00:33:09.519 --> 00:33:12.399
you run into because of um that was one

00:33:12.399 --> 00:33:14.159
thing. Also the whole usage like

00:33:14.159 --> 00:33:16.880
dashboard that I just showed you it

00:33:16.880 --> 00:33:19.519
didn't work because the way that Kro ACP

00:33:19.519 --> 00:33:22.559
um sends back usage information is

00:33:22.559 --> 00:33:24.320
different from the way that Codex does.

00:33:24.320 --> 00:33:26.080
So I had to like fable highlight it

00:33:26.080 --> 00:33:27.919
exactly like where the problem lies like

00:33:27.919 --> 00:33:29.679
the the the Codex back end is sending

00:33:29.679 --> 00:33:32.240
the usage with every message and I said

00:33:32.240 --> 00:33:35.200
like how about we we create like um um

00:33:35.200 --> 00:33:37.039
what do you call this a facade or like

00:33:37.039 --> 00:33:39.679
an interface on top of it. So I'm I'm

00:33:39.679 --> 00:33:42.399
using Codex ACP under the hood but I'm

00:33:42.399 --> 00:33:44.000
exposing it as with the same

00:33:44.000 --> 00:33:46.080
capabilities as KO like Codex says I

00:33:46.080 --> 00:33:47.760
can't show you usage. So when you go to

00:33:47.760 --> 00:33:50.399
usage the dashboard is empty but I said

00:33:50.399 --> 00:33:52.720
hey CEX there is usage information just

00:33:52.720 --> 00:33:55.279
present yourself as the hero usage and

00:33:55.279 --> 00:33:57.600
and under the hood it actually you know

00:33:57.600 --> 00:34:00.000
when codex sends them the message it it

00:34:00.000 --> 00:34:01.519
captures the usage information and

00:34:01.519 --> 00:34:04.240
exposes it uh as a side channel so it

00:34:04.240 --> 00:34:06.240
still works. Yeah. So that's the kind

00:34:06.240 --> 00:34:07.679
the kind of cool stuff that I learned

00:34:07.679 --> 00:34:09.359
doing this, right? Because yeah,

00:34:09.359 --> 00:34:10.639
obviously I don't want to maintain a

00:34:10.639 --> 00:34:12.720
fork like I was actually thinking I set

00:34:12.720 --> 00:34:14.960
up this KU instance to to automatically

00:34:14.960 --> 00:34:16.720
maintain its own fork and every time

00:34:16.720 --> 00:34:19.359
rebases on top of the upstream uh and

00:34:19.359 --> 00:34:21.679
and and and things like that. But uh

00:34:21.679 --> 00:34:24.320
thankfully the the AWS team is actually

00:34:24.320 --> 00:34:26.879
living up to their promise and um and

00:34:26.879 --> 00:34:28.960
they're like you know evaluating because

00:34:28.960 --> 00:34:30.879
they're also saying why don't we just if

00:34:30.879 --> 00:34:32.560
you do the Codex authentication you get

00:34:32.560 --> 00:34:34.960
a bearer token. What if we just like use

00:34:34.960 --> 00:34:37.839
Akira ACP um but you use a Codex like

00:34:37.839 --> 00:34:40.000
bearer token that you get through the O

00:34:40.000 --> 00:34:42.639
flow um and that might be another thing

00:34:42.639 --> 00:34:43.919
that you could do instead of just

00:34:43.919 --> 00:34:46.079
replacing the whole ECP back end. I just

00:34:46.079 --> 00:34:47.919
did it for my like my own exploration

00:34:47.919 --> 00:34:49.839
like what is HCP? What is the impact?

00:34:49.839 --> 00:34:52.240
How does that you know manifest itself?

00:34:52.240 --> 00:34:54.159
It was very cool.

00:34:54.159 --> 00:34:55.599
>> Yeah, this this sounds really

00:34:55.599 --> 00:34:57.119
interesting. Yeah, thanks for

00:34:57.119 --> 00:34:58.640
introducing me to agent decline

00:34:58.640 --> 00:35:00.720
protocol. But wait but wait a minute.

