00:00hi everyone welcome to the a6 & Z
00:02podcast I'm sonal today's episode is
00:04about the tech that comes into our lives
00:06in unexpected ways but more specifically
00:08we talk about the engineering that's
00:10hidden behind and that drives consumer
00:13products that people use every day like
00:15Airbnb which people use tuba combs
00:17vacation rentals and experiences and
00:19Pinterest which people use to discover
00:21ideas or inspiration for things to try
00:23from recipes and home decor to style and
00:25more joining us to have this
00:26conversation we have Lee fan head of
00:28engineering at Pinterest and Mike Curtis
00:31VP of engineering at Airbnb in this
00:34hallway style conversation we cover
00:35everything from what it takes to manage
00:37engineering teams and the myth or the
00:40reality of the 10x engineer to
00:42recommendation systems data and machine
00:44learning to the camera as a vector for
00:46understanding intent especially given
00:48the challenge of connecting the digital
00:50and the physical so you both uniquely
00:53connect to the offline world online to
00:54offline and then back and forth it's not
00:56just unidirectional so a question I have
00:58for you is you know what people are
01:00actually doing in the physical world
01:02isn't the biggest challenge that they
01:03essentially lose data when the person
01:06walks physically you don't have any
01:08freaking clue what they're doing so how
01:10do you solve that problem technically I
01:12think it is a it is an interesting
01:13challenge you know the experience you
01:15have offline is like your true
01:16engagement with the product right yeah
01:17everybody built really really aren't
01:19this review system right an ability to
01:20be able to say like what kind of
01:22experience did you have and what
01:23happened and one of the one of the areas
01:26that were experimenting with now is
01:27being able to use the reviews that were
01:29given on a place as a predictive signal
01:31in the matching model all the way back
01:33so we're able to feel why does a magic
01:35meaning the prediction for like if you
01:37booked this listing how likely are you
01:38to give it a good review so we can
01:40actually sort of collect some amount of
01:42offline data from the actual experience
01:44that you want to have right and then use
01:45that as a feedback mechanism into our
01:47ranking model in terms of what listings
01:49we're going to show you next time
01:50because again if like your objective
01:52function for ranking is really like how
01:53good of an experience you're gonna have
01:55then all this data you can get about
01:56what kind of experience did you actually
01:58have out there in the world can be then
01:59used to your next booking but also to
02:01like be able to look at using that data
02:04towards what how other people are going
02:06to book those reviews are not just
02:07reviews they're much more they're a
02:09vector to more data for you to be able
02:10to do different things in some way we
02:13advantage that user when they use
02:16Pinterest a lot of them start with
02:18explicit action they ping something they
02:20like those explicit signal user give us
02:23is a unique assets and this is
02:25fascinating data said our engineer can
02:28work with so that we know one out of ten
02:30times the unlikely to go down deeper and
02:32a click by click and sometimes went out
02:35to say twenty they were by some see you
02:37know you guys are both talk about
02:38something so fascinating which is the
02:40complexity is not just of search and
02:41matching and relevance but that a lot of
02:44people don't explicitly express what
02:46their intent I'll have a Georgia Nelson
02:48desk and a vintage French country bench
02:52and those two categories do not go
02:54together if you were to do an explicit
02:55search one would do mid-century modern
02:59and a different category for country or
03:01cottage chic but now you have to infer
03:04this collection this this cluster of
03:06traits of what people are interested in
03:08you guys talk to me about some of the
03:10challenges of doing this I'm first a
03:12difficulty or challenges don't know what
03:14they don't know right for me like this
03:16for example very hard to describe it is
03:18contemporary classic traditional and
03:20also you don't know whether this is the
03:22ultimate to say you want a match upon
03:25maybe this you should put in the tip to
03:27different rooms right so I think one how
03:29for saying we hope use it you to do is
03:33choose these are some user have the same
03:35struggle with and the short part is
03:37maybe that user is he in Scotland and
03:40that they use different languages
03:43that's why lens can be powerful because
03:44you don't know how to describe instead
03:47of asking user to input attacks as a
03:50query take your camera pointed to
03:53whatever you want you understand and we
03:56will give you ideas that what are the
03:58related ideas so you're actually saying
04:00that people because people don't
04:02actually always have the language and
04:03even when they do there's different
04:04language is different like camera and
04:06image can actually level exactly and I
04:09also you think you are looking for
04:11contemporary frankly speaking you're one
