0:00

Anaconda is a popular platform for data

0:02

science, machine learning, and AI. It

0:06

provides trusted repositories of Python and R

0:08

packages and has over 35 million users

0:11

worldwide. Rob Fertrik is

0:13

the CTO at Anaconda, and he joins

0:15

the show to talk about the platform,

0:17

the concept of an OS for AI,

0:19

and more. This episode

0:21

is hosted by Lee Acheson. Lee

0:24

Acheson is a software architect,

0:26

author, and thought leader on

0:28

cloud computing and application modernization.

0:31

His bestselling book, Architecting for Scale,

0:33

is an essential resource for technical

0:35

teams looking to maintain high availability

0:37

and manage risk in their cloud

0:39

environments. Lee is the

0:41

host of his podcast, Modern Digital

0:44

Business, produced for people looking to

0:46

build and grow their digital business,

0:48

listen at mdb.fm. Follow

0:51

Lee at softwarearchitectureinsights.com and

0:53

see all his content

0:55

at leeacheson.com. Rob,

1:10

welcome to Software Engineering Daily. Thank

1:12

you, Lee. It is very exciting to be here. I

1:15

actually have listened to this podcast in the past, and

1:17

so it's a little bit of extra excitement in addition

1:19

to looking forward to a great conversation. So

1:22

why don't you start out by just

1:24

telling me, you know, what is Anaconda?

1:26

I'm sure most people on our podcast

1:28

have at least heard of it, but

1:31

I'm sure there's several, there's probably many

1:33

people who haven't yet. So let's start

1:35

out with setting the ground straight and

1:37

telling everyone exactly what Anaconda is. Oh

1:40

man, there's quite a few different answers to

1:42

that question. I guess at its core, you

1:44

know, Anaconda is a company that is really

1:46

focused on helping people innovate and doing that

1:48

by giving them a way of connecting to

1:51

the broader open source ecosystem specifically around Python,

1:53

but not just Python. And

1:55

Anaconda was originally named Continuum Analytics when

1:58

it was founded. They did a

2:00

lot of work in the data science space,

2:02

recognized the need to really empower Python programmers,

2:05

specifically in data science and other areas. And

2:07

in order to do that, basically they were

2:10

solving their own problems. These were Python developers

2:12

that were actually trying to help enterprises run

2:14

their numerical computing workloads, run other kinds of

2:16

workloads. And they realized, especially in like Windows

2:18

environments and other environments, that there was a

2:20

need to standardize how people got

2:23

access to all the broader open source ecosystems,

2:25

the Python packages that millions of people around

2:27

the world were producing. And so they created

2:29

this distribution of Python packages and Python

2:31

libraries, not all of them written in Python, by the way,

2:33

that's kind of the reason why a lot of this stuff's

2:35

written in Fortran, C++, and other languages, and

2:38

getting the right bits onto people's computers, helping

2:40

them do that simply, manageably, et cetera, they

2:42

came up with the Anaconda distribution, which is

2:45

where actually the name came from. You want

2:47

a big distribution of Python, oh, big Python, Anaconda.

2:50

And then suddenly everybody knows continuing

2:52

analytics through the Anaconda distribution. So

2:54

that's how the name got changed

2:56

to Anaconda. And so at

2:58

its heart, we produce trusted repositories of

3:01

Python and our packages, actually. We

3:03

provide products and services around security

3:05

and governance of those packages.

3:08

And again, that connection to the broader

3:10

open source ecosystem. And we actually have

3:12

kind of enterprise products, as you'd imagine,

3:14

around AI, data science, et cetera. That's

3:16

our workbench product and others. And we

3:18

also do a tremendous amount around the

3:20

open source ecosystem. So we either actively

3:22

participate in, or find or develop many

3:24

different open source projects, everything from Numba

3:26

to Bware to Panel to on and

3:29

on and on. And we donate quite

3:31

a bit of money to the various

3:33

open source projects as well. But

3:35

again, the real goal of the company

3:37

is to connect scientists, data

3:39

scientists, engineers, programmers, knowledge workers, actually.

3:41

We can talk about that later,

3:43

but Excel from Microsoft last year

3:45

announced that you are now gonna

3:47

be actually use the Python programming

3:50

language inside of Excel. That's

3:52

really exciting to hear. Yeah. I

3:54

remember hearing that. When

3:56

you have the broader knowledge workers, the kind of advanced

3:58

knowledge workers that want to have... Again, not

4:00

just to the Python programming language, but

4:03

really to that broader ecosystem and all

4:05

the innovations those packages and community provides.

4:08

And Anaconda wants to connect people to

4:10

that. Yeah, I know that one of

4:12

the problems that lots of languages have,

4:14

I'm more familiar with the problem of

4:16

how packages work in Ruby and Rails,

4:18

especially when there are third party packages

4:20

that are not all in Ruby. And

4:22

so, and how it has to build

4:24

in the background and the

4:26

installation of some of those packages can be quite

4:28

problemsome. Python is a lot easier in general,

4:31

but a lot of that is because of

4:33

the Anaconda Packaging System. Is

4:35

that correct? Is that the best way to think of it? I think

4:37

so, yeah. And actually there are open source

4:40

efforts that Anaconda created Conda specifically,

4:42

that that's where the majority of the user base

4:44

uses actually. And actually, I think it's funny, our

4:46

last measurement, it was actually 45 million

4:48

users, which I think a lot of that comes from

4:50

the explosion in AI and Python kind of

4:52

being the lingua franca of AI and

4:55

using Conda to set up your Python environment and

4:57

not just get the right bits on your computer

4:59

in terms of, you know, is this

5:01

NumPy, is this SciPy, is this SciKit Learn, is

5:03

this Keras or PyTorch, but getting all the dependencies,

5:06

all the things, those things rely on and doing

5:08

that solving in order to make sure those environments

5:10

run. Yeah, that's exactly the problem you're talking

5:12

about. So quickly deploy small

5:15

scale apps to share. I think I got that

5:17

phrasing from your website and that sounds a little

5:19

bit like what you're describing, but it's probably, it

5:21

is a lot more than that. But I know

5:24

one of the major use cases for your platform,

5:26

and you just mentioned it briefly here now,

5:29

is the growing field of artificial

5:31

intelligence and AI tooling and

5:34

data science in particular. You've

5:37

kind of taken that a step further and

5:39

you're actually pushing a concept

5:41

that you call OS for AI. Can

5:43

you talk a little bit about that? You

5:46

meaning out of Conda, I'm sorry. Yeah.