00:35:00.720 --> 00:35:03.760
Um, sorry Vince, I was just daydreaming

00:35:03.760 --> 00:35:06.800
there in about using it myself. So you

00:35:06.800 --> 00:35:08.720
said you didn't want to maintain a fork.

00:35:08.720 --> 00:35:11.119
So So going forward here, if I wanted to

00:35:11.119 --> 00:35:15.599
get started using my own um Codex or

00:35:15.599 --> 00:35:18.240
clawed subscription, what what changes

00:35:18.240 --> 00:35:21.920
would I need to make on on the on on Kro

00:35:21.920 --> 00:35:24.400
to unlock that?

00:35:24.400 --> 00:35:27.520
Can I just patch it using your thing or

00:35:27.520 --> 00:35:29.920
do I just launch Fable and ask it to

00:35:29.920 --> 00:35:31.359
make a fork?

00:35:31.359 --> 00:35:33.599
>> So, I have a public fork um that I've

00:35:33.599 --> 00:35:36.320
shared with the team and um shown them

00:35:36.320 --> 00:35:39.680
what works. I identified the gaps um you

00:35:39.680 --> 00:35:42.240
know where where the Codex ACP protocol

00:35:42.240 --> 00:35:44.320
and the Kiro ACP protocol did not match

00:35:44.320 --> 00:35:46.400
up and the ways that I worked around it

00:35:46.400 --> 00:35:49.599
so that I make things work and um my

00:35:49.599 --> 00:35:51.520
fork currently is public so anyone can

00:35:51.520 --> 00:35:54.720
just build a copy of Kiro Crew and run

00:35:54.720 --> 00:35:58.400
it with with Codex uh if they want um I

00:35:58.400 --> 00:36:00.480
would not recommend building a co cloud

00:36:00.480 --> 00:36:03.200
code ACP in fact there is there is a

00:36:03.200 --> 00:36:05.040
bunch of code stripped out so internally

00:36:05.040 --> 00:36:07.359
within AWS they they run kir crew with

00:36:07.359 --> 00:36:10.079
Claude Code ACP I believe. Um but they've

00:36:10.079 --> 00:36:11.280
stripped it out because if you do that

00:36:11.280 --> 00:36:13.119
on a subscription you will get your

00:36:13.119 --> 00:36:16.640
account banned. Um so but but OpenAI it

00:36:16.640 --> 00:36:18.079
tells you you can do it. So if you want

00:36:18.079 --> 00:36:20.800
to run Kira crew with Codex ACP today

00:36:20.800 --> 00:36:23.440
you can use my fork build your own

00:36:23.440 --> 00:36:26.480
instance and use it. Hopefully in the

00:36:26.480 --> 00:36:28.880
next few weeks if if they if they do

00:36:28.880 --> 00:36:30.800
believe this is a good addition it might

00:36:30.800 --> 00:36:33.280
get merged upstream right and then you

00:36:33.280 --> 00:36:34.640
just have a single Kiro crew and then

00:36:34.640 --> 00:36:37.200
you can choose either Codex or Kiro

00:36:37.200 --> 00:36:38.640
>> there's definitely a gap for the market

00:36:38.640 --> 00:36:41.359
I mean I was always kind of assuming

00:36:41.359 --> 00:36:43.599
that open code or some somebody else was

00:36:43.599 --> 00:36:46.800
kind of filling it up but like if if AWS

00:36:46.800 --> 00:36:50.560
is is pushing is is publishing good

00:36:50.560 --> 00:36:53.920
stuff with the Apache 2 license

00:36:53.920 --> 00:36:55.839
I can't help but I think that it's it's

00:36:55.839 --> 00:36:59.599
on it's it's a very good game plan.

00:36:59.599 --> 00:37:00.160
Very good.

00:37:00.160 --> 00:37:01.760
>> I want to show you this elicitation

00:37:01.760 --> 00:37:05.200
thing um because this the fix was very

00:37:05.200 --> 00:37:06.800
interesting actually and you can see it

00:37:06.800 --> 00:37:08.960
at work here

00:37:08.960 --> 00:37:10.880
>> rather.

