►
From YouTube: DDL - Episode 1: Digging into Airbnb's Data Culture
Description
DDL: Decoding Data Leadership - Episode 1
Former leaders from Airbnb's data organization discuss approaches to data democratization, open source project rollouts, practical advice about building your data team/team structure, how to build political capital, and forecasts for the future of data management.
A
Talking
about
the
big
challenges
that
we
face,
how
we
tackle
them
and
the
lessons
we
learn
along
the
way
in
this
first
episode,
we'll
be
hearing
from
former
leaders
of
airbnb's
data
organization,
we'll
hear
everything
about
how
they
approach
data
democratization
to
rolling
out
open
source
projects
like
airflow
and
superset,
also
get
some
practical
advice
about
building
a
data
team,
how
to
think
through
kind
of
structuring
the
team
within
an
organization
and
also
how
to
like
start
to
build
or
gain
some
political
capital.
So
you
can
go
on
to
solve
some
really
big
data
challenges.
B
Hello:
everyone,
I'm
swarup,
I'm
the
co-founder
of
a
company
called
actual
data,
and
I've
had
the
pleasure
of
working
with
this
amazing
set
of
folks
here
who
were
all
at
airbnb
at
a
time
when
the
data
culture
was
really
built,
and
many
great
things
happened
in
terms
of
how
data
was
actually
used
for
business
value
in
terms
of
the
amazing
open
source
projects
that
came
out
of
it
in
terms
of
how
data
education
was
pioneered,
and
you
know
best
practices
and
experimentation.
B
B
So
that's
the
overall
goal
here
right,
let's
start
with
quick
intros,
I
did
did
a
brief
intro
about
myself
just
to
complete
that
a
little
bit
actual
data
drives
the
open
source
data
project
in
collaboration
with
linkedin
data
hub
is
deployed
in
hundreds
of
companies.
Now
out
there.
We
also
win
the
sas
product
to
commercialize.
It
I'll
keep
it
brief
for
now
and
I'll
just
go
in
the
order
of
the
faces
that
I
see
on
my
trade
here.
C
All
right,
hey,
y'all,
good,
seeing
you
again.
My
name
is
jay.
I'm
ceo
of
epo,
where
we're
building
next
gen
experimentation
framework
heavily
inspired
by
the
tool
we
had
at
airbnb
and
the
other
similar
types
of
tools
we've
seen.
Otherwise,
so
I'm
sure
we'll
be
getting
a
lot
more
into
experimentation
and
infra
and
why
you
do
it
later
on
so
I'll
just
pause
here
and
let
you
all
introduce
yourself.
D
All
right,
I'm
I'm
max,
I'm
co-founder
of
preset.
I'm
really
excited
to
be
sitting
in
the
same.
You
know
virtual
room
as
the
folks
here
today
I
think,
is
the
first
time
we're
all
together,
since
probably
2017
at
least
I
don't
think
we've
we've
all
chatted.
You
know
together
since
that
time.
D
We've
all
think
we've
all
like
stayed
connected
and
had
a
bunch
of
conversation
individually,
but
but
it's
good
to
see
these
people
in
the
in
the
room
together,
I'm
the
arsenal
creator
of
apache
air
force,
I
joined
airbnb
2014
and
I
started
working
on
airflow
right
at
that
time.
D
About
a
year
later
I
started
apache
superset,
and
then
you
know
since
then
about
a
few
years
back,
I
started
a
company
called
preset
so
where
we
we
offer
managed
service
around
apache
superset,
which
is
a
data
visualization
data
consumption
tool
very
much
like
the
visualize
visualization
layer
for
the
modern
data
stack
so
who's
up
next
lindsay.
You
want
to
go.
E
Sure
can
you
guys
hear
me?
Okay,
I've
got
some
wi-fi
issues,
we're
good
awesome,
great
yeah,
so
awesome
to
see
everyone.
My
name
is
lindsay
ceo
and
founder
of
a
company
called
iggy,
which
was
very
much
inspired
by
things
we
didn't
have
at
airbnb.
While
I
was
there
so
what
iggy
does
is
we
help
folks
build
build
better
products
and
models
with
location,
data
and
yeah?
I'm
sure
I'll
talk
more
about
it,
but
yeah
fantastic
to
see
everyone
in
one
place
and
yeah.
F
Wow,
just
to
echo
what
I
think
everybody
said
it
is
so
good
to
see
you
all.
I
have
such
fond
memories
of
working
with
each
of
you
and
it's
it's
pretty
amazing,
to
see
how
many
of
us
went
on
to
start
companies
or
become
you
know,
founders
of
companies
across
the
data
space.
F
So
I
am
ceo
and
co-founder
of
transform
data.
We
are
a
metrics
store,
I'm
sure
we'll
talk
more
about
what
that
means
and
the
inspiration
for
really
consolidating
much
of
the
work
on
data
and
infrastructure
at
airbnb
towards
metrics
and
how
that
actually
played
out
in
the
long
term,
the
good,
the
bad
and
the
ugly
of
those
decisions.
So
I
feel
like
I
also
helped
recruit
max
to
join
airbnb
and
I
feel
like
I
also
helped
get
swarup
signed
up
on
the
door.
E
E
I
I
remember
my
first
conversation
with
jha.
I
don't
know
if
you
remember,
but
I
was
like
being
recruited
out
of
insight
and
we
were
up
in
888
on
the
fourth
floor,
like
having
drinks
and
yeah.
Definitely
remember
that
conversation.
So
it's
interesting
right
in
terms
of
the
culture
that
we
all
became
a
part
of,
and
I'm
sure
which
we've
like
brought
to
our
own
companies
awesome.
B
So
one
thing
that
you
know
I
fondly
look
back
on
is,
of
course,
the
people
and
the
amazing
things
that
we
did,
but
the
actual
business
value
that
we
delivered
from
data
was
tremendous
when
you
actually
look
back
now,
meaning
in
the
middle
of
it
all.
Maybe
it
didn't
feel
like
that.
B
Even
so,
I
just
want
us
to
talk
through
these
layers
of
the
cake,
like,
let's
start
with
the
you
know,
the
foundation,
the
culture
and
the
and
the
the
people
and
what
was
needed
to
achieve
that
data,
democracy
and
just
I'll
just
randomly
start
lecture.
If
you
want
to
start
yeah.
C
Yeah
because
I
was
going
to
say,
I
actually
think
a
starting
point
is
the
company's
bones
right,
like
the
business
model,
travel
marketplace
right
that
has
a
lot
of
characteristics
that
lends
itself
to
certain
types
of
data
work.
Probably
a
lot
more
economics
minded
with
a
supply,
demand
concept
and
matching,
and
all
that
and
also
its
founders
founded
by
designers.
Right,
that's
novel.
I
I
think
that's
pretty
unusual
in
terms
of
the
you
know
we
weren't
just
building
stuff.
C
We
also
have
to
win
over
a
culture,
so
I
think
that
that's
kind
of
some
background
setting.
You
know,
I
think
the
data
teams
start
mostly
as
functions
of
people.
You
hire-
and
I
think
you
know,
as
we
all
know,
riley
was
the
first
hire
and
that
kind
of
had
a
lot
of
kind
of
cascading
effects
downstream.
To
me,
the
great
greatest
thing
about
that
is
that
riley
was
intensely
focused
on
impact.
C
You
know
he
might
not
like
have
like
run
large
data
in
for
teams
at
a
place
like
facebook
yet,
but
he
was
very
focused
on
like
actually
delivering
business
value
and
picking
up
stuff
along
the
way,
and
you
know,
even
as
we
fought
through
different
challenges.
I
think
that
you
know
james.
I
don't
know
if
he
was
the
one
who
recruited
you
but
like
I
think
he
like
found
the
right
people
to
kind
of
make
the
division
come
live
especially
right
around
that
kind
of
2013
2014
pivot
point.
C
D
I
mean
one
big
thing
too,
is
adding
the
resources
right
like
you,
need
the
cash
and
the
venture
and
the
capital
to
go
and
hire.
You
know
to
like
just
grow
the
scale
of
a
data
platform,
team
data
engineering
team
to
the
scale
that
we
had
to
so
key
ingredient.
There
is
like
success
of
airbnb
at
the
time
too,
and
the
capacity
to
go
and
hire
the
the
folks
from
a
bunch
of
places
and
give
them
like
pretty
loose
guidelines.
D
Like
you
know,
I
remember
a
fair
amount
of
freedom,
like
you
guys
know,
like
riley,
didn't
necessarily
know
very
much
exactly
what
we
had
to
build.
It
was
like
you
guys
know
better.
You
came
from
bigger,
greater
places
that
have
you
know,
you've
seen
it
before
you
see
in
the
future.
