Listen now | Recording of AURA - Tim Frazer & Shane Gibson on 1st July 2026

WEBVTT
Kind: captions
Language: en
00:00:01.760 --> 00:00:03.750
the uh inaugural I don't even know what
00:00:03.760 --> 00:00:04.870
I'm going to call it. What's it like?
00:00:04.880 --> 00:00:05.670
Aura.
00:00:05.680 --> 00:00:07.190
>> Aura. Aura.
00:00:07.200 --> 00:00:08.629
>> Like an aura ring.
00:00:08.639 --> 00:00:10.870
>> Aura ring. Yeah. It's kind of a take on
00:00:10.880 --> 00:00:13.030
Aurora, right? I was like, yeah. And
00:00:13.040 --> 00:00:15.509
obviously the image was vibe coded. So I
00:00:15.519 --> 00:00:17.830
was like, uh, yeah. I can't remember if
00:00:17.840 --> 00:00:20.150
I think I used nano banana for it or
00:00:20.160 --> 00:00:22.310
could have been claimed. I can't
00:00:22.320 --> 00:00:23.750
remember. It's one of the one of those
00:00:23.760 --> 00:00:25.189
things. But anyway, thanks for coming on
00:00:25.199 --> 00:00:28.470
and being uh the first first of the aura
00:00:28.480 --> 00:00:31.029
forome. Um,
00:00:31.039 --> 00:00:32.790
>> the irony is we have these conversations
00:00:32.800 --> 00:00:34.310
anyway and we're like, why don't we just
00:00:34.320 --> 00:00:36.709
like record it?
00:00:36.719 --> 00:00:38.709
>> We do. Although there'll probably be a
00:00:38.719 --> 00:00:41.670
little bit less swearing or maybe not.
00:00:41.680 --> 00:00:43.350
>> Yeah, I'll try to be nice. I'll try to
00:00:43.360 --> 00:00:45.510
be I'll try try to be nicer around my
00:00:45.520 --> 00:00:47.270
swearing. Sorry. I have a background in
00:00:47.280 --> 00:00:49.190
operations, logistics, working on trucks
00:00:49.200 --> 00:00:51.670
and shipping, so
00:00:51.680 --> 00:00:54.470
it uh it comes through strong at times.
00:00:54.480 --> 00:00:55.910
>> Yeah. And I just come from New Zealand,
00:00:55.920 --> 00:00:58.709
so uh we're all potty mouths over there.
00:00:58.719 --> 00:01:00.229
because I'm the Canadian in the group by
00:01:00.239 --> 00:01:01.990
the way, half Canada Day to anybody
00:01:02.000 --> 00:01:04.229
listening. Um
00:01:04.239 --> 00:01:05.990
that that I wouldn't wouldn't swear on.
00:01:06.000 --> 00:01:07.429
I'd be like super polite. But I think
00:01:07.439 --> 00:01:09.590
you know me long enough now that plate's
00:01:09.600 --> 00:01:12.310
not I mean like I'm I'm kind but maybe
00:01:12.320 --> 00:01:14.149
polite's not my story.
00:01:14.159 --> 00:01:15.670
>> Uh but the question is when you're vibe
00:01:15.680 --> 00:01:17.270
coding, are you polite to the machine
00:01:17.280 --> 00:01:21.670
just in case the the AI?
00:01:21.680 --> 00:01:25.429
>> Yeah, actually I'm nicer than I I than I
00:01:25.439 --> 00:01:27.270
probably should be. I think my language
00:01:27.280 --> 00:01:29.990
is really like courteous.
00:01:30.000 --> 00:01:32.149
I don't know why. I'm like I know like a
00:01:32.159 --> 00:01:34.310
it's not a sentient thing but like you
00:01:34.320 --> 00:01:35.910
know
00:01:35.920 --> 00:01:37.990
uh you're just like fooled into it
00:01:38.000 --> 00:01:39.429
because the way that it and you notice
00:01:39.439 --> 00:01:40.550
is like remember when you were first
00:01:40.560 --> 00:01:42.230
using like chat GBT like 2 or whatever
00:01:42.240 --> 00:01:44.789
and like how like just horrible it was
00:01:44.799 --> 00:01:46.789
like sorry like this hasn't changed so
00:01:46.799 --> 00:01:49.190
much that I actually am super annoyed
00:01:49.200 --> 00:01:50.789
with the xanthanropic one because it's
00:01:50.799 --> 00:01:55.270
so uh wordy which is like I don't want
00:01:55.280 --> 00:01:57.670
to date myself. I already speak way too
00:01:57.680 --> 00:01:58.870
much. And it's like I'm having this
00:01:58.880 --> 00:02:01.190
conversation. I'm like, "Okay, fewer
00:02:01.200 --> 00:02:03.590
words, please." Like under 50 words. And
00:02:03.600 --> 00:02:06.630
if you don't like and also don't blow
00:02:06.640 --> 00:02:10.150
smoke up my uh armpits, I guess. Um like
00:02:10.160 --> 00:02:13.750
really just be direct as you can and you
00:02:13.760 --> 00:02:15.589
know don't stop. But then of course it
00:02:15.599 --> 00:02:16.869
ignores me like three prompts later
00:02:16.879 --> 00:02:17.910
anyway.
00:02:17.920 --> 00:02:19.430
>> Yeah. Open my harness. I've got a whole
00:02:19.440 --> 00:02:20.790
lot of prompts in there about what I
00:02:20.800 --> 00:02:22.390
like and what I don't like and it kind
00:02:22.400 --> 00:02:24.630
of refers to them. Um I don't know. I
00:02:24.640 --> 00:02:26.229
keep looking at things like Caveman and
00:02:26.239 --> 00:02:28.229
going, "Oh, should I?" I I don't care
00:02:28.239 --> 00:02:29.830
about because I'm on, you know, a max
00:02:29.840 --> 00:02:31.190
plan, so I don't care that much about
00:02:31.200 --> 00:02:34.309
tokens yet. Uh will at some stage. Um
00:02:34.319 --> 00:02:35.910
but I'm thinking, well, yeah, if you're
00:02:35.920 --> 00:02:38.150
less verbose, uh it might be useful
00:02:38.160 --> 00:02:40.070
because I don't read half of it anyway.
00:02:40.080 --> 00:02:42.229
Um but yeah, the politeness. Yeah, I'm
00:02:42.239 --> 00:02:43.910
I'm still polite. I don't know why. Even
00:02:43.920 --> 00:02:45.670
even when I'm yelling at it, right, I
00:02:45.680 --> 00:02:47.830
kind of like, yeah, you just you know,
00:02:47.840 --> 00:02:50.390
you lied to me. It was um funny today. I
00:02:50.400 --> 00:02:52.630
was getting one Claude session to write
00:02:52.640 --> 00:02:54.550
a prompt for another Claude session. So
00:02:54.560 --> 00:02:56.790
basically a person my personal OS
00:02:56.800 --> 00:03:00.710
talking to work harness and um and and
00:03:00.720 --> 00:03:04.390
the personal OS one it wrote some pretty
00:03:04.400 --> 00:03:07.190
pretty gnarly prompt. It was like I want
00:03:07.200 --> 00:03:09.270
you to do this and I don't want you to
00:03:09.280 --> 00:03:11.110
pretend. I want you to actually go and
00:03:11.120 --> 00:03:14.470
do the work. And so it was like
00:03:14.480 --> 00:03:16.470
basically saying it knows Claude lies
00:03:16.480 --> 00:03:18.070
even though it's Claude. And it gave me
00:03:18.080 --> 00:03:19.990
the prompt and I went posted into the
00:03:20.000 --> 00:03:21.990
other session and then moved back again.
00:03:22.000 --> 00:03:25.350
So I was kind of that was kind of funny.
00:03:25.360 --> 00:03:26.710
>> Yeah, that's yeah that's really
00:03:26.720 --> 00:03:29.110
interesting. Uh I was going to like
00:03:29.120 --> 00:03:30.390
diverge from this conversation a little
00:03:30.400 --> 00:03:32.229
bit like talk about like so I was at the
00:03:32.239 --> 00:03:33.750
snowflake summit. We could talk I think
00:03:33.760 --> 00:03:34.949
we should talk about a little bit like
00:03:34.959 --> 00:03:37.110
that because obviously you know you do a
00:03:37.120 --> 00:03:39.990
lot of work in like GCP. Um I you know
00:03:40.000 --> 00:03:41.910
I've worked in there uh haven't done a
00:03:41.920 --> 00:03:43.190
ton of work in data bricks and like
00:03:43.200 --> 00:03:44.949
Azure and done some in Azure and
00:03:44.959 --> 00:03:46.149
everything like that. But the one thing
00:03:46.159 --> 00:03:48.229
I noticed is like and I don't think this
00:03:48.239 --> 00:03:51.190
is like a hot take necessarily but just
00:03:51.200 --> 00:03:53.430
said I can't even use that word anymore.
00:03:53.440 --> 00:03:55.509
Um every data vendor is like
00:03:55.519 --> 00:03:57.910
consolidating at like a crazy level. it
00:03:57.920 --> 00:03:59.990
speeds that I you know a year ago I
00:04:00.000 --> 00:04:01.350
probably personally wouldn't have
00:04:01.360 --> 00:04:03.830
predicted quite how fast it is but like
00:04:03.840 --> 00:04:07.190
you look at like warehouse pipelines BI
00:04:07.200 --> 00:04:09.030
semantic later they're all like folding
00:04:09.040 --> 00:04:12.789
in at like in in very intense speeds
00:04:12.799 --> 00:04:15.910
and the if you walk into any conference
00:04:15.920 --> 00:04:17.909
floor the only thing you're going to
00:04:17.919 --> 00:04:19.990
hear is context
00:04:20.000 --> 00:04:22.469
um I mean this is this year's like the
00:04:22.479 --> 00:04:24.790
whatever ontologies context context
00:04:24.800 --> 00:04:27.590
semantic layers uh and The one thing I
00:04:27.600 --> 00:04:29.350
had really, really, really hard time
00:04:29.360 --> 00:04:32.070
when I was at Snowflake is competitive
00:04:32.080 --> 00:04:34.629
advantage showing amongst any of the
00:04:34.639 --> 00:04:36.230
vendors.
00:04:36.240 --> 00:04:39.270
Like it was
00:04:39.280 --> 00:04:41.670
every investment that I see right now in
00:04:41.680 --> 00:04:44.070
any medium-sized vendors is like a huge
00:04:44.080 --> 00:04:46.550
risk because like if you do a one-year
00:04:46.560 --> 00:04:48.870
investment, Snowflake could very much
00:04:48.880 --> 00:04:51.430
release something that's maybe 80% of
00:04:51.440 --> 00:04:53.749
good within the next three months. Same
00:04:53.759 --> 00:04:54.950
with data bricks. like you could just
00:04:54.960 --> 00:04:57.670
see them just chomping through. Anybody
00:04:57.680 --> 00:04:59.670
that says chat with your data, I I have
00:04:59.680 --> 00:05:03.189
a very hard time seeing how they're
00:05:03.199 --> 00:05:05.909
better than what the big guys are.
00:05:05.919 --> 00:05:07.510
>> Yeah. I think I think it's the fact that
00:05:07.520 --> 00:05:09.830
basically they're um veneering over the
00:05:09.840 --> 00:05:11.510
top, right? So, they're taking a
00:05:11.520 --> 00:05:13.029
foundational model, they're taking a use
00:05:13.039 --> 00:05:15.670
case, and then they're kind of making it
00:05:15.680 --> 00:05:17.029
easy for you to solve that use case
00:05:17.039 --> 00:05:19.430
using the foundational model. But as you
00:05:19.440 --> 00:05:21.510
said, the the big players are watching
00:05:21.520 --> 00:05:24.070
what you're doing and uh and they're
00:05:24.080 --> 00:05:25.430
effectively building that into the
00:05:25.440 --> 00:05:26.870
platform, right? You if you look at what
00:05:26.880 --> 00:05:29.270
Claude codes done to Kurser, just
00:05:29.280 --> 00:05:30.870
listened to a podcast on today, you
00:05:30.880 --> 00:05:34.310
know, with um SpaceX, XAI, whatever it's
00:05:34.320 --> 00:05:36.629
called this week, um taking it out with
00:05:36.639 --> 00:05:38.870
a shear swap. Uh you know, they
00:05:38.880 --> 00:05:41.749
basically said Curser was leading. Uh
00:05:41.759 --> 00:05:43.909
but what they had done is done an IDE
00:05:43.919 --> 00:05:46.550
over the top of a foundational model and
00:05:46.560 --> 00:05:48.390
Anthropic just came along and went good
00:05:48.400 --> 00:05:50.870
use case. Um you look at the way
00:05:50.880 --> 00:05:53.110
Anthropic have published that article on
00:05:53.120 --> 00:05:54.469
their data agents.
00:05:54.479 --> 00:05:54.790
>> Yeah.
00:05:54.800 --> 00:05:56.230
>> Sure as [ __ ] they're going to come after
00:05:56.240 --> 00:05:57.749
the data market, right?
00:05:57.759 --> 00:05:59.510
>> If you think about complex problems that
00:05:59.520 --> 00:06:01.430
need to be solved, they're going to come
00:06:01.440 --> 00:06:03.430
after this one hard because it's a
00:06:03.440 --> 00:06:05.430
massive it's a massive TAM for them,
00:06:05.440 --> 00:06:06.950
right? target market. Oh yeah.
00:06:06.960 --> 00:06:08.950
>> Um and and complex problems to be solved
00:06:08.960 --> 00:06:10.629
and they got a [ __ ] ton of data to help
00:06:10.639 --> 00:06:11.510
solve it.
00:06:11.520 --> 00:06:13.110
>> I mean did you hear about Ripling
00:06:13.120 --> 00:06:15.510
starting up their whole data play?
00:06:15.520 --> 00:06:17.590
>> Oh my Yeah. The full they're going full
00:06:17.600 --> 00:06:21.430
lakehouse play because their perspective
00:06:21.440 --> 00:06:23.029
uh is that and they're not wrong is that
00:06:23.039 --> 00:06:26.150
in the HRIS systems payments everything
00:06:26.160 --> 00:06:27.830
a lot of data a lot of things that
00:06:27.840 --> 00:06:29.749
you're reporting on are actually coming
00:06:29.759 --> 00:06:32.070
from those systems. So they've launched
00:06:32.080 --> 00:06:33.670
their whole like a whole new data
00:06:33.680 --> 00:06:35.350
platform. I don't have specifics. I just
00:06:35.360 --> 00:06:37.430
read that and I was like this is crazy.
00:06:37.440 --> 00:06:40.790
This is like what is going on and I was
00:06:40.800 --> 00:06:42.390
like well maybe we're going to start
00:06:42.400 --> 00:06:43.830
seeing more of these where people are
00:06:43.840 --> 00:06:45.830
getting these like open source blocks
00:06:45.840 --> 00:06:47.510
and then yeah they can just stand up on
00:06:47.520 --> 00:06:50.550
iceberg and just like computation layer
00:06:50.560 --> 00:06:53.189
is now like commoditized.
00:06:53.199 --> 00:06:54.710
>> Yeah. Yeah, but we kind of, you know, we
00:06:54.720 --> 00:06:57.270
saw it before where the the
00:06:57.280 --> 00:06:58.550
transactional systems, you know,
00:06:58.560 --> 00:07:00.629
Salesforce, SAP
00:07:00.639 --> 00:07:03.749
>> tried to own the data space and, you
00:07:03.759 --> 00:07:05.990
know, they kind of suck at it. Uh, don't
00:07:06.000 --> 00:07:06.469
know why.
00:07:06.479 --> 00:07:07.510
>> I didn't say they were going to be good
00:07:07.520 --> 00:07:09.110
at it. Definitely didn't say that. And
00:07:09.120 --> 00:07:11.270
yes, I do remember like, you know, like
00:07:11.280 --> 00:07:13.110
Netswuite and like these guys like kind
00:07:13.120 --> 00:07:15.029
of going into the spaces to try to do
00:07:15.039 --> 00:07:16.950
stuff and then, you know, that's called
00:07:16.960 --> 00:07:18.309
by by Oracle. So, it's just a giant
00:07:18.319 --> 00:07:21.029
Oracle database anyway. And and then the
00:07:21.039 --> 00:07:22.950
question is even if they do make it work
00:07:22.960 --> 00:07:25.510
for their own product, uh their own
00:07:25.520 --> 00:07:27.830
product only holds half, you know,
00:07:27.840 --> 00:07:29.830
really holds more than half the context
00:07:29.840 --> 00:07:32.309
and half the data of the organization.