00:37:10.880 --> 00:37:13.440
>> Yeah. So instead of showing a card here

00:37:13.440 --> 00:37:15.599
that that this is the realitation. So

00:37:15.599 --> 00:37:18.160
normally Codex would reply with like a

00:37:18.160 --> 00:37:20.160
um a schema of what should be in the

00:37:20.160 --> 00:37:22.640
form. It should show you the options and

00:37:22.640 --> 00:37:24.560
then the user can click on one of those

00:37:24.560 --> 00:37:26.240
options and then the option gets fed

00:37:26.240 --> 00:37:27.920
back into the gateway gets goes back to

00:37:27.920 --> 00:37:30.720
Codex. But it's not supported. So

00:37:30.720 --> 00:37:32.320
what's what's happening right now? I

00:37:32.320 --> 00:37:34.079
smell something weird like [snorts]

00:37:34.079 --> 00:37:36.720
smoke. I hope I'm not on fire.

00:37:36.720 --> 00:37:39.599
>> Um anyway, I'll go check in a minute.

00:37:39.599 --> 00:37:40.720
It's [laughter] like this it's like this

00:37:40.720 --> 00:37:42.480
meme where the guy's on fire and he's

00:37:42.480 --> 00:37:44.880
like sending message fire.

00:37:44.880 --> 00:37:47.440
>> It's it's always a good idea to nip the

00:37:47.440 --> 00:37:49.119
fire in the bud. Just go and have a

00:37:49.119 --> 00:37:51.119
look. Christ. Maybe it's the car that

00:37:51.119 --> 00:37:53.040
was on that the car alarm and maybe it

00:37:53.040 --> 00:37:54.560
was on fire. Maybe that's why the car

00:37:54.560 --> 00:37:57.440
alarm was going off. [laughter]

00:37:57.440 --> 00:37:59.920
>> It's fine. Yeah. Anyway, so it's fine.

00:37:59.920 --> 00:38:02.240
>> What what what's happened the fix that

00:38:02.240 --> 00:38:04.560
that Fable suggested was just for the

00:38:04.560 --> 00:38:06.079
gateway, you know, remember the gateway

00:38:06.079 --> 00:38:08.480
is the client here to just not clearly

00:38:08.480 --> 00:38:10.160
advertise that it does not support

00:38:10.160 --> 00:38:12.960
elicitation. And when Codex knows that

00:38:12.960 --> 00:38:14.720
it does not support elicitation, but it

00:38:14.720 --> 00:38:16.960
wants to get feedback, it actually sends

00:38:16.960 --> 00:38:19.920
back a structured response um with just

00:38:19.920 --> 00:38:21.839
a list of options. And that's what

00:38:21.839 --> 00:38:23.839
happened here. So I was like, I want to

00:38:23.839 --> 00:38:26.160
build ask us a question like I want to

00:38:26.160 --> 00:38:29.200
have that nice interactivity of options

00:38:29.200 --> 00:38:31.839
>> and uh and then what I realized is

00:38:31.839 --> 00:38:34.320
>> I didn't have to build it because the

00:38:34.320 --> 00:38:36.720
gateway is able to show these options

00:38:36.720 --> 00:38:38.960
right here as little chips above my

00:38:38.960 --> 00:38:40.400
chat. Can you click it and and add

00:38:40.400 --> 00:38:41.920
notes? Oh, I see.

00:38:41.920 --> 00:38:43.119
>> So when you when you click it, you can

00:38:43.119 --> 00:38:45.119
even click both of them and then it go

00:38:45.119 --> 00:38:46.560
they go inside here and then you can

00:38:46.560 --> 00:38:47.119
just modify.

00:38:47.119 --> 00:38:48.560
>> Interesting implementation.

00:38:48.560 --> 00:38:49.920
>> It's a very Yeah. And I thought like

00:38:49.920 --> 00:38:51.920
this is better than than than ask user

00:38:51.920 --> 00:38:53.760
question, right? I was trying to build

00:38:53.760 --> 00:38:55.599
something I knew. Um but there was a

00:38:55.599 --> 00:38:57.839
better pattern that already worked. So

00:38:57.839 --> 00:38:59.280
that's kind of also something I learned,

00:38:59.280 --> 00:39:01.839
right? Sometimes you got to really use

00:39:01.839 --> 00:39:03.760
the system and understand how it works

00:39:03.760 --> 00:39:05.680
before you try to make it work exactly

00:39:05.680 --> 00:39:07.440
the way that you know.

