00:00hi and welcome to the a 16z podcast I'm
00:02Hannah and we're talking today with Adam
00:04Bree co-founder and CEO of sky do and a
00:0716 zj neural partner Chris Dixon about
00:09what happens now that we're reaching
00:10truly autonomous drones in this case
00:12cell flying cameras what it means when
00:15you take the pilot out of the loop and
00:16what becomes possible that wasn't
00:18possible before we take a deep dive
00:20under the hood into the tech that makes
00:22this work from propellers to perception
00:24then talk about what autonomy will
00:25enable and what we can build on top of
00:27it to finally how our relationship and
00:30interactions with computers of all kinds
00:32will change as they become increasingly
00:33powered by a AI let's start by maybe
00:36talking a little bit about the evolution
00:38of drone technology where are we now
00:40what did it take for us to get here in a
00:42lot of ways the products that we have
00:44today essentially grew out of RC
00:45airplanes and RC helicopters this is
00:47stuff that's been around for a long time
00:4930 or 40 years Electric Power was a huge
00:52transition point and then the
00:54combination of cell phone sensors with
00:56an imaging device just turned out to be
00:58very powerful and that's what's caused
00:59kind of the recent spike in attention in
01:01the space so the Hara dime with existing
01:03drones that are available today is
01:06essentially they're manually flown
01:07devices but still it required a pilot
01:10about yeah there needs to be a human
01:12learning how to use it and directing it
01:14yeah exactly you buy the thing you hold
01:16joysticks you're essentially the pilot
01:18and you're responsible for flying and
01:20controlling it and if you're an expert
01:22if you're proficient at that you can do
01:23some pretty cool things but it's a very
01:25difficult thing to master and I would
01:27say a typical experience for a lot of
01:28people is they take it out of the box
01:30and they crash it into a tree so it
01:32feels like an industry that is still
01:34very very early compared to what's
01:36physically possible there's all these
01:38great notions of what you might want to
01:39do with a drone you know on the consumer
01:41side having this camera that makes it
01:42easy to capture amazing video and for
01:45commercial applications like inspection
01:47mapping monitoring security but most of
01:49them we think at their core in order to
01:51really work and in order to really scale
01:53need an autonomous foundation and so our
01:56focus is on giving the drone the ability
01:59to fly itself intelligently which is a
02:01very simple idea but it's a very
02:02technically challenging thing to deliver
02:04on can you explain why the autonomy is
02:07so important what the human failing is
02:09that makes that change everything I mean
02:12I can use an analogy
02:13imagine if on your iPhone if you wanted
02:16to take a picture with it and you push
02:18the wrong button at the wrong time it
02:20would irreparably destroy itself I think
02:22we'd all agree that that would be like a
02:24pretty big barrier not really workable
02:26to using iPhones and there probably
02:28wouldn't be nearly as many of them out
02:29there and people probably wouldn't like
02:30them as much as they do and that's
02:32basically where we are in the drone
02:34industry now we think autonomy is a huge
02:37enabler just at that basic level of
02:39being able to trust these devices to do
02:41reasonable things in every situation it
02:44removes the need for the operator to be
02:47paying attention to it or flying it at
02:48all times which depending on what you're
02:50doing can be pretty huge I also think
02:52going back to our iphone analogy one
02:53thing the iPhone enabled because it
02:54abstracted away you know things like GPS
02:57and accelerometers and made it really
02:59easy for software developers to access
03:01those things it allowed for example an
03:04uber and lyft an Instagram and all sorts
03:06of things to focus just on the kind of
03:07application development side of things
03:08so a loud app developers to be app
03:10developers mm-hmm and provided the phone
03:12operating system provided all the
03:14functionality of the box and so
03:15in my view that you know sort of
03:17autonomy knowing the drone will avoid
03:19obstacles be able to do kind of
03:21high-level functionality like identify
03:23people and objects will eventually allow
03:26software developers to build
03:28applications in the same way they did on
03:29the phone yeah exactly the iPhone in a
03:32lot of ways it's basically the big
03:34touchscreen that you can put whatever
03:35you want on and it abstracts away the
03:37hardware and makes it possible to just
03:38use that screen to interface with the
03:40user for a bunch of different
03:41applications with drones autonomy is
03:44likely to be sort of that key layer that
03:46separates sort of the underlying physics
03:48worrying about collisions worrying about
03:50a lot of the nitty-gritty robotics
03:51navigation problems with the application