04:12adamant content you take a picture we
04:14know okay maybe you like those styles
04:17you think this is exactly you it's not
04:19necessary you have to describe in that
04:21in the English tax right you give me the
04:23image I will tell you like South and the
04:27match your cabinet with this one or
04:29match this sofa with the other table or
04:31something like that you may not think
04:33your white sofa match is a red table
04:35like outrageous but when you see the
04:37image you say wow that looks good we
04:40want to give you that inspiration in
04:41some way this is different from Google
04:43there's no right or wrong this is not
04:45the classic Google rabbit hole of you
04:47click on a bunch of page cuz you're
04:48interested in one topic for me it's
04:49fascinating cuz it's actually
04:50understanding yourself in a weird way
04:51Computers augmenting humans I'll give
04:54you a concrete kind of dumb example but
04:55this literally happening in Pinterest in
04:57the early days I had no idea I loved the
05:00combination of dark green black and gray
05:03so whenever I did searches on shopping
05:05sites I would always put my favorite
05:06colors but I never knew I loved that
05:08combo and then I noticed one day that
05:10one of my Pinterest boards for dresses
05:11was all dark green gray and black and I
05:14was like oh my god I love this combo and
05:16in a weird way the system kind of taught
05:17me what I like I love that idea in like
05:20the sort of act of like helping inspire
05:23somebody for what they might like that
05:25they don't necessarily know that they're
05:26looking for I think is really
05:27fascinating and it's it's you know it's
05:29actually kind of analogous so like the
05:30challenge that we have to solve on
05:32Airbnb because you think well I mean
05:34you've got millions of travelers who are
05:36traveling you've got millions of homes
05:38that they could potentially stand every
05:40home is completely unique right like
05:41totally different like we're not selling
05:43like a block of hotel rooms right
05:45there's so much variety in the style
05:46even every single one is unique and so
05:48you know a similar problem for us is
05:49like how can we look through you know
05:51like the click behavior of somebody
05:53who's going around and looking to stay
05:54in a certain area for thinking about you
05:56know staying in Paris next weekend you
05:57might have to sort through like 50,000
05:59different places like obviously we can
06:01narrow some of that down with filtering
06:02but then we can also look at what are
06:04the similarity and characteristics
06:06between the listings that you're
06:07clicking on or the ones that you're
06:09expressing interested in and then the
06:10deeper you go into it the more we can be
06:12rear anka and surfacing other things
06:13that share some of those attributes some
06:16of the areas that we're exploring now is
06:17like how can we you know find other
06:19embeddings and those images that can
06:20take the unstructured data the image
06:22turn it into something that can actually
06:24be tagged and labeled and then used in
06:26that ranking algorithm like maybe you
06:28particularly like places where you have
06:30views of trees from the window and all
06:32the listings happen to have those
06:35I like that dude how you technically
06:38solve that challenge of moving from one
06:39structure to structured data in that
06:41case how do you now extract that data
06:43feed it in in the same way like you know
06:45you can hold up your your camera and
06:46it'll detect your face right there's
06:48like on snapchat right filters that kind
06:51of technology can be used to detect all
06:53kinds of different objects that could
06:55potentially be like correlated with what
06:56would be interesting for you and so it's
06:58that same kind of computer vision
06:59techniques that can be brought in to
07:01bring like that unstructured data
07:03forward and turn it into something you
07:04can actually what is it like move what's
07:07the pipeline first is you start with
07:08data and I think a Bose company have
07:10massive interesting data that user
07:13expressed they shared their interest
07:15right you click on your book the first
07:17step is really understand that data we
07:19have computer vision technology to
07:20identify there's a sofa
07:22there's a table and we can train label
07:25those orders like a millions actually
07:26millions of machines we will have a
07:27subset of labeling data say this is a
07:30white sofa and the computer will pick up
07:32all those labeling and start to learn
07:34this is a white sofa this is a leather
07:35sofa this is a table with a brown color
07:38things like that once they learn you
07:40will then connect your user data and say
07:43okay this user lie to click for example
07:45the open loft a feeling of house and
07:48these are the things connection now
07:50you'll learn the crashing you say we can
07:52do personalization for you now you go to
07:54say search the same query for example
07:57red shoe we know if you are male you am
07:59likely to want to have maybe a running
08:02shoe instead of a high-heeled pump for
08:04women so those are the things we have to
08:06solve personalize your favorite thing