00:39:07.440 --> 00:39:10.240
>> Yeah. Yeah, I mean it is strange to

00:39:10.240 --> 00:39:11.920
think I mean we've talked about ask

00:39:11.920 --> 00:39:14.000
question tool

00:39:14.000 --> 00:39:16.160
a number of times

00:39:16.160 --> 00:39:19.040
but it's one of those things that we we

00:39:19.040 --> 00:39:21.520
will keep on revisiting I suppose

00:39:21.520 --> 00:39:25.359
because the way you use it has so many

00:39:25.359 --> 00:39:28.560
nuances and uh I don't think we've even

00:39:28.560 --> 00:39:31.119
touched on the multiplayer

00:39:31.119 --> 00:39:34.160
u aspect really because ultimately the

00:39:34.160 --> 00:39:36.320
the ask user tool probably is like one

00:39:36.320 --> 00:39:38.720
of the major the

00:39:38.720 --> 00:39:41.359
decision points in your decision tree

00:39:41.359 --> 00:39:44.079
about how you get to your end product or

00:39:44.079 --> 00:39:46.000
somehow

00:39:46.000 --> 00:39:48.079
it is quite fascinating to see where

00:39:48.079 --> 00:39:51.760
those those turns were made. Yeah. So

00:39:51.760 --> 00:39:54.240
here is the interaction and maybe you

00:39:54.240 --> 00:39:55.680
can say well this is like what the chat

00:39:55.680 --> 00:39:58.079
GPT app gives me right but again it runs

00:39:58.079 --> 00:39:59.839
on your box your memory system all your

00:39:59.839 --> 00:40:02.000
data. The second thing is you have the

00:40:02.000 --> 00:40:04.640
context window, you know, tracer here.

00:40:04.640 --> 00:40:06.240
You can choose the models and because I

00:40:06.240 --> 00:40:08.160
build codecs, I have all of the CHP

00:40:08.160 --> 00:40:08.640
models.

00:40:08.640 --> 00:40:10.000
>> Can you show the tracer? I'm just

00:40:10.000 --> 00:40:12.240
curious because I' often find the

00:40:12.240 --> 00:40:14.960
tracing implementation is

00:40:14.960 --> 00:40:15.599
>> what tracing

00:40:15.599 --> 00:40:18.240
>> sometimes not what I expected to be.

00:40:18.240 --> 00:40:19.920
>> What is the tracing?

00:40:19.920 --> 00:40:22.480
>> Oh, like when you see the the actual

00:40:22.480 --> 00:40:25.119
calls and uh the thinking and things

00:40:25.119 --> 00:40:26.800
like that.

00:40:26.800 --> 00:40:28.400
Oh, maybe it's not there. Ah, you mean

00:40:28.400 --> 00:40:30.000
the thinking the thinking steps that it

00:40:30.000 --> 00:40:31.520
did? Okay. Sorry, I thought you wanted

00:40:31.520 --> 00:40:34.160
to see like telemetry. So, is there some

00:40:34.160 --> 00:40:35.920
thinking? Okay, there's no there's been

00:40:35.920 --> 00:40:37.680
no thinking on this chat, but let's go.

00:40:37.680 --> 00:40:39.200
What should the agent elicit from the

00:40:39.200 --> 00:40:44.160
user uh buying a car options? And let's

00:40:44.160 --> 00:40:45.440
stop with the questions and have it

00:40:45.440 --> 00:40:47.520
build an artifact.

00:40:47.520 --> 00:40:49.200
Should the walk through expose the

00:40:49.200 --> 00:40:52.720
protocol? Enough questions.

00:40:52.720 --> 00:40:56.933
Build the artifact. Oh my god.

00:40:56.933 --> 00:40:59.040
>> [laughter]

00:40:59.040 --> 00:41:02.000
>> So yeah, so here's a thinking right. So

00:41:02.000 --> 00:41:03.599
you can see the thinking box at the top

00:41:03.599 --> 00:41:04.319
here.