03:54so that if you wanted to do roof
03:55inspection you could be an expert in how
03:58roof inspection works and the workflows
03:59there and right for that rather than
04:02having to worry about what the drones
04:03doing for commercial applications you
04:06don't actually want a pilot flying every
04:08drone like you just want to push a
04:09button have the thing do its mission or
04:11have many of them do their missions and
04:12then get the data back that you care
04:14about and that's what autonomy will
04:16enable sort of taking that core layer
04:18and making it possible to do a lot of
04:21different kinds of things on top of it
04:22well let's talk about the autonomy is it
04:24the same kind of autonomy that goes into
04:27self-driving cars how is it different
04:28from different modes of autonomous
04:31robotics across-the-board so in order to
04:34make a fully autonomous system we've
04:35really designed it for autonomy from the
04:37ground up so it's not sort of like an
04:39afterthought where we strapped on some
04:41extra sensors there are a lot of
04:42similarities to self-driving cars and we
04:45call our first product a self flying
04:46camera for that reason because it's
04:48taking control itself but the way that
04:51the algorithms are designed are
04:52specifically tailored for flight so
04:54where as a car relies on the structure
04:56of the road and staying in its lane
04:58right there are certain Givens yeah
05:00there's there's a lot of Givens on the
05:02road that you can't take for granted in
05:03the air with a car your goal is
05:05basically just to follow the rules the
05:07perfect self-driving car is just gonna
05:08follow the rules and be predictable
05:10whereas with drones we do a lot of just
05:12sort of general 3d reasoning
05:14understanding the scene in every
05:15direction in any time and it's kind of a
05:18different set of challenges in some ways
05:21it's harder in some ways it's easier but
05:23it's also a regime that sort of allows
05:25for a lot more inherent freedom and
05:26there's just sort of more of a creative
05:28aspect to it it's interesting when you
05:30talk about the Givens on the road you
05:32know that you think about like a static
05:33road that you're I mean do you guys have
05:35to factor in things like wind and
05:37breezes or is that just a much older
05:38obvious technology because it's
05:40basically like a helicopter you know how
05:42do you think about moving through space
05:43like that yeah you're right I mean there
05:45are a lot of variables in play and
05:47there's a lot of things that are sort of
05:49taken for granted in manually flown
05:51drones that oh if that comes up the
05:53pilots just going to be able to deal
05:54with like what like a giant gust of wind
05:56for example or so you feel it you can
05:58see it yeah yeah if you're flying it
06:00manually in a big gust of wind comes up
06:02you'll see it and you'll respond
06:03appropriately and if you're next to a
06:05building you'll probably fly away from
06:06that building to make sure you don't get
06:07blown into it and you know if you couple
06:10that with you have a low battery or
06:11something else like there's there's just
06:13a lot of things happening and one of the
06:15challenges has been building a system
06:17that can reason intelligently in all
06:19these different situations and are you
06:21using cameras to do that too or what are
06:23the other ways you collect instant data
06:25about physical movement so actually when
06:27we register through a sensor called an
06:29IMU which is an inertial measurement
06:31unit so this is an example of where
06:33phone technology has kind of transferred
06:35over into robotics your phone has an IMU
06:37in it so it can tell which way is north
06:39and it can tell what it's orientation
06:41and so we use basically exactly the same
06:43component and you can actually use that
06:45to sense the force that's being applied
06:47to the vehicle externally and so it's
06:49constantly seeing the force that it's
06:51experiencing from the world and then
06:53using that to estimate what's causing
06:55that and what it should do about that
06:56and the drones are constrained on the
06:58sensors in a different way than with
07:00yeah there's much stricter limits around
07:01the sensing and compute than you can
07:03carry and that's because of the size of
07:05it because of the size and the weight
07:06everything that you're carrying you have
07:07to have propulsion to lift off of the
07:09ground so for example lidar is very
07:11popular with the self-driving cars and
07:12it's not feasible for drones well it's
07:14feasible for drones but it's not gonna
07:15work very well we did a lot of work with
07:17lidar but the vehicles were super heavy
07:19super expensive so when we started sky
07:22do we made a big bet on vision because
07:24we felt like the progress that's
07:26happening in computer vision now
07:27especially with deep learning but even
07:29in sort of traditional geometric vision
07:32is just super fast and the amount of
07:35information that's in images is