5:49

Okay. I

5:51

guess it's more of a conceptual or philosophical

5:53

point than like a literal operating system. And

5:55

that does confuse some people, but the idea

5:57

here is You have

5:59

this. The operating system you know and

6:01

of takes all the hardware abyss. You.

6:04

Know especially if you go back to the

6:06

seventies and you'll have like a similar similar

6:08

of the Peter Steele up until this day.

6:10

but you have you back and and you

6:12

needed the operating system in order to take

6:14

all of that complexity and actually makes as

6:16

platform that Dental continues to be Job applications

6:18

for people actually use the hardware and kind

6:20

of and able to change and conceptually speaking

6:22

that's what we feel we live. By.

6:24

Providing that feel, providing that that middle

6:26

layer their that connects people to that

6:29

broader open source ecosystem makes it simple

6:31

and easy to develop your applications. Everything

6:35

from science to modern A I

6:38

had models and applications. For.

6:40

Births are specific to I know

6:42

by desserts are there specific things

6:45

you do. Towards. The a

6:47

I use case that make his last. Yeah, absolutely. So

6:49

it's all an extension of what we get. Sub you

6:51

look at it and accountants. I hate. I get my

6:53

Python package is from. You get my packages and you

6:55

provide ways of doing this. You know if the individual

6:58

user that's what they care about, I want to get

7:00

my environment that up only get access to what I

7:02

need so I can train a model I can to

7:04

play model I can. I can leverage that model inside

7:06

of my applications. Then you have the organization's employees, people

7:08

that wanna make sure that they have some. Control.

7:11

Over the Wild West. That. That other uses

7:13

our our interview with a meters. There's hundreds of

7:15

thousands if not more different packages the could be

7:17

that can be used and it's not of are

7:19

secure world. And so seriously that

7:21

was. That was the data sites space.

7:24

and now. If. You to sneak okay

7:26

well as anacondas really going to getting bits on

7:28

your computer? The. Right? That's for you,

7:30

simply sincerely, audibly, etc. But we change

7:32

the definition. Those bits snow under a

7:34

Python and are packaged now. Maybe it's

7:36

datasets, That. You want to use in

7:38

order to train your models are finding your mouse

7:40

or the models themselves that you need to get

7:42

installed and get up and running or easily again

7:45

for your specific environment. Are you running some kind

7:47

of the Linux server? Are you running a Windows

7:49

desktop? Know what kind of steep you you have?

7:51

You even have a Gp You is that of

7:53

sound. Mobile. Device as computing and so

7:56

if you can abroad would Anaconda does say it's

7:58

not just Python and our packages, it's game. That

8:00

those bits happen to be models, they have

8:02

to be datasets, etc. Then you can see

8:04

why. the to kind of ai explosion is

8:06

a natural extension of what encoded us. So.

8:09

You also the data delivery for

8:12

use by a on. I

8:14

would say some of the stuff is

8:16

today. Some the stuff will be ocean

8:18

kernels. Yeah so it's. A lot

8:21

of the. Rapidly. Changing improvements

8:23

means to right now which are be honest I think

8:25

I'm ready to release when but say which is I

8:27

can't even keep on top of it. At All

8:29

is that says. We. Have the

8:31

millions of users we have people understand how we are,

8:33

you help them get their and barnes up and and

8:35

twelve applications each other more not and so that's a

8:37

very natural extension for them to come to us as

8:39

they will help us of these problems to and for

8:41

see them. So yes some of the susceptible to today

8:43

some this stuff will do in the near future and

8:46

some stuff is is gonna be coming. To. The

8:48

some those datasets millward Giant are.

8:50

so are you still packaging those

8:52

for delivery or are you also

8:54

looking at tooling to get access

8:56

to data system, other locations and

8:59

maniacs to the whole world of

9:01

possibilities air and whole world. We

9:03

address Valve. And it's not

9:05

just access to the data itself, it's everything

9:07

around the actual province video to see. Have

9:09

that, the ongoing copyright discussions, and kind of,

9:11

do I have access to the right data

9:14

drive? Do I even know. That. I'm

9:16

using that kind of data to train that I have

9:18

a license to train on. Or. That I

9:20

have the right to train on and so helping

9:22

people saw those problems. It is a D or

9:24

it's It's not just us, it's a very common

9:26

refrain that people are holding off on Sundays Technologies

9:28

to. They don't understand the exposure, They don't understand

9:30

what they can do and how they can do

9:33

it. And then you'll even hear things about like

9:35

what data should even be twenty on with it

9:37

is to be using. And so actually I went

9:39

Attended a very interesting presentation on the size model

9:41

for Microsoft at Gdc. and they were

9:43

making a point that saito sometimes trend is models as

9:45

a bit a garbage in garbage out know you want

9:47

to use high quality data so it's wanting to train

9:49

your data and all your models are on all the

9:51

date on the internet it's another thing to say work

9:53

and actually took a subset of data and and get

9:55

better output because we've we don't we don't need to

9:57

have every possible piece of san section that was ever

10:00

written, incorporated into our model when we really want

10:02

it to be an expert in, you know, I

10:04

don't know, science or physics or something else. It's

10:06

not just about the size of the data sets.

10:08

It's about provenance. It's about security. It's about auditability,

10:10

reproducibility, and then collaboration and sharing. Cool.

10:13

Cool. So two different topics and just taking

10:15

from words that you said to words that

10:17

you said, and I like to talk about

10:19

them independently, we could probably have a whole

10:21

conversation on both of them, but one

10:24

of them is secure and the other one

10:26

was copyright. So let's

10:28

take whichever one you want to cover first,

10:30

but I'd like to talk about both of

10:32

those areas of what you do. I

10:34

mean, flip a coin. They're both tremendously interesting. Then

10:37

let's start with security. All right.

10:39

So you, you mentioned securely deliver

10:41

to the desktop, the package,

10:44

whatever, whether the package is data code,

10:46

whatever. So that implies

10:48

a lot of things and perhaps a

10:50

lot more than what you actually do.

10:53

I'm not exactly sure it can

10:55

be a wide spectrum of different things that

10:57

securely delivered to your desktop can mean. Why

10:59

don't you tell me what you mean by

11:01

that? Yeah, again, it could absolutely be

11:03

a lot of things, but I guess at its heart,

11:05

it's, you know, they talk about a lot about the

11:07

software bill of materials. You want to think of it

11:09

as like chain of custody. Like you watch, you know,

11:12

law and order and either kind of police procedurals and

11:14

you know, you want to make sure that, you know,

11:16

that evidence, you understand what happened to it every step

11:18

of the way so that there's nothing that was corrupted,

11:20

there was nothing, no kind of broken trust or misuse.