00:41:04.319 --> 00:41:05.280
>> Pretty cool.

00:41:05.280 --> 00:41:07.119
>> Yeah. And then this is the this is one

00:41:07.119 --> 00:41:09.119
of the main features uh from Kiro Crew

00:41:09.119 --> 00:41:11.359
as well uh which is like security.

00:41:11.359 --> 00:41:12.880
There's this embedding model and

00:41:12.880 --> 00:41:14.960
autoclassifier. Sorry not embedding it's

00:41:14.960 --> 00:41:17.520
an local autoclassifier. So right now

00:41:17.520 --> 00:41:21.280
it's it's um it says you can say trust

00:41:21.280 --> 00:41:23.680
this particular action or trust all. I'm

00:41:23.680 --> 00:41:25.280
just going to go by the most secure way,

00:41:25.280 --> 00:41:28.373
which is trust everything. [laughter]

00:41:28.373 --> 00:41:29.680
[snorts]

00:41:29.680 --> 00:41:31.359
>> This is interesting. It's really

00:41:31.359 --> 00:41:32.400
interesting.

00:41:32.400 --> 00:41:35.200
>> Yeah, but like if you only managed like

00:41:35.200 --> 00:41:37.760
if you only used um Hermes through

00:41:37.760 --> 00:41:40.079
Discord like I have, this is a breath of

00:41:40.079 --> 00:41:41.680
fresh air. And I haven't even touched

00:41:41.680 --> 00:41:42.880
and like because of all of the

00:41:42.880 --> 00:41:44.240
capabilities that it comes with, right,

00:41:44.240 --> 00:41:46.160
with the taskr runner, the the knowledge

00:41:46.160 --> 00:41:48.079
base, the schedule, and then you have

00:41:48.079 --> 00:41:50.960
the whole apps ecosystem. So they have a

00:41:50.960 --> 00:41:53.200
whole SDK and I haven't tried any of

00:41:53.200 --> 00:41:57.599
these but you can have um like there's

00:41:57.599 --> 00:42:00.480
extensions where you have more than one

00:42:00.480 --> 00:42:03.440
agent and so you can have a whole um

00:42:03.440 --> 00:42:05.359
animation and a world where you see like

00:42:05.359 --> 00:42:07.440
oh suddenly your review agent woke up

00:42:07.440 --> 00:42:09.359
because a new issue has been raised or a

00:42:09.359 --> 00:42:10.960
new pull request has been raised and you

00:42:10.960 --> 00:42:13.040
see that agent moving towards a desk to

00:42:13.040 --> 00:42:14.720
start reviewing the things like those

00:42:14.720 --> 00:42:16.240
are the things that people show and I

00:42:16.240 --> 00:42:18.160
think this is a bit stupid but it looks

00:42:18.160 --> 00:42:20.160
cute right and it's great great visual

00:42:20.160 --> 00:42:22.240
And it's good advertisement, right? But

00:42:22.240 --> 00:42:24.480
I haven't enabled any of that. Like not

00:42:24.480 --> 00:42:25.520
yet.

00:42:25.520 --> 00:42:26.560
>> Not at this point.

00:42:26.560 --> 00:42:30.319
>> Yeah, I've got Oh, this man. Yeah, this

00:42:30.319 --> 00:42:31.920
is this is a great little development.

00:42:31.920 --> 00:42:35.040
So, the title of this podcast should

00:42:35.040 --> 00:42:38.240
definitely just be agent context agent

00:42:38.240 --> 00:42:40.400
client protocol. Sorry.