08:09and all those of how the massive data
08:11set the understanding of the image
08:13understanding connection between image
08:15and a user and using machine learning
08:17technology to make those connection
08:19okay so you have let's say you have
08:20users that are clicking what they're
08:21interested in their browsing you know
08:23room listings or browsing pins how do
08:25you distinguish between the aspirational
08:27and actual outcomes with both you can
08:29have people who are pinning things
08:30because they want to have their dream
08:32house one day or this is a type of
08:34Airbnb you're staying and because you
08:35can't afford to live in a beautiful
08:37industrial loft but hey when you go away
08:40you can at least do it for a day or two
08:41and that's a great way to get this
08:42experience do you actually weigh what
08:45they booked or actually pinned and
08:48highly than the things that they might
08:50just be browsing and clicking through
08:51because one of the tricky things is
08:53differentiating intent when it's
08:55aspirational versus actual and sometimes
08:58you don't know is get an outcome that
08:59you can link to the choice that they've
09:00made because you might not know until
09:02weeks later that they bought that dress
09:04after oh yeah yeah exactly have the same
09:06happen I would say it depends on user
09:09state of mind sometimes user are ready
09:11to book or purchase sometimes they're
09:14just explore right I plan my vacation
09:16miles ahead and I start to explore at
09:18that time what I need is a creative
09:21inspirational ideas what's possibilities
09:23that's how the mind works like things
09:25sit in the back exactly at that time I
09:27think what we could do as a product to
09:29expand your horizon and help you to
09:31discover new interesting ideas we don't
09:34have to push you to in too deep and say
09:36purchase days or book this but then
09:39later we can tell users behavior they're
09:42going narrow in a narrow now we I have
09:44this living room I do clean looks like
09:47you are going down and down where is
09:49this sofa no I wanted there leather
09:51I wanna get a la have the recliner
09:52things like that now we see those
09:54signals we know ok you probably have the
09:56right intent you're ready to do
09:58something more real then we were happy
10:01to drill down all those who need a
10:03computer to understand the user intent
10:05and the which state you are not everyone
10:08is ready to purchase right away you know
10:10similarly when you think about
10:11differentiate between aspiration and an
10:13actual intent to do something there's
10:16also another dimension of like feelings
10:18when you think about what in the early
10:20days of Facebook you could only like or
10:22not like something and now you can
10:23express a variety of things we have
10:25emoji as a new communication form and so
10:27we're essentially making emotions more
10:29machine readable and that's a really
10:31useful thing but you don't get those
10:32signals actually because you're talking
10:34about people both in the physical world
10:36and maybe I don't know if you guys
10:37actually have the range of emoji not
10:39just a star for how you liked or not
10:40like something or pinned or not been
10:42something how do you guys think about
10:43that dimension oh I think I think
10:45there's actually a lot of facets to this
10:46could be like you know I'm looking for
10:48something that's a little bit more
10:49secluded and out of the way or I'm
10:51actually looking for something that's
10:52like in the middle of the night life and
10:53like we can start picking up on those
10:55signals again based on like how you're
10:56searching and browsing through it and
10:57then so that's sort of at the front end
11:00but then you know all the way at the
11:03again with that review information
11:04there's some of it that is structured
11:06right like give a star rating but then
11:08there's also the content of the review
11:10itself so we can do sentiment analysis
11:12and you know some natural language
11:14processing on that to sort of suss out
11:15which aspects of this like what feelings
11:18did it evoke you might have given it a
11:20five uncleanliness but maybe you felt
11:22like oh it wasn't really like the right
11:24neighborhood it lacked energy or didn't
11:26really have a warmth to it like those
11:27are words that are interesting people
11:29have such emotional connection with
11:31these experiences that they have
11:32traveling and and I think like as a
11:34result of that the reviews that people
11:36write in the comments that they have in
11:38everything are very rich and like filled
11:40with emotion about these experiences
11:42that they had don't you this problem
11:43which is typical with all review
11:45behavior I think where you have this
11:47sampling from the extremes you have this
11:49natural skew where only the people who
11:52are most extremely motivated because
11:53they loved it so much or extremely
11:56frustrated god I hated that place you
11:58progressed to the mean when you sample
12:00from the extremes how do you think about
12:02that you know there's there's such a
12:04personal connection when you're staying
12:05in somebody's home that I think there's
12:07there's sort of almost like a social