11:22

You can think of packages and whatnot, the same thing.

11:24

So how do I know that this open source code

11:26

that I'm pulling into my own, you

11:28

know, my own computer and I'm trying to use,

11:31

or I'm trying to share with others doesn't have

11:33

anything nefarious in it. It wasn't the wrong code.

11:35

It wasn't something that somebody, you know, some malicious

11:37

actor injected something bad into it. You know, there've

11:39

been plenty of very high profile security and incidents

11:41

over the years where maybe a company like, well,

11:44

I don't name any names, but you can look

11:46

them up, but lots of big name companies had

11:48

a supplier that provided them software, provided them

11:50

something. And the attack actually came through the

11:52

supplier to that major company. And so

11:54

didn't talk much about my background, but I spent

11:56

17 years in high performance computing space. So I've

11:59

started several companies. of the years and

12:01

the last one that I founded was acquired by Microsoft in

12:03

2017 and so I joined Azure

12:06

and I led their HPC and AI software infrastructure

12:08

team from the product side actually, it's supposed to

12:11

be the dev side and then that's where I

12:13

left and then joined Anaconda. And so, you know,

12:15

while at Microsoft it was a very, very, very

12:17

major concern that anything we're pulling in from anywhere

12:19

else that we understood exactly what we were doing

12:21

because Microsoft was such a huge target for various

12:24

obvious reasons and our clients were such huge targets

12:26

and the last thing we wanted is for people

12:28

to be able to kind of backdoor into

12:30

our clients through us, through our suppliers and so on.

12:33

All those same concepts apply to getting bits

12:35

onto your computer and so when it's Python

12:37

packages, whether it's a model that comes from

12:39

somewhere, whether it's a data set that's used

12:41

them, you want that same chain of custody

12:43

knowledge, you want that same problem, you want

12:45

that same control. That could be for security

12:47

reasons, like I said, in terms of malicious

12:49

actors, but it can also even just be

12:51

for reproducibility later. If you're going to use

12:53

this stuff in pharmaceutical industries or financial industries

12:55

or other industries that have regulatory needs or

12:57

other kinds of, you know, especially life or

12:59

death situations, you know, if your model

13:02

makes certain decisions, if your system causes certain things

13:04

and something goes wrong, they're going to understand why

13:06

you better be able to reproduce your results. So

13:08

even just being able to track your data, track

13:11

your models, track your packages and control them over

13:13

time is very important. This

13:20

episode of Software Engineering Daily is

13:23

brought to you by HookDeck. Building

13:25

event-driven applications just got significantly easier

13:27

with HookDeck, your go-to event gateway

13:29

for managing webhooks and asynchronous messaging

13:32

between first and third party APIs

13:34

and services. With HookDeck, you

13:36

can receive, transform, and filter webhooks

13:38

from third party services and throttle

13:40

the delivery to your own infrastructure.

13:42

You can securely send webhooks triggered

13:45

from your own platform to your

13:47

customers' endpoints. Ingest events at

13:49

scale from IoT devices or SDKs

13:51

and use HookDeck as your asynchronous

13:53

API infrastructure. No matter your use

13:55

case, HookDeck is built to support

13:58

your full software development. Like

14:00

all you see: Hooked Xcl I to

14:02

receive events on your local hosts, automate

14:05

death staging and prod environment creation using

14:07

Be Hooked that a P I or

14:09

Terraform provider and gainful visibility of all

14:11

events using Be Hooked that logging and

14:14

metrics and he hooked up Dashboard. Start.

14:16

Building reliable and scale a bull

14:18

Avenger of in applications today. Visit

14:21

hopethat.com/as Eat daily and sign up

14:23

to get a three month trial

14:25

of the hook That team plan

14:27

for free. From.

14:35

From putting him since on what

14:37

you do in one of his

14:39

don't do what you do you

14:41

do for Nord Security validates in

14:44

Michigan Too costly of the bits

14:46

is correct and validated so that

14:48

people to never served become com

14:50

is where you intended to be.

14:53

sports are not do the things

14:55

like packaging the code into server

14:57

computers and running and virtual environments

14:59

and that sort of not managing

15:02

back process. Not necessarily. yeah that's

15:04

sensory more what are usually do Hollywood of the

15:06

plight of stuff you can take that as we

15:08

are. We do things like Cd duration and some

15:10

of our A for products so we can give

15:12

people the ability to say looks. these packages of

15:14

these other do a lot of it's they have

15:16

known vulnerabilities that we're okay with a voter though.

15:18

These other things we can mitigate or they don't

15:20

apply to us or you're we understand them. It's

15:22

ago had use those but these other cities these

15:25

other problems do not use those and discuss facts

15:27

you getting organizations the ability to does have a

15:29

bit of control over the wild west and so

15:31

they want to power their users. Going to bother.

15:33

Scientists and engineers but they have to have

15:35

some level of control over that. And so

15:37

yes it's everything from building the source as

15:39

you know building it ourselves to the we

15:42

know that what you're getting is what we

15:44

have A to wheat we've actually produced the

15:46

artifacts. from the source code itself to city

15:48

to ration to mickey. Source of integrates with

15:50

other security providers tools. As a general philosophy

15:52

I very much like to force myself to

15:55

nazis competitively or not sentence your somewhere in

15:57

So anytime the center by go what is

15:59

that a person hospice thing that we like

16:01

to use do is break. How can we

16:03

work together to give you a better experience,

16:05

a better solution. And so when it comes

16:08

to things like security and other capabilities I

16:10

don't view what anacondas as as having to.

16:12

I don't view people who part of the counter who

16:15

are you so much that anacondas threats he said it's

16:17

it's kind of growing the pie for everyone so we

16:19

do look and say we don't need to every possible

16:21

thing for security we to see to make sure that

16:23

the pieces that we handle we handle very very well.

16:25

No we integrate that other schooling so that people can.

16:27

people can get the solution of any. Good.