00:42:40.400 --> 00:42:43.200
>> Yeah, agent client protocol. It's a

00:42:43.200 --> 00:42:45.520
thing and it's really interesting. Well,

00:42:45.520 --> 00:42:47.440
it's a thing that powers Kiro and it's

00:42:47.440 --> 00:42:49.760
a major rewrite that AWS did on how they

00:42:49.760 --> 00:42:51.680
built Kiro because originally they had

00:42:51.680 --> 00:42:54.000
three surfaces. One was the Kiro IDE

00:42:54.000 --> 00:42:55.839
which is like a VS code IDE plus

00:42:55.839 --> 00:42:58.000
specdriven development. Then they had

00:42:58.000 --> 00:43:01.200
the Kiro CLI which is a renamed QLI and

00:43:01.200 --> 00:43:03.520
then they had the cloud Kiro um

00:43:03.520 --> 00:43:06.160
offering like a a cloud service and each

00:43:06.160 --> 00:43:08.000
one of them had to had a separate

00:43:08.000 --> 00:43:10.079
codebase and had to be redeveloped. So

00:43:10.079 --> 00:43:11.520
what they did is they rerollled all of

00:43:11.520 --> 00:43:14.160
it into a single harness um that

00:43:14.160 --> 00:43:16.400
supports the agent client protocol and

00:43:16.400 --> 00:43:17.920
then they built different interfaces on

00:43:17.920 --> 00:43:20.240
top of it. So they built this um Kirk

00:43:20.240 --> 00:43:21.599
crew gateway on top of it. They built

00:43:21.599 --> 00:43:24.240
the Kiro CLI as one of the you know

00:43:24.240 --> 00:43:26.240
interfaces on top of the harness and

00:43:26.240 --> 00:43:28.400
they're also rewriting the um you know

00:43:28.400 --> 00:43:31.520
other capabilities there. So I do hope

00:43:31.520 --> 00:43:33.480
that's all public information.

00:43:33.480 --> 00:43:35.200
[laughter] I do think because they

00:43:35.200 --> 00:43:36.880
announced this uh they made this public

00:43:36.880 --> 00:43:38.800
and they're advertising. Uh, no. I've

00:43:38.800 --> 00:43:40.880
definitely seen the LinkedIn posts.

00:43:40.880 --> 00:43:43.040
>> Okay,

00:43:43.040 --> 00:43:44.000
>> let's end [clears throat] the podcast

00:43:44.000 --> 00:43:45.680
there because I I need to get on with

00:43:45.680 --> 00:43:46.400
some other

00:43:46.400 --> 00:43:48.800
>> Wait, you need to see the artifact.

00:43:48.800 --> 00:43:51.760
It has created the artifact for you.

00:43:51.760 --> 00:43:53.920
>> And then and then of course I got to put

00:43:53.920 --> 00:43:57.119
my security hat on and work out

00:43:57.119 --> 00:43:59.599
>> the wor so hard to create an ACP

00:43:59.599 --> 00:44:02.079
explanation for you using

00:44:02.079 --> 00:44:03.839
>> what the multiplayer what the

00:44:03.839 --> 00:44:07.280
multiplayer story is. So I really love

00:44:07.280 --> 00:44:09.040
this interactivity here. Like when you

00:44:09.040 --> 00:44:11.599
select it, you can just go quote. Um

00:44:11.599 --> 00:44:13.280
then the artifact has been built. You

00:44:13.280 --> 00:44:15.440
click on it, it opens on the side. You

00:44:15.440 --> 00:44:17.119
can make it full screen. You can leave

00:44:17.119 --> 00:44:17.920
some comments.

00:44:17.920 --> 00:44:21.280
>> It reminds me of claw claude code.

00:44:21.280 --> 00:44:22.960
>> I mean not claw code.

00:44:22.960 --> 00:44:24.400
>> Cloud desktop. Yeah.

00:44:24.400 --> 00:44:26.319
>> Yeah. Maybe maybe some

00:44:26.319 --> 00:44:28.720
>> Do you not use claw desktop?

00:44:28.720 --> 00:44:32.480
>> Not so much. No. No. I sometimes like I

00:44:32.480 --> 00:44:34.720
ask it to like fill out a PDF for me

00:44:34.720 --> 00:44:36.880
like

00:44:36.880 --> 00:44:38.319
>> Well, there's still there's still

00:44:38.319 --> 00:44:40.240
there's definitely value in an

00:44:40.240 --> 00:44:42.400
open-source Claude Code in my opinion.

00:44:42.400 --> 00:44:42.880
Definitely.

00:44:42.880 --> 00:44:44.400
>> This is pretty cool. Look at this. It

00:44:44.400 --> 00:44:48.480
created this like users start to turn

00:44:48.480 --> 00:44:50.960
artifact. Oh, I haven't seen actually.