00:07:32.319 --> 00:07:33.749
Um so you're still going to need
00:07:33.759 --> 00:07:35.029
something else. So are they thinking
00:07:35.039 --> 00:07:36.790
that you're going to go move the rest of
00:07:36.800 --> 00:07:39.029
your data or make it available and then
00:07:39.039 --> 00:07:40.710
you're going to go pump all your context
00:07:40.720 --> 00:07:43.350
into their their platform?
00:07:43.360 --> 00:07:44.870
>> Of course, another platform for
00:07:44.880 --> 00:07:47.589
platforms. the um but I think one of the
00:07:47.599 --> 00:07:48.309
things that we were talking about
00:07:48.319 --> 00:07:52.390
earlier is um and it's not that uncommon
00:07:52.400 --> 00:07:54.950
but I think Amazon called them out for
00:07:54.960 --> 00:07:58.070
this is probably the most infamous for
00:07:58.080 --> 00:08:00.950
this play which is like where snowflake
00:08:00.960 --> 00:08:02.629
and data bricks are the landlord and
00:08:02.639 --> 00:08:04.309
they become the tenant. So like, you
00:08:04.319 --> 00:08:06.070
know, think about like the Sigmas, the
00:08:06.080 --> 00:08:08.309
Omnies, the Lookers, the Tableau and
00:08:08.319 --> 00:08:09.589
stuff like that where they're like
00:08:09.599 --> 00:08:13.589
they're their partners for now and then,
00:08:13.599 --> 00:08:15.430
you know, they do CLA classic platform
00:08:15.440 --> 00:08:17.029
trap now. It's just like, oh, you could
00:08:17.039 --> 00:08:19.990
get 80% of these BI tools, but like you
00:08:20.000 --> 00:08:21.430
know what, 20% of the cost because
00:08:21.440 --> 00:08:22.550
you're already on our layers. So we'll
00:08:22.560 --> 00:08:24.710
give it to you like super super cheap.
00:08:24.720 --> 00:08:26.790
Same thing like you were just saying SAP
00:08:26.800 --> 00:08:28.710
and all those other guys playing this,
00:08:28.720 --> 00:08:30.790
but like
00:08:30.800 --> 00:08:33.029
now you see like Snowflake is doubling
00:08:33.039 --> 00:08:34.949
down on the iceberg play, right? So
00:08:34.959 --> 00:08:36.469
they're doubling down on their ability
00:08:36.479 --> 00:08:38.630
to move data and you see them launch
00:08:38.640 --> 00:08:40.149
something called Coco. I think it's
00:08:40.159 --> 00:08:44.389
called Coco or maybe Co-work now. Um
00:08:44.399 --> 00:08:46.150
I can't remember what it's now. It's the
00:08:46.160 --> 00:08:47.910
name the name's changed so many times
00:08:47.920 --> 00:08:50.470
similar to Salesforce. But the gist is
00:08:50.480 --> 00:08:53.190
is like if it's a chat interface like
00:08:53.200 --> 00:08:56.070
claw like open AI on your data and
00:08:56.080 --> 00:08:58.550
snowflake has the computation and the
00:08:58.560 --> 00:09:00.870
money to train these models themselves
00:09:00.880 --> 00:09:02.630
maybe their own models only and they do
00:09:02.640 --> 00:09:05.269
have one 90% as good as the frontier
00:09:05.279 --> 00:09:07.030
frontier models but then they start
00:09:07.040 --> 00:09:08.790
swapping them in and you don't really
00:09:08.800 --> 00:09:10.630
most time notice the difference and they
00:09:10.640 --> 00:09:11.910
start getting you more and more into
00:09:11.920 --> 00:09:13.670
this space and same with data bricks
00:09:13.680 --> 00:09:15.590
like I wouldn't say there's like a force
00:09:15.600 --> 00:09:17.350
adoption between Delta and iceberg but
00:09:17.360 --> 00:09:19.030
you see Delta and iceberg converging for
00:09:19.040 --> 00:09:22.550
this open table format and now
00:09:22.560 --> 00:09:23.590
everybody's going to start controlling
00:09:23.600 --> 00:09:25.190
these standards.
00:09:25.200 --> 00:09:26.389
>> Yeah. So
00:09:26.399 --> 00:09:29.190
>> standards. Yeah. Oh, it's like open
00:09:29.200 --> 00:09:31.990
semantic interface standard.
00:09:32.000 --> 00:09:34.710
>> Right. It's like um I remember remember
00:09:34.720 --> 00:09:36.790
there's like this time this era when you
00:09:36.800 --> 00:09:38.389
could sell on Facebook marketplace.
00:09:38.399 --> 00:09:47.190
Thank you, Nick.
00:09:47.200 --> 00:09:51.030
Where you going?
00:09:51.040 --> 00:09:54.070
>> Literally no idea. I just had a
00:09:54.080 --> 00:09:56.550
>> Mac takeover. All right, I'm back now.
00:09:56.560 --> 00:09:59.030
Um, that was weird. I've never had that
00:09:59.040 --> 00:10:01.829
happen before. The uh where Amazon's
00:10:01.839 --> 00:10:03.430
like come to our Facebook or not
00:10:03.440 --> 00:10:06.389
Facebook marketplace, Amazon marketplace
00:10:06.399 --> 00:10:08.230
and sell your batteries, sell your
00:10:08.240 --> 00:10:09.829
whatever your goods. And my one of the
00:10:09.839 --> 00:10:11.350
companies I worked for did that. And
00:10:11.360 --> 00:10:13.670
then one year we were selling and
00:10:13.680 --> 00:10:15.829
there's like this pedometer. It looked
00:10:15.839 --> 00:10:19.750
exactly the same and it was like it it
00:10:19.760 --> 00:10:21.509
was like one of Amazon's like subbrands
00:10:21.519 --> 00:10:23.990
or whatever and we were but it was like
00:10:24.000 --> 00:10:25.590
not called Amazon was like like because
00:10:25.600 --> 00:10:28.710
they had so many like other brands. Um
00:10:28.720 --> 00:10:29.750
but then they sell in the same
00:10:29.760 --> 00:10:31.030
marketplace and compete against us and
00:10:31.040 --> 00:10:32.870
it was super we we were like well you
00:10:32.880 --> 00:10:33.990
actually know our margin. You know how
00:10:34.000 --> 00:10:35.670
much we're like selling to you for doing
00:10:35.680 --> 00:10:36.870
all these things. you have all this
00:10:36.880 --> 00:10:38.710
insider information and now you're
00:10:38.720 --> 00:10:41.910
cannibalizing us but it's Amazon so you
00:10:41.920 --> 00:10:43.430
couldn't argue you couldn't really argue
00:10:43.440 --> 00:10:46.230
with them like what file file a zenes
00:10:46.240 --> 00:10:48.069
ticket or order Amazon for those things
00:10:48.079 --> 00:10:50.150
so like you know is data versus
00:10:50.160 --> 00:10:52.069
snowflake doing the same thing this sort
00:10:52.079 --> 00:10:52.389
of thing
00:10:52.399 --> 00:10:54.310
>> I don't know I don't know I mean for
00:10:54.320 --> 00:10:55.829
agile data we're on Google right we've
00:10:55.839 --> 00:10:57.190
been on Google cloud for seven eight
00:10:57.200 --> 00:10:59.110
years now uh one of the things we like
00:10:59.120 --> 00:11:01.990
is in theory uh all our customer data
00:11:02.000 --> 00:11:05.110
and all our behavior is isolated to our
00:11:05.120 --> 00:11:07.350
projects our tenants gencies and Google
00:11:07.360 --> 00:11:09.110
don't look at them. That's what the
00:11:09.120 --> 00:11:11.430
terms and conditions say. Um I I've
00:11:11.440 --> 00:11:12.710
never read the Snowflake terms and
00:11:12.720 --> 00:11:14.710
conditions or the data brick ones.
00:11:14.720 --> 00:11:16.069
There's an interesting thing. Uh I
00:11:16.079 --> 00:11:18.150
listened to a L podcast. Yeah, I
00:11:18.160 --> 00:11:20.069
listened to a Lenny podcast ages ago. Uh
00:11:20.079 --> 00:11:22.310
Brian Belfer was talking about the the
00:11:22.320 --> 00:11:24.630
kind of cycle of these big companies and
00:11:24.640 --> 00:11:26.310
he talked about marketplaces and he was
00:11:26.320 --> 00:11:27.430
kind of talking about who's going to
00:11:27.440 --> 00:11:30.470
dominate the AI agent marketplace. Um
00:11:30.480 --> 00:11:32.310
but he had this playbook and it was like
00:11:32.320 --> 00:11:33.910
perfect. It was like you start out
00:11:33.920 --> 00:11:35.350
early. Yeah. And then you get early
00:11:35.360 --> 00:11:37.350
adopter partners on and you're all
00:11:37.360 --> 00:11:38.550
friendly to them and you give them a
00:11:38.560 --> 00:11:40.470
shitload of money and free credits and
00:11:40.480 --> 00:11:42.470
you know a lot of loving. Uh and then
00:11:42.480 --> 00:11:44.870
effectively, you know, once you got some
00:11:44.880 --> 00:11:46.710
success, you then start having partner
00:11:46.720 --> 00:11:49.910
tiers, you know, bronze, silver, gold.
00:11:49.920 --> 00:11:52.150
Um and then after that, you you open up
00:11:52.160 --> 00:11:53.590
the marketplace and everybody starts,
00:11:53.600 --> 00:11:55.190
you know, putting their stuff up to sell
00:11:55.200 --> 00:11:56.630
and then you start carving out the
00:11:56.640 --> 00:11:57.670
products that are really really
00:11:57.680 --> 00:11:59.110
successful, right? And you start
00:11:59.120 --> 00:12:00.550
replicating them. I don't know if you
00:12:00.560 --> 00:12:02.550
ever saw the story about what Amazon did
00:12:02.560 --> 00:12:04.949
to Trifactor, but they effectively
00:12:04.959 --> 00:12:08.710
Amazon Glue or whatever it was called um
00:12:08.720 --> 00:12:11.430
was a complete scrape of Trifactor's
00:12:11.440 --> 00:12:14.389
interface and the way it worked. Um what
00:12:14.399 --> 00:12:15.750
was the other one that the open source
00:12:15.760 --> 00:12:17.750
one that they then picked up and
00:12:17.760 --> 00:12:22.710
actually embedded on AWS? uh
00:12:22.720 --> 00:12:24.310
>> oh was it one of the ones
00:12:24.320 --> 00:12:25.829
>> for licensing and then obviously then
00:12:25.839 --> 00:12:27.190
they didn't like that because then
00:12:27.200 --> 00:12:29.990
Amazon could couldn't just like suck 90%
00:12:30.000 --> 00:12:32.230
of the of the engineering for free.
00:12:32.240 --> 00:12:33.990
>> Yeah. But the thing that interests me is
00:12:34.000 --> 00:12:36.310
is how few data professionals or data
00:12:36.320 --> 00:12:38.389
practitioners actually look at the
00:12:38.399 --> 00:12:40.870
market and where the vendors are and
00:12:40.880 --> 00:12:42.710
bring that into their strategy, right?
00:12:42.720 --> 00:12:44.949
Like if you can if you can go fully on
00:12:44.959 --> 00:12:47.509
Snowflake as a full end toend platform
00:12:47.519 --> 00:12:50.069
now and they're cost competitive, right?
00:12:50.079 --> 00:12:52.230
and they're constantly adding features
00:12:52.240 --> 00:12:54.150
uh and they're not going to go and sell
00:12:54.160 --> 00:12:57.590
out to Salesforce or get get acquired,
00:12:57.600 --> 00:12:59.590
then it's probably a good strategy. If
00:12:59.600 --> 00:13:01.590
you go to somebody that's veneering, you
00:13:01.600 --> 00:13:04.710
know, uh one use case on top of a
00:13:04.720 --> 00:13:07.509
foundational model, you got a you got a
00:13:07.519 --> 00:13:08.710
risk that that puppy's going to get
00:13:08.720 --> 00:13:10.230
built bought, right? And we've seen the
00:13:10.240 --> 00:13:12.470
waves of, you know, start a company, get
00:13:12.480 --> 00:13:15.269
to market share, and then get bought
00:13:15.279 --> 00:13:16.710
out. And we've seen the consolidation
00:13:16.720 --> 00:13:18.389
from the modern data stack. always
00:13:18.399 --> 00:13:20.150
surprises me how many people when they
00:13:20.160 --> 00:13:22.550
kind of invest in these things don't
00:13:22.560 --> 00:13:24.790
actually look at uh the life cycle of
00:13:24.800 --> 00:13:26.389
companies and what it means to their
00:13:26.399 --> 00:13:27.509
company.
00:13:27.519 --> 00:13:28.710
>> Well, I mean I guess your hope is that
00:13:28.720 --> 00:13:31.990
your your company outlast the platform
00:13:32.000 --> 00:13:35.509
the um Yeah, I mean it's really Yeah, I
00:13:35.519 --> 00:13:37.190
mean it's really really interesting to
00:13:37.200 --> 00:13:38.949
me. It's just like and also like I think
00:13:38.959 --> 00:13:40.150
a lot of people underleverage the
00:13:40.160 --> 00:13:41.509
platforms, but then they're adding so
00:13:41.519 --> 00:13:43.190
many new features and then we've had
00:13:43.200 --> 00:13:45.750
conversations before around like the
00:13:45.760 --> 00:13:47.110
quality of those features just because
00:13:47.120 --> 00:13:48.949
more is not always better. Like you go
00:13:48.959 --> 00:13:51.030
to use it and then it's like did you
00:13:51.040 --> 00:13:53.430
read the fine print? Did you do like you
00:13:53.440 --> 00:13:54.710
know I've been using trying to use
00:13:54.720 --> 00:13:57.430
Snowflake Evals and I would say betaware
00:13:57.440 --> 00:14:00.470
at best. Um
00:14:00.480 --> 00:14:03.110
and then I'm like okay but like how long
00:14:03.120 --> 00:14:04.629
is that going to stay beta right at the
00:14:04.639 --> 00:14:06.069
speed of acceleration that we're seeing
00:14:06.079 --> 00:14:09.990
today? Um but what like I don't know
00:14:10.000 --> 00:14:12.949
what do you think like um I don't know
00:14:12.959 --> 00:14:14.310
the thinner slice ones. So you have
00:14:14.320 --> 00:14:15.910
these the bigger platforms you could say
00:14:15.920 --> 00:14:17.829
like there's what we say like there's
00:14:17.839 --> 00:14:19.430
click house is almost like a dark horse
00:14:19.440 --> 00:14:20.550
right now but I actually don't think
00:14:20.560 --> 00:14:21.910
it's as dark horse people you think it
00:14:21.920 --> 00:14:24.710
is. I think it's it's it's growing
00:14:24.720 --> 00:14:28.949
growing quite fast. Um I think that you
00:14:28.959 --> 00:14:31.430
know then you have Google always sneaky
00:14:31.440 --> 00:14:33.030
sneaky. It's so funny because you use
00:14:33.040 --> 00:14:35.350
Google. I've used it before in BigQuery,
00:14:35.360 --> 00:14:37.590
but like
00:14:37.600 --> 00:14:39.670
honestly 90% of the interviews and
00:14:39.680 --> 00:14:40.870
companies I've talked to in the past,
00:14:40.880 --> 00:14:42.389
like they use it for Google Analytics
00:14:42.399 --> 00:14:44.389
and that's like it.