07:36incredibly rich extracting that
07:38information is challenging but the tools
07:40to do that keep getting better and we're
07:42really riding that wave we have 13
07:44cameras on the device they see in every
07:45direction and then at its core it has a
07:48super powerful computer we're using the
07:49Nvidia tx1 which is basically like a
07:53deep learning supercomputer that uses
07:55the same GPU designed the same
07:57architecture that's found in their cloud
07:58compute systems and all of that is
08:01necessary to run the software and flight
08:03algorithms that give it the autonomous
08:05behavior so break it down for what does
08:07that actually mean in sky dio's case
08:09you're using cameras plus intelligence
08:11instead of lidar right what are the
08:13layers that make that work we've
08:15developed what we call the sky do
08:16autonomy engine and that's basically the
08:18complete system that does all the
08:20perception decision-making and control
08:23of the vehicle so it's a technically
08:25very complex thing but it's actually
08:27fairly intuitive to understand what it's
08:30doing because it's similar in a lot of
08:31ways to what people do like it basically
08:33processes visual information it uses the
08:35visual information to figure out the 3d
08:37structure of the scene so where
08:39everything is and then it builds up sort
08:41of a deeper understanding of what the
08:43different kinds of objects are so in
08:44particular we care about people we care
08:46about objects that we might run into and
08:47then all of that information goes into a
08:50planning system that balances a bunch of
08:52different objectives about what the
08:53drones trying to a cop
08:54it's trying to capture some video you
08:57definitely don't want to run into things
08:58we need to obey the maneuvering limits
09:00of the device so all of these things are
09:02constantly being traded off for it to
09:04decide what it should do thinking about
09:06how this tool is navigating when it's
09:08following you around and mapping its way
09:10through space a different kind of space
09:12right because it's Aaron how does it
09:13actually do that how does that work how
09:15does it choose where it's going so one
09:17of the keys to getting intelligent
09:19behavior is you actually have to predict
09:20what's going to happen because if you
09:22don't predict what's going to happen you
09:23end up being purely myopic and you just
09:25react to whatever just happened the path
09:28planning algorithm is actually
09:30predicting about four seconds into the
09:31future so it's looking at what you're
09:33doing it's looking at the environment
09:36it's reasoning and then it's using all
09:38of these sort of future possibilities to
09:39figure out what it should do and even
09:42though it's predicting four seconds into
09:43the future it's not waiting to the end
09:45of those four seconds to then decide
09:48what to do it's constantly doing that so
09:50many times a second it's making these
09:51predictions with the latest information
09:54about what you're doing and what the
09:55environment around it is doing and then
09:57updating its notion of what it should do
09:59based on that how about some in the
10:01design choices you're making as you're
10:02combining these different types of
10:03technology right the propeller aspect
10:05and the weight and the cameras and like
10:07how they're all fitting together the
10:08whole stack what were your
10:09considerations so I think one of the
10:11overall things here this general idea of
10:14like the RC helicopter that's gotten
10:15better and better but then there's this
10:17other sort of paradigm of like what if
10:18this thing's a flying computer and we've
10:20kind of come at it more from that
10:22direction and that's reflected in a lot
10:23of the design choices we made it so we
10:25really do think about it more like a
10:27flying computer and we've tried to make
10:29aesthetic design choices that reflect
10:30that so one example of this is that we
10:34built in this perimeter blade guard
10:36structure as a first-class citizen of
10:38the device so it's made out of carbon
10:40fiber to be super light and stiff and it
10:42means that you don't have exposed
10:44propellers around the under chop off
10:46your finger yeah so it gives it a
10:48significant safety benefit for flying
10:50around people but it also I think is
10:52just closer to the kinds of things that
10:54we're used to sliding into and out of
10:55backpacks because it's just sort of this
10:57single compact thing that you can like
11:00hold anywhere you want and is easy to
11:02manage yourself as opposed to having
11:04propeller blades sticking off and
11:06landing gear hanging down and things
11:08so we've talked a bunch about the
11:09under-the-hood technology how about on a
11:11more basic level what the personal
11:13experience is like of how the technology
11:14actually works when you first begin
11:16using it so like the goal is to make it
11:18really easy to capture amazing videos
11:21from this dynamic perspective that you