Bring
that
view
that
future
here
and
they
were
very
supportive.
D
I
think
of
open
source
too,
which
I
think
played
that
was
one
of
the
key
ingredients
there
in
terms
of
just
my
my
own,
like
drive
and
motivation,
came
largely
from
the
fact
that
the
impact
was
not
limited
to
airbnb.
It
was
just
like
limited
to
the
world.
Tusa
was
like
full
speed
like
very,
very
excited
to
come
and
work
on
open
source.
B
Lindsey,
I
think
you
kind
of
were
in
growth
at
a
very
interesting
time
when
that's
that
was
probably
the
most
data
intensive
parts
of
the
company
where
experimentation
was
huge
and
like.
Maybe
you
can
talk
us
through
the
culture
aspects
that
related
to
that
before
we
get
into
tooling
and
all
that.
E
Sure
yeah,
so
I
let's
see
I
got
hired
at
the
end
of
2014,
started
2015.
So
super
exciting
time
right
erf
was,
you
know,
had
been
kicked
off,
but
I
think
the
the
usage
of
it
was
was
like
kind
of
scattered
and
clustered
right
and.
E
Yeah
right
who
knew
right,
we
never
used
it.
I
mean
we
used
it.
Just
not.
We
never
used
all
the
words
yeah,
so
I
mean
god
it.
It
was
wild
at
the
time
I
so
this
is
my
first
industry
job.
I
came
straight
from
a
phd
in
politics.
Actually,
so
you
know
had
sort
of
exposure
to
the
econometric
side
of
things,
and
you
know
no.
No
one
told
me
about
erf.
I
didn't
know
that.
E
That's
what
I'd
be
you
know
able
to
to
an
airflow
at
the
same
time,
right
sort
of
separate
conversation,
but
you
know
we.
We
did
a
number
of
things
on
the
growth
team
that
I
think
really
like
jump-started,
how
experimentation
took
off.
So
we
did
fun.
Things
like
we
started,
like
experiment
review,
a
weekly
experiment,
review
on
on
the
growth
team.
Where
you
know
we'd
have
an
hour
an
hour
and
a
half
folks
would
sign
up
to
discuss
an
experiment
right.
E
It
always
had
to
be
really
hypothesis
driven
right,
they'd
lead
with
a
hypothesis
they'd
bring
screenshots
of
erf
they'd
talk
about,
you
know,
surprises
or
bugs
or
challenges,
and
you
know
they're
always
sort
of
like
action
items
like
you
know,
we're
gonna,
stop
this
experiment,
we're
gonna,
move
forward,
here's
how
we're
gonna
change
it
and
man
people
just
loved
it
right.
It
was
like
an
awesome
way
to
empower,
particularly,
I
think
young
engineers
to
you
know,
take
ownership
and
socialize
what
they
were
working
on
to
get
feedback
so
yeah.
E
That
was
you
know
that
was
my
first
year
at
the
company
and
it
was
just
so
exciting
to
see
the
enthusiasm
like
a
couple
months
in
the
experiment.
Review
sessions
would
just
be
booked
out
right,
like,
like
everyone,
wanted
to
talk
about
their
experiments,
and
I
think
you
know
that's
that's
a
sign
that
you
know
there
was.
There
was
so
much
potential,
so
much
support
and-
and
then
you
know
you
guys
can
talk
more
closely
about
this.
E
But
then
you
know
the
it
invited
interesting
challenges
of
scale
right,
like
you
know,
I
remember
like
losing
sleep
one
night,
because
you
used
to
be
able
to
take
down
erf
with
like
a
bad
metric.
You
know
the
whole
pipeline
would
stop
and
I
would
just
be
like
max.
I
think
I
did
it
again.
E
B
James,
I
want
to
get
to
like
the
the
tooling
and
metrics
layer
and
all
of
that
in
a
bit,
but
I
actually
wanted
to
talk
to
you
about
the
data
education
piece
which
you
drove
earlier
on,
which
is
actually
an
underappreciated
aspect
of
what
happened
at
airbnb
in
terms
of
the
multitude
of
stakeholders,
actually
using
all
the
amazing
tools
that
we
built.
You
want
to
talk
about
that.
A
bit.
F
Yeah
yeah
sure-
and
I
I
can't
take
all
the
credit
for
this-
I
actually
think
a
lot
more
credit
goes
to
riley
and
the
data
scientists
who
who
push
forward
a
lot
of
the
data,
education
and
jeff
fung
as
well.
Oh
yeah,
max
your
shirt
was
blurry,
but
it
showed
up
there
for
a
second.
You
know
my
whole
thing
was
yeah
aaron
kaufman
for
sure
huge,
huge
part
of
that
aaron.
F
She
was
a
natural
just
like
a
naturally
gifted
teacher,
which
I
think
made
that
made
that
even
better
for
folks
and
the
education
for
me
was
if
you
built
a
tool
me
as
a
product
manager-
and
I
know
I
pushed
everyone
on
this
on
this
call
in
different
ways
on
this,
and
I
always
came
back
to
like
if
you
built
a
tool,
are
people
actually
using
it?
How
often
are
they
using
it?
What
is
your
weekly
active
user
percentage
of
total
addressable
market?
You
know
internally
and
I've.
F
I
think
that
as
a
pm
in
the
technical
space,
it
was
really
my
job
to
be
trying
to
trying
to
get
everyone
into
that
user-oriented
mindset
and
that
would
absolutely
not
have
been
possible
without
the
education
efforts.
Just
handing
somebody
a
tool
that
is
an
interface
to
a
sql
database
right
is
not
enough.
Go
teach
them
how
to
use
it,
teach
them
how
to
export
teach
them
how
to
take
a
csv
file
and
turn
it
into
excel
and
make
it
useful.
F
F
This
is
our
tool
and
we're
going
to
use
it
we're
going
to
educate
each
other
on
how
to
use
it
and
we're
going
to
make
this
like
a
part
of
our
product
development
life
cycle
connecting
that
all
the
way
through
it
was
almost
like
the
cust.
You
know
in
it
now
that
I'm
learning
about
this
from
the
from
the
you
know
business
side.
This
is
almost
like
your
customer
success.
F
You
can
build
a
great
tool,
but
you
have
to
have
customer
success,
to
teach
and
train
and
educate
and
get
your
users
to
provide
feedback
and
to
improve
the
tools
and
have
a
virtuous
cycle
there
and
then
also
like
the
the
job
isn't
done
at
the
moment.
The
tool
is
created.
The
job
is
only
done
when
people
are
using
it
and
and
finding
true
value
in
it.
That's
that's
really.
What
I
learned
internally
from
that
that
I
then
brought
externally
in
into
my
new
goal.
I
bet.
D
You
guys
did
too
yeah.
I
think
a
big
thing
was,
you
know,
kind
of
second
shot
effect
or
like
third
or
four
shots.
I
think,
like
we
had
seen
good
from
different
places
before
so
people
coming
from
linkedin
facebook
other
places
in
the
valley.
I'd
seen,
you
know,
tools
that
have
been
successful
in
those
places,
and
then
we
took
what
what
was
best
from
these
ideas
and
then
reinvented
those
in
a
better
way
right,
maybe
the
first
experiment.
D
The
second
experiment
is
not
always
like
the
perfect
tool,
but
then
we
all
had
a
shot
to
do
it
one
one
more
time,
and
I
think
that
was
a
big
thing.
One
thing
I
wanted
to
mention
too,
that
was
like
magical
things
happening
to
on
the
culture
side
like
first
there
was
cookies
every
day,
there's
a
little
card
coming
with
fresh,
baked
cookies
that
were
baked
by
the
culinary
team.
I
think
it
was
like,
like
3
p.m.
D
Every
day,
little
happy
joy,
joy,
cart
of
cookies,
that
that
helps
them
the
morale
of
the
troops,
and
then
I
remember
this
thing
and
I've
never
seen
anything
like
that.
In
other
work
environments
like
before,
after
that,
people
would
just
like
suddenly
start
cheering
and
clapping
for
unknown
reasons,
and
it
would
go
like
it
would
storm
around
the
office.
The
office
was
gigantic
and
you
know
kind
of
open
space
but
kind
of
a
big
like
square
with
a
hole
in
the
middle.
D
But
then
you
would
just
hear
this
like
cheer
roaring,
like
kind
of
coming
from
another
department.
Maybe
a
team
was
celebrating
something
a
birthday
of
product
achievement.
It
was
like
who
knows
what
it
was
just
like.
Go
like
a
wave
through
the
office
and
people
would
stand
up
and
start
clapping,
and
then
you
know,
go
back
to
work.
B
D
Know
yeah:
have
you
seen
that
anywhere
else?