12:09contract a little bit the percentage of
12:10trips to get and get reviewed are
12:12incredibly high by the way you might
12:13have a unique benefit here to which is
12:15that you have the boat size your
12:16marketplace hosts can also be supply and
12:19supply can be demand in the sense that
12:21people who are hosts are also guests and
12:23people yes can also be hosts we get a
12:25high review rate from the host site as
12:27well and I think that's also part of
12:28being a host and hosting right one thing
12:30we were worried about a while ago was
12:32that the reviews were really nice you
12:33know one of the reasons for that of
12:34course could be like fear of retribution
12:36right like if I reviewed you the host
12:38and said like oh yeah this place wasn't
12:40that good then you say well this guest
12:42how'd you address it so so one of the
12:44things that we did was we changed it so
12:45that we have a simultaneous reveal of
12:47the reviews so both the host and the
12:49guest have two weeks to write a review
12:50and then they're revealed at the same
12:51time so you don't have to worry about
12:53like the retribution they can review
12:55asynchronously but the review is
12:56synchronous so therefore it's they don't
12:58know until they're right exactly and the
13:00reason that we went down that path again
13:01is also to make sure that we have good
13:03accuracy in the data because there's a
13:04lot of information in that review that
13:06can come through and be useful signal to
13:08us later on I would say paint is unique
13:10in that way because when we design we
13:12really design an app or a product for
13:14user to feel positive yeah
13:16operation should be a positive thing use
13:21Pinterest and main amazing.i I figure
13:23that's every pain you see in like a so
13:25beautiful looking right and this is
13:27something actually surprising enough we
13:29train computer to learn why this image
13:32of the same livingroom you you take a
13:35picture of this way and that way they
13:37look so different when this looks so
13:39inspirational the other maybe just
13:41boring and the computer will start you
13:43to learn those skills and they already
13:45learned certain colors a yellow is more
13:47you know inspirational a happy than
13:49other this reminds me of the old days of
13:51like color coding moods and color coding
13:53design but it's actually amazing because
13:54you can you you can't actually model so
13:59what are the images it's more
14:01inspirational than the other of course
14:02it's not science it's actually art we
14:04learned for example if a same dress in
14:06in a stock very boring white background
14:10didn't sell we also want to add a
14:13personalization signal we have some
14:16experimenting house if you you take a
14:19picture of yourself and we learn your
14:21skin tone but by the way we can also
14:23learn from the pings you like and we
14:25know there's certain style certain shape
14:28of a model you like to see and usually
14:30it's people that are it's similar that
14:32you've related so then then we will know
14:34those are type of dress or something you
14:36look for right it's not super fancy for
14:39example for me it's more kind of
14:40businesslike that I can wear so I think
14:43all those you know signals we can pick
14:45up and we also I think one luxury we
14:47have is a user super engaged with our
14:49platform so sometimes they're willing to
14:51give more signals take a picture of your
14:54dish take a picture of your living room
14:55take a picture of your your kids artwork
14:59just you think about the future where
15:00we're going with this connection between
15:02digital into physical there is a world
15:04of sin certification happening all
15:06around us whereas more and more sensors
15:07are embedded in our environment you
15:09don't only have to rely on your
15:10smartphone to then do that that example
15:12of colors is fascinating because you can
15:14see colors you can see sofas white
15:16leather the style this type of thing but
15:19one of the funniest means that makes its
15:20way on the internet and I love this meme
15:22and it's actually a fun new genre is can
15:24computer sell the difference between a
15:25dog and a bagel or the latest one I saw
15:27I was like Kentucky Fried Chicken and
15:29like these little like who
15:30rules that have the same kind of texture
15:31and a human can instantly spot the
15:34difference and this is one of the
15:35challenges and deep learning because
15:37colors are easy mm-hmm but some of this
15:39is not I would say if you give a classic
15:43picture of dog versus hospital of a
15:46bagel the computer can solve at this
15:48point right however if user generated
15:51the random pictures with a dark
15:53background or with the dog in a very
15:55weird shape or whatever it is hard
15:58remember computer learn things from the
16:00data we train them right if you have a
16:03millions of pictures of a dog and it
16:05with different shape a different color
16:07I'm sure it will get there right now I
16:10was a lot of domain is limited by the
16:12data if you only have a limited data to
16:15teach let's say fashion how can we know
16:18this is a fashion that are high-end and
16:21more for the wrong way instead of a