11:23wouldn't otherwise be able to get of you
11:25doing your favorite activities so
11:26there's a lot of complex technology
11:28under the hood to make it all possible
11:29but the end-user experience is pretty
11:31simple the way this works is you take it
11:34out of your backpack you turn it on and
11:36you're controlling it from your
11:37smartphone so you can hold it in your
11:39hand and swipe up to take off and it'll
11:42take off from your hand it'll fly away
11:43from you and turn around and look at you
11:45we have a deep neural network that's
11:47been trained against a bunch of
11:48different data sets some of it our own
11:50some of it open source data sets to
11:51recognize people robustly so when it
11:54looks at you it'll know that you're a
11:55person it builds up a unique visual
11:57identifier of what you look like based
11:59on your appearance your clothing things
12:01like this and it'll use that to tell you
12:02apart from other people in the scene and
12:04after that you can put your phone away
12:07and you're done and it'll just follow
12:08you it'll avoid obstacles it sees in
12:11every direction and you can go for a run
12:12you can go hiking biking skiing things
12:15like this and it will move itself in a
12:17nice smooth way and give you video that
12:19literally looks like it was filmed by a
12:21professional film crew no longer have to
12:23be staring up at it yeah I mean it looks
12:25like you have a Hollywood crew there
12:27with like booms and dollies and these
12:28things like moving a camera around but
12:30it's just doing all of it itself through
12:31software it's like a two-step magic
12:34trick or the first step
12:35people see the drone fly around and
12:37follow you and when the drone does that
12:38it's moving a lot right to navigate with
12:41all the stabilization both hardware and
12:42software stabilization people are
12:44shocked about the video looks afterwards
12:46it's not just the tracking it's the
12:47actual quality of the image that you're
12:49getting how do you think storytelling
12:51starts to change when we're now enabled
12:54to be at the star of your own movie like
12:56that you know when the point of view
12:57just becomes sort of you and your life
12:59and it's technologically possible to
13:01just let that unfold how do you think
13:02that starts to change I think it's
13:04really exciting I mean I think we're
13:06still very early days on all this stuff
13:07but if you just think about how
13:09fundamental cameras are to our daily
13:12lives now like the major use case for
13:13smartphones is taking pictures and
13:15videos this is the feature that gets the
13:17most attention there we have you know
13:19huge social networks that are built up
13:22on top of sharing picture and video and
13:24the ability of a camera to understand
13:26what it's looking at and move itself in
13:28autonomously capture footage I think is
13:31going to be like a really powerful
13:34ingredient in the world in a lot of
13:36different ways it's a totally new tool
13:38yeah and you know the most interesting
13:40pictures and video are generally of
13:42people over the last six months of
13:44development we've had a bunch of
13:45prototypes around the company we've done
13:46a lot of internal testing we've also had
13:48external beta testers and one of the
13:50exciting things to us is seeing the
13:52footage that comes back there's these
13:54wacky creative ones work yeah what stuff
13:56do people do so we had an employee who
13:58basically made a music video in his
14:00we're like he had the camera flying he
14:02was dancing around and there was this
14:04interaction between him and the device
14:06that resulted in this video that's kind
14:08of captivating to watch that you really
14:10wouldn't get any other way like it's not
14:11like he was gonna fire a film crew to
14:13come and film this thing in his driveway
14:15but with the device with almost no
14:17pre-planning he could just put it in the
14:19air and get this this kind of amazing
14:21thing I want to talk about like how this
14:23actually feels you know how its first
14:26recognizes you and the relationship you
14:28kind of develop with your drone right
14:30because it's really now a relationship
14:31thing how does that play out what does
14:34it feel like I think it's impossible not
14:36to personify it really and we like that
14:38I mean it responds to you it acts
14:40intelligently based on what you do and
14:42what's around it and so there is this
14:44kind of like fun aspect of it's like
14:46it's sort of your companion the real
14:48element of the product experience well
14:51it also makes it part of your life in a
14:52different way fraid you're not just a
14:53bug under a microscope it's also
14:55interesting when you talk about the
14:56point of view changing right there's a
14:58perceptual distance there's a
14:59perspective there's a perspective shift
15:01it seems to me like that's an
15:03interesting kind of like mental
15:04consciousness shift that happens when
15:06you start seeing yourself from that
15:08other perspective the broader trend is