00:14:44.399 --> 00:14:45.829
>> Yeah. I mean, they got their Trojan,
00:14:45.839 --> 00:14:48.870
right? You know, their G4 uh integration
00:14:48.880 --> 00:14:51.670
into BigQuery, uh their free data
00:14:51.680 --> 00:14:53.430
studio, sorry, Looker Studio, sorry,
00:14:53.440 --> 00:14:54.710
data studio because it's just been
00:14:54.720 --> 00:14:56.870
renamed. But, you know, Data Studio has
00:14:56.880 --> 00:14:58.870
no cost, right? No license cost. You
00:14:58.880 --> 00:15:00.710
know, freeers in puppies, not freeers in
00:15:00.720 --> 00:15:04.069
beer. Um, but I I hop because we we used
00:15:04.079 --> 00:15:05.670
to use it as our front end, right, for
00:15:05.680 --> 00:15:06.790
our customers because there was no
00:15:06.800 --> 00:15:08.150
license. It was fully integrated to
00:15:08.160 --> 00:15:09.750
BigQuery. We got really good performance
00:15:09.760 --> 00:15:11.910
out of it. Um, and now we're kind of
00:15:11.920 --> 00:15:13.430
moving to more building our own
00:15:13.440 --> 00:15:15.269
information products. But I hopped into
00:15:15.279 --> 00:15:17.509
data studio the other day and now
00:15:17.519 --> 00:15:19.189
there's conversational analytics
00:15:19.199 --> 00:15:22.310
embedded there at no cost. No cost as in
00:15:22.320 --> 00:15:24.470
freeers and puppies, right? Um,
00:15:24.480 --> 00:15:27.269
>> and so I'm like, excellent. click on the
00:15:27.279 --> 00:15:29.110
button and then it goes, "Oh, no. Now
00:15:29.120 --> 00:15:31.350
you've got to go into GCP console and
00:15:31.360 --> 00:15:33.750
set up your agents." And you know, I
00:15:33.760 --> 00:15:35.509
love Google's technology. I love their
00:15:35.519 --> 00:15:37.110
pricing models because you can run it on
00:15:37.120 --> 00:15:39.910
a smell of an oily rag. Uh but it's a
00:15:39.920 --> 00:15:42.949
technical tool and uh you've got to
00:15:42.959 --> 00:15:45.269
cobble it all together. And so, you
00:15:45.279 --> 00:15:47.030
know, they've come out with their
00:15:47.040 --> 00:15:48.710
context graph or their knowledge,
00:15:48.720 --> 00:15:50.949
whatever it is, which is part of data,
00:15:50.959 --> 00:15:52.389
which was part of something else, which
00:15:52.399 --> 00:15:54.949
was part of something else. Um it's
00:15:54.959 --> 00:15:56.949
incredibly ugly. It's still hard to use,
00:15:56.959 --> 00:15:59.509
but if you're an engineer, you can put
00:15:59.519 --> 00:16:02.710
it all together. Um, the thing is great
00:16:02.720 --> 00:16:04.629
technology, [ __ ] marketing, [ __ ]
00:16:04.639 --> 00:16:07.749
partnering. Um, which is why, you know,
00:16:07.759 --> 00:16:09.910
all you ever hear is data bricks and,
00:16:09.920 --> 00:16:11.749
uh, snowflake, right? Maybe a bit of
00:16:11.759 --> 00:16:13.509
fabric, you know, if somebody's keen on
00:16:13.519 --> 00:16:15.590
PowerPoint,
00:16:15.600 --> 00:16:17.990
you know, I hardly hear I mean, I hear a
00:16:18.000 --> 00:16:19.910
bit of Amazon bedrock. You never hear
00:16:19.920 --> 00:16:21.990
red shift anymore. Like, you don't
00:16:22.000 --> 00:16:24.389
really hear Amazon coming to this. So
00:16:24.399 --> 00:16:25.990
it's definitely a race of snowflake
00:16:26.000 --> 00:16:27.749
versus data bricks. Um
00:16:27.759 --> 00:16:29.269
>> I mean every conversation I completely
00:16:29.279 --> 00:16:30.710
agree with that like I would say they
00:16:30.720 --> 00:16:34.550
dominate the I have you know in LA I
00:16:34.560 --> 00:16:35.910
would say when I I was doing some
00:16:35.920 --> 00:16:37.670
interviews last year uh before my
00:16:37.680 --> 00:16:41.509
current role and I would say I think I
00:16:41.519 --> 00:16:43.350
did about 10 solid interviews for
00:16:43.360 --> 00:16:44.389
position trying to figure where I wanted
00:16:44.399 --> 00:16:46.710
to go done the consulting having kids so
00:16:46.720 --> 00:16:48.550
I was just I need more more stable
00:16:48.560 --> 00:16:50.710
lifestyle. I don't know what that
00:16:50.720 --> 00:16:52.389
actually means in the startup world
00:16:52.399 --> 00:16:56.470
these days or any business work. Um, but
00:16:56.480 --> 00:16:58.790
so many people so many people were
00:16:58.800 --> 00:17:01.269
talking about moving from Snowflakes to
00:17:01.279 --> 00:17:03.269
data bricks because it was going to be
00:17:03.279 --> 00:17:06.710
cheaper and I just I had such a hard
00:17:06.720 --> 00:17:08.949
time with that and I I couldn't get like
00:17:08.959 --> 00:17:10.549
a straight answer because so many people
00:17:10.559 --> 00:17:12.470
at the time that I was talking to that
00:17:12.480 --> 00:17:14.150
was the move. I was like, "So, you're
00:17:14.160 --> 00:17:16.390
going to move from like, you know,
00:17:16.400 --> 00:17:19.029
Snowflake compute engine to a smart
00:17:19.039 --> 00:17:21.029
comput engine and it's going to be
00:17:21.039 --> 00:17:23.189
radically different."
00:17:23.199 --> 00:17:25.270
Maybe
00:17:25.280 --> 00:17:26.789
maybe
00:17:26.799 --> 00:17:29.669
um I just had a lot of like but there's
00:17:29.679 --> 00:17:31.669
like so much other cost around operating
00:17:31.679 --> 00:17:33.830
these these systems and I was like
00:17:33.840 --> 00:17:35.190
there's probably other ways you can do
00:17:35.200 --> 00:17:37.190
in your process. It's far simpler than
00:17:37.200 --> 00:17:39.590
moving a whole platform play. And it was
00:17:39.600 --> 00:17:42.870
not like, you know, I mean, I can bar I
00:17:42.880 --> 00:17:43.990
barely hear anybody talking about
00:17:44.000 --> 00:17:45.750
Redshift these days. And if I do, it's
00:17:45.760 --> 00:17:47.350
usually how are we moving on from
00:17:47.360 --> 00:17:50.230
Redshift. Uh
00:17:50.240 --> 00:17:53.110
so yeah, Amazon from this conversation,
00:17:53.120 --> 00:17:54.710
at least from my perspective, has mostly
00:17:54.720 --> 00:17:57.350
disappeared and then Microsoft in this
00:17:57.360 --> 00:17:59.029
conversation,
00:17:59.039 --> 00:18:01.669
I mean, I hear about fabric, but I
00:18:01.679 --> 00:18:03.190
personally haven't seen a ton of it
00:18:03.200 --> 00:18:05.909
deployed in my conversations. There's a
00:18:05.919 --> 00:18:08.470
lot of people talking about it, but even
00:18:08.480 --> 00:18:10.710
the people that I know that are on Azure
00:18:10.720 --> 00:18:12.630
uh that I've worked with, they use data
00:18:12.640 --> 00:18:15.350
bricks on Azure, which honestly I
00:18:15.360 --> 00:18:16.870
thought you two two three years ago that
00:18:16.880 --> 00:18:18.150
Microsoft was going to buy data bricks
00:18:18.160 --> 00:18:18.630
like I was like
00:18:18.640 --> 00:18:20.150
>> yeah I thought so too. I mean it's going
00:18:20.160 --> 00:18:21.110
to be interesting to see what happens to
00:18:21.120 --> 00:18:23.510
PowerBI given the you know what's
00:18:23.520 --> 00:18:25.669
happening in the BI reporting space and
00:18:25.679 --> 00:18:27.270
the dashboarding space. It's going to be
00:18:27.280 --> 00:18:29.430
really interesting to see whether things
00:18:29.440 --> 00:18:31.430
like streamlet take out, you know, if
00:18:31.440 --> 00:18:33.750
you've got a PowerBI constant, you know,
00:18:33.760 --> 00:18:36.470
consume uh last mile sitting on top of
00:18:36.480 --> 00:18:38.630
Snowflake, how many people are going to
00:18:38.640 --> 00:18:40.549
go to building out their own stream
00:18:40.559 --> 00:18:41.990
apps, which is obviously where
00:18:42.000 --> 00:18:43.590
Snowflake's going. I mean, the other
00:18:43.600 --> 00:18:45.029
thing that confuses me is when I talk to
00:18:45.039 --> 00:18:46.950
organizations that have Snowflake and
00:18:46.960 --> 00:18:49.750
data bricks and I'm like, well, is this
00:18:49.760 --> 00:18:51.909
a layered architecture? you know, you've
00:18:51.919 --> 00:18:54.230
got three layers, you know, effectively
00:18:54.240 --> 00:18:55.909
what we call history. So, some kind of
00:18:55.919 --> 00:18:58.070
raw staging, persistent stage, and then
00:18:58.080 --> 00:18:59.669
the place you do all the heavy lifting,
00:18:59.679 --> 00:19:01.190
your design, whatever you want to call
00:19:01.200 --> 00:19:03.430
it, and then your consumption layer, you
00:19:03.440 --> 00:19:05.430
know, a snowflake, your consumption, and
00:19:05.440 --> 00:19:09.110
data bricks, your hidden mirror. Um, but
00:19:09.120 --> 00:19:11.270
yeah, I I I never quite you have to go
00:19:11.280 --> 00:19:12.789
see their architecture diagrams if they
00:19:12.799 --> 00:19:15.110
exist. Um, we've got a comment here from
00:19:15.120 --> 00:19:17.270
uh Dalen Webb. Excuse everybody if I
00:19:17.280 --> 00:19:18.390
ever pronounce your name wrong. It's
00:19:18.400 --> 00:19:20.470
going to happen every time. Uh my mor my
00:19:20.480 --> 00:19:22.470
org is exploring a move from fabric
00:19:22.480 --> 00:19:24.549
powerbi to data bricks genie spaces.
00:19:24.559 --> 00:19:27.669
PowerBI is sticky through enlarge orgs.
00:19:27.679 --> 00:19:30.549
>> Oh yeah it's like MS teams problem.
00:19:30.559 --> 00:19:32.470
Remember everybody was on Slack remember
00:19:32.480 --> 00:19:34.390
before the Salesforce acquisition and
00:19:34.400 --> 00:19:36.230
then Microsoft that did the Internet
00:19:36.240 --> 00:19:38.230
Explorer play again. I was like didn't
00:19:38.240 --> 00:19:39.830
you get like caught out from like the
00:19:39.840 --> 00:19:42.150
monopoly play on this and it's just like
00:19:42.160 --> 00:19:43.830
you've tied everything to you tie it all
00:19:43.840 --> 00:19:45.830
together and you're like well you get
00:19:45.840 --> 00:19:47.669
free PowerBI you get free this you get
00:19:47.679 --> 00:19:49.350
free that. You get free Microsoft Excel,
00:19:49.360 --> 00:19:50.390
you get free like all these things
00:19:50.400 --> 00:19:54.630
versus one low time fee your soul. Um,
00:19:54.640 --> 00:19:57.830
but before Slack, I used to use Yammer.
00:19:57.840 --> 00:20:00.070
Yama, I used to love it. IT WAS GREAT
00:20:00.080 --> 00:20:01.510
BECAUSE I COULD FOLLOW THE BLOODY
00:20:01.520 --> 00:20:04.150
threads and then the bastards Microsoft
00:20:04.160 --> 00:20:05.990
bought it and they killed it and I had
00:20:06.000 --> 00:20:07.590
to go to Slack like everybody else and I
00:20:07.600 --> 00:20:09.669
still can't find any threads. I, you
00:20:09.679 --> 00:20:11.190
know, the data bricks genie thing is
00:20:11.200 --> 00:20:12.549
really interesting and we got Chris
00:20:12.559 --> 00:20:14.070
coming on in a week's time and he's been
00:20:14.080 --> 00:20:15.590
to the data bricks conference, right? So
00:20:15.600 --> 00:20:16.549
actually at some stage we should
00:20:16.559 --> 00:20:17.909
probably have both you on and have a
00:20:17.919 --> 00:20:20.630
punch up about the conference. But um
00:20:20.640 --> 00:20:22.549
yeah they were announcing this whole
00:20:22.559 --> 00:20:26.549
genieontology and graphy thing and was
00:20:26.559 --> 00:20:30.630
it 4.5 million ontology snippets so it
00:20:30.640 --> 00:20:32.310
can work out how your organization
00:20:32.320 --> 00:20:35.750
works. But I suppose my question is if
00:20:35.760 --> 00:20:37.510
you're going to run data bicks genie and
00:20:37.520 --> 00:20:40.149
you're going to run PowerBI like how do
00:20:40.159 --> 00:20:41.909
they integrate right? Like if I'm going
00:20:41.919 --> 00:20:45.110
to go do chat, you know, ask AI, do I go
00:20:45.120 --> 00:20:47.110
to data bricks? And if I'm going to go
00:20:47.120 --> 00:20:48.710
traditional dashboarding, am I going to
00:20:48.720 --> 00:20:50.549
go to PowerBI? Are you going to embed
00:20:50.559 --> 00:20:54.149
the data bricks genie in PowerBI?
00:20:54.159 --> 00:20:57.190
Like this integration piece. And yeah, I
00:20:57.200 --> 00:20:59.110
still don't believe that OSI is going to
00:20:59.120 --> 00:21:01.669
give us uh one standard to rule them all
00:21:01.679 --> 00:21:04.070
for the BI semantic layer, let alone the
00:21:04.080 --> 00:21:06.789
context plane, right? Um, so you got to
00:21:06.799 --> 00:21:08.470
couple like do you couple those two
00:21:08.480 --> 00:21:10.390
together? Like do you build a new
00:21:10.400 --> 00:21:13.669
interface for the consumer that then in
00:21:13.679 --> 00:21:15.350
the background talks to each of those?
00:21:15.360 --> 00:21:17.190
Like how do you architect this stuff?
00:21:17.200 --> 00:21:18.710
And more importantly, how do you
00:21:18.720 --> 00:21:20.630
architect it today knowing that in six
00:21:20.640 --> 00:21:22.310
months time it's all going to completely
00:21:22.320 --> 00:21:24.789
change? Like are you just going to be
00:21:24.799 --> 00:21:27.110
getting good at rebuilding everything?
00:21:27.120 --> 00:21:28.950
>> Get really good at foundations. Put your
00:21:28.960 --> 00:21:32.310
semantics in. I mean, uh, people I know
00:21:32.320 --> 00:21:34.070
like, uh, I think Bow Plan had a
00:21:34.080 --> 00:21:35.270
conversation about this recently, and I
00:21:35.280 --> 00:21:36.470
was like, "Yeah, they're not too wrong
00:21:36.480 --> 00:21:38.950
about this, but like it's in your code.
00:21:38.960 --> 00:21:40.390
Keep it as close to your code. Keep it
00:21:40.400 --> 00:21:42.710
as raw text as possible and then model
00:21:42.720 --> 00:21:44.070
on top of it." Like we've talked about
00:21:44.080 --> 00:21:45.830
that if you're like in text and you're
00:21:45.840 --> 00:21:48.149
in like practically speaking like in Git
00:21:48.159 --> 00:21:50.710
with your like markdown files with SQL
00:21:50.720 --> 00:21:52.070
with code, everything's kind of
00:21:52.080 --> 00:21:53.590
together, then it really doesn't matter
00:21:53.600 --> 00:21:55.110
where you're moving around because you
00:21:55.120 --> 00:21:56.549
could just like transform it with all
00:21:56.559 --> 00:21:58.630
these large language models. But like
00:21:58.640 --> 00:22:00.470
the further away you go into any
00:22:00.480 --> 00:22:03.270
vertical on these vendors, they have
00:22:03.280 --> 00:22:06.149
like I don't know their special special
00:22:06.159 --> 00:22:08.070
sauce and then yeah, you're trapped,
00:22:08.080 --> 00:22:09.510
right? Like either you're trapped in
00:22:09.520 --> 00:22:11.029
data works. Not that that's wrong. I
00:22:11.039 --> 00:22:12.149
think a lot of people avoid being
00:22:12.159 --> 00:22:13.430
trapped in a vendor at times because
00:22:13.440 --> 00:22:14.789
they're like, "Oh my god, it's the the
00:22:14.799 --> 00:22:16.310
worst thing ever." I think there's just
00:22:16.320 --> 00:22:18.710
like sometimes bigger things to fry out
00:22:18.720 --> 00:22:21.430
there than like getting trapped in.
00:22:21.440 --> 00:22:23.510
>> Yeah. I mean, it's like this open source
00:22:23.520 --> 00:22:24.950
[ __ ] where people tell me they do it
00:22:24.960 --> 00:22:26.549
because it's transportable, right? And
00:22:26.559 --> 00:22:29.990
I'm like, ah, it may cost you nothing to
00:22:30.000 --> 00:22:32.149
move to it, but the cost of the person,
00:22:32.159 --> 00:22:33.669
right, the cost of the people to
00:22:33.679 --> 00:22:36.230
replplum your entire stack onto the open
00:22:36.240 --> 00:22:38.149
source stuff or the other way, that's
00:22:38.159 --> 00:22:39.909
where it really hurts, right? Rebuilding
00:22:39.919 --> 00:22:42.230
it and then retesting it. So testing it
00:22:42.240 --> 00:22:43.990
always kills us. And maybe with some of
00:22:44.000 --> 00:22:46.149
the AI geni tooling now, we're getting
00:22:46.159 --> 00:22:47.909
better automated testing. Kind of
00:22:47.919 --> 00:22:49.350
impressed with that.