16:24daily it's a lot of data cuz it is the
16:26subtleness it's a very subjective era
16:28when vo differently and then you have a
16:31computer really get there it would take
16:33a while I think that the breakthrough
16:35recently in the deep learning is that we
16:37can learn more so fix together way but
16:39it's a fascinating moves so fast yeah I
16:41mean at this particular point in time
16:42and like on the spectrum of point of
16:44time like you know a year from now two
16:46years from now or five years from now if
16:47you look at like the advances that have
16:49been happening just even really in the
16:51last couple of years in deep learning
16:52and AI technology it is like an
16:54explosion over the 20 years before that
16:56it's like these two huge forces sort of
16:59coming together and reaching a tipping
17:00point the availability of massive
17:03and then the processing power to be able
17:05to actually train models on them and in
17:06a time reasonable way like the broad
17:08availability of GPUs so I think what
17:10we're seeing is that we're at this
17:11moment of you know potential like
17:13exponential growth in this field that
17:15has been promised multiple times before
17:17in history only to kind of not really
17:19have the ingredients there the
17:21applications of this type of AI
17:23technology can go to every facet of how
17:25we live and work and you know how people
17:26sort of exist in the world and what
17:29technology companies are really going to
17:30be great in the futures are the ones
17:32that think about this as this needs to
17:34be core to what we do and we're not just
17:36utilizing this technology but we're
17:37figuring out ways to push it forward
17:39because we're going to be half that
17:40we're all going to have to be in leading
17:42positions in this technology in order
17:44to be competitive and the in the world
17:45that's coming in short order that's so
17:47true and I agree with you let's shift
17:48gears and talk about that so you guys
17:50essentially are startups
17:52you're still a startup technically which
17:53is really weird to think about given how
17:55big you are and one of the biggest
17:56challenges is obviously getting to scale
17:58and when you have that scale you grow
18:00very fast you do you think about
18:01balancing building this kind of
18:03competency as heads of engineering when
18:06in a lot of traditional engineering jobs
18:08I would imagine that you kind of know
18:09your road map already and here in this
18:11kind of business you shift direction how
18:13do you think about building this
18:14organizationally and operationally how
18:16do you hire people there's such high
18:18demand for people who have skills around
18:20you know computer science data you know
18:24and machine learning the technical
18:26challenge is so important and that has
18:27to be there it's like the foundation for
18:29as an engineer am I going to be
18:30fulfilled by my work but when you can
18:33satisfy that at multiple places because
18:35you know we all have interesting
18:36technology challenges to solve then it
18:38starts coming down to like what is the
18:39purpose behind my work people more and
18:41more are choosing their work based on
18:44the mission behind the work and the
18:45purpose of networking as opposed to like
18:47the specific thing that they're going to
18:49work on do you guys have your face and
18:51you know there's a lot of these funny
18:51religious basic programming languages
18:53are like how you build your stack like
18:55open source versus like an open source
18:58and the own stack and a sheer stack or
19:00announced your stack you know open
19:01source it we can be kind of a lightning
19:03rod on some of these things but I think
19:04we've we've basically just taken a
19:06position on it and said like you know
19:08that this is the way we're going to
19:09manage it we are strong believers in
19:11open source we regularly open source our
19:13own internal technologies particularly
19:14around data like everything we do for
19:16managing data pipelines it's how we do
19:17in data analysis we pushed out to open
19:19source and then internally we kind of
19:21have the philosophy that if there's a
19:23great open source tool that can solve
19:25this problem that we're trying to solve
19:26let's look to use it before we look to
19:28build our own because we want we want a
19:30higher percentage of the hours like
19:32engineering hours and thought and
19:34creativity at work that happens here
19:36being focused on things that are very
19:37unique to our business it's a
19:39pro-wrestler right because I seen
19:41computing in my early days a career like
19:43I'm almost all the big internet
19:44companies do their own stack and then we
19:47have a tremendous progress in the
19:49open-source community then you can say
19:51that principle without hurting your
19:53business efficiency a lot of engineer
19:55was they believe they can build better
19:58judgment called you to encourage that or
20:00say let's look carefully I totally agree
20:02when there is a open source tool that
20:04you can leverage let's pay attention and
20:07it's considered sometimes I smile when
20:10this question come up because I feel
20:12like engineer by nature they are so