00:44:50.960 --> 00:44:53.760
No, you can do that with claw desktop.

00:44:53.760 --> 00:44:55.760
>> It's really cool. I can see why you're

00:44:55.760 --> 00:44:58.079
excited about it.

00:44:58.079 --> 00:45:01.920
And then ideally I can publish this uh

00:45:01.920 --> 00:45:03.920
if I because right now this runs on a on

00:45:03.920 --> 00:45:06.880
a VM on my my um desktop at home and it

00:45:06.880 --> 00:45:08.960
doesn't have AWS credentials but if it

00:45:08.960 --> 00:45:10.960
runs on an EC2 box and that box has the

00:45:10.960 --> 00:45:12.720
right permissions it will publish this

00:45:12.720 --> 00:45:14.319
and makes this available through

00:45:14.319 --> 00:45:16.400
whatever like do you have a blog and do

00:45:16.400 --> 00:45:18.240
you you know host it on S3 with

00:45:18.240 --> 00:45:20.000
CloudFront then you can just directly

00:45:20.000 --> 00:45:22.400
publish your you know visualizations and

00:45:22.400 --> 00:45:23.599
share with other people right there.

00:45:23.599 --> 00:45:28.000
There's definitely a some some

00:45:28.000 --> 00:45:30.160
value a lot of value in having it open

00:45:30.160 --> 00:45:31.359
and

00:45:31.359 --> 00:45:32.720
>> this is quite exciting. Well, thanks

00:45:32.720 --> 00:45:36.560
thanks for letting showing me and

00:45:36.560 --> 00:45:38.480
I'm going to get back to my holiday now.

00:45:38.480 --> 00:45:41.040
I are you are you not taking any holiday

00:45:41.040 --> 00:45:42.640
in August or September?

00:45:42.640 --> 00:45:44.640
>> At the end of August and this is only

00:45:44.640 --> 00:45:46.240
one part of it, right? I I only showed

00:45:46.240 --> 00:45:48.640
you the artifacts. I haven't even

00:45:48.640 --> 00:45:50.079
started with all the extensions

00:45:50.079 --> 00:45:51.920
capabilities or the schedule or the

00:45:51.920 --> 00:45:53.680
taskr runner. I haven't even given it

00:45:53.680 --> 00:45:55.520
the GitHub identity yet. Like this is

00:45:55.520 --> 00:45:57.280
supposed to replace my Harness box,

00:45:57.280 --> 00:45:58.880
right? And there's so much more that I

00:45:58.880 --> 00:46:01.520
have to do. Uh but I've had so much fun

00:46:01.520 --> 00:46:03.760
already with it. Um but anyway,

00:46:03.760 --> 00:46:06.160
>> yeah. Um read I'm going to post more

00:46:06.160 --> 00:46:09.280
blogs and and share them with you. It's

00:46:09.280 --> 00:46:11.520
going to break down.

00:46:11.520 --> 00:46:13.040
>> Yeah, I'm pretty I'm pretty happy with

00:46:13.040 --> 00:46:18.880
Hermes Hermes agent. So, but um but I

00:46:18.880 --> 00:46:21.920
definitely will play around with this.

00:46:21.920 --> 00:46:26.400
Definitely. Cool. All right. Um,

00:46:26.400 --> 00:46:28.640
thanks for listening this far. If you

00:46:28.640 --> 00:46:32.000
got this far, like the video,

00:46:32.000 --> 00:46:34.400
>> do all that stuff. Otherwise, I wish you

00:46:34.400 --> 00:46:36.560
a good summer holiday and take a break

00:46:36.560 --> 00:46:39.839
like I'm trying to do now. Take a break.

00:46:39.839 --> 00:46:41.760
>> So, I guess I won't see you. Will I see

00:46:41.760 --> 00:46:42.160
you next week?

00:46:42.160 --> 00:46:44.800
>> Stand still. Never look back.

00:46:44.800 --> 00:46:47.760
Don't look back in anger. Anyway, okay.

00:46:47.760 --> 00:46:50.000
Take a break. See you, man. Bye.

00:46:50.000 --> 00:46:52.640
>> Bye.