00:22:49.360 --> 00:22:52.630
>> But my view is is figure out what's
00:22:52.640 --> 00:22:54.710
cattle and what's a pet. Yeah. Take the
00:22:54.720 --> 00:22:57.430
take the DevOps approach. So for me the
00:22:57.440 --> 00:22:59.590
context right the text about your
00:22:59.600 --> 00:23:02.470
organization is your pet. So protect
00:23:02.480 --> 00:23:04.070
that and put it somewhere where you own
00:23:04.080 --> 00:23:04.710
it
00:23:04.720 --> 00:23:06.789
>> and then hydrate it to those other
00:23:06.799 --> 00:23:08.149
platforms as cattle. So if you're going
00:23:08.159 --> 00:23:10.070
to use data ops genie
00:23:10.080 --> 00:23:11.990
>> master that context somewhere else and
00:23:12.000 --> 00:23:13.830
then just push it into genie and then if
00:23:13.840 --> 00:23:15.669
you have to later on you can push it
00:23:15.679 --> 00:23:18.470
into coco or you can push it into
00:23:18.480 --> 00:23:19.190
anything else
00:23:19.200 --> 00:23:21.190
>> Google knowledge whatever. Isn't that
00:23:21.200 --> 00:23:24.870
great branding? Right. So I we we use
00:23:24.880 --> 00:23:27.990
Google Cloud every day and I know I can
00:23:28.000 --> 00:23:30.789
say data datab bricks Genie and
00:23:30.799 --> 00:23:33.430
Snowflake Coco more often than I could
00:23:33.440 --> 00:23:35.350
try and remember what the context
00:23:35.360 --> 00:23:38.230
knowledge graph thing that Google has
00:23:38.240 --> 00:23:40.310
natively.
00:23:40.320 --> 00:23:42.070
>> Yeah. Wow. I know they have it. It's
00:23:42.080 --> 00:23:43.190
like one of those things that just like
00:23:43.200 --> 00:23:45.350
slips from my mind. So wild. I had to go
00:23:45.360 --> 00:23:48.630
into that Google BigQuery uh the other
00:23:48.640 --> 00:23:50.070
day for some like Google Analytics stuff
00:23:50.080 --> 00:23:51.990
that I was working on. Um, but it was
00:23:52.000 --> 00:23:53.350
just like the interface had changed
00:23:53.360 --> 00:23:55.110
again. I was like, what happened? I've
00:23:55.120 --> 00:23:56.630
only been out of here for like 90 days
00:23:56.640 --> 00:23:58.789
and I was like, what is going on here?
00:23:58.799 --> 00:24:01.350
Um, I was like, how do I query a table?
00:24:01.360 --> 00:24:03.110
And I don't want to use tokens. I just
00:24:03.120 --> 00:24:05.190
wanted to write like a select. I know,
00:24:05.200 --> 00:24:07.590
forgive me, select star from, but like I
00:24:07.600 --> 00:24:09.590
just wanted to see see it. I know that
00:24:09.600 --> 00:24:11.590
like the limiter is not it's like fake.
00:24:11.600 --> 00:24:14.149
It's like cosplay. Uh, but like I just
00:24:14.159 --> 00:24:14.950
wanted to see that
00:24:14.960 --> 00:24:16.870
>> limit limit limits the data that comes
00:24:16.880 --> 00:24:17.590
back. It does not
00:24:17.600 --> 00:24:19.190
>> Yeah, it limits the data, not the size
00:24:19.200 --> 00:24:22.230
of the query. Uh, and I'm just like,
00:24:22.240 --> 00:24:24.070
where'd you go? Where did you put my
00:24:24.080 --> 00:24:25.190
like, you know, where'd you put my beer?
00:24:25.200 --> 00:24:27.669
I can't find it. I think that the like,
00:24:27.679 --> 00:24:29.430
but there's all these like other thesis,
00:24:29.440 --> 00:24:31.110
the capture thesis, I would say, for
00:24:31.120 --> 00:24:32.549
some of these other players. Like, if
00:24:32.559 --> 00:24:35.669
you think about like um these the people
00:24:35.679 --> 00:24:37.190
that are aren't tenants, but they're
00:24:37.200 --> 00:24:39.269
trying to dissolve the platform. So, for
00:24:39.279 --> 00:24:41.590
example, look at uh you got Mother Duck,
00:24:41.600 --> 00:24:43.110
right? So Lar like some of the you have
00:24:43.120 --> 00:24:44.390
the duck lake but essentially they're
00:24:44.400 --> 00:24:46.789
pushing people I think to improve
00:24:46.799 --> 00:24:48.710
iceberg in like a mother duck and their
00:24:48.720 --> 00:24:52.230
play is like a swappable compute engine
00:24:52.240 --> 00:24:54.149
not a fortress like completely the
00:24:54.159 --> 00:24:56.470
opposite of MPP right it's just like a
00:24:56.480 --> 00:24:58.149
very different play that they're going
00:24:58.159 --> 00:25:01.269
for and marketing is brilliant like
00:25:01.279 --> 00:25:03.590
everything seems to do is just like
00:25:03.600 --> 00:25:07.430
clearly they've done it well big fan
00:25:07.440 --> 00:25:08.870
they they really know how to tell a
00:25:08.880 --> 00:25:11.510
story shockingly good using ducks
00:25:11.520 --> 00:25:15.750
Um, and but then it's but it's so
00:25:15.760 --> 00:25:18.470
interesting because of data bricks and
00:25:18.480 --> 00:25:21.669
snowflake almost force adopting iceberg
00:25:21.679 --> 00:25:23.190
and everybody's converging on these open
00:25:23.200 --> 00:25:24.390
tape performance. So then everybody's
00:25:24.400 --> 00:25:26.390
having to move upstream like even even
00:25:26.400 --> 00:25:27.830
Google is starting to allow me to play
00:25:27.840 --> 00:25:29.830
on iceberg and they really didn't want
00:25:29.840 --> 00:25:32.149
me to to do that stuff. Yeah. And it's
00:25:32.159 --> 00:25:33.350
getting better.
00:25:33.360 --> 00:25:36.549
>> Um, but where each of these pets
00:25:36.559 --> 00:25:38.149
>> but I'm seeing ma duck change like I've
00:25:38.159 --> 00:25:39.350
just seen them change their messaging,
00:25:39.360 --> 00:25:41.350
right? So yes, they were, you know,
00:25:41.360 --> 00:25:44.230
around duct DB, around edge, around
00:25:44.240 --> 00:25:46.549
small data, fast data, you know, that
00:25:46.559 --> 00:25:49.110
that kind of stuff. But I think they've
00:25:49.120 --> 00:25:50.710
flipped to a platform play now. They
00:25:50.720 --> 00:25:53.430
bought in the the diet.
00:25:53.440 --> 00:25:56.149
>> Yep. Um, and actually they've just come
00:25:56.159 --> 00:25:59.029
out with cookbooks, right? which is, you
00:25:59.039 --> 00:26:00.950
know, everybody's tried this every every
00:26:00.960 --> 00:26:03.669
wave, every decade, we all try to come
00:26:03.679 --> 00:26:06.549
out with uh cookbooks that are reusable
00:26:06.559 --> 00:26:08.710
and can be, you know, 80% of it and then
00:26:08.720 --> 00:26:11.909
20% can be uh tweaked, right? And I've
00:26:11.919 --> 00:26:13.909
never seen it work, right, commercially.
00:26:13.919 --> 00:26:16.070
Uh be interesting to see what happens
00:26:16.080 --> 00:26:17.830
now, right? Because if we can provide
00:26:17.840 --> 00:26:20.230
that context as a cookbook to claude
00:26:20.240 --> 00:26:23.269
code or whatever and it can do 80% of
00:26:23.279 --> 00:26:26.230
the work for us, uh that's really really
00:26:26.240 --> 00:26:27.909
useful, right? And the interface is
00:26:27.919 --> 00:26:29.350
different to how we used to have to do
00:26:29.360 --> 00:26:31.669
it. But it's like that poll I put on
00:26:31.679 --> 00:26:33.430
LinkedIn the other day. You know what's
00:26:33.440 --> 00:26:35.269
harder to define, customer or
00:26:35.279 --> 00:26:37.190
subscription?
00:26:37.200 --> 00:26:39.750
>> Well, I'll be honest. I was I was I was
00:26:39.760 --> 00:26:43.269
I was club uh subscription because I've
00:26:43.279 --> 00:26:45.190
had to design so many of them over time.
00:26:45.200 --> 00:26:47.750
And to me, it's like it gets like this
00:26:47.760 --> 00:26:48.950
especially when you get into corporate
00:26:48.960 --> 00:26:51.029
finance, it's like what is a
00:26:51.039 --> 00:26:53.110
subscription? We get paid upfront with
00:26:53.120 --> 00:26:54.470
this customer and this customer pays
00:26:54.480 --> 00:26:56.549
later. How do you advertise? It just
00:26:56.559 --> 00:26:59.350
gets so nuts. I would say customer too
00:26:59.360 --> 00:27:01.110
could get like pretty gnarly because
00:27:01.120 --> 00:27:02.230
then you're trying to get the customer
00:27:02.240 --> 00:27:04.870
to define even the stages of a customer.
00:27:04.880 --> 00:27:07.110
Uh but yeah, I was doing some team.
00:27:07.120 --> 00:27:08.390
>> Yeah. And so how does that work with a
00:27:08.400 --> 00:27:10.070
cookbook, right? So I can get a cookbook
00:27:10.080 --> 00:27:12.310
for Stripe. I can get a cookbook for
00:27:12.320 --> 00:27:14.070
Charify.
00:27:14.080 --> 00:27:16.230
But that's not going to give me the
00:27:16.240 --> 00:27:17.990
definition of subscription for any
00:27:18.000 --> 00:27:20.070
organization that I've ever done it for
00:27:20.080 --> 00:27:22.390
because there's always the if then but
00:27:22.400 --> 00:27:25.029
else, right? Oh well, we do trials and
00:27:25.039 --> 00:27:26.789
actually we didn't really want to do
00:27:26.799 --> 00:27:28.789
them properly. So what we do is yada
00:27:28.799 --> 00:27:31.510
yada yada yada, right? And Stripe and
00:27:31.520 --> 00:27:33.750
that's the context that's almost
00:27:33.760 --> 00:27:36.390
impossible uh to put into a cookbook. Or
00:27:36.400 --> 00:27:39.190
maybe it's not. Maybe, you know, every
00:27:39.200 --> 00:27:41.029
platform that's seeing that behavior
00:27:41.039 --> 00:27:44.149
will then strip out the context of the
00:27:44.159 --> 00:27:46.149
five different ways everybody does it.
00:27:46.159 --> 00:27:47.750
Because you know in theory I'm assuming
00:27:47.760 --> 00:27:50.390
there's five to 10 to 20 patterns across
00:27:50.400 --> 00:27:51.990
every subscription company on how they
00:27:52.000 --> 00:27:54.710
deal with all those not edge cases but
00:27:54.720 --> 00:27:56.630
all those cases
00:27:56.640 --> 00:27:59.350
>> you know you know a subscription or a
00:27:59.360 --> 00:28:01.750
customer the irony is every business is
00:28:01.760 --> 00:28:03.350
like a snowflake the devil's in the
00:28:03.360 --> 00:28:06.149
details like all those things is so wild
00:28:06.159 --> 00:28:08.630
it's so wild like every person I ever
00:28:08.640 --> 00:28:11.269
consulted or done work for every time I
00:28:11.279 --> 00:28:14.149
think I'm about to apply a pattern or a
00:28:14.159 --> 00:28:16.470
certain set of rules it's just like it
00:28:16.480 --> 00:28:18.389
has to be unique to that customer or
00:28:18.399 --> 00:28:21.350
that company. And I was like,
00:28:21.360 --> 00:28:24.710
okay, why why did you have to redefine
00:28:24.720 --> 00:28:27.190
this this thing? And I mean, maybe it's
00:28:27.200 --> 00:28:28.789
important for some from some parts of
00:28:28.799 --> 00:28:30.149
the business and maybe that's like how
00:28:30.159 --> 00:28:33.350
they they define it. But um
00:28:33.360 --> 00:28:34.549
>> but I think there's some value in the
00:28:34.559 --> 00:28:36.230
cookbook idea if it tells me the
00:28:36.240 --> 00:28:37.669
patterns I got to watch out for, right?
00:28:37.679 --> 00:28:39.430
So, you know, first time I rocked up to
00:28:39.440 --> 00:28:41.750
a company that wanted me to go and help
00:28:41.760 --> 00:28:43.350
codif you know, physically codify
00:28:43.360 --> 00:28:45.029
subscriptions and count them. I'm like,
00:28:45.039 --> 00:28:46.389
"Yeah, subscriptions, baby." Like,
00:28:46.399 --> 00:28:48.070
"We'll do customer last because that's
00:28:48.080 --> 00:28:50.070
hard. How hard can a subscription be?"
00:28:50.080 --> 00:28:51.909
And then I learned. And so I wonder
00:28:51.919 --> 00:28:54.310
whether actually uh those cookbooks,
00:28:54.320 --> 00:28:56.710
right, where it says 80% of, you know,
00:28:56.720 --> 00:28:58.549
of the subscription process looks like
00:28:58.559 --> 00:29:01.029
this. Everybody pretty much follows it.
00:29:01.039 --> 00:29:03.110
Then here's all the edge cases or the
00:29:03.120 --> 00:29:04.789
things that everybody does various
00:29:04.799 --> 00:29:08.470
patterns for. So ask them about trials.
00:29:08.480 --> 00:29:11.269
Uh ask them about prepayment or, you
00:29:11.279 --> 00:29:12.870
know, uh ask them what happens about
00:29:12.880 --> 00:29:15.430
cancellations. ask them what happens
00:29:15.440 --> 00:29:17.909
when somebody subscribes, cancels, and
00:29:17.919 --> 00:29:20.630
then resubscribes, right? That that that
00:29:20.640 --> 00:29:22.470
>> that funnel because there's a bunch of
00:29:22.480 --> 00:29:24.950
use cases that really kill you. And then
00:29:24.960 --> 00:29:26.630
maybe that's where the value is. So when
00:29:26.640 --> 00:29:27.830
you're going to that domain for the
00:29:27.840 --> 00:29:28.950
first time,
00:29:28.960 --> 00:29:30.389
>> you're kind of getting the hints, right,
00:29:30.399 --> 00:29:31.990
of things to look for and how to look
00:29:32.000 --> 00:29:33.830
for them. Maybe that's where the first
00:29:33.840 --> 00:29:36.310
round of value of those patent cookbooks
00:29:36.320 --> 00:29:36.950
will be.
