How to Survive the Next Tech Revolution (w/ Brett Gibson)

00:00

technology it's changing the world and

00:03

it's the people who can build something

00:05

from scratch something that never

00:07

existed before

00:09

who will go on to change that world

00:12

but then what happens for v2 the

00:15

rewrites always better who wins in that

00:17

world it's a question playing out in AI

00:21

today so that's why we're sitting down

00:23

with my multi-time co-founder and friend

00:25

Brett Gibson he runs initialized now

00:27

alongside Jen Wolfe now that I'm back at

00:31

YC what's the future gonna look like it

00:34

won't be the same as the past but it

00:36

will rhyme let's get started

00:41

[Applause]

00:43

all right

00:48

[Music]

00:53

Brett and I worked on a startup named

00:55

posterous together with our other

00:57

co-founder Sachin Agarwal

01:00

one of the new things we did was a new

01:02

kind of authentication by email header

01:05

you could post your blog where we would

01:08

use your email to figure out who you

01:12

were so then you could just send email

01:14

to postpostress.com and we'd figure it

01:17

out that was the thing that had not

01:19

really been done before in a consumer

01:21

product and to do it it required a

01:24

different kind of engineering mindset a

01:27

big part of an engineer's day-to-day

01:29

problem is figuring out the right tools

01:32

for the job figuring out what they

01:33

should be using and figuring out how it

01:36

ties back to user value because you know

01:39

I think that I come from this you know

01:42

this default engineering mentality where

01:44

I really want things to be precise and

01:48

deterministic and you know oftentimes

01:51

the engineering challenges you face to

01:54

get to get the product working the way

01:56

you want it for your users are much more

01:58

probabilistic because it hadn't been

02:01

done before we had to do some pretty

02:03

unusual things to get it right you know

02:06

we ended up writing our own uh parts of

02:08

our own SMTP client which I don't I

02:10

don't really recommend doing a lot of

02:13

analysis on the routing that email went

02:15

through so that we could you know have

02:17

some probabilistic score about How

02:19

likely it was to actually be from you

02:21

and then the postaven experience was

02:23

night and day Preposterous later sold to

02:25

Twitter for 20 million dollars but a few

02:28

years later they decided to shut it down

02:31

since the real reason why they bought it

02:33

was for the team

02:34

when it shut down Brett and I got back

02:37

together to write it again this was

02:40

called post Haven and this time when we

02:43

wrote it again this post by email

02:46

authentication that we had to write for

02:48

the first time had been commodified

02:50

there was a product it was mail gun it

02:52

was ready it worked off the shelf for 90

02:55

of what we needed had we been too

02:58

grounded in fighting the last battle we

03:00

would have missed that we would have

03:01

wasted a lot of time we would have built

03:03

a lot of software that we didn't need to

03:04

build what we're seeing today in 2023 is

03:07

similar to what we found rewriting

03:09

posturous again as post-haven even just

03:12

four years later first it's a lot easier

03:14

to rewrite once you've written it once

03:16

but second tooling an open source and

03:19

freely available apis gets so much

03:22

better in just the course of a few years

03:24

for AI now that set of years sometimes

03:29

is months or weeks

03:31

but know that this isn't a new thing

03:34

what is new is how much more powerful

03:37

these capabilities in ml and large

03:40

language models have become now everyone

03:42

has this new primitive of generative Ai

03:45

and especially large language models and

03:49

they're trying to figure out exactly how

03:51

to use them and the moment is actually

03:52

very strange because there's a lot of

03:55

interesting Tech on how to do it you

03:58

know there's a lot of best practices

03:59

around stuff like vector databases and

04:02

embeddings so that you can get your

04:04

prompts correct but it's it's

04:07

exacerbated because large language

04:09

modules are fundamentally

04:11

non-deterministic systems

04:13

um that's you know the the lack of

04:15

determinism is is in part what makes

04:17

them work so well and seems so magic and

04:20

so the the class of I guess you know

04:23

first generation hacks to get them

04:25

working in ways that deliver value to

04:28

users is is perhaps more interesting

04:31

than some of the things we've seen in

04:32

the past along the same lines this alien

04:34

intelligence we have access to now it's

04:37

a strange one and we're just in the

04:39

first Innings of the tooling the

04:41

foundational models and especially the

04:44

open source it's interesting because I

04:46

see a lot of parallels in what's

04:48

happening in AI right now and what

04:50

happened in crypto over the last few

04:53

years because there was a moment you

04:55

know there's so much infrastructure

04:56

around wallets and building up

05:00

you know ability to be consumer facing

05:02

in crypto that if you were building in

05:04

2018 you just had to build from scratch