16:36

This episode of Software Engineering Daily

16:39

at Forty Five Vantage. You

16:41

know your father will be for this

16:43

month. So many companies cloud pass or

16:45

the number two line item in our

16:47

budget and the number one fastest growing

16:49

category. A Spend. Vantage. Hop

16:51

See, get a handle on your

16:53

cloud does that says said reports

16:55

and dashboard still for engineers, finance

16:57

and operation seems. With. Vantage, you

16:59

can put costs in the hands

17:01

of the service owners and managers

17:04

who generate them, giving them budgets,

17:06

alerts, anomaly detection, and granular visibility

17:08

into every dollar. With native filling

17:10

integrations with dozens of cloud services

17:12

including eight of us Azure, Tcp,

17:14

Data Dog, Snowflake and To Win

17:16

at Ease, Vantage is the one

17:19

in Op's platform to monitor and

17:21

reduce all your cloud bells. To.

17:23

Get started! Had to vantage that

17:25

sh connect your accounts and get

17:27

a free savings estimate as part

17:30

of a fourteen day free trial.

17:38

Was looks google assistant were than an

17:41

ambiguously babies are more closely related Burma

17:43

I miss we thought or copyright you

17:45

imply the you do things to help

17:48

with. Copyright Reuters I imagine there's

17:50

you're talking with the same sorts of

17:52

things as far as trusting were. Source:

17:54

Of information or not, I'm a

17:57

conservative, comes from the original source,

17:59

and is. Credited. Back to

18:01

the original source, but for can you elaborate a

18:03

little bit more what you meant by copyright. Yeah,

18:05

so we don't Today you know if you

18:08

go get a cast of the that's not

18:10

something we're doing, it's more some the were

18:12

very very interested in leaning into an axial

18:14

couple saying it rak to solve that problem,

18:16

but we want to be part of the

18:18

solution that com. It's really to see

18:20

think if we went how people innovative and connecting

18:22

to open source and helping do it simply streamers

18:24

of than we have to alberta arm as well

18:26

and so he actually recently went. To

18:29

his ears know. Cofounder of Anaconda and the

18:31

the City Oh for a long time again

18:33

big shoes some since here and now it

18:35

was Ceo see recently actually stepped over into

18:38

the role of she's a I officer and

18:40

is below with the copyright and time data

18:42

provenance and making sure that we're all doing

18:44

the right thing with how we use intellectual

18:47

property and data from the trainees models and

18:49

whatnot. that is kind of a passionate subject

18:51

from is actually one of the reason why

18:53

you can kinda anaconda was listen to him

18:56

talk about those with such intensity and twenty

18:58

platform us. And so she is also eating

19:00

some kind of a i specific technology and

19:02

other developments with an anaconda. And so a

19:04

lot of this is just around. Being. Aware

19:06

of that problem says and leaning and everything from us

19:08

during the Ike A and a kind of joined the

19:11

Be Beacon A recently formed A I Consortium. To.

19:13

Even collaborate with people, illegal industry and others to again

19:16

be aware of his problem and and help it evolve

19:18

in a direction that we the going to be best

19:20

for innovation. I don't think it's right to look at

19:22

the since a lock it down because we don't know

19:24

how to handle it at the same point. You.

19:27

Have to you this responsibly that we can

19:29

keep this kind of commons around February to

19:31

benefit from. As

19:38

a listener software Engineering daily you

19:40

understand the impact of generative ai.

19:43

And the podcast we've covered many

19:45

exciting aspects of Cheniere technologies as

19:47

well as a new vulnerabilities and

19:49

risks they bring. Hatter. Ones

19:51

A I read teaming addresses the

19:53

novel challenge as if air safety

19:55

and security for businesses. Lansing new

19:57

Ai opponents. The. a place in

19:59

that sense testing AI models and deployments

20:02

to make sure they can't be tricked

20:04

into providing information beyond their intended use,

20:06

and that security flaws can't be exploited

20:08

to access confidential data or systems. Within

20:11

the HackerOne community, over 750 active

20:14

hackers specialize in prompt hacking and

20:16

other AI security and safety testing.

20:19

In a single recent engagement, a team of

20:21

18 HackerOne hackers quickly identified

20:23

26 valid findings within

20:25

the initial 24 hours and accumulated

20:28

over 100 valid findings in the two-week

20:31

engagement. HackerOne offers strategic

20:33

flexibility, rapid deployment, and

20:35

a hybrid talent strategy.

20:38

Learn more at hackerone.com/AI.

20:40

That's hackerone.com/AI.

20:52

So, it is safe to say that

20:54

you are, as a

20:57

company, skewing towards AI

20:59

versus the general package

21:01

delivery process. Is

21:03

that a fair statement? Yes. Okay.

21:07

So, our existing business, our existing community, our

21:09

existing efforts in user base, like that is

21:11

not being abandoned in any way, shape, or

21:13

form. A lot of this is, as I

21:16

mentioned, a natural extension of what our users have been

21:18

asking us to help themselves, or in many cases, what

21:20

we've been helping themselves. So, I know you know this,

21:22

and a lot of the listeners are going to know

21:24

this, but AI reminds me of those bands that

21:27

have been toiling for 10 or 15 years. And

21:29

then again, hit album, and people are like, whoa, where did

21:31

this overnight success come from? And it's like, this

21:33

has been going on for a long time. And

21:36

we've been doing machine learning and artificial

21:38

intelligence almost from the beginning of the

21:40

company. It's just the new

21:43

kind of impact that LLMs

21:45

have had, and some of the advances they've

21:47

enabled in conversational programming, and chat interfaces, and

21:49

that kind of revolution. I think it's broadened

21:51

what people can apply it to. It's broadened

21:53

the number of scenarios people can actually, the

21:55

problems people can solve with the stuff. And

21:58

so, it feels a bit like an overnight shift. And it

22:00

is in the sense of those breakthroughs, but it's not

22:02

like the field of AI hasn't existed for decades. In

22:05

1997, I worked as a summer intern

22:07

for a company called Biocomp Systems in

22:09

Seattle that used neural nets to help

22:11

people do industrial optimization. People used it

22:13

for like financial stuff, which the CEO

22:16

was really unhappy with. And

22:18

I remember leaving that job and thinking, oh, neural nets are so

22:20

cool, I'll never see those again. Obviously

22:22

very wrong about that. But anyway, so this stuff's been

22:24

around for a while and Anaconda has been working on

22:26

it and helping people with these problems for a very

22:28

long time. And so it's really just, I think,

22:31

the explosion of interest, the kind of addressable

22:33

market, if you will, the number of problems

22:35

makes it seem like we're, I guess, you

22:37

know, moving over there. And I

22:39

think instead it's really more adding these capabilities

22:41

and bringing them to our community and then

22:43

adding all that energy and kind of growing

22:45

it for everyone. I got a

22:47

story to go along with your early AI story.