15:11kind of a new thing in computing where
15:13basically for the entire history of
15:15computers they've been these static
15:16objects that you like type on or tap on
15:19and it you know at the end of the day
15:20almost all computing results and just
15:22showing you an image on a screen that
15:24then you react to and so the ability of
15:26a computer to sense its environment and
15:28act in the physical world and that's
15:31basically what a robot is but we're
15:33entering this new sort of world where
15:35kinds of things are possible it's so
15:36much more dynamic relationship just from
15:38a technology point of view that people
15:40started to see it's like Alexa and
15:41products like this and generally they're
15:43sort of you know there's been all of
15:45this incredible progress in AI in the
15:47last a couple years but I think still a
15:49lot of it hasn't reached you know
15:51production products yet but we're gonna
15:53see more and more over the next couple
15:54of years computing devices that that do
15:57feel like active agents like active
15:59agents that's a really good way to put
16:01it and most public is thinking now of AI
16:03is this like dialogue but the
16:04relationship with you is silent you're
16:06not talking to it you're not telling it
16:07to follow you it's all like implicit and
16:10silent yeah I mean it's kind of like
16:11physical like it's in your space and
16:14it's responding to what you do I mean in
16:15a lot of ways you control it through
16:17your motion like not through gestures
16:19but through how you're actually
16:20physically moving so right now the drone
16:23is designed primarily for consumer use
16:25cases imagining down the road what do
16:27you think other possibly use cases for
16:29autonomous drones are there are a lot of
16:31exciting things on the commercial side
16:33and in a lot of ways I would say it's
16:35kind of less mature and wide open than
16:38consumer even general possibilities of
16:40using drones to automate collecting data
16:42that either isn't collected today but
16:44should be or is kind of difficult and
16:46manual and slow to get so we mentioned
16:48roof inspection there's this fairly
16:50manual slow painful process where people
16:52actually have to get on a ladder climb
16:54onto a roof look for physical damage to
16:56inventory it for insurance purposes you
16:59know it's not a particularly efficient
17:00thing it's pretty dangerous to the
17:01people to do it this could be done super
17:03efficiently by a drone but flying it
17:05manually to do that would be hard
17:06people's roofs have trees and powerlines
17:08and things like this around and so you
17:11don't want like the world the best drone
17:12pilot to have to go out to your house to
17:13safely fly this thing around doesn't
17:15scale to make it work there's a lot of
17:17other things like that where autonomy
17:19could have a really big impact
17:21construction sites have become actually
17:23pop today's case for drones the drone
17:25flies over and looks at the state of the
17:27construction make sure things are put
17:29you know sees how the progress is going
17:31check the inventory levels yeah you know
17:34infrastructure inspection mining bridges
17:36you know make sure it's not gonna fall
17:38these things are very hard to inspect
17:40but humans a lot of these things are
17:41basically like efficiently digitizing
17:44the physical world in some ways that
17:46that data can be like tracked over time
17:48and issues and errors can be found
17:51mm-hmm there's something in the order of
17:53like 10 to 20 million jobs in the US
17:55where people have to climb things like
17:56for example the most dangerous job in
17:58America is climbing cell towers to
17:59inspect them they have to be inspected
18:01every like six months or something to
18:02see if the equipment's working and yeah
18:04it's like for safety issues things like
18:06this there's a hundred thousand rigs in
18:08the world and the salt water can corrode
18:10the steel and then you have you know gas
18:12and fire and you can imagine the bad
18:14things that can happen if you don't
18:15inspect it properly and today it's a
18:17very you know manual intense process
18:19I met a company recently using drones to
18:21do that in all these cases they're all
18:22using kind of manually piloted drones
18:24and so the big constraint is they have
18:25to have these expert pilots that's right
18:27because imagine if obviously if you hit
18:29something it could be catastrophic it's
18:31precisely for all the things that the
18:33human scale is not enough but we're so
18:35we've still until this point been held
18:36back by the human scale now suddenly
18:38that's out of the picture so I'm just
18:39gonna ask when those humans are out of
18:41the picture in this new autonomous drone
18:44ecosystem where are the roles for the
18:46humans and that I think they're still
18:48humans it's just the human doesn't have
18:49to do the dangerous job and you can do
18:50it more efficiently but you're still
18:52going to need all sorts of things around