I
haven't
seen
that
that
was
great,
but
there
was
like
a
lot
of
cool
stuff
happening.
It
just
felt
great
to
be
there
and
you'd.
Wake
up
like
excited,
go
to
work
and
work
on
things
and
then
with
really
good.
Like
co-workers,
smart
people
like
driven
it
was
awesome
times.
E
Yeah
and
one
thing
I
wanted
to
just
add,
there
sorry
was-
and
I
know
jay
you
can
you
know
I'm
interested
to
hear
your
thoughts
on
this,
but,
like
you
know
how
much
we
were
all
learning
together
right,
like
okay,
so
when
erf
launched
it
was
oh,
we
can
do
two-tailed
tests
and
then
you
know
eventually
we
were
doing.
E
I
never
know
what
to
call
it
cupid
right,
which
you
know
very
introduction
techniques,
and
we
were
doing
like
you
know
in
erf,
dimensional
splits
of
our
data
right
to
see
breakdowns
in
certain
regions
or
certain
devices.
You
know.
Initially,
we
had
to
write
these
scripts
and
r
and
do
this
like
ugly
stuff
right
to
report
to
our
product
teams.
E
You
know
I
worked
on
the
china
team.
So
what
was
going
on
in
china
right
like
we
didn't,
have
a
cut
for
country
level.
We
didn't
have
a
cut
for
device
level,
and
you
know
it
was.
It
was
exciting
at
the
time
a
little
frustrating
right,
because
folks
would
be
like
what's
a
p-value
right,
you
know
to
walk
them
through
and
you
know
people
were
peaking
and
like
announcing
victory
on
things
and
like
it.
E
B
I
want
to
talk
about
a
few
more
concrete
things
as
we
move
up
the
cake
there.
First,
of
course
we
built
a
very
scalable
platform.
You
know,
blog
posts
have
been
written
about
it
and
streaming
and
all
that
good
stuff
happened.
You
know
we
we
actually
built
a
really
solid
foundation,
and
then
we,
you
know,
built
amazing
products
like
airflow,
superset,
experimentation,
minerva,
and
I
want
us
to
talk
about
some
of
these
tools
and
I'm
going
to
obviously
pick
on
the
respective
people
for
those
and
lindsay.
B
I
want
us
to
discuss
a
little
bit
about,
like
you,
know,
usage
of
jio
data
in
pricing
and
ranking,
and
so
on,
so
we'll
get
to
that
in
a
bit,
but
obviously
che.
Please
talk
us
through
all
the
amazing
things
we
did
with
experimentation.
B
C
All
perfect,
and
especially
on
the
time
scale,
you
know,
companies
founded
what
2008
riley
hired
2010
I
joined,
2012
like
erf,
can
beat
what
2014
or
something
right.
So
there's
a
lot
of
years
to
get
there,
and
by
the
time
we
got
to
erf,
we
had
basically
already
won
a
cultural
battle
to
some
extent,
if
we
like
have
justified
a
whole
data,
org,
like
I
think.
C
C
Person
data
teams
out
there
like
who
got
started
because
of
basic
reporting
needs
and
are
trying
to
figure
out
how
to
get
this
developed
sophistication
and
without
regard
I
feel
like
and
under
until
30
james.
I
know
you
remember
this
part
of
the
reason
we
built
urf
is
because
an
executive
from
booking.com
came
by
airbnb
and
like
on
with
like
a
sort
of
nba
thing
that
made
him
be
careful
of
language
and
basically
in
front
of
joebot
in
front
of
the
whole
product
team
was
like
hypothetically.
C
If
I
was
in
charge
of
airbnb,
I
would
build
an
incredibly
sophisticated,
powerful
experimentation
framework
and
that's
what
I
would
do
hypothetically
or
whatever
right,
and
that
was
like
the
big
political
bias
to
go,
invest
a
ton
here,
in
addition
to
like
yan
and
the
search
and
marketplace,
jan
and
bar
and
search
marketplace
team
having
some
really
success.
So
I
feel
like
the
I,
when
I
talk
to
other
data
leaders,
just
really
on
a
like.
How
do
you
operate
a
data
leader?
It's
like
you
have
to
deliver.
C
C
So
you
know
what
I'll
say
is
what
I
think
we
did
really
well
at
airbnb
was
one
hire
really
great
communicators
early
on,
because
if
your
infrastructure
is
really
bad
early
on,
it's
good
that
you
can
just
have
someone
that
you
really
respect
that
can
communicate
you
with
you
really
well
and
kind
of
be
a
trusted
ally
even
amidst
some
infra
issues.
C
That
was
a
true
before
after
of
the
team
and
like
just
had
really
entrepreneurial
people,
because
the
first
five
ten
data
scientists
like
we
didn't,
we
weren't,
directed
to
what,
where
the
the
oil
in
the
ground
was
right,
like
you
have
to
go
out
and
find
it
get
a
bunch
of
wins,
and
then
you
get
to
hire
like
data
infra
teams
and
build
your
ethanol,
some
stuff.
C
B
Cool,
let's
keep
moving
so
max
talk
us
through
the
genesis
of
I
mean
now
famous
tools
but
like
back
in
the
day,
I
remember
four
of
us
standing
and
deciding
on
the
name
of
airflow,
workflow
and
airbnb
should
be
called
airflow.
That's
right!
That's
how
I
think
we
ended
up,
but
let's
talk
us
through
how
it's
grown
beyond
your
wildest
yeah
imagination.
D
Yeah,
it's
like
when
I
joined
too,
like
I'm
trying
to
remember.
Like
my
impressions
when
I
first
joined
it
was
very
small.
It
was
fairly
small
team,
so
data
platform
was
pretty
small
and
then
data
engineering
was
a
handful
of
people.
It
was
not
called
data
engineering,
they
called
themselves
athleians
etl
ends,
and
then
they
were
working
on
something
they
called
core
data
which
was
like
you
know
this
effort
around
like.
Oh,
let's,
you
know
organize
our
core
data
sets
in
one
place
so
that
we
can
all
agree
on.
D
You
know
metrics
and
dimensions,
and
it
was
like
some
some
some
star
schema
type
approach
to
data
engineering
and
what
I
remember
is
coming
from
facebook.
I
had
just
thought
that,
like
the
data
was
so
small,
I
was
like.
Oh
my
god.
This
is
like
I
felt
like
I
had
the
wings.
You
know
like
yeah,
I've
been
training,
you
know
a
really
tough,
like
high
gravity,
environment
and
now
you're
in
a
zero
gravity
environment.
D
We
could
rebuild
the
entire
warehouse
in
like
a
few
hours
and
you
know,
and
the
platform
was
solid
for
the
amount
of
data
that
we
had
because
airbnb
is
like
we
don't
there's
not
that
many
bookings
and
the
traffic
on
the
website's,
just
not
that
crazy,
compare
at
least
like
compared
to
facebook
coming
from
facebook.
D
But
then
there
was
not
a
lot
of
glue
and
stuff
was
not
working
super
well,
you
know
when
I
joined
so
like
one
thing.
As
a
data
engineer,
we
had
something
called
chronos
that
was
a
scheduler
on
top
of
mesos
and
I
was
kind
of
hard
to
use
and
that's
coming
from
from
facebook.
I
was
a
big
user
of
data
swarm.
That
was
an
internal
tool
that
you
know
I
used
and
had
contributed
to
and
kind
of
pushed
its
boundaries
too.
D
D
It
was
not
going
to
be
like
I
didn't
get
hired
to
work
on
airflow,
but
it
was
going
to
be
okay
for
me
to
work
on
airflow,
and
then
I
started
working
on
it
in
between
jobs
and
put
it
on
my
own
open
source
like
my
own
github,
so
no
one
could
kind
of
question
the
decision
I
wouldn't
have
to
jump
through.
It's
like.
D
I
already
started
already
it's
already
kind
of
working,
so
I
came
in
and
I
already
had
a
prototype
after,
like
a
two
weeks
break
where
I've
been
working
on
the
thing
and-
and
then
I
worked
very
closely
with
like
the
adoption-
was
really
easy,
because
people
there
was
some
pain
with
the
current
system,
but
also
like
people
wanted
this
renewal,
which
had
this
mindset
of
like
hey,
we're,
building
everything
new
and
I'm
gonna
use
your
thing
use
my
thing
and
surely
we
can
do
that,
we're
a
little
delusional
in
a
really
good
way
too
right.
D
I
think
you
need
that
the
history
of
invention
is
paved
with
people
who
are
delusional
right,
like
you
need
to
be
crazy
enough
to
think
you
can
do
better,
then,
whatever
exists
or
whatever
doesn't
exist.
So
you
know
so
that
was
kind
of
story.