00:29:36.960 --> 00:29:38.549
>> Yeah. Patent cookbooks is also like
00:29:38.559 --> 00:29:40.549
honestly why I like doing data detective
00:29:40.559 --> 00:29:43.190
work, right? And even like you know
00:29:43.200 --> 00:29:44.230
think about you think about the
00:29:44.240 --> 00:29:45.350
cookbooks but like even how you
00:29:45.360 --> 00:29:47.350
investigate like data structures on
00:29:47.360 --> 00:29:48.789
companies today is very different. I
00:29:48.799 --> 00:29:50.470
mean obviously depends on the client or
00:29:50.480 --> 00:29:51.909
depends on the company working on the
00:29:51.919 --> 00:29:55.590
access to AI stuff but like building um
00:29:55.600 --> 00:29:58.070
data sense maker right and like how do
00:29:58.080 --> 00:29:59.669
you make sense of company and then
00:29:59.679 --> 00:30:01.190
you're like looking at the data and then
00:30:01.200 --> 00:30:02.950
you're looking at the you're looking at
00:30:02.960 --> 00:30:05.110
the reports and these are two different
00:30:05.120 --> 00:30:07.190
things like I don't understand like
00:30:07.200 --> 00:30:08.789
where where's the context and you
00:30:08.799 --> 00:30:10.310
realize it's like in Bob's head and
00:30:10.320 --> 00:30:11.990
you're like okay Bob I can look at your
00:30:12.000 --> 00:30:14.549
sequel and now like thankfully you know
00:30:14.559 --> 00:30:16.149
I can you sit down with plaud or
00:30:16.159 --> 00:30:18.230
whatever LLM of choice this week and
00:30:18.240 --> 00:30:20.310
like kind of reverse engineer it a bit
00:30:20.320 --> 00:30:21.590
to get a better pretty good
00:30:21.600 --> 00:30:23.669
understanding. But then of course the
00:30:23.679 --> 00:30:24.789
hardest part is like now you got to
00:30:24.799 --> 00:30:28.149
change Bob's logic and you're like okay
00:30:28.159 --> 00:30:29.750
>> I tell you one of the things I do when I
00:30:29.760 --> 00:30:31.590
coach a team is at some stage I'll ask
00:30:31.600 --> 00:30:33.350
them how the company makes money because
00:30:33.360 --> 00:30:34.630
I'm kind of interested if I don't know
00:30:34.640 --> 00:30:36.149
already but I'm also interested if they
00:30:36.159 --> 00:30:38.549
know like how does a company make money
00:30:38.559 --> 00:30:40.389
right because that's kind of important
00:30:40.399 --> 00:30:43.029
if you're a data team. Um I had this the
00:30:43.039 --> 00:30:44.870
other day somebody was saying oh their
00:30:44.880 --> 00:30:47.029
team's structured here and and they kind
00:30:47.039 --> 00:30:49.510
of I think it was in it under the CTO or
00:30:49.520 --> 00:30:51.269
whatever and they they were looking to
00:30:51.279 --> 00:30:53.269
move you know should they move the team
00:30:53.279 --> 00:30:54.549
to another part of the business and if
00:30:54.559 --> 00:30:56.950
so how do they determine um which part
00:30:56.960 --> 00:30:59.830
and answer to them was which part of the
00:30:59.840 --> 00:31:00.870
business you know how does the business
00:31:00.880 --> 00:31:03.669
make money and they told me I said
00:31:03.679 --> 00:31:05.269
>> who who's the most senior person in the
00:31:05.279 --> 00:31:06.549
organization apart from the chief
00:31:06.559 --> 00:31:08.549
executive that runs that part of the
00:31:08.559 --> 00:31:10.149
business that makes money and I think
00:31:10.159 --> 00:31:11.830
they said chief chief marketing officer
00:31:11.840 --> 00:31:13.909
or something and I was like that's who
00:31:13.919 --> 00:31:14.950
you want to work for, right? Because
00:31:14.960 --> 00:31:17.190
they've got the problems that actually
00:31:17.200 --> 00:31:18.549
will make a difference to the
00:31:18.559 --> 00:31:21.750
organization using data. So because I
00:31:21.760 --> 00:31:24.789
asked teams that I wonder if actually if
00:31:24.799 --> 00:31:27.510
I asked an LLM that's been running in an
00:31:27.520 --> 00:31:29.190
organization for a while, how does a
00:31:29.200 --> 00:31:30.470
company make money?
00:31:30.480 --> 00:31:31.990
>> That would be interesting, right? Like
00:31:32.000 --> 00:31:34.549
could it actually detect where that that
00:31:34.559 --> 00:31:36.710
value is?
00:31:36.720 --> 00:31:38.789
Um,
00:31:38.799 --> 00:31:40.470
yeah, some of them, some of them, I
00:31:40.480 --> 00:31:42.710
mean, it just really depends on how good
00:31:42.720 --> 00:31:44.230
how good the player is, right? The
00:31:44.240 --> 00:31:45.590
person like gets in command of these
00:31:45.600 --> 00:31:47.750
harnesses and and stuff that you we've
00:31:47.760 --> 00:31:50.310
talked about before. Um, I think that's
00:31:50.320 --> 00:31:52.950
like really really important of
00:31:52.960 --> 00:31:55.350
basically how
00:31:55.360 --> 00:31:57.190
um
00:31:57.200 --> 00:31:59.430
you're creating like almost like you
00:31:59.440 --> 00:32:01.590
know wiki, everybody had like those wiks
00:32:01.600 --> 00:32:02.950
like everybody had to have a wiki that
00:32:02.960 --> 00:32:04.389
nobody maintained. We all said we're
00:32:04.399 --> 00:32:07.590
going to do it. know it. Um it's crazy
00:32:07.600 --> 00:32:10.950
on how important that has become. It's
00:32:10.960 --> 00:32:13.029
like the downside is that then if you go
00:32:13.039 --> 00:32:14.870
and of course use that old wiki to go
00:32:14.880 --> 00:32:17.190
into your new data models, it's so wrong
00:32:17.200 --> 00:32:18.789
because then none of them got updated.
00:32:18.799 --> 00:32:20.310
So it's like you know everybody says
00:32:20.320 --> 00:32:22.149
like what was it like everybody should
00:32:22.159 --> 00:32:24.230
document but nobody wants to and thus it
00:32:24.240 --> 00:32:25.909
continues and now you have AI helping
00:32:25.919 --> 00:32:28.389
you document it. So I mean maybe that
00:32:28.399 --> 00:32:31.669
will uh that will get better now that we
00:32:31.679 --> 00:32:32.870
can all be lazier about our
00:32:32.880 --> 00:32:35.110
documentation. I mean, okay, hang on,
00:32:35.120 --> 00:32:36.789
we'll go blast the past. Maybe Okay,
00:32:36.799 --> 00:32:39.190
remember Dragon Naturally Speaking?
00:32:39.200 --> 00:32:39.830
>> Yeah.
00:32:39.840 --> 00:32:41.509
>> Yeah. Like, I remember using so many
00:32:41.519 --> 00:32:43.590
versions and I like, you know, this is
00:32:43.600 --> 00:32:46.630
the like early 2000s and I was like,
00:32:46.640 --> 00:32:49.029
save so much money to go buy this $200
00:32:49.039 --> 00:32:50.950
piece of software, get the DVD, probably
00:32:50.960 --> 00:32:53.590
CD at the time, install it, use it, and
00:32:53.600 --> 00:32:54.870
you're like, you see all these
00:32:54.880 --> 00:32:57.669
commercials or magazine things about how
00:32:57.679 --> 00:33:00.630
good it's going to be. Uh, Dragon
00:33:00.640 --> 00:33:02.149
Naturally Speaking. Going back here,
00:33:02.159 --> 00:33:06.549
Mark. Uh, and it was so bad. It was so
00:33:06.559 --> 00:33:09.830
bad, but the ad in PC magazine was so
00:33:09.840 --> 00:33:14.549
good. But oh boy. Um, but now now the
00:33:14.559 --> 00:33:17.269
language models like dictating like I
00:33:17.279 --> 00:33:19.269
use I use Super Whisper for privacy. I
00:33:19.279 --> 00:33:20.470
like it. Whisper flow. I know it's
00:33:20.480 --> 00:33:21.509
everybody's vibe, but I like Super
00:33:21.519 --> 00:33:23.750
Whisper. And its translation and
00:33:23.760 --> 00:33:27.269
dictation is wild. I recently was
00:33:27.279 --> 00:33:29.830
yelling at Siri because it just could
00:33:29.840 --> 00:33:32.310
not understand me and my Canadian accent
00:33:32.320 --> 00:33:33.830
is not that big. I'm sorry. Or at least
00:33:33.840 --> 00:33:34.470
I
00:33:34.480 --> 00:33:36.149
>> No, try trying to be a Kiwi where you
00:33:36.159 --> 00:33:37.750
know everything's fuss and chops. Hey,
00:33:37.760 --> 00:33:39.269
just a shout out to Mark Freeman for
00:33:39.279 --> 00:33:40.870
joining us and saying hello. Anybody
00:33:40.880 --> 00:33:42.310
who's on actually wants to ask a
00:33:42.320 --> 00:33:44.389
question. It is ask us anything. But as
00:33:44.399 --> 00:33:46.630
I said in the in the promos, otherwise
00:33:46.640 --> 00:33:49.669
we'll just talk [ __ ] for a while. Um
00:33:49.679 --> 00:33:51.590
that dictation one's really
00:33:51.600 --> 00:33:53.990
>> That was our first I managed to dodge a
00:33:54.000 --> 00:33:56.149
few of them. Okay. What's up?
00:33:56.159 --> 00:33:58.389
>> Weird.
00:33:58.399 --> 00:34:01.110
>> Not where I come from. Um, like one of
00:34:01.120 --> 00:34:03.029
the people, one of the the teams I was
00:34:03.039 --> 00:34:04.630
working with a while ago, we we
00:34:04.640 --> 00:34:07.669
experimented with uh the developers, the
00:34:07.679 --> 00:34:10.389
engineers actually writing uh the
00:34:10.399 --> 00:34:12.389
context of what they were doing as they
00:34:12.399 --> 00:34:13.349
were coding
00:34:13.359 --> 00:34:15.589
>> uh separately, right? So, this idea of
00:34:15.599 --> 00:34:18.069
just write your thought stream uh no
00:34:18.079 --> 00:34:19.669
structure, no template, right? try and
00:34:19.679 --> 00:34:22.149
make it really really easy uh and then
00:34:22.159 --> 00:34:24.069
grab all that context over time, whack
00:34:24.079 --> 00:34:26.950
it into an LLM and see how much insight
00:34:26.960 --> 00:34:28.470
it gave. And that was really
00:34:28.480 --> 00:34:29.750
interesting. And and I kind of thought,
00:34:29.760 --> 00:34:31.510
well, actually, why don't we just
00:34:31.520 --> 00:34:33.430
dictate it? I mean, I don't I'm not a
00:34:33.440 --> 00:34:37.990
dictator, right? Um yeah. Uh so I don't
00:34:38.000 --> 00:34:40.470
I don't think that way. I always yet I'm
00:34:40.480 --> 00:34:42.310
going to go try and change it. Um but
00:34:42.320 --> 00:34:43.669
maybe that's what we should be doing,
00:34:43.679 --> 00:34:45.270
right? as we're coding, as we're
00:34:45.280 --> 00:34:46.869
figuring, well, as we're telling Paul to
00:34:46.879 --> 00:34:49.750
code, um, we should be kind of dictating
00:34:49.760 --> 00:34:51.669
why we're doing it. And remember there
00:34:51.679 --> 00:34:54.550
was that guy Nico and somebody else that
00:34:54.560 --> 00:34:58.790
uh did uh what was their product called?
00:34:58.800 --> 00:35:01.030
They did like a version that was uh
00:35:01.040 --> 00:35:03.109
their first version of the company was
00:35:03.119 --> 00:35:05.990
uh orchestrating the modern data stack
00:35:06.000 --> 00:35:08.069
uh kind of thing, but their second one
00:35:08.079 --> 00:35:11.190
was uh analyst journal. And so basically
00:35:11.200 --> 00:35:13.670
as an analyst you were working in this
00:35:13.680 --> 00:35:15.670
IDE where you were kind of quering the
00:35:15.680 --> 00:35:17.589
data, writing code and looping through
00:35:17.599 --> 00:35:20.230
it all. And at on the side you were
00:35:20.240 --> 00:35:22.390
keeping notes uh about why you did that
00:35:22.400 --> 00:35:24.390
and and their their idea was when you
00:35:24.400 --> 00:35:25.829
come back six months later and you're
00:35:25.839 --> 00:35:27.829
like why the hell did I put that where
00:35:27.839 --> 00:35:30.150
clause in there? You had your journal
00:35:30.160 --> 00:35:32.550
notes on the right hand side, right? Uh
00:35:32.560 --> 00:35:34.790
so it's highly valuable to you. Maybe
00:35:34.800 --> 00:35:35.829
you'd share it with the rest of your
00:35:35.839 --> 00:35:37.430
team depending on how you feel about job
00:35:37.440 --> 00:35:40.390
security. Uh and and so my theory was
00:35:40.400 --> 00:35:43.109
like wow that was great product too
00:35:43.119 --> 00:35:45.349
early because right now if if you were
00:35:45.359 --> 00:35:47.589
doing that that context of why you're
00:35:47.599 --> 00:35:50.230
writing that code uh is really really
00:35:50.240 --> 00:35:52.150
valuable. So yeah, maybe try just using
00:35:52.160 --> 00:35:55.990
whisper. But going back to your pet,
00:35:56.000 --> 00:35:59.589
>> Nico, I miss your open Nico. I forgot
00:35:59.599 --> 00:36:01.589
his co-founder's name. Sorry. But like
00:36:01.599 --> 00:36:04.390
reconfigured and it was I was I was a
00:36:04.400 --> 00:36:05.670
pain customer.
00:36:05.680 --> 00:36:08.230
>> A you are so early adopter. I reckon
00:36:08.240 --> 00:36:09.750
every time we do this every month, we're
00:36:09.760 --> 00:36:12.630
going to have to have Tim's top pick for
00:36:12.640 --> 00:36:14.630
bleeding edge [ __ ] to play with this
00:36:14.640 --> 00:36:15.190
month.
00:36:15.200 --> 00:36:17.589
>> Oh man, I'm like shout out to Mark
00:36:17.599 --> 00:36:19.349
Freeman because I've been using Petri.
00:36:19.359 --> 00:36:21.109
Sorry man. I'm so confused. I'll figure
00:36:21.119 --> 00:36:22.390
it out. I swear.
00:36:22.400 --> 00:36:23.750
>> Okay. So, that's the one though. Is that
00:36:23.760 --> 00:36:26.390
Is that our top Tim's top tip for next
00:36:26.400 --> 00:36:28.470
this month to because you're you're at
00:36:28.480 --> 00:36:29.670
the beginning of the month, right?
00:36:29.680 --> 00:36:30.069
>> Yeah.
00:36:30.079 --> 00:36:32.470
>> Leading edge tool to play with Petri.
00:36:32.480 --> 00:36:35.030
>> Uh God, what have I got today? I'm using
00:36:35.040 --> 00:36:35.910
>> No sponsorship.
00:36:35.920 --> 00:36:37.270
>> Oh, I got one. No sponsorship by the
00:36:37.280 --> 00:36:39.030
way. Um but I'm using one called human
00:36:39.040 --> 00:36:41.589
layer. Oh man, if you love deep coding
00:36:41.599 --> 00:36:43.670
and like like harnesses and all the
00:36:43.680 --> 00:36:45.430
concept stuff, but like we were talking
00:36:45.440 --> 00:36:47.349
about it and working me working through
00:36:47.359 --> 00:36:49.030
it and you're like, "Oh my god, that's
00:36:49.040 --> 00:36:51.349
so much like frontloading and it's like
00:36:51.359 --> 00:36:54.069
the whole structured process and it's
00:36:54.079 --> 00:36:57.589
like it's like BMAD on steroids."
00:36:57.599 --> 00:36:59.109
>> Yeah. Let's talk about that, right?
00:36:59.119 --> 00:37:00.950
Because um I tried to use superpowers
00:37:00.960 --> 00:37:02.630
and I it kept asking me too many
00:37:02.640 --> 00:37:04.150
questions and I'm like just shut up and
00:37:04.160 --> 00:37:06.230
go away and then I deleted the skill.
00:37:06.240 --> 00:37:08.310
It's a I have this theory that actually
00:37:08.320 --> 00:37:09.750
because you know vibe coding allows
00:37:09.760 --> 00:37:10.950
people like me that are completely
00:37:10.960 --> 00:37:13.430
unstructured and can't code for [ __ ] to
00:37:13.440 --> 00:37:16.230
do really cool or dangerous things. And
00:37:16.240 --> 00:37:17.829
so I think there's kind of three people,
00:37:17.839 --> 00:37:20.230
right? I think there's people like you
00:37:20.240 --> 00:37:22.230
that come from the background of doing
00:37:22.240 --> 00:37:24.470
it as a as a practitioner and are very
00:37:24.480 --> 00:37:26.230
structured, right? So you don't mind
00:37:26.240 --> 00:37:29.430
writing a quite detailed spec uh with
00:37:29.440 --> 00:37:31.990
acceptance criteria up front and then
00:37:32.000 --> 00:37:33.829
let the engines run off and actually
00:37:33.839 --> 00:37:36.550
build it, right? and then then test it.