05:07

and the underlying ecosystem has come a

05:09

long way and you're probably having to

05:11

throw away some of that custom code and

05:14

in AI it's it's the same thing but

05:17

perhaps even moving faster because not

05:19

only do you have the day-to-day building

05:22

out of of the of the types of tooling

05:25

that would make your life easier you we

05:28

have these step function improvements in

05:30

the underlying foundational models the

05:32

change the requirements I think that the

05:35

North Star is always backing it out from

05:37

what what experience you want to give to

05:38

the user because if you make sure you

05:41

get that right then you can change your

05:44

technology behind the scenes

05:46

um but you know trying to try to have

05:48

the trade-off of uh how long it takes to

05:51

deliver that versus competitors that is

05:54

the lesson from software engineering and

05:56

it's the same one Brett and I learned in

05:58

2008 2013 and we're learning again today

06:02

in 2023 whether it was just some new

06:05

capability in the early days of social

06:07

to what's happening today in AI first

06:09

you have to be willing to build

06:11

something that doesn't exist yet and do

06:13

some pretty unusual things to do it but

06:16

then later you need to be able to do

06:19

rewrites and stay current with open

06:21

source and new Stacks capabilities and

06:24

apis all while being connected to the

06:28

needs of your customers and users make

06:30

stuff people want and that is what makes

06:33

products sticky and what makes a durable

06:36

business being a domain expert in

06:38

tooling

06:40

really helps gives you a lot of Leverage

06:42

toward that goal but it's not the goal

06:45

in its own right

06:48

thank you Brett's one of the best

06:51

Engineers I've ever had a chance to work

06:52

with across our Year's writing code for

06:54

Preposterous y combinator post Haven and

06:57

initialized and he's always been super

06:59

pragmatic and I've learned something new

07:01

from him regularly he's got a new

07:03

podcast called the high bit where he

07:06

sits down with some of the best

07:07

technical Founders to talk Tech

07:09

engineering and the idea maze kind of

07:11

what we just did in this episode one of

07:14

those companies was bison Trails which

07:16

was acquired by coinbase and post IPO

07:18

the notional value of that acquisition

07:19

was nearly a billion dollars that's

07:22

because it became coinbase Cloud one of

07:24

the key SAS Revenue generators for

07:26

coinbase

07:28

Brett invested in bison Trails when it

07:30

was just a few people very early when

07:32

proof of stake was little more than an

07:34

idea something that did not actively run

07:38

the world's largest blockchains like it

07:40

does now but that's precisely how you

07:43

should invest when you want to create

07:45

the future when I met bison Trails there

07:48

was there were little or no major

07:49

networks running proof of stake and and

07:52

I still had doubts in my mind as to the

07:55

viability of widespread shift to proof

07:58

of stake networks especially for already

07:59

running networks like ethereum now it's

08:02

very clear that it's it's demodually

08:04

possible if not optimal in many contexts

08:07

with the Bryson Trails Founders

08:08

recognized was that relative to the

08:11

current proof of work mining that was

08:13

happening running validators on proof of

08:15

stake networks was a novel and Rel and

08:19

difficult problem they were able to

08:22

build something that had never been

08:23

built before because of two things first

08:25

they were generalist focusing on a wide

08:29

skill set here's Aaron Henshaw

08:31

co-founder of Bison Trails on how hiring

08:34

was a big piece of their success the

08:36

best thing that we did was we've

08:38

actually brought on like generalists at

08:39

the beginning more people who could both

08:41

like right JavaScript and do terraform

08:42

that's a really wide band between those

08:45

two things but that's what we needed at

08:47

the beginning because we had JavaScript

08:49

right we actually had like a front end

08:51

we had an API to build so we had a

08:53

middle and then we had this whole back

08:54

end you're lucky if you go far enough to

08:57

be able to hire the people that you need

08:59

to fix your problems that you've created

09:00

for yourself and those people you know

09:03

they come in wide open but they like

09:04

they're so good at it they're so they're

09:06

so good at these specific problem

09:07

solving and like you're just like in awe

09:10

that they can come in and just like

09:12

change it all and make it work like 10x

09:15

better second you've really got to

09:17

invest into software testing so that

09:18

when the chips are on the line and they

09:21

always are in crypto custody and staking

09:23

you have the right engineering infra

09:25

some of the things that we did well

09:27

early was like we did not try to over

09:29

engineer things there was probably some

09:32

opportunities to use like slightly more

09:35

off-the-shelf CI CD and bake in a little

09:38

bit more testing early on that I think

09:40

like might have slightly slowed us down

09:42

but actually sped us up this may be like