20:14practical but sometimes they go to the
20:16extreme of philosophical discussion
20:18engineers are secretly poets yeah and
20:20I'm like this let's talk about this case
20:22right instead of say are you in this or
20:25not you're saying that instead of
20:27letting it be an abstract fight you just
20:28go right down to the create my strategy
20:31when you go down to concrete scenes and
20:33I always tell shared contest share the
20:35information you have what most likely
20:37happened is when you shared exactly same
20:39information your conclusion would be
20:41this but it takes a lot to force them to
20:44spend the time to communicate to each
20:46other and to spend time to understand
20:49you know that there's not so big a
20:52difference des you expected it so we did
20:54a couple things relatively early on to
20:56try to head some of that off the first
20:58is that we standardized are supported
21:00technical stacks and said like we're
21:02just going to operate with within these
21:04stacks and not deviate from them except
21:06for in exceptional cases and we also got
21:08you know actually a pretty broad
21:10engineering effort a couple of years ago
21:11to come up with from an infrastructure
21:13standpoint like what are the tenants by
21:15which we want to do development and
21:17they're basically just guidelines and
21:18the idea is if you're an engineer if
21:20you're working within the architectural
21:21tenants for how we how we develop
21:23software then you're pretty much good to
21:25and like just just a within it if you
21:26want to go outside here's the group of
21:28people that you can talk to and I think
21:29that that has like sort of calm down a
21:32fair amount of that chatter don't get me
21:35yeah I think that's definitely a good
21:39principle it's just the constant debater
21:42you probably agreed is really to
21:44encourage them to understand the cons
21:46the benefit of a consistency and the
21:49company principle versus their own
21:51creativity right I think the engineer
21:53always want they're like I have
21:55basically but yes exactly and you just
21:57have to constantly balance I'm gonna ask
22:00you guys a question about technical debt
22:01big topic you've had to grow very
22:03quickly in buildings very fast and it
22:05means you leave this legacy of crap
22:06behind in front frankly and I want to
22:08know how you guys think I just firmly
22:09believe that really we
22:11you know in some way this is engineering
22:13excellence right you have tribute a
22:15assistant or infrastructure that lasts
22:18long time right because it's ultimate
22:21efficiency however there would be a lot
22:23of product features we iterate and we
22:25don't know whether it was take for three
22:27months right and maybe this features
22:29user don't want maybe something we will
22:31admit like it's a failure right so
22:33it's okay so we want to be I use a
22:36Hamilton code a lot like the reason you
22:40know young scrappy hungry it's moving
22:42fast at the same time this is why we
22:45need a senior engineer leaders in-house
22:46they look at this is the system we plan
22:49like I say machine learning framework
22:51for a company we plan to use four to
22:53three years that's something like you
22:56need a little Washington to balance out
22:57the Hamilton and also historically you
23:02accumulated all those that and let's
23:03face it and a dome pretend its theory
23:07and and not face it we set aside time
23:09say fix it week I guess it's the only
23:11way you'll actually make things happen
23:12you to carve out otherwise principle the
23:16engineer leads just want to ship in new
23:19things this is a human nature right you
23:21wanna have fascinating news is coming
23:23out and look at that's like a hoe system
23:25like but you need them you don't want
23:27that the reliability will hurt you in
23:29the future the great engineering leader
23:30finds the right balance like between
23:31time spent fixing things that came from
23:34before and doing for for development
23:35because neither end of that spectrum is
23:37correct like if you spent all of your
23:38time having like a technically perfect
23:40you probably didn't do anything to
23:41further and vice versa so I think like
23:45there definitely has to be has to be a
23:47spectrum of time the way we've been
23:49thinking about it and the way that we
23:51try to do it now is that the business
23:52leaders who are responsible for
23:54furthering the business also have goals
23:55that are associated with bugs
23:57performance like because an engineering
24:04leader should be an owner of the
24:05business right thinking about like
24:07balancing these types of trade-offs and
24:08the other angle that we think about
24:10technical debt on is really about what
24:12is our overall capacity as a technical
24:15organization to to move like how much
24:17can we produce in any given amount of
24:19time and you have to be watching that
24:21because if technical that crops up what
24:23happens is your overall through
24:25what Goes Down and the worst thing that
24:26you can do at that point is say like
24:27well we'll solve it by hiring more
24:29people right yeah and like you can't
24:31hire your way out of that every person
24:33you hire just ends up becoming less and