00:37:36.560 --> 00:37:38.069
That's one extreme, one side. And then
00:37:38.079 --> 00:37:39.349
there's people like me where I'm just
00:37:39.359 --> 00:37:41.349
going to go natural language is [ __ ] and
00:37:41.359 --> 00:37:43.190
then test it and just keep iterating as
00:37:43.200 --> 00:37:45.030
fast as I can until it actually does
00:37:45.040 --> 00:37:46.950
something that I like. And then I think
00:37:46.960 --> 00:37:48.230
there's people in the middle, right,
00:37:48.240 --> 00:37:50.950
that are kind of not willing to write it
00:37:50.960 --> 00:37:52.950
all up front, but not willing to just
00:37:52.960 --> 00:37:56.230
vibe it, you know, to to yolo it, uh, or
00:37:56.240 --> 00:37:58.950
beer back it. Um, and so they're kind of
00:37:58.960 --> 00:37:59.990
stuck in the middle. And I'm not sure
00:38:00.000 --> 00:38:01.270
I'd want to be the person in the middle,
00:38:01.280 --> 00:38:02.630
right? I think you want to be one of the
00:38:02.640 --> 00:38:05.109
two. Um, and there's a whole lot of
00:38:05.119 --> 00:38:06.790
harnesses now, right? So, you know,
00:38:06.800 --> 00:38:08.310
we've talked a while ago, we talked
00:38:08.320 --> 00:38:11.030
about now and KTX, and we talk about
00:38:11.040 --> 00:38:12.710
them because they're doing a brilliant
00:38:12.720 --> 00:38:14.790
in paid influencer thing at the moment,
00:38:14.800 --> 00:38:16.230
right? They're kind of popping up in the
00:38:16.240 --> 00:38:18.390
feed all over the place. Um, there's
00:38:18.400 --> 00:38:21.030
another one which is uh open UI, open
00:38:21.040 --> 00:38:23.510
web UI, web UI, which isn't for data
00:38:23.520 --> 00:38:25.750
people, but uh is another harnessy
00:38:25.760 --> 00:38:28.550
framework um for putting in your prompts
00:38:28.560 --> 00:38:30.470
and and all your aentic context and all
00:38:30.480 --> 00:38:33.589
the other context. Um, so before we do
00:38:33.599 --> 00:38:35.510
that, do you want to have a go at
00:38:35.520 --> 00:38:41.109
explaining what a harness is?
00:38:41.119 --> 00:38:42.950
>> I mean, I could try. I feel like every
00:38:42.960 --> 00:38:45.030
time I try to do it, I'm like I'm like I
00:38:45.040 --> 00:38:45.990
don't
00:38:46.000 --> 00:38:48.790
>> get it. Um,
00:38:48.800 --> 00:38:50.950
I mean, a hardest to me is like an or
00:38:50.960 --> 00:38:52.950
it's like an orchestration of markdown
00:38:52.960 --> 00:38:55.670
files. It's like the least it's just
00:38:55.680 --> 00:38:57.990
like almost like a giant if then then
00:38:58.000 --> 00:39:01.990
else component of it. like a harness to
00:39:02.000 --> 00:39:04.069
me and I'm probably wrong here folks and
00:39:04.079 --> 00:39:05.510
despite reading like a hundred articles
00:39:05.520 --> 00:39:10.150
on this stuff. Um
00:39:10.160 --> 00:39:12.470
I mean it's almost like you know a mini
00:39:12.480 --> 00:39:17.030
airflow or a mini dag
00:39:17.040 --> 00:39:19.030
what's the difference between loop
00:39:19.040 --> 00:39:20.310
engineer?
00:39:20.320 --> 00:39:22.950
>> Uh so my answer to that would be don't
00:39:22.960 --> 00:39:24.790
loop engineer until you have a harness.
00:39:24.800 --> 00:39:27.589
So the way I think about harness is it's
00:39:27.599 --> 00:39:29.750
a bunch of guardrails
00:39:29.760 --> 00:39:30.310
and
00:39:30.320 --> 00:39:32.470
>> either way I mean you think of state as
00:39:32.480 --> 00:39:33.589
a guardrail but sure.
00:39:33.599 --> 00:39:35.430
>> Yeah. Yeah. So the mechanics is a bunch
00:39:35.440 --> 00:39:37.990
of MD files or a bunch of if statements.
00:39:38.000 --> 00:39:41.510
Uh I tend to I tend to take my coaching
00:39:41.520 --> 00:39:43.510
harness and apply it to the LLM is kind
00:39:43.520 --> 00:39:44.550
of interesting. So I have a
00:39:44.560 --> 00:39:46.390
>> Is it only caps though? Do you like keep
00:39:46.400 --> 00:39:48.310
them?
00:39:48.320 --> 00:39:50.550
What what's the one where you do like f
00:39:50.560 --> 00:39:52.390
you put two words together and the first
00:39:52.400 --> 00:39:55.589
letter of each one is caps.
00:39:55.599 --> 00:39:56.630
>> Camel case.
00:39:56.640 --> 00:39:58.710
>> Yeah. Yeah. I came everything where I
00:39:58.720 --> 00:40:00.790
don't underscore it and I don't know if
00:40:00.800 --> 00:40:01.990
I'm a tabs or
00:40:02.000 --> 00:40:03.270
>> burn some tokens
00:40:03.280 --> 00:40:05.829
>> spaces. Um
00:40:05.839 --> 00:40:09.430
so so I have a a persona file which
00:40:09.440 --> 00:40:11.349
tells the agent how I want it to behave
00:40:11.359 --> 00:40:13.589
and who it is. I have a policy file
00:40:13.599 --> 00:40:15.270
which is the rules it can't break,
00:40:15.280 --> 00:40:17.430
right? very few of them but it's like
00:40:17.440 --> 00:40:20.150
you cannot do these things. I have a way
00:40:20.160 --> 00:40:22.950
of working file which basically uh says
00:40:22.960 --> 00:40:24.710
this is how we work and how we do tasks
00:40:24.720 --> 00:40:26.710
together and all those kind of things.
00:40:26.720 --> 00:40:28.950
Uh and then I have a retro file where
00:40:28.960 --> 00:40:31.829
I'm just working through um some kind of
00:40:31.839 --> 00:40:34.550
self self looping at the moment where uh
00:40:34.560 --> 00:40:37.190
I have a red blue team and a judge
00:40:37.200 --> 00:40:39.270
reading all my logs, all our chat logs
00:40:39.280 --> 00:40:41.349
and then putting stuff into the retro
00:40:41.359 --> 00:40:43.589
that it thinks it should update. And I'm
00:40:43.599 --> 00:40:45.349
just finishing off the bit where there's
00:40:45.359 --> 00:40:46.790
another task that goes through and reads
00:40:46.800 --> 00:40:48.310
the retros and then updates those other
00:40:48.320 --> 00:40:50.630
three files. And so and then of course
00:40:50.640 --> 00:40:52.310
you're using code there's skills right
00:40:52.320 --> 00:40:53.910
which makes it even slightly harder
00:40:53.920 --> 00:40:55.670
because that's a combination. And
00:40:55.680 --> 00:40:57.670
there's claude and then there's your
00:40:57.680 --> 00:40:59.829
memory right which is a persistent store
00:40:59.839 --> 00:41:01.109
of all the things you've done when you
00:41:01.119 --> 00:41:03.990
start new session. So all those things
00:41:04.000 --> 00:41:07.030
are there to allow you to start a new
00:41:07.040 --> 00:41:09.030
session and it remember where you were
00:41:09.040 --> 00:41:11.349
and not lose the context of all the work
00:41:11.359 --> 00:41:13.910
you've done with the LLM before. And
00:41:13.920 --> 00:41:15.510
they're there to stop it doing things
00:41:15.520 --> 00:41:17.910
you don't like or things you do like you
00:41:17.920 --> 00:41:20.870
know and that's it. loop engineering
00:41:20.880 --> 00:41:22.230
because I've only started reading on
00:41:22.240 --> 00:41:24.630
this is this idea that actually I can
00:41:24.640 --> 00:41:29.589
now give an LLM a goal and then it using
00:41:29.599 --> 00:41:31.589
my guard rails it will then just keep
00:41:31.599 --> 00:41:34.710
looping itself around and around until
00:41:34.720 --> 00:41:37.510
it achieves that goal which if we go
00:41:37.520 --> 00:41:39.990
back to me that means I have to write
00:41:40.000 --> 00:41:42.069
acceptance criteria and eval criteria
00:41:42.079 --> 00:41:44.309
what success looks like so I'm never
00:41:44.319 --> 00:41:46.790
going to loop successfully for you yeah
00:41:46.800 --> 00:41:48.550
you could probably go this is what this
00:41:48.560 --> 00:41:49.990
is what good looks like right when you
00:41:50.000 --> 00:41:52.470
get to hear finished. Um, but that's how
00:41:52.480 --> 00:41:53.670
I see the difference. What do you think
00:41:53.680 --> 00:41:54.069
of that?
00:41:54.079 --> 00:41:55.910
>> Well, yeah. I mean, that's that's very
00:41:55.920 --> 00:41:59.030
good. I was saying like, you know, uh
00:41:59.040 --> 00:42:00.790
Bart still on the line if you still here
00:42:00.800 --> 00:42:02.309
because he's got a scientist background.
00:42:02.319 --> 00:42:03.910
You can see Petri like it's like built
00:42:03.920 --> 00:42:05.829
on that kind of like looping from what I
00:42:05.839 --> 00:42:08.150
can understand. And but you look at like
00:42:08.160 --> 00:42:10.150
um Carpathy like he did like he's done
00:42:10.160 --> 00:42:11.349
so many interesting things and one of
00:42:11.359 --> 00:42:12.950
the ones was like his wiki was really
00:42:12.960 --> 00:42:15.829
interesting. Um but the other thing was
00:42:15.839 --> 00:42:17.750
this like auto research which is
00:42:17.760 --> 00:42:20.150
basically it was just a big loop
00:42:20.160 --> 00:42:21.510
constantly going over it giving it the
00:42:21.520 --> 00:42:23.430
tools it's 100 lines of Python and it
00:42:23.440 --> 00:42:25.430
was like same thing what's you know
00:42:25.440 --> 00:42:26.630
ultimately what's the goal that you're
00:42:26.640 --> 00:42:28.069
trying to achieve here I think one of
00:42:28.079 --> 00:42:29.750
the fun things as the models have got
00:42:29.760 --> 00:42:31.349
improved and you can test obviously you
00:42:31.359 --> 00:42:33.190
can attest against this is like the
00:42:33.200 --> 00:42:35.030
weird wiggles that all the all the
00:42:35.040 --> 00:42:37.829
models have when you're trying to like
00:42:37.839 --> 00:42:40.069
ascertain a goal and like how they all
00:42:40.079 --> 00:42:43.109
get to that and also how crazy different
00:42:43.119 --> 00:42:44.470
they can
00:42:44.480 --> 00:42:47.270
like
00:42:47.280 --> 00:42:49.430
>> I think the other thing is
00:42:49.440 --> 00:42:51.430
>> when we talk about harnesses so we'll
00:42:51.440 --> 00:42:53.349
codify some guard rails or some behavior
00:42:53.359 --> 00:42:55.670
in our harness and then when the next
00:42:55.680 --> 00:42:57.750
model comes out it's actually better at
00:42:57.760 --> 00:42:59.910
dealing with that problem than the
00:42:59.920 --> 00:43:03.270
codified way we did it and so the if you
00:43:03.280 --> 00:43:05.670
don't remove it out of your harness then
00:43:05.680 --> 00:43:07.589
actually the model is not as good as it
00:43:07.599 --> 00:43:09.670
could be and so I don't know how you do
00:43:09.680 --> 00:43:11.349
that right how do you go back and say to
00:43:11.359 --> 00:43:13.910
it well look at the latest model cuz I
00:43:13.920 --> 00:43:15.829
noticed that um Fable 5's just been
00:43:15.839 --> 00:43:17.990
released been coming back on slowly.
00:43:18.000 --> 00:43:20.630
Don't have it in my one, but um Sonet 5
00:43:20.640 --> 00:43:23.030
is out now, right? So, how do I get it
00:43:23.040 --> 00:43:25.670
to look at Sonnet 5, figure out what
00:43:25.680 --> 00:43:28.950
it's better at than I than my harness is
00:43:28.960 --> 00:43:31.349
and then remove that from my harness
00:43:31.359 --> 00:43:32.870
because the NATS foundational model
00:43:32.880 --> 00:43:34.550
deals with it.
00:43:34.560 --> 00:43:36.150
>> Like there's too many moving parts now,
00:43:36.160 --> 00:43:38.630
right? Too many moving parts. the famous
00:43:38.640 --> 00:43:40.710
the famous what like 12-month role or
00:43:40.720 --> 00:43:42.390
14month role of prompt engineering is
00:43:42.400 --> 00:43:44.150
already at Swanson.
00:43:44.160 --> 00:43:49.270
>> Um the prompt engineer rules damn it. Um
00:43:49.280 --> 00:43:50.870
it's really interesting. So I'm like I'm
00:43:50.880 --> 00:43:52.470
working that I'm like you know working
00:43:52.480 --> 00:43:55.190
in this space obviously. Um
00:43:55.200 --> 00:43:57.190
>> and I'm still very much like hands-on
00:43:57.200 --> 00:43:59.510
coding. Well, let's be honest. 90% AI
00:43:59.520 --> 00:44:01.910
coding, 10% practical coding. I did I
00:44:01.920 --> 00:44:03.910
did teach myself Rust by hand without
00:44:03.920 --> 00:44:06.630
using too much AI. That was brutal. is
00:44:06.640 --> 00:44:08.630
so hard. So hard than I thought it was
00:44:08.640 --> 00:44:10.470
going to be. I was like, "God, only two
00:44:10.480 --> 00:44:13.109
years ago, this like it would have been
00:44:13.119 --> 00:44:14.069
I'm not gonna say it would have been
00:44:14.079 --> 00:44:15.190
easier to learn, but I would have
00:44:15.200 --> 00:44:16.710
probably been like less addicted to just
00:44:16.720 --> 00:44:19.910
asking a quick question." Um,
00:44:19.920 --> 00:44:22.309
and uh, God, we got to talk about that
00:44:22.319 --> 00:44:23.990
like the ability to understand stuff
00:44:24.000 --> 00:44:27.589
with all this AI offloading, but um,
00:44:27.599 --> 00:44:29.510
yeah, like getting consistent output. So
00:44:29.520 --> 00:44:30.950
like how I started working on some stuff
00:44:30.960 --> 00:44:33.349
is like I I work on the testing. So I
00:44:33.359 --> 00:44:35.510
use a coding language called Yeah. not a
00:44:35.520 --> 00:44:37.670
vendor spiel, guys, but BAML, they're
00:44:37.680 --> 00:44:40.470
out of Seattle. Uh, and it's like a
00:44:40.480 --> 00:44:42.230
coding language for LLMs. And I've
00:44:42.240 --> 00:44:43.910
really liked using it because it's like
00:44:43.920 --> 00:44:46.309
I'm putting in kind of like my tests
00:44:46.319 --> 00:44:48.309
around it and like expected outputs and
00:44:48.319 --> 00:44:50.230
I can kind of let the LM go off and do
00:44:50.240 --> 00:44:51.670
its like crazy whatever it's going to
00:44:51.680 --> 00:44:54.230
do, but I'm like no, I need this JSON
00:44:54.240 --> 00:44:56.470
blob or this output of a thing to be
00:44:56.480 --> 00:44:58.550
have this consistency and the model in
00:44:58.560 --> 00:45:00.710
the background can do all its crazy
00:45:00.720 --> 00:45:02.630
stuff and I can have my prompts going
00:45:02.640 --> 00:45:04.550
through it and then I can know
00:45:04.560 --> 00:45:06.870
relatively like how this like you know
00:45:06.880 --> 00:45:08.790
non-deterministic thing is doing all
00:45:08.800 --> 00:45:11.670
these outputs. Um, and still like let's
00:45:11.680 --> 00:45:13.670
be honest, most of us want deterministic
00:45:13.680 --> 00:45:15.430
outputs on a lot of the things that we
00:45:15.440 --> 00:45:18.470
did today. You know, I want to know if
00:45:18.480 --> 00:45:20.069
uh I'm doing image recognition on
00:45:20.079 --> 00:45:22.470
something like a cup is a cup and if it
00:45:22.480 --> 00:45:24.550
like it's a donkey, you probably want it
00:45:24.560 --> 00:45:26.550
to be like because there's a donkey on
00:45:26.560 --> 00:45:27.910
the mug, right? So, it's like which
00:45:27.920 --> 00:45:29.430
one's more important. And these large
00:45:29.440 --> 00:45:30.950
large language models like I was using
00:45:30.960 --> 00:45:33.030
some older ones recently uh just to see
00:45:33.040 --> 00:45:36.630
what it was like and it was wild the
00:45:36.640 --> 00:45:38.390
quality changes I've seen. It's just
00:45:38.400 --> 00:45:41.510
like this crazy curve
00:45:41.520 --> 00:45:43.030
um
00:45:43.040 --> 00:45:45.190
in in possibly both.