The
story
of
airflow
is
like
very
quickly
like
within
months.
D
We
had
things
in
production
and
we
used
it
for
core
data
and
then
course
cs,
which
is
the
customer
success
stuff
and
then
really
just
like
really
good
adoption,
and
then
tableau
was
going
to
pay
in
the
butt
because
we
didn't
have
a
presto
driver
for
it.
It
was
really
hard
to
like
we.
I
made
a
huge
investment
in
presto
and
hive
and
we're
making
an
investment
in
druid,
and
we
just
couldn't
connect
tableau
directly
to
it.
D
Live
mode,
didn't
work
and
then
those
extracts
is
just
like
a
really
shitty
database,
or
at
least
it
was
at
the
time
so
so
so
decided
to
start
building
superset
as
something
that
would
connect
directly
and
natively
to
these
databases
and
just
be
lightweight,
and
also
like
largely
inspired
by
similar
tools
that
had
been
built
at
facebook.
Where
I
came
from,
it
was
something
called
uni
dash
or
something
actually
james
was
directly.
F
D
Argus
argus
high
pal
and
then
uni
dash
was
kind
of
taking
shape,
as
I
left
as
I
left
facebook,
but
like
this
delusion,
it
didn't
seem
that
delusional
to
build
something
like
that,
because
we
were
coming
from
a
place
where
things
like
that
got
built
every
day.
Like
multiple
of
these
things
got
built,
you
know
every
quarters
and
then
some
things
would
emerge.
My
dogs
are
barking.
It's
probably
the
sign
for
me
to
stop
talking
and
pass
it
over
to
the
next
person,
but
like
yeah,
these
things
like
took
off
internally.
D
There
was
such
a
hunger
for
data.
I
think,
based
on
the
people
we
hired
and
the
sponsorship
and
the
executive
sponsorship,
the
education
program
that
these
these
tools
were
just
like
taken
off
internally,
like
adoption,
seemed
like
a
breeze,
though
we
did.
I
remember
I
would
give
sessions
on
air
like
we
had
data
universities
that
every
two
weeks
or
every
month
we
had
a
cohort
of
new
employees
getting
trained
at
data
university.
D
So
I
remember
doing
a
lot
of
teaching.
I
remember
doing
a
lot
of
like
data
doctors
too,
so
kind
of
office
hours
where
you
have
like.
We
all
put
like
white
gowns
like
we're
working
a
hospital,
and
then
it
would
be
like
little
signs
on
the
table.
Here's
the
airflow
table
the
superset
table
the
erf
table
and
we'd
all
go
there
and
kind
of
help.
Each
other
and
people
would
come
and
visit
with
our
use
case
and
their
questions
and
we'd
unblock
them.
I
thought
that
was
super
great.
B
Awesome
so
there
were
a
lot
of
great
things
about
airbnb
culture,
but
one
thing
I
did
not
like
is
the
engineering
promotion
culture
and
I
can't
believe
how
long
it
took
me
to
get
max
promoted
to
the
highest
levels.
In
spite
of
all
this
amazing
impact,
I
just
hated
that
part
about
airbnb
engineering
culture,
but
anyway,
moving
on
lindsay,
you
saw
a
lot
of
first-hand.
You
know
use
of
high
value,
jio
data
being
used
very
strategically
in
pricing
and
ranking.
B
E
E
I
mean
the
this
is
what
really
pushed
me
to
to
found.
My
company
was,
like
you
know
the
amount
of
energy
well,
taking
a
step
back
right,
like
airbnb
sort
of
native
data
models
were
rooted
in
in
the
listing
right
like
which
made
sense
it's
what
it's
the
data
that
the
hosts
have.
You
know
that
that
that
no
one
really
had
access
to,
and
I
don't
think
anyone
at
airbnb
knew
about
the
mls
like
the
multiple
listing
service
which
which
is
used
in
real
estate.
E
You
know
so
you
know
we
had
all
this
great,
like
user
generated
content
and
buried
in
reviews
and
photos
and
listing
descriptions
was
like
a
lot
of
great
information
about.
You
know
not
just
listings
themselves,
but
like
the
context
of
of
the
listing-
and
you
know
there
were,
there-
are
various
projects.
I
worked
on
to
like
kind
of
surface
that
data
right
like
if
I
wanted
to
run
a
marketing
campaign
on
the
growth
team
to
all
of
the
hosts
who
had
airbnbs
on
the
beach
like
that,
didn't
exist
right.
E
We
had
and
we
had
user
generated
content.
So
we
could
like
kind
of
dig
it
out,
but
you
know
massive.
I
don't
know
I
kind
of
saw
it
as
a
massive
problem
and
opportunity,
and
you
know
when
I
left
airbnb
as
we
all
did
right.
The
first
thing
is
like
well
is
this
idea
crazy?
You
know
got
to
do
my
my
customer
discovery
and
see
like.
Is
this
a
a
problem
and
an
opportunity
that
exists
elsewhere?
And
you
know
indeed
it
is
right.
E
It
took
five
minutes
into
conversations
with
you
know,
folks,
at
zillow
and
trulia
and
redfin
and
booking.com,
and
you
know
all
these
great
companies
and
travel
and
real
estate
and
realizing
that,
like
you,
know,
sure,
every
once
in
a
while,
you
get
entrepreneurial
folks
who
are,
who
are
you
know,
kind
of
unearthing,
these,
these
insights
and
geospatial
data,
but
the
the
amount
of
work
it
takes
is.
Is
you
know
it's?
E
It's
excruciating
right,
like
sourcing
like
sourcing
this
data
independently
right,
we
had
some
in
reviews,
but
again
it
was
selected
data,
so
we
didn't
have
the
coverage
that
that
we
needed
and
and
yeah
it's
a
really
common
problem
and
and,
as
I
keep
saying
like,
you
know,
a
pretty
interesting
opportunity,
I
think,
for
you
know
anyone
in
in
real
estate
and
travel
to
leverage
more
outside
data
sources.
I
think
the
irony
of
so
many
data
teams
is
that
they're
so
focused
on
this.
E
Maybe
it's
not
an
irony
but
like
they're,
so
focused
on
their
data,
their
reporting
right
about
their
product
about
their
revenue,
and
all
of
that
makes
sense.
But
I
think,
particularly
if
you're
in
you
know
the
kind
of
business
where
location
matters
which
for
like
real
estate
and
travel
like
location,
is
number
one
right
there.
E
There
aren't
necessarily
good
tools
that
exists
to
serve
those
use
cases
and
so
yeah
I
mean,
I
think
it's
I
don't
know
if
you
all
have
been
following,
but
andrew
from
you
know,
former
coursera,
google
et
cetera,
he's
on
this
big
kick
now
for
like
data
centric,
ai
right.
It's
like.
E
Let's
move
away
from
the
hyper,
like
all
the
tuning
we've
gotten,
that
taken
care
of
right,
like
you
can
figure
that
out,
but
let's
actually
return
to
like
data
quality,
and
I
mean
I'm
sure,
that's
something
we
all
have
opinions
on,
but
that's
one
way
that
I
think
about
the
work
that
we're
doing
with
iggy
is,
like
you
know,
just
making
more
data
accessible
to
more
product
teams
and
ml
teams,
because
you
know
airbnb
is
doing
great
things
with
with
their
category
search.
But
you
know
the
ability
to
do.
D
The
state
of
open
data
is
not
what
it
should
be
right
like
if
you
want
to
like
run
a
query
of
like
population
per
zip
code
somewhere,
maybe
now
on
like
open
data
sets
on
that,
bigquery
is
putting
forward
like
maybe
some
of
the
super
basic
like
demographic
stuff.
Is
there,
but
it's
crazy.
D
How
open
source
code
is
so
advanced,
like
you
think,
of
the
number
of
like
libraries
and
products
and
frameworks
that
exist,
and
then
you
think
about
like
how
hard
it
is
to
just
get
like
population
by
city,
or,
I
don't
know
just
weather
data
like
through
not
api
but
like
through,
like
scans
too.
So
like
that,
you
know
we.
I
called
it
the
denominator
problem
to
you
on,
like
geospatials,
like
lift
two.
We
had
this
issue
of
like
hey.
D
E
And
you
hit,
I
think
you
hit
the
nail
on
the
head
in
that
there's
a
lot
of
like-
and
this
is
what
I
did
at
airbnb-
it's
a
lot
of
like
one-off,
analytics
stuff
right
like
that
that
you
can
do,
but
actually
also
bring
this
data
into
production.
It's
a
whole
different
challenge.
Right,
like
that
to
me,
was
one
of
the
the
best
things
about
working
at
airbnb
was
exposure
to
production
systems
right.
So
I
think
that
there,
I
think,
there's
a
lot
of
opportunity
there.