24:34less productive and so you have to be
24:36able to look at like what is your
24:37contribution per engineer or per
24:39engineering our work and seeing that
24:41that is like increasing over time
24:43through retirement of technical debt
24:45because then the organization has more
24:47throughput as a whole one of our general
24:48partners Martine Posada wrote a post on
24:50why you have to hire a VP of engineering
24:52early on and the big point he makes cuz
24:55frankly no offense to you guys not
24:57knowing this world I thought of it as
24:58roadmaps and deadlines and keeping track
25:00of schedules and you don't realize
25:01there's just big trade-off between the
25:03big picture versus individual pieces
25:04even if they're empowered owners by
25:06definition a lot of individual people or
25:08groups don't own that big picture and so
25:10how do you sort of connect those dots in
25:11this way and including making the
25:12decisions about priorities and
25:14trade-offs because he gave this great
25:15example of how a way to accrue technical
25:17debt is that someone will actually go to
25:18an individual engineer and say how long
25:20would it take to build out this feature
25:21for this customer and of course that
25:23person will give them an estimate and
25:24say well I think it'd take two weeks not
25:25realizing that actually that's an
25:27estimate in isolation not trading off
25:29the bigger picture and under you need to
25:31take into account who this engineer yes
25:34what do you mean meaning that engineer
25:36versus Imagineer this ripple can be ten
25:38eggs are you all right yes really
25:40believe in this I mean is it a myth of
25:41the 10x Engineering I believe in I also
25:43believe that we should reward according
25:45to oh you could really reward that
25:48throughput I love that idea I want to be
25:49a 10x editor but it is I think the
25:52leader to decide like a whether there is
25:54a 10 X or 2x dear friends this is the
25:56beauty and why this field is so
25:58fascinating a good engineer the amount
26:01of a and the throughput or the
26:03contributing they can bring on table
26:04sometimes there's a beyond of whether
26:06you can measure or imagine my thinking
26:09on that is so slightly different in that
26:10I think that you know people have
26:13periods of time where they can be like
26:15producing 10x what their peers might be
26:18and it's the magical moment when they're
26:20aligned with the right project with the
26:21right skill set and like the right
26:23personal energy around it I love that
26:25you said that throughput is this
26:26connotation from the semiconductor
26:27industry it's actually bigger than
26:29efficiency because it's about the social
26:31cultural context that actually supports
26:33this idea and I love it because it
26:35reminds me of creativity you have
26:36moments where you can do like 10 edits
26:37again you have a moment
26:38when you can maybe do one video a lot of
26:40yeah and that's yeah that's exactly how
26:42engineers and like everybody works
26:46motivated not to make a difference this
26:48is like one of the most core fundamental
26:51premises behind engineering management
26:52is like understanding the person what
26:55motivates them what are their skills
26:56what are they trying to develop that
26:58magic moment that I talked about like
27:00being able to find those moments for
27:01people and that being a core part of the
27:03responsibility that a manager has cuz
27:04everything good happens then I have to
27:07you know you both have founders and
27:09co-founders who are not necessarily tech
27:11quote native came from design product
27:13but not classic tech is that sometimes a
27:15thing that you have to explain like
27:17what's that light I really enjoy it
27:19because Bryan brings this incredible
27:20design sense and it's not just to like
27:22how the pixels look on the screen is to
27:24like how are things going to work how
27:26does this ultimately pan out down the
27:27road and I actually think that that in
27:29combination with the technical
27:30discussion of me super powerful a mic on
27:32that I actually really enjoyed because
27:33they fascinated about the challenges and
27:36a very supportive when you have every
27:39daily conversation with tech stack like
27:41a people and you have formed a certain
27:44way because like included a certain
27:46knowledge when you talk to your designer
27:47as a questions like wow I never thought
27:49about this and that the process was
27:51thinking about how to answer this and I
27:53discover there's some hole in my logic
27:55or discover look maybe we were to you
27:58know go down this path and maybe we
28:00should sink you so we encourage you to
28:02think out of box I think this is a way
28:04with the future in fact because there's
28:05been this traditional adversarial
28:07portrayal and this idea that like
28:09engineers don't collaborate well with
28:10designers and designers and always
28:11collaborate with engineers I think
28:12that's changed a lot in the last 10
28:14years as you have a lot of these
28:15products that examples we're talking
28:16about today that you're touching lies
28:18but there's deep deep tech behind them
28:20well thank you guys for joining the a
28:21six in Z podcast thanks for having us