18:53like an insurance case around kind of
18:55analyzing the claim and paying it out
18:58and doing all the kind of work and
18:59bringing a drone out there I think it's
19:01all leverage so it means that like the
19:03people who are involved or working at a
19:04much higher level of abstraction and
19:06commanding a lot of resources to get the
19:09data in an efficient way it's about the
19:10strategy of data kind of yeah yeah and
19:14and then making decisions afterwards
19:15about like what do we need to do like
19:18based on whatever we've learned the
19:20history of technology is you know it's
19:22there's a sort of fundamental asymmetry
19:23where it's easy to imagine the jobs that
19:25go away and hard to imagine the ones
19:26that are created yeah but the history of
19:28it shows that you know with every new
19:29technology wave yeah there are things
19:31that were done before that become
19:33obviated but there's a whole new set of
19:34things that come along yeah
19:36drone pilot wasn't even a job
19:37description five years ago and
19:39fundamentally I think drones are likely
19:41to enable a lot more new areas of
19:43creativity new kinds of value to be
19:45created if one of the big limits to the
19:47drone space which we're now achieving is
19:50autonomy what's the next state of the
19:52art what are we pushing up against next
19:54I mean I think you can look at birds as
19:56an example of what's possible like how
19:58far is a drone that you buy today from
19:59what a bird is capable of I would say
20:02these things are kind of like bulky and
20:04difficult to manage and autonomy and
20:07intelligence seems like one key element
20:08of that but a lot of the hardware design
20:11aspects also I mean I think are like
20:14relatively early and immature and it
20:16hasn't been possible until now to build
20:18a small lightweight device with a bunch
20:21of onboard intelligence a bunch of
20:22sensing and electric power system all
20:24packed into one there are some really
20:27powerful new combinations of those
20:28things that we're gonna start to see
20:30emerge over the next few years this is
20:32part of a broader wave which is if you
20:35just go back and look at the history of
20:36computing we had the mainframes URI
20:38hesita for every 10,000 people or
20:41hundred thousand people then you had pcs
20:42sort of one for every ten people and
20:44then you had you know smartphones we
20:46have three billion smartphones sort of a
20:47computer per person or we're gonna get
20:49to that point and now we're gonna start
20:50to see kind of 10 plus computers per
20:53person and that means you know computers
20:55embedded around your house at the office
20:57in the air and drones in your car you
21:03I think we're on the cusp of this kind
21:04of Cambrian explosion of computing
21:06devices a little galaxy around hm yeah a
21:08little galaxy around all empowered by AI
21:10is the critical ingredient which lets
21:13these devices understand and interact
21:16with the real world and new interfaces
21:18like speech and gestures and yeah the
21:21ability just to walk around and have it
21:22follow you jester's you may not even be
21:24aware of great that's right and then
21:25just and then just the ability to
21:27understand the environment and the
21:28second-order implications of this I
21:30think are really profound which is you
21:31know we don't even know yet what people
21:33are going to do with these devices and
21:34all the applications you know if you
21:36said to somebody in 2005 there's going
21:39to be these amazing smart phones I don't
21:41think people would have predicted some
21:42of the applications that people develop
21:44so for example ride-sharing it wasn't a
21:46widespread prediction right and so I
21:47think there'll be things like that once
21:49developers and creative people have
21:51drones in their hands drones that are
21:53fully programmable sort of flying
21:54computers and what are all the new
21:56things they come up with it's going to
21:58be I think a really exciting time of the
22:00next three to five years as we discover
22:02that well if you had to take a wild
22:04guess and say okay this is the human
22:05behavior that I think is gonna change
22:07what would you say one of the first ways
22:09we're gonna start seeing human behavior
22:10change because of this self flying
22:12camera people call me an optimist but I
22:14think a lot of what computing does
22:15democratizes things that were only
22:17available previously to you know super
22:19wealthy people today someone with an
22:21iPhone and Google has more information
22:22on the president's tates did 20 years
22:24ago right similarly like you you could
22:26film things like you could do ask I do
22:28but you had to have a Hollywood film
22:29crew right now you're democratizing that
22:31in the same way that you know iMovie
22:33sort of democratize film editing what
22:35happens then right what happens sort of
22:37second order to that well I guess we're
22:38on the way to soon finding out thank you
22:40both so much for joining us on the a 16z