It's
not
like
it's
not
just
access
to
the
data.
E
It's
you
know,
because
we
give
away
a
lot
of
data
for
free,
but
once
you
have
it
like,
how
do
you
actually
get
it
to
speak
to
your
bi
tools
right?
How
do
you
get
it
to
speak?
To
you
know
your
ml
system,
so
you
can
actually
take
it.
You
know
into
account
when
you're
building,
so
lots
of
opportunity
left.
B
One
thing
I
wanted
to
discuss
is
we
had
this
air
slash
data
thing
data
portal,
you
know
which
really
democratized
everything
in
terms
of
how
anyone
can
access
anything.
We
had
experimentation
framework,
but
I
think
what
really
brought
all
this
alive
in
terms
of
people
discovering
business
critical
data
isn't
like
the
metrics.
B
You
know
layer
that
came
about
the
the
minerva
thing.
You
know
I
mean
obviously
lots
of
learnings
from
other
companies
in
terms
of
how
they
had
done
it,
but
I
felt
like
data
portal
came
alive
after
matrix
was
introduced
in
it
and
then
experimentation
also
became
a
lot
more
reliable
after
you
had
you
know
a
consistent
layer,
and
you
know
superset
also
became
a
lot
more.
You
know
exciting,
like
people
can
just
spin
up
like
various
cuts,
so
james,
maybe
you
can
talk
us
through
the
evolution
of
that
you
know.
F
F
There's
a
lot
of
intelligence
and
learning
in
in
this
room-
and
I
was
I
was
just
thinking
through
before
you
prompted
this
question.
What
is
what
is
the
practical
advice
we
can
offer
to
someone
they've
created
some?
They
work
at
some
company
that
company
has
created
some
data
assets.
C
Yeah,
I
think
I
I
love
this
line
of
question
and
just
as
one
baseline
for
our
starting
point
like
we're
living
in
a
dbt
world
and
snowflake
like
that,
so
people
don't
have
to
figure
out
the
high
provisioning
whatever
right.
So
you
got
a
cluster,
it's
elastic,
you're,
modeling,
stuff,
you're,
probably
delivering
poor
bi
reports.
C
B
I
I
want
to
get
to
that
in
a
bit
like,
but
just
want
to
complete
that
airbnb
story
a
little
bit
in
terms
of
you
know,
rounding
out
the
picture,
and
then
I
want
to
get
to
the
you
know.
So.
James
okay,
yeah.
F
Yeah
sure
sure
that
that's
totally
fair
so
where
we
got
to
was
we
put
together,
we
put
together
some
core
pieces
of
infrastructure.
We
got
to
a
place
where
experimentation
was
a
big
deal
and
people
had
really
rallied
around
metrics
for
experimentation.
F
Similar
story
at
facebook
before
that
metrics
were
incredibly
useful
for
experimentation
for
having
one
lens
onto
the
success
and
failure
of
an
experiment,
then
from
there
what
else
are
metrics
useful
for
stop
having
to
define
them
two
separate
places
or
three
separate
places,
bring
them
all
together.
So
the
same
metrics
you're
using
for
experimentation
are
the
same
metrics
that
you're
using
for
validating
your
business
performance
are
the
same
metrics
you're
using
for
you
know
your
dashboards
and
superset
are
the
same
metrics
you're
using
for
our
forays
into
anomaly.
F
Detection
are
the
same
metrics
that
you're
using
for
executive
reporting
and
daily
flash
emails.
Creating
consistency
in
that
metrics
layer,
I
think,
was
a
late
stage,
evolution
that
we
witnessed
as
the
logical
conclusion
to
many
of
these
other
distributed
efforts,
and
I
think
that
was
an
important
story.
Then,
right
to
your
point,
how
do
you
show
that
to
people?
The
data
portal
was
incredibly
useful,
but
it
got
way
more
useful
when
metrics
came
to
be
because
most
folks
do
not
think
of
data
in
terms
of
tables.
F
They
do
not
think
of
core
data
bookings
table
and
what
is
the
associated
metadata
for
that
table?
They
just
want
to
know
how
many
bookings
happened.
How
many
bookings
happen
in
china,
and
that
is
the
core
question
that
I
think
data
portal
ran
into
and
you
can
speak
to
that
even
much
better
than
I
can.
B
I
want
to
be
conscious
of
the
time,
so
let's
keep
moving
and
bring
up
the
question
that
che
brought
up
and
we
can
chat
all
day
about.
You
know
data
portal,
james,
so
so
now,
let's
switch
to
the
the
modern
data
stack
or
post
modern
data
stack
or
whatever
you
want
to
call
it.
So
you
know
you
know
that
we
didn't
get
everything
right
airbnb.
B
You
know
we
kind
of
evolved
into
a
decentralized
structure.
Almost
by
force
like
we
ran
into
multiple
business
units.
The
technology
stack
consolidated.
I
think
that's
still
true.
It's
there's
a
little
bit
of
fragmentation
and
this
little
bit
of
consolidation.
I
feel
like
that's
always
going
to
be
the
case
because
can't
stop
innovation.
B
You
know,
we've
seen
this
with
oracles
of
the
world
and
whatnot.
So
the
the
key
question
is
you
just
have
to
embrace
decentralization
and
there's
a
there's,
a
starter
pack
for
data
teams,
if
you're
just
starting
out
and
then
there
are
like
big
companies
which
are
grappling
with
huge
amounts
of
fragmentation.
So
I
want
to
tackle
like
a
few
questions
related
to
that.
So
first
thing
is:
you
know,
ownership?
B
Let's
stop
there.
I
think
we
ran
into
huge
issues
there
and-
and
you
know,
change,
why
don't
you
talk
about
it?
I
think
you
have
a
position
for
the
starter
pack
as
it
were
or,
and
there
are
like
bigger
company
problems
too
yeah.
E
C
Good
tools
out
there
and
you
know
now,
there's
a
lot
of
tools
out
there,
but
yeah
just
being
able
to
spin
up
basic
reporting
with
a
small
skeleton
group
is,
I
think
you
know
the
default
mode
out
there,
but
I
think
the
the
thing
that
I
find
myself
taking
a
lot
of
advice
on
is
like
yeah
once
you
spin
up
a
basic
bi
dashboard
like
how
do
you
get
people
thinking
in
terms
of
data
and
metrics,
which
gets
into
gets
into
both
the
infrared,
invest
in
and
organizationally
how
your
set
yourself
up?
C
I
mean
I'm
wondering
what
you
all
think,
I'm
a
big
believer
in
the
kind
of
embedded,
centralized
reporting
thing
so,
like
embedded
on
teams
centralized,
you
know,
reporting
structures.
I
think
the
one
one
of
the
things
I
I
think
is
like
a
very
underrated
is
having
all
data
functions
report
through
one
data
org.
We
almost
have
that
at
airbnb.
C
You
know
we
had
a
lot
of
data
functions
report
through
it,
but
we
did
not
have
like
data
infra
report
through
it
and
we
did
not
have
like
core
ml
necessarily
go
through
the
data
org
and
so
those
cleavages
kind
of
you
know
never,
I
think,
led
to
a
lot
of
downstream
consequences
and
I
think
there
are
other
teams
of
like
way
worse.
C
Organizational
setups,
which
is
like
all
the
bi
and
analytics,
is
a
wholly
separate
place
than
like
analytics
engineering
and
data
and
for
whatever
and
that
stuff
really
matters,
and
so
I
would
say,
start
off
there
and
then
two
is
like
make
your
early
hires,
like
essentially
entrepreneurial
product
people
or
just
heavily
indexed
on
whatever
mechanism
you're
trying
to
maximize
maybe
they're
marketing
people
whatever.
C
But
the
point
is
that,
like
they
need
to
be
almost
as
much
practitioner
as
just
data
specialists
early
on
just
to
like
get
wins,
understand
the
business
make
the
right
choices
and
then
that
kind
of
gets
you
the
political
will,
the
investment
everything
else
to
then
get
more
sophisticated
stuff.
B
So
agree
with
that,
I
think
there's
one
painful
learning
that
I
have.
I
think
you
know
bunch
of
us
went
through
that
too.
I
think
you
know
core
data
like
the
core
foundational
tables
got
splintered
and
kind
of
moved
into
various
business
units
and
people
these
days
call
it
data
mesh
and
data
products.
B
What
not
so,
but
you
know
we
try
to
resist
that,
but
it
was
just
the
force
was
too
much
and
ultimately
you
know
the
business
units
ended
up
owning
what
was
closer
to
their
logic,
but
we
still
needed
some
central
core
data
tables
I
feel
like
as
an
industry.
We
haven't
really
solved
that
problem.