22:50

In 89, I was working for Hewlett-Packard. And

22:52

my boss at the time, along with his boss and

22:55

a couple other people from the group I was working

22:57

at, had moved not long

22:59

ago from working on a Lisp

23:01

machine, which was part of the HP

23:04

AI strategy. And that was, it was

23:06

all Lisp. Lisp was the AI language

23:08

of choice. That was back in the

23:10

80s. And it wasn't long where you

23:12

didn't hear about that anymore. And it

23:14

was like, well, AI obviously isn't gonna

23:16

happen. This is just, no, it's never

23:18

gonna come. This is unrelated to

23:21

anything that has any value anytime in

23:23

the future. And then bam, it hasn't

23:25

been a bam. It's just been worked

23:27

on little by little in the background.

23:29

And now suddenly it's come into its

23:31

stride. And it's really has

23:33

made some great progress in recent years. A

23:35

good friend of mine, a gentleman named Ian

23:37

Fender, he actually has a Lisp machine. Really?

23:40

Oh, wow. Ian collects computers

23:43

and he has a mind blowing collection

23:45

of technology that he has. I've

23:47

worked with him for years previously and yeah, good friend.

23:49

But yeah, the Lisp world is definitely world I'm familiar

23:51

with. And one thing I try to keep in mind

23:54

with anything is things are never quite as good as

23:56

they seem, but they're never as bad as they seem

23:58

either. And with hype cycles, it's... It's the same

24:00

thing. And so, you know, AI is going

24:02

to take over the world. You know, sure, my, I

24:04

would say super worried about, you know, the impending doom

24:06

or the impending transformation. No, I don't think it's going

24:08

to be as good or as bad, but I think

24:10

you're right. People get caught up in those cycles, like

24:12

they did in the seventies and the eighties and the

24:15

nineties and the disappointment that it isn't as good as

24:17

they thought it was going to be leads to that

24:19

disillusionment. When in reality, it's just a long, steady march

24:21

towards progress. Exactly. Yep. Yep. And

24:23

AI is going to be a major part

24:25

of that, just like other technology has been,

24:27

but no more, no less than that. Eventually.

24:29

We just don't know exactly where yet. And

24:31

so I was speaking to a group

24:34

of interns the other day, and

24:36

their number one question was, will we have

24:38

jobs anymore? Software interns. Software interns. I said,

24:40

well, a couple of things to keep in

24:42

mind. One, AI is going to

24:44

change jobs, but it's not going to eliminate

24:46

jobs. It's going to change jobs. And just

24:48

like any other piece of technology, it's going

24:50

to change jobs. But the other thing

24:52

is, if any job is going

24:54

to be even more important than it

24:56

was before, when in the world of

24:58

AI, it's software developers. So don't worry.

25:00

And that actually helped a lot, I

25:02

think, but I think it's amazing how

25:04

many people are actually worried about AI

25:06

at this point. And not

25:08

really for valid reasons. Well, they're valid reasons,

25:11

of course, but they're not reasons that are

25:13

going to come to fruition. We just don't

25:15

know yet. It's just too early to

25:17

tell for sure what's going to happen. But

25:19

history has shown us what will likely happen.

25:22

I think it's that uncertainty that is what

25:24

drives the worry. If you have a high

25:26

trust environment, if you can rely

25:29

on your community, your friends, your coworkers, whatever

25:31

group you're thinking about, and you have that

25:33

support and you have that trust, I think

25:35

you can face that uncertainty and that adversity

25:37

together when it removes some of the worry. I

25:39

think when you have a low trust environment, then you feel much more

25:42

responsible for yourself to solve those problems. And I think it leads

25:44

to that anxiety of, what if I do lose my job? How

25:46

am I going to pay my mortgage and feed my family? But

25:48

I do try to maintain a positive attitude that, technology,

25:52

industrial, societal progress Has

25:55

made the world better for everyone. I Would not

25:57

trade my life today to go back and be

25:59

a Roman emperor. I would absolutely not. Absolutely

26:01

no. As so as a result. Again, I try

26:03

to keep in mind that yes, there's going to

26:05

be change, but we're going to navigate together and

26:07

my own purpose. You know, my own kind ago

26:10

my sister make the world a better place and

26:12

and hopefully capture a bit about betterment from school.

26:14

I am capitalists, but I do. We make the

26:16

world better place in that means helping. The.

26:18

Nazis grandiose but helping to roll through the say it's like

26:21

trying to make it to as we got his our revolution

26:23

it is better for everyone. But. Your point?

26:25

Yeah, I absolutely disagree. job the future. And one thing

26:27

that I that I made a mistake early on. I

26:30

took a lot of pride in being like a

26:32

hardcore C Plus Plus developer and being like really

26:34

low level knowledge in August and arse of an

26:36

icy side away from Python for a time. For

26:38

me it was like if the tool wasn't hard

26:40

was it a real tool Sand I had kind

26:42

of lost the purpose which is to solve a

26:44

problem like why are you doing what you're doing

26:46

and all the sun was like hey Russia chance

26:48

of the the problems ice with the best tool

26:51

for the job and so his ai is going

26:53

to remove grunt work. If it's gonna be this

26:55

network of experts that I can have helping me

26:57

solve problems and answer questions and broaden my creativity,

26:59

it's learn. Things like how is that not better

27:01

for me. And so yeah, I think as long

27:03

as you. View what you're doing is

27:05

helping people solve problems, address challenges, do things

27:07

like that and you don't get caught up

27:10

into specific. Skill. Set of the

27:12

tool or your knowledge you know kind of esoteric

27:14

knowledge or something is be the reason why you

27:16

have value. Then you'll be able to adapt. You'll

27:18

be able to learn you know whatever those to

27:20

list. learned how people see some from. From.

27:23

Think we have a similar background. I.

27:26

Spend. Most my early career in C

27:28

Plus Plus as well and I was

27:30

viewed oxford and C Plus last modified.

27:32

I was actually the someone wrote an

27:34

article about from the work we were

27:36

doing and it's P C post Western

27:38

or because we want the first ones

27:40

to use C Plus Plus code in

27:42

a Unix kernel. Is like that

27:45

was revolutionary back. This is not that

27:47

hard you know exists. This is how

27:49

this works. That of course it's all

27:51

pretty common nowadays, but it was kind

27:53

of interest to him. Always imagined it

27:55

would never move away from C Plus

27:57

Plus. Other thing, the mood me away

27:59

some C Plus Plus. The project that

28:01

ended up point me into learning Roby.