00:45:45.200 --> 00:45:46.470
>> I think that's a good point though is
00:45:46.480 --> 00:45:48.710
that my view is right now if you're
00:45:48.720 --> 00:45:50.790
entering into this space you should be
00:45:50.800 --> 00:45:52.550
using the LLMs to help you write
00:45:52.560 --> 00:45:55.910
deterministic code not to write it to do
00:45:55.920 --> 00:45:57.829
it without you. So you know one of the
00:45:57.839 --> 00:45:59.270
things we got for an information product
00:45:59.280 --> 00:46:00.870
is is we're building at the moment is
00:46:00.880 --> 00:46:04.150
we'll do a brief um so customer will
00:46:04.160 --> 00:46:06.470
pick a thing let's just say product and
00:46:06.480 --> 00:46:08.950
then it will give them a a bit of text
00:46:08.960 --> 00:46:10.630
right with numbers embedded in it which
00:46:10.640 --> 00:46:12.069
is the brief around that product you
00:46:12.079 --> 00:46:14.390
know this product is your best seller
00:46:14.400 --> 00:46:15.910
you know had a 12 over the last 12
00:46:15.920 --> 00:46:18.230
months it increased by yada yada yada
00:46:18.240 --> 00:46:20.790
your margins so much right and your main
00:46:20.800 --> 00:46:23.349
channel is blah blah blah and so the
00:46:23.359 --> 00:46:24.790
first question the customer said is is
00:46:24.800 --> 00:46:26.710
the LLM generating that and it's like
00:46:26.720 --> 00:46:31.030
No, the almium generated the code right
00:46:31.040 --> 00:46:33.589
that we use but whenever that code's
00:46:33.599 --> 00:46:35.030
running it's now deterministic right
00:46:35.040 --> 00:46:36.950
it's running a select statement to say
00:46:36.960 --> 00:46:39.190
you know top selling product and here's
00:46:39.200 --> 00:46:40.710
the logic and then it's running one
00:46:40.720 --> 00:46:42.550
saying what's its increase over the last
00:46:42.560 --> 00:46:45.430
12 months and so the every time it runs
00:46:45.440 --> 00:46:47.670
it it's running deterministic
00:46:47.680 --> 00:46:49.910
but we used LLM to help us write that
00:46:49.920 --> 00:46:52.309
code well I did anyway um I want to play
00:46:52.319 --> 00:46:53.670
with the next one which is putting
00:46:53.680 --> 00:46:55.270
another section in that information
00:46:55.280 --> 00:46:56.470
product page
00:46:56.480 --> 00:46:59.750
that is did you know in bright yellow
00:46:59.760 --> 00:47:02.150
and then actually let the LLM just kind
00:47:02.160 --> 00:47:03.670
of guess at some stuff that might be
00:47:03.680 --> 00:47:05.430
interesting, right? But putting big
00:47:05.440 --> 00:47:07.109
warnings around it about that made this
00:47:07.119 --> 00:47:09.430
[ __ ] up half the time. Um, so I think if
00:47:09.440 --> 00:47:10.390
you're going to start out, that's what
00:47:10.400 --> 00:47:11.430
you should be doing, right? Is you
00:47:11.440 --> 00:47:13.829
should be looking at using the tools to
00:47:13.839 --> 00:47:15.349
make you a little bit faster. And Nick's
00:47:15.359 --> 00:47:17.349
got a um point in there. He's saying,
00:47:17.359 --> 00:47:19.589
you know, I was at a CTO CPO event last
00:47:19.599 --> 00:47:21.510
night and one of the lines that stayed
00:47:21.520 --> 00:47:23.750
with me from a CTO was 90% of my skills
00:47:23.760 --> 00:47:25.990
are now worth zero. But the remaining
00:47:26.000 --> 00:47:28.950
10% are worth a thousand times more. And
00:47:28.960 --> 00:47:31.910
so this comes back to this argument. Uh
00:47:31.920 --> 00:47:35.510
is LLMs and GI going to kill juniors or
00:47:35.520 --> 00:47:38.630
seniors? Uh I think what we're going to
00:47:38.640 --> 00:47:41.510
end up with is 80 to 90% of everything
00:47:41.520 --> 00:47:43.589
that's done is going to be commodified
00:47:43.599 --> 00:47:48.550
GI slop to a degree and 5% of it is
00:47:48.560 --> 00:47:51.030
going to be massively high quality
00:47:51.040 --> 00:47:53.190
differentiated. And if you're working in
00:47:53.200 --> 00:47:54.790
the space, you need to understand which
00:47:54.800 --> 00:47:56.550
one do you want to play in. Yeah. Are
00:47:56.560 --> 00:47:58.309
you going to be the top 5% of your
00:47:58.319 --> 00:48:00.630
domain uh or are you going to be really
00:48:00.640 --> 00:48:03.349
good at using those tools and you know
00:48:03.359 --> 00:48:06.550
doing throughput or token maxing or
00:48:06.560 --> 00:48:08.230
whatever the metric for your company is
00:48:08.240 --> 00:48:11.190
that uh apart from you know better
00:48:11.200 --> 00:48:12.870
business, making more money, serving
00:48:12.880 --> 00:48:15.430
better customers, uh shareholder growth,
00:48:15.440 --> 00:48:16.630
what what are the other metrics that
00:48:16.640 --> 00:48:18.550
you've got to gamify? I mean what's your
00:48:18.560 --> 00:48:20.870
thought? Yeah, I mean it's not going to
00:48:20.880 --> 00:48:23.829
be too too different um in perspective.
00:48:23.839 --> 00:48:25.270
Yeah, I gonna say there's, you know,
00:48:25.280 --> 00:48:26.470
there's going to be a lot of people that
00:48:26.480 --> 00:48:27.990
build the jets, but there's only be very
00:48:28.000 --> 00:48:29.670
few people that are that to fly them,
00:48:29.680 --> 00:48:31.270
right? So, you got to figure out kind of
00:48:31.280 --> 00:48:33.190
where you're going to be. And like most
00:48:33.200 --> 00:48:36.390
people shouldn't and can't
00:48:36.400 --> 00:48:39.190
fly jets, right? Or are going to be able
00:48:39.200 --> 00:48:41.030
to be in that top 5%. It's just it's
00:48:41.040 --> 00:48:43.670
going to require such a crazy like I
00:48:43.680 --> 00:48:44.950
mean you can see that like the people
00:48:44.960 --> 00:48:46.790
that are are exceptional at using a lot
00:48:46.800 --> 00:48:49.030
of these tools are like good at multiple
00:48:49.040 --> 00:48:51.109
domains like across like there's this
00:48:51.119 --> 00:48:53.030
old TED talk about being a multi-
00:48:53.040 --> 00:48:55.670
potentialite and I think that's you know
00:48:55.680 --> 00:48:58.549
ADHD uh and I was just like it's now
00:48:58.559 --> 00:49:00.390
almost like a superpower like your
00:49:00.400 --> 00:49:03.109
ability to be a master of so many
00:49:03.119 --> 00:49:05.109
different things. Before it used to be
00:49:05.119 --> 00:49:06.549
everybody's kryptonite. Now you're like
00:49:06.559 --> 00:49:09.430
well actually I I understand the
00:49:09.440 --> 00:49:11.030
business. I understand the technical. I
00:49:11.040 --> 00:49:12.390
understand the the architecture. I
00:49:12.400 --> 00:49:14.549
understand, you know, all these
00:49:14.559 --> 00:49:17.349
different like components that make up a
00:49:17.359 --> 00:49:18.950
business. And now the people that have
00:49:18.960 --> 00:49:20.309
those skills, I think they're going to
00:49:20.319 --> 00:49:22.309
be very good. But then like you've got
00:49:22.319 --> 00:49:25.349
to have some depth there. Like you can't
00:49:25.359 --> 00:49:27.589
just be service level on everything.
00:49:27.599 --> 00:49:29.510
>> And you got to be able to multitask
00:49:29.520 --> 00:49:31.510
because I don't know about you, but you
00:49:31.520 --> 00:49:33.349
know, I'm time slicing all the time now.
00:49:33.359 --> 00:49:35.270
I'm running multiple things and trying
00:49:35.280 --> 00:49:36.870
to do some other work because I'm not
00:49:36.880 --> 00:49:38.230
sitting there waiting for the thing to
00:49:38.240 --> 00:49:39.829
finish, right? I'm like, "Yeah, go do
00:49:39.839 --> 00:49:41.190
this. It's the next step. Then I'll go
00:49:41.200 --> 00:49:43.270
do something else. Then I'll forget."
00:49:43.280 --> 00:49:45.349
Like I actually got one of them actually
00:49:45.359 --> 00:49:46.790
sends me a signal message when it's
00:49:46.800 --> 00:49:48.230
waiting for me now
00:49:48.240 --> 00:49:49.910
>> because it's on my mini Mini and I never
00:49:49.920 --> 00:49:52.150
see the screen and it's like it sits
00:49:52.160 --> 00:49:54.069
there for too long. Like I'm like, "Oh,
00:49:54.079 --> 00:49:55.349
now it sends a message saying I'm
00:49:55.359 --> 00:49:56.710
ready."
00:49:56.720 --> 00:49:59.270
>> That just feeds the beast even more. I I
00:49:59.280 --> 00:50:01.270
mean I'm so bad with that stuff. Like I
00:50:01.280 --> 00:50:02.710
actually think focus is so important.
00:50:02.720 --> 00:50:04.630
have like an actual physical hourglass
00:50:04.640 --> 00:50:07.349
on my desk that's like 15 minutes
00:50:07.359 --> 00:50:09.750
because if I can't like and that's why
00:50:09.760 --> 00:50:11.510
probably why I spend so much time on a
00:50:11.520 --> 00:50:13.670
PRD or you know or specs or something
00:50:13.680 --> 00:50:15.109
like that and I think really long and I
00:50:15.119 --> 00:50:18.069
dictate on it and I'll like focus
00:50:18.079 --> 00:50:20.470
which is not my strength and trained
00:50:20.480 --> 00:50:23.349
over years um to focus on the thing and
00:50:23.359 --> 00:50:25.030
so I have clarity of exactly what the
00:50:25.040 --> 00:50:27.910
mission's going to be and then I go off
00:50:27.920 --> 00:50:29.510
and yeah of course I have like three or
00:50:29.520 --> 00:50:31.030
four of my like what do you call them
00:50:31.040 --> 00:50:33.750
cattle going up and doing my things and
00:50:33.760 --> 00:50:35.829
getting slaughtered in the field. Um
00:50:35.839 --> 00:50:47.990
because like Oh, please
00:50:48.000 --> 00:50:49.829
gone. What's going on?
00:50:49.839 --> 00:50:51.109
>> Out of there.
00:50:51.119 --> 00:50:52.549
>> I don't know. It's just like I touched
00:50:52.559 --> 00:50:54.630
my desk. Maybe my power is going off.
00:50:54.640 --> 00:50:56.710
Mind you, it's super super hot here. So,
00:50:56.720 --> 00:50:58.390
my electrical in my house built in
00:50:58.400 --> 00:51:01.990
1800s. It's not doing great. Um
00:51:02.000 --> 00:51:04.950
the um but anyway yeah going off my
00:51:04.960 --> 00:51:06.950
agents to go off and do the work. It's
00:51:06.960 --> 00:51:08.390
just been really interesting as like as
00:51:08.400 --> 00:51:10.069
we work through these processes over the
00:51:10.079 --> 00:51:12.630
past couple years just like how dtorial
00:51:12.640 --> 00:51:15.190
we get. Also like the one thing I do do
00:51:15.200 --> 00:51:17.589
um is I actually read the code like not
00:51:17.599 --> 00:51:19.349
all the time. It depends what to like
00:51:19.359 --> 00:51:21.109
what level I get. And then I'll like,
00:51:21.119 --> 00:51:22.390
you know, sometimes I'll scan it and
00:51:22.400 --> 00:51:23.750
sometimes I'll like get to a place and
00:51:23.760 --> 00:51:25.589
then I'll have it like explain certain
00:51:25.599 --> 00:51:27.510
things so I understand what is actually
00:51:27.520 --> 00:51:29.270
performing and what is actually going
00:51:29.280 --> 00:51:33.829
on. Um can't say inspect y very much
00:51:33.839 --> 00:51:36.150
like actual logic. I I'll actually want
00:51:36.160 --> 00:51:37.910
to understand it because I consider my
00:51:37.920 --> 00:51:40.069
responsibility as an engineer. Um what I
00:51:40.079 --> 00:51:41.829
deploy into the field I'm absolutely
00:51:41.839 --> 00:51:43.670
have to own it. And I guess like the
00:51:43.680 --> 00:51:44.950
agents are going to get obviously
00:51:44.960 --> 00:51:46.710
getting are getting better, but I would
00:51:46.720 --> 00:51:51.270
say I have a a healthy healthy um
00:51:51.280 --> 00:51:52.870
suspicion
00:51:52.880 --> 00:51:57.030
about all the different QA agent coding
00:51:57.040 --> 00:51:59.030
skills.
00:51:59.040 --> 00:52:00.950
>> Yeah. Yeah. I mean, I read something the
00:52:00.960 --> 00:52:02.470
other day that you should never ask the
00:52:02.480 --> 00:52:04.390
agent, the LM your work session you're
00:52:04.400 --> 00:52:06.309
working in to peerreview the work it's
00:52:06.319 --> 00:52:07.910
doing. you should always ask another
00:52:07.920 --> 00:52:10.069
session and then you should always ask
00:52:10.079 --> 00:52:12.069
that session to peer review it with a
00:52:12.079 --> 00:52:14.150
different model. So if you're Yeah.
00:52:14.160 --> 00:52:15.510
Yeah. So if you're kind of going along
00:52:15.520 --> 00:52:18.710
and doing uh you know sonnet then you
00:52:18.720 --> 00:52:20.950
should do or whatever it's called. Yeah.
00:52:20.960 --> 00:52:22.870
You should change the model. Um if not
00:52:22.880 --> 00:52:24.549
the vendor because that's the only way
00:52:24.559 --> 00:52:26.710
you're going to get a true review. Um
00:52:26.720 --> 00:52:29.270
>> yeah. Yeah. I agree. I use a tool called
00:52:29.280 --> 00:52:32.230
Robo Rev. Um it's the guy that created
00:52:32.240 --> 00:52:33.829
um pandas and and
00:52:33.839 --> 00:52:35.030
>> that's Wes, right?
00:52:35.040 --> 00:52:36.549
>> Yeah. Wes. Yep.
00:52:36.559 --> 00:52:38.309
>> Very. And so I've been I'm pretty
00:52:38.319 --> 00:52:40.230
durable on that one. And I Yeah, I use
00:52:40.240 --> 00:52:42.069
obviously I use different usually I do
00:52:42.079 --> 00:52:43.109
use different language models to
00:52:43.119 --> 00:52:45.349
inspect. I find uh open as one is very
00:52:45.359 --> 00:52:47.829
pedantic. Uh sometimes Gemini if it
00:52:47.839 --> 00:52:51.109
doesn't break uh
00:52:51.119 --> 00:52:53.270
it also and then it allows me to have
00:52:53.280 --> 00:52:55.430
like form up different like different
00:52:55.440 --> 00:52:57.030
pieces. I also like like breaking up my
00:52:57.040 --> 00:52:58.790
work like really really small chunks.
00:52:58.800 --> 00:53:00.470
Um, I figured like same with you, it's
00:53:00.480 --> 00:53:01.990
like as soon as your contacts or like
00:53:02.000 --> 00:53:04.549
your workstream like any PR for any
00:53:04.559 --> 00:53:07.430
person gets too big, funny enough, it
00:53:07.440 --> 00:53:08.870
gets really squirly, but if you like
00:53:08.880 --> 00:53:11.510
chunk it up like really small and tight,
00:53:11.520 --> 00:53:13.750
um, I have found better um,
00:53:13.760 --> 00:53:14.710
capabilities. And by the way, I'm
00:53:14.720 --> 00:53:17.030
nowhere close to being in the top 5% of
00:53:17.040 --> 00:53:19.190
people yet. I've still
00:53:19.200 --> 00:53:21.030
>> I just sometimes I find the tools change
00:53:21.040 --> 00:53:23.270
so much so often that every time I think
00:53:23.280 --> 00:53:26.069
I've like nailed it, another one comes
00:53:26.079 --> 00:53:27.510
out. And you know me, I'm like always
00:53:27.520 --> 00:53:29.589
like, you know, playing with the toys
00:53:29.599 --> 00:53:31.030
>> and and everything changes. Um, you
00:53:31.040 --> 00:53:33.589
know, that PR review one. So when we So
00:53:33.599 --> 00:53:35.990
when I get Claude to submit the PR, then
00:53:36.000 --> 00:53:38.069
Gemini automatically reviews it on the
00:53:38.079 --> 00:53:40.710
on the GitHuby thing. Um, and they were
00:53:40.720 --> 00:53:42.549
having a fight today because I had a
00:53:42.559 --> 00:53:47.109
list of values and uh there's like 451
00:53:47.119 --> 00:53:48.710
unique variables in that list of values
00:53:48.720 --> 00:53:50.150
is a drop down on the information
00:53:50.160 --> 00:53:52.630
product. Uh, and so I'm like, yeah,
00:53:52.640 --> 00:53:54.870
that's tested it. Performance was okay.