It's
all
it's
all
nice
to
say
that
domains
should
own
their.
B
You
know
their
data
products,
but
you
know
what
happens
to
certain
key
dimensions
that
matter
to
all
business
units
who
takes
care
of
backfilling
them.
Who,
like
you
know
these
problems,
don't
magically
go
away.
So
I'm
particularly
interested
in
hearing
max
and
james,
like
you
and,
of
course,
lindsay
too
so,
like
is
this
a
salt
problem
today
or
is
it?
Is
this?
Are
people
just
papering
over
this
painful
problem.
D
I
mean
it's
a
big
company
problem
right
like
the
the
mesh,
and
I
I'm
not
a
big
proponent,
of
like
the
data
mesh.
You
know
paper
or
philosophy,
like
I
think,
there's
some
really
good
ideas
and
some
good
problem
statements
in
there
too.
For
me
like
working
at
a
much
smaller
company.
D
Now,
I
would
say
like
if
you
can
centralize,
like
the
data
asset
creation,
into
like
a
kind
of
a
crafty,
small
and
mighty
team
that
can
manage
this
stuff
for
a
while,
like
push
that
do
that
as
long
as
you
can,
until
the
forces
of
like
decentralization
becomes
stronger.
But
one
thing
is
for
there's
a
lot
of
learnings
in
the
data
world
from
the
software
engineering
world
and
like
the
micro,
the
micro
service
idea
does
not
translate
well
into
the
data
world,
because
stuff
is
like
much
more
much.
D
There's
much
more
gravitational
pull
to
bring
everything
together.
Everything
needs
to
be
joined
and
union
together.
So
it's
much
harder
to
say,
like
hey,
we're,
gonna
carve
out
a
data
product
or
a
service
or
a
microservice,
and
then
you
know
take
it
out.
So
I
think
it's
that
part
like
decentralizing
or
meshing.
Data
is
extremely
challenging
and
you
need
to
have
like
distinction
between
the
centralization
right,
like
people.
Work
who
have
the
same
skills
want
to
work
together
and
be
centralized,
but
then
the
the
product
teams
want
to
pull
out.
D
One
thing:
that's
kind
of
a
different
question
I
wanted
to
explore
is
like
what,
if,
in
2014
when
I
joined
airbnb
or
like
right
around
the
time
where
we
invested
big
in
data?
What
if
we
had
everything
we
had
to,
we
have
today
what
if
we
had
snowflake
bigquery?
What
if
we
had?
You
know
airflow
and
dbt
like
laying
around
what,
if
we
could,
you
know
just
become
customers
of
preset
transform
and
all
the
companies
that
acro
like
other
companies
who
are
what
would
we
have
done?
Then
right?
D
Is
it
really
interesting
questions
and
I
don't
think
we
would
have
invested
as
much
and
did
a
platform
that
would
have
been
just
like
you
know
things
we
would
have
purchased
and
then
I
think
we
would
have
gone
through
product
teams
to
help
them
like
the
data
people
like
I
would
have
gone.
D
I
would
have
picked
a
team
like
the
bookings
team
or
the
you
know
like
pick
a
vertical
team
and
get
closer
to
them
and
do
some
data
work
with
them
and
for
them
with
like
driving
business
outcomes,
as
opposed
to
like
working
on
centralizing
infrastructure
pieces
to
support
the
rest
of
the
company.
But
it's
it's
kind
of
interesting
question
like
what
would
we
do
if
we
had
all
these
tools
around?
You
know
I'll.
B
Tell
you
I'll
tell
you
something
that
happened
after
all.
You
folks
left,
and
I
was
the
last
remaining
person
at
airbnb.
I
I
lived
through.
You
know,
pre-ipo
readiness
very
close
to
it
and
I
lived
through
overweight
hitting
the
company.
B
You
know
the
problems
of
getting
ready
for
compliance
gdpr
and
when
you
know
it
happened,
the
amount
of
attention
that
cost
needed.
You
know
these
problems
were
not
magically
going
away.
If
you
purchase
all
the
tools
that
you
know,
you
still
need
approaches
to
automate
these
things
at
scale
and
still
not
you
know,
burden
the
day-to-day
practitioners
of
data.
I
think
that's,
that's.
B
The
nature
of
the
platform
keeps
changing
it's
not
about
streaming
and
scale
or
whatnot,
but
you
always
need
something
which
evolves
and
you
know,
cuts
across
all
the
tools
and
it
gives
you
these
horizontal
superpowers
so
that
the
day-to-day
practitioners,
you
know
if
you
burden
everyone
with
this-
is
how
you
control
cloud
costs.
It's
not
going
to
happen.
So
that
was
my
learning
anyway.
Let
me
not.
I
want
to
get
back
to
the
the
core
data
learning
from
james.
I
think
you
have
some
insights
there.
F
I'm
wondering
about
how
many
fewer
gray
hairs
would
have
popped
out
of
my
head.
There's
an
interesting
corollary
there
between
what
was
it
like
for
airbnb
to
start
running
their
business
on
aws
instead
of
having
to
hire
a
bunch
of
infra
people
to
like
run
my
sequel,
sweaty
warehouse
farms.
I
think
there's
a
there's
like
this
new
wave
of
what
problems
got
solved
and
then
to
syrup's
point.
What
problems
still
remain,
what
are
the
low
value
problems
to
stall?
F
I
don't
see
hardly
any
new
companies,
spinning
up
and
being
like
hey
we're
going
with
a
on-prem
hadoop
stack
like
I
just
don't
see
that
right,
it's
all
snowflake,
it's
all
bigquery,
it's
all!
Even
microsoft
data
azure,
you
know
data
warehouse
kind
of
stuff.
Even
that
stuff
feels
like
it's
it's
there,
but
the
problems
of
that
last
mile
will
always
be
there.
How
do
you
make
experimentation
actually
useful
for
my
experiments?
F
F
How
do
you
take
geospatial
data
and
make
it
useful
for
the
questions
I
need
to
ask
and
the
models
I
need
to
build,
and
I
think
that
going
off
of
max's
question
is,
if
you
can
abstract
away
those
hard
infrastructure
and
tooling
challenges.
You
are
left
with
that
last
mile
of
connecting
the
business
problems
to
the
people
who
are
trying
to
solve
those
problems,
and
I
think
that's
like
a
much
more
interesting
place
to
be
than
like
tuning
a
hadoop
cluster.
I
don't
know
what
do
you
all
think,
but.
D
Building
tools,
building
those
are
super
fun,
though
right
like
that,
or
at
least
that's
my
thing
like
I
don't
know
what
it
used
to.
I
want
to
build
tools.
Maybe
then
it's
like
you
join
like
a
company
like
one
of
ours
right
like
if
you're
a
tool
builder,
you
join
the
places
where
the
tools
are
made,
but
practitioner
is
also
really
cool.
You
get
to
solve
a
real
business
problem
with
people.
C
B
Just
to
be
clear,
I
want
to
like
temper
the
enthusiasm
a
little
bit.
I
think
we
used
to
worry
about
shuffles
and
whatnot
in
hadoop.
I
think
now
you
have
to
look
at
spill
percentage
on
snowflake
and
see
if
you
have
to
move
it
to
an
extra
large
warehouse.
Those
problems
don't
go
away.
Man,
I
think
just
you
know,
as
as
you
do
more
demanding
things
they
just
don't
go
away
anyway,
back
to
back
to
learnings
in
the
modern
data.
B
Stack
lindsay
just
want
to
kind
of
take,
get
your
take
on
it
too,
like
if
someone
has
a
starter
pack,
as
it
were,
what
are
your
learnings
in
terms
of
what
we
should
be
doing
as
an
industry.
E
I
mean
honestly,
I'm
so
on
board
with,
what's
already
been
said
right.
It's
such
a
provocative
question
and
you
know
it's
sort
of
like
the
hierarchy
of
needs
right,
like
the
the
the
needs
that
are
served
now
are
more
easily
served
they're,
not
without
challenge,
as
you
note,
but
yeah
I'm
a
big
fan
of
of
I
mean
it's
like
what
you
know.
One
of
our
value
props
as
as
a
company
is
like
you
know,
and
many
of
the
like
dev
tools,
spaces
right.
E
E
I
don't
wanna,
it's
not
creativity
per
se,
but
I
think
there
is
like
there's
a
lot
of
magic
in
like
good
analysts
and
good
data
scientists,
and,
like
I'm
very
intrigued
at
about
like
how
do
you
find
those
people?
How
do
you
grow
those
people
right?
The
people
who
who
make
you
know
whether
they
have
magical
insights
or
they
they
run
experiments
that,
like
change
change,
the
surface
of
your
product,
right,
like
that
to
me,
is
like
mds,
doesn't
really
solve
that
at
all.