28:04

And the alley not Ruby was such a

28:06

sake language until a really high powered and

28:08

it's now have all the languages I used.

28:10

Ruby is by far my favorite just because

28:13

it's. So. Easy to do what you

28:15

want to do. Now it's not the best for a

28:17

lot of environments. Lot of. Projects.

28:19

I work on nowadays but at the time

28:21

was really interesting and but it's the language

28:23

is like your pi thoughts. The i bought

28:26

me away from C plus was yeah I

28:28

totally great guy Rubber I took such pride

28:30

I did a ton of the that template.

28:32

never going. Back. In early aughts and

28:34

I was so proud and in turn is by

28:36

not only say as he did come up with

28:38

the great designed to a particular challenge palmer that

28:40

I was working at time but the I read

28:42

All and Rescues books and all that stuff and

28:44

it was was so much pride in the difficulty

28:46

of it and it really was revelation to say

28:48

that it's just the right tool for that job.

28:50

There are plenty of other tools out there and

28:52

yeah I was is used to to and it

28:54

was fine actually. early on we had Python versus

28:56

a list arguments but then we used Chef. Back.

28:59

When it was still fall Outs code but the Chef

29:01

Automation language and that was my introduction ruby and it

29:03

was the same thing of like oh. Again,

29:05

Right tool for the right job. Really going

29:07

abroad, my perspective. Makes.

29:10

Perfect sense of self, but also

29:12

research for anaconda a sound of

29:14

phrase I'm I sound very enlightening

29:16

that I sense a layman. Very

29:18

interesting that I don't think really

29:20

applies now that we've had this

29:23

conversation. While. I think a

29:25

better phrase might apply setting that context.

29:27

Let me tell you what the phrase

29:29

was, it was Anaconda is low code

29:32

A I Development. Now.

29:34

And I don't see that in what you're

29:36

saying here, but maybe a better phrase? all

29:38

birds you for that same com it might

29:40

be. Sastre. Onramp day

29:42

I get started with have quicker easier

29:44

is that a better example of what's

29:47

you really are doing his I think

29:49

those are two aspects I don't quite

29:51

into the same thing other they are

29:53

related. The low code part is so

29:55

we we are car doesn't year college

29:57

blocks. And it's a get this over them. It's really

29:59

bad. The regime around helping people understand and learn

30:01

Python it's aimed at students and whatnot and Edgewater

30:04

kind of in the names. but the real focus

30:06

their or could have taken distraction out of it

30:08

is low code know code kind of composition of

30:10

capabilities to produce results. And if you look at

30:12

things like I want to have models and I

30:15

want to corporate the my application Michael that's just

30:17

more lego blocks that you're kind of using to

30:19

assemble and build your final structure. but yes faster.

30:21

Honor of the I that is both we're trying

30:23

to you today and I guess a guiding principle.

30:26

Of. The company and as is everything from.

30:29

Making. It simple on when he myself but

30:31

maybe some people incorporate and you're gonna see

30:33

some stuff coming from us later this year.

30:35

I think it'll and body those things much

30:37

more clearly. Good. Time as

30:39

very much looking forward seen other that's

30:42

great. Talk about

30:44

standardization. How the standardization fit

30:46

into your strategy. In.

30:48

What way? Where Do you mean? So.

30:51

Is a eyes closed or

30:53

open? That's

30:55

obviously biased in this prospectus for

30:58

us up or dislike. I

31:00

think open source ecosystems have clearly

31:03

demonstrated over the last couple decades.

31:05

That. They were really drive innovation and

31:07

you know look at a change in

31:09

Microsoft. Look. At them except under eighteen

31:12

and Under Bomber A dazzling and a bomber.

31:14

Open Source is absolutely a feared from there

31:16

and all that kind of or thought of

31:18

the time and the now one of the

31:20

largest about the largest contributor open source projects

31:22

in the world. And it's really it's not

31:25

just in words, it's assets are hope I

31:27

can be seen a nap and I think.

31:29

Being. Able same up. there are billions of people. Miss World.

31:32

And. We want to empower all those people

31:34

you know. Talent is is evenly distributed, opportunity

31:36

is not. So Let's make opportunity be equally

31:38

distributed and let's give people the ability to

31:41

contribute all that innovation. I don't see how

31:43

any close ecosystem any one company could possibly

31:45

help to compete with that as I work

31:48

at a company and with comes to settle

31:50

again. I'm a capitalist, but I think the

31:52

world better place when you have that open

31:54

innovation. and there is absolutely

31:57

a role to play there and

31:59

excited look at this, we say, okay, what

32:01

are the kinds of problems the community doesn't want

32:03

to solve? What are the kinds of problems the

32:06

communities aren't good at solving? What are the kinds

32:08

of problems that we can collaborate with those communities?

32:10

And then what are the kinds of problems those

32:12

communities run into in organizing and collaborating and maturing

32:14

and operating over time? And how can we help

32:16

address those problems? And so when it comes to

32:18

that, I guess I'm a true believer that open

32:20

source and open collaboration, that's just, that's the main

32:22

driver. Standardization, I think really helps when it comes

32:24

to helping to enable that, you

32:27

know, if you want to get together with

32:29

a friend and go to dinner, that's trivial, right? You know,

32:31

now it's maybe a birthday party, and you got 10, 20

32:34

people takes a little bit more effort, might have to schedule

32:36

a time, might have to make a reservation, you have to

32:38

put stuff for that. Now you're doing a wedding, and

32:40

you got 100 people, you got 200 people, and

32:42

like that takes real planning. And then now you

32:44

have like a major concert, you have your Taylor

32:47

Swift, and you want to come to a city

32:49

and take over. And it's just it's just order

32:51

of magnitude more collaboration. And I think when you

32:53

want to have innovation happen at scale, and when

32:55

you want to enable people to build and to,

32:57

again, put those Lego blocks together, you have to

32:59

have some kind of definition of a Lego block,

33:01

right? If you had 50 different kinds of sizes

33:03

and connectors and whatnot, like no one's building, you

33:05

know, whatever the latest hot Lego model. And

33:08

so I think there's a role for standardization, which is really all people just

33:10

coming together and saying, look, let's define the, you know,

33:12

your programmer, so you get it, like you have an

33:14

API, let's define the interfaces. And let's define those interfaces

33:16

that we are all free to innovate within those interfaces.