00:53:54.880 --> 00:53:56.630
Cool. you know, like that's good. Just
00:53:56.640 --> 00:53:58.309
just generate the whole list every time
00:53:58.319 --> 00:54:00.950
they go in there. And then pushed it.
00:54:00.960 --> 00:54:02.470
And then Gemini is like, "Oh, no, that's
00:54:02.480 --> 00:54:03.589
going to have a performance problem.
00:54:03.599 --> 00:54:05.030
It's going to crash their browser. You
00:54:05.040 --> 00:54:07.910
need to limit to 100." And so Claude
00:54:07.920 --> 00:54:09.670
goes, "Oh yeah, I'll limit to 100." And
00:54:09.680 --> 00:54:11.349
I'm like, "No, you won't because I want
00:54:11.359 --> 00:54:14.150
all 451." It's like, "Okay, they pushed
00:54:14.160 --> 00:54:15.750
it back." And then Gemini is like, "No,
00:54:15.760 --> 00:54:17.190
it has to be 100. I'm not letting it
00:54:17.200 --> 00:54:18.950
go." And they were just bloody arguing
00:54:18.960 --> 00:54:21.829
until I overrode them. So just like a
00:54:21.839 --> 00:54:23.510
team, right? It's like two people
00:54:23.520 --> 00:54:25.510
arguing. I think the other thing that
00:54:25.520 --> 00:54:27.750
you mentioned about is like the number
00:54:27.760 --> 00:54:29.430
of times I'm going you'll mention
00:54:29.440 --> 00:54:30.870
something and I'm like, "Oh, it's
00:54:30.880 --> 00:54:32.150
another harness." And then I'll go look
00:54:32.160 --> 00:54:34.150
at it. Oh, I like that bit. And and
00:54:34.160 --> 00:54:36.549
you're just constantly trying to find
00:54:36.559 --> 00:54:38.150
something that just does what you want
00:54:38.160 --> 00:54:40.069
to do perfectly. And you never do,
00:54:40.079 --> 00:54:42.549
right? You find patterns that are useful
00:54:42.559 --> 00:54:44.230
and you should adopt them if they have
00:54:44.240 --> 00:54:46.950
value. Um, but I'd say 80% of the
00:54:46.960 --> 00:54:49.430
harness that I've built for myself, I
00:54:49.440 --> 00:54:51.030
don't actually use or get any value out.
00:54:51.040 --> 00:54:52.950
Like I've built a whole personal OS.
00:54:52.960 --> 00:54:54.710
Nick Nick Zed started me off on this
00:54:54.720 --> 00:54:58.069
one. Thanks, Nick. Um, and it tries to
00:54:58.079 --> 00:55:00.470
task manage me, right? I've spent ages
00:55:00.480 --> 00:55:02.470
trying to help get it to help me manage
00:55:02.480 --> 00:55:03.829
which tasks are the most important to
00:55:03.839 --> 00:55:05.190
work on next.
00:55:05.200 --> 00:55:08.150
>> Of course, I I ignore it because I don't
00:55:08.160 --> 00:55:10.150
work that way. So,
00:55:10.160 --> 00:55:11.829
>> I know, right? You're trying to like get
00:55:11.839 --> 00:55:13.750
things done, get like that whole list
00:55:13.760 --> 00:55:15.829
and you're just like that's just so
00:55:15.839 --> 00:55:18.309
counter to like you as a like a
00:55:18.319 --> 00:55:20.230
personality. You're like, I just like it
00:55:20.240 --> 00:55:21.829
doesn't matter. You're like, "Was like
00:55:21.839 --> 00:55:25.589
task list me to-do list manifesto like
00:55:25.599 --> 00:55:28.390
or like I mean mine is like calendar. I
00:55:28.400 --> 00:55:29.910
put stuff on my calendar to do my that's
00:55:29.920 --> 00:55:31.349
my to-do list is like if it's not on the
00:55:31.359 --> 00:55:33.109
calendar then it's going to be put into
00:55:33.119 --> 00:55:35.430
like the whatever list."
00:55:35.440 --> 00:55:38.150
>> Yep. But but what I do do is in my
00:55:38.160 --> 00:55:41.270
harness I force the LLMs to keep uh
00:55:41.280 --> 00:55:44.309
to-do a doing and a done. Um because
00:55:44.319 --> 00:55:46.549
when I run out of tokens, right, when
00:55:46.559 --> 00:55:47.910
they run out of tokens in the middle of
00:55:47.920 --> 00:55:50.789
a task, uh and 3 hours later I can kick
00:55:50.799 --> 00:55:52.309
them off again, I want them to know
00:55:52.319 --> 00:55:53.750
where they caught up, where they they're
00:55:53.760 --> 00:55:55.430
sitting, right? And so they just go back
00:55:55.440 --> 00:55:57.829
and read the file. Um before I had that,
00:55:57.839 --> 00:55:59.990
it was constantly going back and reading
00:56:00.000 --> 00:56:01.910
all the context memory and trying to
00:56:01.920 --> 00:56:03.430
figure out what it was and then
00:56:03.440 --> 00:56:05.430
forgetting it done half the job and
00:56:05.440 --> 00:56:08.150
rewriting it and yeah. So again, the
00:56:08.160 --> 00:56:09.670
interesting thing for me is I'm bringing
00:56:09.680 --> 00:56:11.910
those human behaviors in and I'm not
00:56:11.920 --> 00:56:13.349
sure that's right. You know, I'm
00:56:13.359 --> 00:56:15.349
treating them like employees and agents.
00:56:15.359 --> 00:56:17.349
I'm going this is how I manage to help a
00:56:17.359 --> 00:56:19.510
team manage themselves.
00:56:19.520 --> 00:56:21.190
I keep going back to them with clear
00:56:21.200 --> 00:56:23.589
sessions to the LLMs going if you were
00:56:23.599 --> 00:56:24.870
going to run yourself, how would you do
00:56:24.880 --> 00:56:26.789
it?
00:56:26.799 --> 00:56:28.630
I never get a good answer. So
00:56:28.640 --> 00:56:31.349
>> but but like but like you know like but
00:56:31.359 --> 00:56:32.630
that's how the marketing machine works
00:56:32.640 --> 00:56:34.390
and that's why it's been so effective,
00:56:34.400 --> 00:56:36.230
right? Like I would argue the reason
00:56:36.240 --> 00:56:38.950
these systems are so effective and have
00:56:38.960 --> 00:56:43.510
gotten so widespread is because they're
00:56:43.520 --> 00:56:45.430
like the touring machine. Like the very
00:56:45.440 --> 00:56:46.549
thing that we always like it won't
00:56:46.559 --> 00:56:47.990
happen, it won't happen. What happened?
00:56:48.000 --> 00:56:49.589
You know, it's happened and you're just
00:56:49.599 --> 00:56:52.309
like, wow. I mean, I think it's incept
00:56:52.319 --> 00:56:54.069
my brain faster than any social media
00:56:54.079 --> 00:56:56.789
ever did. And it's just like, oh,
00:56:56.799 --> 00:56:58.309
whoops. I didn't realize that the
00:56:58.319 --> 00:57:01.270
shortcut is language. And obviously lots
00:57:01.280 --> 00:57:03.589
of researchers knew that but like the
00:57:03.599 --> 00:57:05.990
way that they like incept is so intense
00:57:06.000 --> 00:57:11.270
and uh I think
00:57:11.280 --> 00:57:13.430
in this space and this is where I get
00:57:13.440 --> 00:57:16.150
like worried a little bit is like all
00:57:16.160 --> 00:57:17.670
you're like using claw and you downgrade
00:57:17.680 --> 00:57:19.589
and it goes not that claw anthropic
00:57:19.599 --> 00:57:21.510
claims will never do ads. We'll see.
00:57:21.520 --> 00:57:23.430
Right. And also you're doing that and
00:57:23.440 --> 00:57:25.270
it's like have you thought about Dave
00:57:25.280 --> 00:57:26.789
Brooks
00:57:26.799 --> 00:57:29.349
and you're just like wait a minute. Um
00:57:29.359 --> 00:57:31.589
you know or or maybe you know even worse
00:57:31.599 --> 00:57:33.510
though in these in these systems it
00:57:33.520 --> 00:57:35.190
becomes more subtle. Yeah.
00:57:35.200 --> 00:57:36.150
>> Right.
00:57:36.160 --> 00:57:38.710
>> Unfortunately we can't solve this
00:57:38.720 --> 00:57:41.750
problem because you've chosen an
00:57:41.760 --> 00:57:43.670
inferior system like in fancier
00:57:43.680 --> 00:57:46.390
language. You should think about context
00:57:46.400 --> 00:57:48.150
plane brought to you by whatever. I
00:57:48.160 --> 00:57:48.549
don't really.
00:57:48.559 --> 00:57:49.349
>> Yep.
00:57:49.359 --> 00:57:52.390
>> And you're like, "Damn it."
00:57:52.400 --> 00:57:53.829
>> Oh, and it's got the magic button. Would
00:57:53.839 --> 00:57:55.829
you like me to migrate you all magically
00:57:55.839 --> 00:57:57.589
for a thousand seconds?
00:57:57.599 --> 00:57:58.150
>> Don't worry.
00:57:58.160 --> 00:57:59.190
>> Don't worry about it. We already have
00:57:59.200 --> 00:58:01.910
your credit card on file
00:58:01.920 --> 00:58:04.069
>> and we'll move you over. Um,
00:58:04.079 --> 00:58:04.710
>> yeah.
00:58:04.720 --> 00:58:06.870
>> Yeah. It's, you know, Yeah. Yeah, we're
00:58:06.880 --> 00:58:09.670
almost at time. But you know, that's the
00:58:09.680 --> 00:58:11.750
funniest thing is uh you remember when
00:58:11.760 --> 00:58:13.510
we did like that era and it still exists
00:58:13.520 --> 00:58:15.829
is like those massive migrations like
00:58:15.839 --> 00:58:18.309
moving from say like an ERP system or
00:58:18.319 --> 00:58:21.430
accounting systems like oh my god that
00:58:21.440 --> 00:58:25.349
whole space with these tools now
00:58:25.359 --> 00:58:26.870
it's going to go
00:58:26.880 --> 00:58:28.710
>> it's easier right yeah it's easier to
00:58:28.720 --> 00:58:31.109
map the two systems it's easier to get
00:58:31.119 --> 00:58:32.549
them to help us write the code to do the
00:58:32.559 --> 00:58:35.750
migration and it's much easier to write
00:58:35.760 --> 00:58:38.549
LLM tooling that actually tests it for
00:58:38.559 --> 00:58:40.710
us, right? And picks up that 80% that we
00:58:40.720 --> 00:58:42.309
always miss. All right, we're up on the
00:58:42.319 --> 00:58:45.190
hour. Uh what's the one thing if you're
00:58:45.200 --> 00:58:47.190
if you somebody a data practitioner said
00:58:47.200 --> 00:58:49.430
to you, "What should I do this month to
00:58:49.440 --> 00:58:51.510
kind of get my toe in the water for this
00:58:51.520 --> 00:58:53.430
stuff?" What would you say? What would
00:58:53.440 --> 00:58:56.549
you recommend they they do?
00:58:56.559 --> 00:58:58.470
>> Um
00:58:58.480 --> 00:59:01.750
build your own build your own brain. uh
00:59:01.760 --> 00:59:03.670
like your own memory in your LLMs right
00:59:03.680 --> 00:59:06.390
now like your own proprietary thing for
00:59:06.400 --> 00:59:08.230
your household for whatever it is really
00:59:08.240 --> 00:59:10.789
just get good at that because like then
00:59:10.799 --> 00:59:12.230
it doesn't really and and don't attach
00:59:12.240 --> 00:59:13.670
it to a I know a lot of people who just
00:59:13.680 --> 00:59:14.950
said like who cares if you're on a
00:59:14.960 --> 00:59:16.950
platform but like like be honest like
00:59:16.960 --> 00:59:19.030
keep your information yours to a certain
00:59:19.040 --> 00:59:20.549
extent I think that's a very important
00:59:20.559 --> 00:59:22.950
piece kind of like that CTO CPO piece
00:59:22.960 --> 00:59:24.950
like if you're dictating doing all these
00:59:24.960 --> 00:59:26.710
things like create up your own like it's
00:59:26.720 --> 00:59:30.150
really easy to like do uh uh embedding
00:59:30.160 --> 00:59:32.470
spectra embeddings I mean this data data
00:59:32.480 --> 00:59:33.910
podcast out you probably know how to
00:59:33.920 --> 00:59:35.750
factor embeddings but like take all that
00:59:35.760 --> 00:59:37.190
stuff that context all that information
00:59:37.200 --> 00:59:39.349
that you're taking down every day and
00:59:39.359 --> 00:59:40.630
just start building a memory bank for
00:59:40.640 --> 00:59:41.910
yourself all those articles all those
00:59:41.920 --> 00:59:43.750
things because like ultimately becomes
00:59:43.760 --> 00:59:48.390
your self referential library and it can
00:59:48.400 --> 00:59:50.069
also be in your own voice and then allow
00:59:50.079 --> 00:59:53.589
you to I don't know go against it um I
00:59:53.599 --> 00:59:55.829
think that's that would be my thing and
00:59:55.839 --> 00:59:58.069
I know a lot of people are trying to
00:59:58.079 --> 00:59:59.910
work into that space but I have found
00:59:59.920 --> 01:00:02.630
that personally that the most rewarding.
01:00:02.640 --> 01:00:04.230
>> Yeah. And something like Obsidian makes
01:00:04.240 --> 01:00:06.069
it a little bit easier to store it and
01:00:06.079 --> 01:00:09.270
see it in a UI as well as as code. Uh
01:00:09.280 --> 01:00:11.589
yeah. So my one is go go go read an
01:00:11.599 --> 01:00:14.069
article that talks about some of the
01:00:14.079 --> 01:00:17.349
[ __ ] And if that article sounds
01:00:17.359 --> 01:00:19.270
interesting in Jill's view, go and try
01:00:19.280 --> 01:00:22.069
and build something that is related to
01:00:22.079 --> 01:00:23.910
it. Like even if it's just one small
01:00:23.920 --> 01:00:25.270
part of what that article is talking
01:00:25.280 --> 01:00:27.910
about, right? Just go do it because as
01:00:27.920 --> 01:00:29.589
you play with these things, you'll
01:00:29.599 --> 01:00:31.990
learn. so much more about how they work,
01:00:32.000 --> 01:00:34.549
how they don't work. Um, so just yeah,
01:00:34.559 --> 01:00:36.470
it's time to play. Like, everybody's
01:00:36.480 --> 01:00:38.870
learning. There's no actual answers yet.
01:00:38.880 --> 01:00:41.750
So, uh, yeah, go play with it and learn.
01:00:41.760 --> 01:00:44.069
Like, yeah, enjoy the experience of
01:00:44.079 --> 01:00:46.309
having a complete change of technology
01:00:46.319 --> 01:00:49.910
and way of working, uh, come on us.
01:00:49.920 --> 01:00:51.670
>> I know, right? Man, what a time to be
01:00:51.680 --> 01:00:54.470
alive to have this going on. I super It
01:00:54.480 --> 01:00:55.990
scares crap out of me at the same time.
01:00:56.000 --> 01:00:57.670
It's like, as an engineer, it's like the
01:00:57.680 --> 01:01:00.150
ultimate toy box.
01:01:00.160 --> 01:01:05.109
Yeah. Oh, you got again.
01:01:05.119 --> 01:01:06.150
All right. Talk to you later.
01:01:06.160 --> 01:01:07.510
>> Excellent. All right. We'll talk to you
01:01:07.520 --> 01:01:11.480
next month. Catch you.