09:44

one of the biggest takeaways from the

09:46

whole thing was

09:48

you should invest heavily in tests

09:50

especially even early like don't go

09:54

crazy right like you can't be writing

09:55

tests for ever but you should have core

09:58

tests that cover base cases and that

10:01

like help your whatever it is that

10:02

you're deploying Founders and investors

10:04

both make money and impact the world

10:07

when they believe something nobody else

10:10

believes yet but are right another

10:13

company Brett's invested in is a very

10:16

ambitious one called astroforge they're

10:19

literally doing mining for high value or

10:22

on asteroids at no point was anyone

10:25

really worried about the market

10:27

opportunity you know we're always

10:29

balancing Market risk with technical

10:31

risk in in the startups we fund and

10:35

there's just no question that if astral

10:37

Forge can go to asteroids and mine Rare

10:41

Earth elements and bring them back you

10:43

know the the only real risk in the

10:45

market is that they do it so

10:46

successfully that those metals are worth

10:49

a lot less and so really all we spent

10:52

our time was in was validating is this

10:55

possible and I think specifically if it

10:59

is in fact possible why is it possible

11:01

now suddenly why why did why are we

11:03

seeing this company today because on

11:06

some timeline it seems like something

11:08

that's well within the ability of humans

11:10

to produce but you need to have a view

11:13

that it's going to be it's going to

11:16

happen in time for this company to make

11:18

money off it Hosea Kane co-founder of

11:20

Aster Forge is trying to solve

11:21

engineering problems on literally the

11:23

edge of human capability and the thing

11:27

is you don't have a manual for that

11:29

you've got to write the manual yourself

11:31

the gut reaction was asteroid mining is

11:34

difficult the refinery is difficult it

11:36

couldn't be awkward it couldn't be a

11:37

chord it couldn't be it couldn't be the

11:39

simple pass through the length that we

11:42

went through to solve this problem

11:44

baffles me to this day you're focusing

11:46

on this problem that's difficult glaze

11:48

using over something so simple and and I

11:51

talked about this fishbone diagram

11:52

because these are things that you should

11:54

really actually check you should write

11:56

out great the graphs before agreement do

11:57

we check everything in there and not

11:59

just say like yeah I checked the voltage

12:01

which is just one part of it

12:03

um and we just checked it off like okay

12:05

cool power supply works hard this works

12:07

great even if you are under heavy time

12:09

pressure take a minute and pause slow

12:11

down it's okay to do that and really

12:14

assess the situation and really validate

12:16

all these potential failures to get to

12:19

that Solution that's just a taste of

12:21

what you'll find in Brett's new podcast

12:22

I'm subscribing and I recommend that you

12:25

do too you know the intent is less a

12:27

technical Deep dive and more about this

12:29

this the the story about the art of

12:33

problem solving and what goes into

12:35

day-to-day Engineering in these

12:37

technical domains and how it can be you

12:40

know just interesting in its own right

12:41

you can find high bit by Brett Gibson on

12:45

anywhere you listen to podcasts or

12:47

initialize

12:49

Link in the description below

12:55

initialized is also hiring a new

12:57

investment partner you can't apply to

12:59

jobs you don't know about so that's why

13:01

they're running an open hiring process

13:03

for this role initialize is currently

13:05

hiring a partner role and you know we we

13:10

really prize uh backgrounds in founding

13:14

and working in startups and ability to

13:16

advise startups in in some area of deep

13:20

domain expertise so uh you know high

13:23

level I think we're looking for someone

13:24

who's who's been a Founder before you

13:27

know hopefully gotten a company uh

13:29

through perhaps a series B round and and

13:31

but also it's very helpful if they have

13:34

some investing experience whether that's

13:37

as an angel or within or as part of a a

13:41

venture firm

13:42

um you know a big a big part of the

13:44

applied knowledge of going from founder

13:47

to investor is just getting reps and and

13:50

seeing pitches being an investor to me

13:53

is just an absolute blessing and if you

13:57

ever wanted to join a team doing it

13:59

right and you have some investment

14:01

experience already this is a big

14:04

opportunity

14:05

that's it for this week Brett's an old

14:07

friend of mine and I'm so excited to see

14:09

where he takes initialized Capital links

14:12

in the description to apply to be an

14:13

investor at initialize and to subscribe

14:15

to Brett's new podcast hi bit technology

14:18

is truly reshaping the world and its

14:21

technologists the engineers the PMS the

14:24

designers and the builders themselves

14:26

they're the ones doing it it if you

14:29

watched to the end you must believe this

14:32

too and I'm here to tell you you're on

14:35

the right track see you next time

14:38

[Music]

14:41

thank you

14:43

[Music]