E
Right
that
there's
we
joke
at
my
director
of
data
science,
and
I
joke
about
how
like
oh
man,
you
know.
Sometimes
our
customers
are
not
as
creative
as
they
as
we
want
them
to
be
right,
because
they're
they're
worrying
about
a
million
things.
But,
like
you
know,
that's
the
side
of
data.
I've
always
been
really
really
like
intrigued
by
and
really
excited
about,
which
you
know,
hopefully
we're
facilitating,
but
it's
that
that
that,
like
very
creative
side
of
it,
so
I.
D
D
If
we
had
like
really
good,
you
know,
training
no
training
program
but
like
at
least
you
know,
training
people
on
the
modern
data
stack
on
on
the
practices
that
people
use
today
on
delivering
business
value,
joining
a
company
and,
being
you
know,
just
a
powerful
person
with
like
analytical
skills
and
tools
like
that
would
be
really
great.
For
you
know
the
industry
in
general.
E
Well-
and
I
think
you
know,
there's
a
rise
in
sorry,
one
more
thing
here:
this
is
a
rise.
I
don't
know
if
you
all
have
noticed
in
this
role
of
the
product
engineer
which
is
sort
of
what
che
was
talking
about
before
right
these.
It's
like
it's
not
like
a
back-end
front
end.
It's
like
someone
who
an
engineer
who
you
know
their
primary
concern
is
the
product
right
and
like
how
they
contribute
to
that.
E
I
don't
know
a
data
science
version
of
that
right,
like
there
should
be
a
data
science
version
like
a
product
data,
scientist
right
and
and
as
as
james
suggested.
You
know
I
do
think
inside
data
science
help
helped
a
lot
there,
I'm
I'm
a
grad,
so
a
bit
biased,
but
yeah.
B
All
right,
I
think,
you're
almost
a
time,
but
I
do
want
to
just
go
around
and
get
your
take
on
what
your
outlook
is.
You
know
we
are
in
the
modern
data
stack
era
and
you
know
what
challenges
remain
and
what
do
you
foresee
happening
in
the
next
three
four
years?
And
you
know
you
can
also
pitch
your
product
in
terms
of
how
you're
addressing
those
things
so
jay
I'll
just
start
with
you
again.
C
Yeah,
I
I
think
the
the
type
of
I
repeat
myself
a
little
bit.
I
think
the
type
of
data
teams
that
exist
out
there
right
now,
just
don't
look
a
lot
like
the
air
me.
Data
teams
and
part
of
that
is
that
we
have
a
lot
of
natural
advantages,
including
capital
and
a
business
model
that
lends
itself
to
certain
types
of
data,
and
we
also
were
able
to
hire
incredible
talent.
C
You
know
everyone
gets
to
pick
up
the
cream,
the
crop
from
facebook
or
whatever,
and
so
I
think
right
now
we
have
a
lot
of
people
who
can
do
the
data
modeling
create
you
know,
manifestations
of
what
a
purchase
is
or
whatever,
but
it
I
don't
think
they
necessarily
have
as
much
of
a
theory
of
change
of
like
you
know
what
what
happens
like.
How
do
you
make
metrics
and
data
really
drive
impact
at
the
company,
and
I.
C
Of
changing
a
little
bit
where,
like
you
know,
I
have
this
like
very
pithy
statement.
That's
like,
I
think
it's
funny
that
we
call
them
data
scientists
when
I
think
today,
data
scientists
and
product
managers,
because
I
think
in
today's
world
we
have
like
product
scientists
and
data
managers
like
where
it's
kind
of
like
all
very
data
modeling
stuff,
and
then
you
just
kind
of
plug
it
into
other
things
and
then
the
whole
rest
of
the
org
has
analytical
people
and
doesn't
matter
if
they
know
sql
or
not
or
tables
or
whatever.
C
F
Yeah,
I'm
I'm
trying
to
so.
Where
is
this
all
going?
The
toolkits
that
I
see
most
commonly
are
meant
to
help
abstract
some
of
those
problems.
There
are
such
phenomenally
smart,
talented
people
at
these
little
tiny
companies
that
I've
never
heard
of,
or
maybe
aren't
even
you
know
in
that
echelon
of,
like
top
silicon
valley
things,
but
there
are
some
really
really
smart
people
out
there
trying
to
make
best
use
of
tools
and
best
use
of
data.
Where
is
this
all
going?
I
think
that
I
would
I
would.
F
I
would
give
a
plus
one
to
what
mac
said
earlier,
which
is.
I
really
believe
that
for
most
companies,
smaller
companies
can
you
stay
centralized
on
the
data
management
as
long
as
possible.
Right
just
just
keep
somebody
whose
job
it
is
to
make
sure
that
the
data
sets
that
exist
are
tight
that
are
useful
and
are
helping
to
actually
empower
those
those
people
making
decisions
based
on
data
analyzing,
the
data,
finding
insights
and
becoming
persuasive
sort
of
business
partners
on
the
analytics
side.
F
Keep
that
stuff
centralized
keep
keep
the
tight
reins
on
that
high
quality
data.
As
long
as
you
can
empower
people
and
finish
that
last
mile
of
enablement
and
and
teaching
and
training
and
make
sure
that
the
tools
are
driving
real
business
value,
I
mean
that
feels
like
kind
of
generic
advice.
But
if
the
modern
data
stack
is
about
abstracting
those
hard
problems
that
we
faced
six
years
ago,
seven
years
ago,
making
those
now
tools
that
you
can
manage
and
manipulate
instead
of
having
to
build
stuff
from
scratch.
B
D
Max
all
right,
I
I
there's
this
new
term
I
like
to
use,
as
I
call
a
data
native
companies,
so
we
use
it
to
use
the
term
like
digital
native
or
like
cloud
native
companies,
the
companies
that
were
born
after
aws,
you
know
kind
of
started
existing
and
that
I
think,
as
the
stack
and
the
foundation
changes,
it
fundamentally
changes
the
way
that
companies
operate
and
the
way
that
teams
and
and
specialization
and
how
teams
operates.
I
think
it's
interesting
to
look
into
data
native
companies
and
what
they
do.
D
So
we
see
a
lot
of
five
trend,
dbt
airflow,
coming
in
a
little
bit
later
because,
like
you,
can
go
quite
a
long
way
with
dbt
pre
preset
super
set.
You
know
bit
for
the
visualization
layer
so
for
us
we're
like
simple,
visualization,
straightforward
visualization.
On
top
of
your
data
warehouse,
the
data
warehouse
is
like
bigquery
snowflakes
gonna.
You
know
probably
win
the
market
there
now
like
for
the
companies
that
are
just
born
as
it
is
super
empowering.
You
can
set
up
these
services
like
so
quickly,
including
things
like
high
touch.
D
Yeah
high
touch
5
tran
bigquery,
like
I
could
set
it
up
for
a
new
company.
You
know
in
a
day
you
know
just
create
these
accounts,
connect
them
make
them
work.
So
how
does
that
inform
how
teams
are
working?
So
I
think
the
so
data
modeling
is
still
like
a
really
important
thing
and
a
lost
art
a
little
bit
like
good
data.
D
I
think
people
decided
that
data
pipelines
are
largely
going
to
be.
You
know
expressed
as
sql,
so
that
kind
of
sucks.
I
don't
like
the
idea
of
like
mountains
of
sql,
but
at
the
same
time
it
works
and
there's
a
lot
of
value.
That's
getting
that's
getting
created
there
and
then
there's
like
longer
term
and
like
the
crystal
ball,
like
where
I
would
look,
I
would
look
at
more
parallels
with
the
software
engineering
world
and
the
devops
movement
getting
drawn
over
time.
D
So
that
means
like
data
ops,
is
emerging
in
the
footsteps
of
devops
and
there's
a
lot
of
patterns
there
around
like
managing
stuff
as
code
and
a
lot
of
ci
cds,
I
think
cicd
for
data,
a
lot
of
data
quality
stuff.
You
know
similar
to
like
quality
infrastructure
and
like
stuff
we
have
on
the
software
engineering
side
is
going
to
keep
like
catching
up
on
software
engineering.
D
I
think,
like
one
thing,
I
think
we've
assumed
in
the
past
that
you
would
just
add,
like
a
handful
of
things
and
tools
and
practices
and
framework
on
the
data
world
like
if
I
could
just
add,
like
these
three
things
that
I'm
set.
That's
not
this
way
on
the
software
engineering
world,
like
we
really
embrace
the
diversity
there
like
how
many
languages
like
programming
languages
exist
and
as
anyone
does
anyone
question
like.
D
D
So
I
think
in
the
data
world
we're
gonna
start
to
accept
too,
that
there's
gonna
be
like
this
mesh
and
network
of
tools
and
we're
gonna.