33:18

But now we have those ways of collaborating together. So

33:21

that's where I believe standardization play a role where I

33:23

don't like standardization, I think I go so far as

33:25

to say, I think you and definitely listeners would agree

33:27

to this is when it's used for any kind of

33:29

capture, when it's used for any kind

33:31

of, in the old days, companies were trying into

33:33

standards bodies and make it so that their technology

33:35

was the only compliant one or things like that.

33:37

But again, getting back to supporting open source and

33:40

supporting the commons, I think there are ways of

33:42

having those standards that foster innovation and foster collaboration

33:44

without locking people out. So you

33:46

can imagine open

33:48

interfaces for AI, that

33:51

general comment can apply to a

33:53

couple of different layers within the

33:55

AI ecosystem. It can certainly companies

33:57

like open AI are trying to...

34:00

to create open interfaces to give you

34:02

access to AI. But

34:04

there's also interfaces to consistent uses

34:07

of the same datasets and

34:09

making datasets available for AIs

34:11

and making large language models

34:13

available for multiple use cases in different

34:15

ways. And standardization that can happen

34:17

in those areas or openness, I should say.

34:20

Maybe it's a better term than standardization given the

34:22

language you're using. I'm not sure we can talk

34:25

about that. But there's different layers

34:27

there. Where do you think we

34:29

are in that hierarchy as far as we

34:32

know how to do open software and

34:34

we know how to do open APIs? Do

34:37

we know how to do open datasets? Do we know

34:39

how to do open large language models? Do we know

34:41

how to do whatever the

34:43

next level is there? Are we

34:45

good at that yet? Are we starting out or

34:47

do we just not know where we're going yet?

34:50

That's an interesting question. I've never been asked that

34:52

question with that phrasing and framing, which is really

34:54

cool. I would say, on one hand, I

34:57

think we absolutely do know what to do. And

34:59

what I mean by that is you look at

35:01

how software has been distributed with the different licenses

35:03

and the evolution of how people open up their

35:05

code and the variety of ways people can do

35:08

it. And you say, OK, conceptually, we're probably

35:10

going to do the exact same thing for data. We're going to do

35:12

the exact same thing for models. So I think

35:14

in that sense, there's a bit of, I don't want to

35:16

say it's common sense. I'm not trying to oversimplify. But I

35:18

do think that there is a way of saying, look, these

35:20

problems are solved in this domain. Let's just broaden our perspective

35:22

and say, OK, they're probably going to be solved in a

35:24

similar way. The devil is in the details, though.

35:26

So the part that we haven't gotten right is what do those

35:28

data licenses look like? How do we make

35:31

sure that people who contribute are maybe

35:33

compensated or credited and whatnot? And so

35:35

it's not that those technical challenges and

35:37

those legal challenges and those philosophical challenges

35:39

aren't there to be solved. But I do

35:41

think that we can say we've solved these problems before

35:43

in a different domain, and we can just almost apply

35:45

those approaches in this new domain. I think there's a

35:47

tendency to look at something new and say, this is

35:49

new. And so therefore, it's entirely new. We don't know

35:52

how to do anything in this area. And

35:54

in reality, it's just a different view of the problem you've

35:56

already solved. And so you just kind of have to have

35:58

that adaptation. to take a while

36:00

to shake through all that. And given the, again,

36:03

the massive uncertainty and that kind of changes there,

36:05

and honestly, the economic impact, it's probably going to

36:07

be a complicated discussion. Yeah, I 100%

36:10

agree with everything you said there, but there

36:12

is one aspect with things like

36:14

data and large language models that

36:16

use data that's different, typically, not

36:19

all the time, but typically compared

36:21

to just code. And hate

36:23

saying the word just code, but you know what I mean.

36:26

And that is privacy. And

36:28

whether we're talking about PII

36:30

or whatever, there's

36:33

information in data that is

36:35

specific and valuable just by

36:37

having the data, independent

36:39

of whether you have the right to use it

36:41

or not, having that information here may not be

36:44

appropriate and the privacy aspects that

36:46

go with that. That's different

36:48

than with code, is rarely

36:50

a time where code itself,

36:53

your open source code has to be

36:56

private. No, you're absolutely

36:58

right. Rarely that situation, but that wouldn't

37:00

be a situation with data. Open

37:02

source data might still be private. Yeah,

37:05

that's exciting. The emergent risks or problems

37:07

that come also when you, any one

37:09

data set may be fine, but

37:12

you could put three or four of them together and suddenly

37:14

now you can identify people and suddenly now

37:16

you can do attribution. So there is an issue where,

37:18

you know, you get these emergent things that come out

37:20

of, okay, well, I've released my data set because it's

37:23

fine by itself. There's no PII or there's no way

37:25

of tying it back. But then you get, you know,

37:27

other people release their data sets and suddenly somebody realizes,

37:29

well, if I get data sets, A, B, C and

37:31

D, now I can do horrible things.

37:34

Well, how do you deal with that problem? Because you

37:36

have to source each one of those data sets is

37:38

fine. And so how do you, yeah, that's

37:40

exactly a challenge. I remember the first time I learned

37:42

that, you know, this is obviously even before AI, this

37:45

is just when people were like, Oh, I can take,

37:47

I don't know, IP address information and I can take,

37:49

you know, health information. I can buy data sets from

37:51

credit card companies and, you know, your shopper's club card

37:53

for your grocery store. And I can put these things

37:55

all together and I can learn all kinds of interesting

37:57

stuff about you that you did not intend. But again,

37:59

every day, data set provider in that sense wasn't doing

38:01

anything wrong. Right, right. And so how do you then

38:03

deal with that in this world where the models can

38:05

be trained on that data and then people can actually

38:07

do stuff is a very interesting question. I

38:09

don't have an answer for that one. This is why I love being in

38:11

this space every day. So we actually have a question, not get excited. Yeah,

38:14

no, that's great. Yeah, I think if you

38:17

would agree that this is an important area that has

38:19

to be dealt with in the open

38:21

data. Well, I do. I also

38:23

wonder how much data is actually not going to be

38:25

public. I think there are a couple of guiding principles

38:27

that I think I'll just say I have at the

38:30

moment, always open to evolving them. But one of them

38:32

is that small open source models are going to be

38:34

great, if not perfect for many, many people. But they're

38:36

going to want to be able to control the data.