Stop
thinking
that
one
person
with
one
tool
like
some
some,
some
some
guy
or
some
lady
with
informatica,
you
know
in
the
corner
at
the
office-
can
take
solve
all
the
data
problems
along
with
like
power,
bi
or
whatever.
It
is
right,
like
we're.
Gonna
understand
that
we
need
all
sorts
of
frameworks,
tools,
libraries
roles
too
right.
B
E
Yeah
I
really
like
what
max
was
was
saying
about
software.
You
know
software
engineering
practices
coming
over
data
modeling
right
like
I
think
we
can
all
agree.
That's
that's
particularly
for
the
users
of
you
know,
creators
of
data
products
right
being
able
to
rely
on.
You
know,
consistency
and-
and
you
know
things
like
that,
but
one
thing
I
I
get
really
excited
about
is
people
I
think
are
like
pushing
back.
E
This
is
probably
related
to
max's
comment
but
like
not
as
many
people
are
are
taking.
No.
As
an
answer,
and
by
that
I
mean
like
I
remember
when
I
was
like
learning
pandas,
which
was
like
on
the
job
at
airbnb
and
pandas,
I
hate
it
right,
like
pandas
doesn't
make
like
I.
I
can
never
wrap
my
head
around
the
mental
model
of
pandas,
but
it
was
sort
of
like
well.
This
sequel
won't
work.
E
So
so
you
know
you
got
to
use
this
and
you
know
I
just
I
see
so
many
more
people
and
it
could
be
because
I'm
I'm
like
in
a
founder
community
now
right
talking
to
all
of
you
guys
but
like
it's
like
people,
are
so
much
more
creative
about
the
solutions
that
they
want
and
they
you
know
they
dream
a
little
bit
differently.
It's
like
this
thing
doesn't
work.
E
We
don't
have
to
build
it
right
or
we
can
but
like
someone
else,
can
build
it
and
we
can
find
new
solutions,
and
I
don't
know
I'm
I'm
really
inspired
by
that
by
by
the
idea
that,
like
a
lot
of
things
that
I
just
took
as
status
quo,
you
know
five
years
ago,
six
years
ago,
at
at
airbnb,
you
know
are
being
updated
and
changed,
and
you
know
why
shouldn't
I
be
able
to
to
jump
between
sql
and
python
in
an
analytic
notebook.
E
If
that's,
if
I
know
when
to
use
which
right
like
we're,
seeing
obviously
that
being
done
with
with
hex
so
yeah
overall,
I'm
I'm
super
excited.
I'm
excited
that
folks,
like
you
know,
have
high
standards
and
and
and
you
know,
high
hopes-
and
I
don't
know-
continue
to
get
funded
for
these
crazy
ideas
that
we
all
have
right.
It's
great.
B
Cool,
I
I
kind
of
make
my
own
observations
and
then
we'll
we'll
just
end
with
some
fun
fun
facts.
I
guess
so
in
terms
of
how
I
see
it.
First
of
all,
I'm
super
excited
that
the
budgets
are
just
growing
right.
No
one
resists
data
related
value
questions.
These
days
the
budgets
are
just
growing,
it's
good
for
all
of
us.
B
B
That's
my
prediction
that
you
know
they
have
to
make
do
with
less,
but
they'll
always
be
essential
for
central
concerns,
but
their
role
is
going
to
change
into
more
advisory
and
governance,
and
that
type
of
thing-
and
you
know,
as
people
pointed
out
all
already
like
workload
concurrency
all
these
concerns
are
gone
now,
like
you
know,
snowflake
between
they've
solved
those
things
for
the
most
part,
but
I
think
the
horizontal
concerns
I
I
actually
don't
believe
that
consolidation
will
happen.
I
think
the
fragmentation
is
here
to
stay.
B
There
is
the
databricks
of
the
world,
there's
snowflake,
there's
bigquery
there's
you
know,
there's
looker,
there's
preset.
This
is
the
world
we're
going
to
live
in.
I
think,
and
I
think
there'll
always
be
a
little
bit
of
consolidation
but
followed
quickly
by
fragmentation,
but
I
think
the
concerns
around
how
to
achieve
consistency.
Amidst
all
of
this,
how
do
you
address
things
like
you
know,
compliance?
How
do
you
address
things
like
cost?
B
How
you
address
these
things
are
not
going
to
go
away,
so
you
actually
need
something
that
sits
on
top
in
a
much
more
loose-fitting
way,
not
in
a
taking
control
way,
but
almost
like
as
an
overlay.
On
top
of
all
these
tools,
I
think
is
important
and
I
think
finally,
you
know
just
in
terms
of
all
the
things
that
you
folks
mentioned
already
build
test
deploy.
B
I
I
just
think
more
will
be
demanded
from
the
analytics
engineer
of
today
in
terms
of
rigor
and
even
even
the
end,
product
manager
more
will
be
demanded
in
terms
of
actually
deriving
value.
I
think
people
kind
of
got
away
with
not
being
great
at
data
in
the
past,
but
I
think
that's
with
the
growing
budgets
and
actually
to
justify
that.
I
think
you
will
see
that
more
will
be
demanded
from
the
people
there.
B
So
that's
my
thoughts
anyway
with
that,
let's
round
out
the
the
industry
side
of
this-
and
I
think
it
will
be
good
if
you
can
all
share
like
one
fun
thing
happening
in
our
lives
right
now,
I
know
I'm
kind
of
putting
in
the
spot
a
little
bit
but
che
I'm
going
to.
B
C
I
just
had
a
kid
two
weeks
ago,
so
that's
a
you
know
to
that
question.
So
yeah,
I'm
on
friend,
to
leave
I'm
excited
to.
I
was
it's
great
to
catch
up
with
y'all
along
here.
On
that
note,
I
also
I'm
gonna
have
to
leave
daycare
check
out
my
other
kid
so
great
talking
to
y'all.
D
So,
for
me,
with
the
pandemic,
decided
to
relocate
the
tahoe
back
in
december,
so
I
left
the
bay
area
and
we
decided
to
go,
live
in
the
woods
a
little
bit
more
and
a
tighter,
smaller
community,
and
it's
been
super
great.
Like
I'm
looking
at
the
lake,
it's
not
too
far
from
me,
I
could
be
on
a
kayak
in
10
minutes.
It
might
be,
though
it
is
a
little
cloudy
now,
but
it's
kind
of
interesting
to
be
empowered
to
work
from
anywhere.
D
Now,
and
it's
very
true,
I
mean
people
have
talked
a
lot
about
this
and
it's
very,
I
think
it
is
transformative
for
or
all
the
like
information
workers
and
or
professions.
So
yeah.
I
remember
a
reminder
to
everyone
else
that
you
can
take
advantage
of
that
you
can
probably
work
out
of
anywhere
now.
What
would
you
rather
be
next
week
next
month
next
year,
you.
B
B
E
Yeah,
well,
I
I
don't
know
how
many
of
you
were
still
at
airbnb.
It
might
have
just
been
useful,
but,
right
before
I
left,
I
got
hit
by
a
car
when
I
was
riding
my
bike
and
broke.
E
My
leg
and
biking
was
always
one
of
my
favorite
activities
and
it
took
a
while
for
me
to
get
back
on
a
bike,
but
you
know
bought
a
road
bike
a
couple
months
ago
and
I'm
so
glad
that,
like
you
know,
I've
I've
been
able
to
do
that
and
and
not
let
you
know
a
pretty
bad
injury.
E
Stop
me.
So
it's
been,
it's
been
super
exciting
to
just
you
know,
ride
around
the
bay
go
up
to
sonoma.
I
have
lots
of
lots
of
riding
plans
this
summer.
F
Mayfield,
let's
see
after
dodging
kobit
for
two
and
a
half
years
finally
caught
it
ripped
through
my
whole
family
and
now
the
two-year-old
is
the
last
one
recovering,
and
I
think
his
fever
is
all
better
and
he's
just
starting
to
not
be
a
fussy,
little
mess.
And
that
is
a
very
nice
thing.
In
my
life.
B
What
I
mean
that
is,
that
is
some
kind
of
a
fun
story:
okay,
all
right
I'll.
I'll
conclude
this.
I
have
gotten
back
to
practicing
my
indian
classical
music
comments.
All
the
startup
craziness
I
used
to
do
that
back
in
the
day,
but
somehow
I've
gotten
back
to
it.
Otherwise
I
was
kind
of
going
crazy
with
the
non-stop
startup
grind,
so
that's
what's
happening
in
my
life.
It's
been
amazing.
Connecting
with
you
all
again,
I
think
you
know
we
can
keep
talking
all
day.