38:38

Like I think the AI revolution

38:41

has really, I think, changed people's

38:43

view of the data, right? Like you see people putting

38:45

all their content behind paywalls or locking it behind agreements that

38:47

you're not going to use to train models because they

38:49

suddenly realize, hey, all that data is actually incredibly valuable in

38:51

a way that it was, it was not that it

38:53

wasn't before, but that, you know, there wasn't that direct

38:55

connection. And so I think you're going to have people keep

38:57

those data sets private, and they're going to want to

38:59

train their models internally. They're going to

39:01

want to govern them internally and probably run them

39:04

privately or at the edge or in some hybrid

39:06

fashion. And so I think that is a change.

39:08

I think you're going to see what people previously

39:10

kind of gave away, or at least it didn't

39:12

necessarily govern, and they aren't going to do that anymore.

39:15

And so that will actually hinder innovation. It'll be very

39:17

interesting to see, like, does that mean that only companies

39:20

that have massive resources like OpenAI and others and Microsoft

39:22

and the Googles of the world that can license content

39:24

are going to be able to train? Or,

39:26

you know, how is that going to evolve over time? I don't know. Yeah,

39:29

it's a great question. I think

39:31

that's one of the fundamental questions that's going to

39:33

come with the whole AI revolution.

39:36

Is AI going to be, in

39:39

general, giant data sets, or

39:41

is it going to be many, many, many

39:43

small data sets and training that comes from

39:45

that? I think there's a big

39:47

world in the

39:49

small data set AI. Let

39:52

me give you a simple example. I would love

39:54

more than anything. I write a lot of content.

39:56

I've got books and articles and everything that I've

39:58

written. I would love. I love nothing

40:00

more than to take all my content, put

40:03

it into an AI, and have a

40:05

chat bot be on my website to

40:07

be able to have people ask questions

40:10

and respond with things that I

40:12

know. Now, I know I can

40:14

do this today. I just haven't taken the time to do

40:16

it. And I know there's companies that are

40:18

looking or have started to do that sort of things, but

40:21

that's really what I want. That's not a large dataset

40:24

problem. It's a large learning problem,

40:26

but it's not a large dataset

40:28

problem. Should we be

40:30

separating the learning and the ability

40:32

to learn from an AI model from

40:35

data to be able to do

40:37

things like that easier? If I mentioned

40:39

any question, I think so, because I like

40:41

to think of these models as almost like

40:43

expert assistants, a grad student that

40:45

maybe is fresh out of college or even just

40:47

a, I don't know, computer science students fresh out

40:49

of college. If I'm building, okay, I'm a technical

40:51

co-founder of a startup and I'm trying to build

40:54

a team out, I don't look for one person

40:56

that can be the product manager, the technical lead,

40:58

the system architect, the IT person, security person, the

41:00

front end developer, back end developer, et cetera. I

41:02

look for a team of people and

41:04

I combine them together and it's the, some is

41:06

greater than the individual parts, that's the power there.

41:08

And I feel the same way about AI models.

41:10

And so I think you're right, being able to

41:12

say, hello, we have these, you'll see that through

41:14

a mixture of experts and other systems that say,

41:16

let's take these individual pieces and put them together

41:18

and actually generate it that way. As I mentioned

41:20

earlier, you train your model in the entirety of

41:22

data on the internet and you're gonna get every

41:25

piece of fan fiction, you're gonna get celebrity birthdays

41:27

and celebrity obituaries and you name

41:29

it, all in that model, do I need

41:31

those in order to ask it to help

41:33

me how to improve the flow of my

41:35

story or the grammar or to explain physics

41:37

concepts to me or I'm a big

41:40

fan of Khan Academy. If I'm trying to brush up

41:42

on different math concepts, I don't need all that. And

41:44

so I think you're exactly right, taking small data sets and small

41:46

models and putting them together so I have this team of experts

41:48

that really enable me and empower me, I do think that that

41:51

is a large part of the future. That's

41:53

great. So normally about this

41:55

time in an interview, I ask, so what's next

41:58

or what's the future? But we've been talking. a

42:00

lot about that, but maybe the best

42:02

way for me to rephrase that question this

42:04

time is to say we've talked a lot

42:06

about where AI is going, but where

42:09

is Anaconda going next? Perfect.

42:12

Yeah. So we did touch on this earlier, but it's

42:15

taking what we've traditionally done, getting the right

42:17

bits on your computer, making it manageable and

42:19

governable, and then helping people solve kind of

42:21

higher level AI, machine learning, data science problems,

42:23

and expand that into data and models. I

42:25

mentioned serverless Python, things like PyScript, and whatnot

42:27

that allow you to actually execute this stuff

42:29

using the cheapest hardware, which is the hardware

42:31

you already own, i.e. your laptop or your

42:33

phone, things like that. We're also very, very

42:36

heavily focused on high performance Python. People

42:38

often talk about, oh, Python's not as fast as C++

42:40

or it's inefficient or whatever. And I think getting back

42:42

to the earlier point about why, what are you trying

42:45

to solve? Why is this not the right tool for

42:47

the job forever? That wasn't the limiting factor. The

42:49

limiting factor was getting it into people's hands, making

42:51

it understandable. Again, collaboration, secure, all this stuff that

42:54

Anaconda traditionally focused on. But

42:56

Python's the lingua franca of AI, and AI is

42:58

central to the world. And that Nvidia GPU isn't

43:00

cheap. And they're

43:02

in short supply. And so helping people get

43:04

the most out of their investment

43:07

in their infrastructure is actually a core concern.

43:09

And I think Anaconda is uniquely suited to

43:11

help solve Python performance problems. And so it's

43:13

a category of problems and it's a category

43:15

of technologies and approaches. So you're going to

43:17

see a lot of stuff from Anaconda around

43:20

that, both directly from Anaconda, but also what we're going

43:22

to foster in the open source ecosystem and the Python

43:25

ecosystem. Everything from the interpreter to the language to

43:27

the libraries, you name it. And

43:29

then you kind of combine those all together and

43:31

it's really about making it easy for

43:33

people to build these models, to

43:35

incorporate these models, to deploy these applications fast,

43:38

efficiently, effectively at scale. This

43:41

has been a great conversation, Rob. I really

43:43

appreciate it. We're so close to being out

43:45

of time. But I want to thank you

43:47

so much for coming on. This has been

43:49

a great conversation. Thank you. My

43:51

guest today has been Rob Futrich, the

43:54

CTO, not EVP of engineering, but

43:56

the CTO at Anaconda. Rob, thank

43:58

you for joining me. and software

44:00

engineering daily. Thank you, Lee. I

44:02

always love an exciting conversation. And thank you for

44:04

absolutely providing one. It's been fantastic talking to you.