00:05Keller welcome to no priors thanks let's
00:08start with the basics tell us about
00:10yourself what you were like as a kid
00:12what you did right out of school have
00:13you always been working on crazy
00:20I feel like life is a winding path I uh
00:24in college I built computers made of RNA
00:26and DNA that operate within human cells
00:28uh you know the goal was to build these
00:31molecular automata or molecular doctors
00:33that could recognize cancer on a
00:35cellular basis and and cure it
00:37so I was really interested in
00:38biotechnology I also got to build a
00:41climbing wall in college I was a
00:44professional rock climber right after
00:45graduating for a year and a half and
00:49obsessed with robotics you know it felt
00:52like there was a lot of really cool
00:53technology coming out of academic labs
00:55and no one was really making that
00:57technology work in the real world in a
00:59way that would be reliable enough that
01:01millions or tens of millions of people
01:03could really depend on it
01:06the more we learned about Robotics and
01:09automation the more we got excited about
01:11logistics it seemed like what you really
01:13want in in robotics is
01:18repetitive task and Logistics is about
01:21as boring and repetitive as it gets and
01:24then the more we learned about logistics
01:25the more we understood that you know it
01:27really serves the golden billion people
01:28on Earth well but it does a very bad job
01:31of serving the people outside of the
01:33golden billion and for a hundred years
01:36we've been making excuses for
01:39you know why Logistics is so unevenly
01:43we you know as a result of of that five
01:46and a half million kids lose their lives
01:48every year due to lack of access to
01:50basic medical products and we pretend
01:52like this problem is unavoidable or
01:55somehow excusable and we just felt like
01:57it was neither of those things it was
01:59if we're going to build a new kind of
02:00logistics system the transform Logistics
02:03toward you know automation zero emission
02:0710 times as fast let's also build the
02:09first Logistics system that serves all
02:11people equally and that was ultimately
02:12the vision that they created zipline
02:15uh it's it's an amazing Vision I want to
02:17get to how you went from rock climbing
02:20to Logistics but for clarification first
02:23what do you mean by Golden billion
02:25the richest billion people on earth
02:28got it and you got obsessed with
02:31robotics you weren't working on
02:33Logistics at the beginning uh can you
02:35talk a little bit about remotiv
02:37yeah I mean when we you know the tricky
02:40thing and maybe this is true for more
02:42startups than you realize the tricky
02:44thing is like we didn't know we were
02:46starting a company when you know all I
02:49knew was me and my co-founders didn't
02:51have jobs and it you know it seemed like
02:53a cool thing to do to build some robots
02:55we put something on Kickstarter we ended
02:57up selling 150 000 worth of robots on
02:59Kickstarter which was a huge amount of
03:01money to us at the time and we ended up
03:03building those robots in uh the
03:06apartment of it was it was my apartment
03:08at the time but I mean it was
03:09technically really Tony Shea's apartment
03:11you know Tony was kind of a mentor and
03:14inspiration to me I had just read his
03:18um he lived in the exact same dorm I did
03:21in college just 10 years ahead of me and
03:23so seeing like what he had built and
03:26yeah I didn't even know building a
03:27startup was a thing that you could do
03:28but you know we he had all these
03:30apartments in Las Vegas he gave us a
03:32bunch of them we started building robots
03:34in those apartments and shipping them to
03:36people all over the world and these were
03:38really simple right because we had no
03:39money no credibility we weren't that
03:42good at building robots at the time so
03:44these were really simple they were
03:46laser cut out of acrylic and you could
03:49attach your phone to it and become a
03:50little autonomous roving platform that
03:52you could use you know to teach kids
03:54programming or to do telepresence
03:57um it wasn't that good of an idea in
03:58retrospect but it was
04:00um you know it was the thing that
04:02ultimately uh enabled us to to actually
04:06you know ship something make some money
04:09um and and Learn and Grow the company as
04:13and how did you begin to get curious
04:15about logistics from these little roving
04:20I think that what I realized because the
04:22original vision for what we were doing
04:23is we were building things you know we
04:25were building robots that would operate
04:26inside a house and like interact with
04:28people and it didn't really work very
04:31well like in the first year people were
04:33buying them but they ended up going on a
04:35shelf after like a month or two and so I
04:40you know okay we could we could continue
04:42investing money here and and
04:45building this and even potentially
04:47selling the product but I didn't think
04:48it was going to be a very sticky product
04:50in the long run and as I mentioned you
04:52know what you what you really want is
04:54like very boring very repetitive tasks
04:56in controlled environments
04:58and the home is not a controlled
05:00environment and there aren't very many
05:01like really boring repetitive tasks it's
05:03all like a bunch of one at one-off tasks
05:05like people always ask why can't you you
05:07know design a robot to fold my clothes
05:09and if you think about like all the
05:11tasks that are required to fold clothes
05:12that's like not repetitive at all it's a
05:15really actually a really hard thing to
05:16get a robot to do well and around the
05:18same time I actually remember meeting
05:20the founder of a company called Kiva
05:22which was uh sold to Amazon I think in
05:25in like 2011 or 2012. yeah and they had
05:28designed these orange robots so it run
05:30all over the warehouse so basically go
05:32and lift up a bunch of shelves and bring
05:35the shelves to a human where the human
05:36could then grab things off the shelf and
05:38pack them and so as a way of automating
05:40warehouses moving things around inside
05:42warehouses and I remember seeing that
05:43thinking wow you know
05:46someone is going to design Kiva for
05:48outside the warehouse and that's going
05:50to be like a world-changing company
05:52because if you could just move things
05:54back and forth outside
05:56you know that would change the way the
06:00and uh so it's you know a simple idea
06:02turns out complex in execution and also
06:06I guess yet the last thing is that
06:08you know during that first year or two
06:09it became clear that if we were going to
06:10make this work like we'd have to commit
06:12Our Lives to it for 10 plus years and so
06:14we started asking ourselves like what
06:15problem would be important enough that
06:17we would be very excited to spend a
06:18decade by the way a decade has ended up
06:21being you know an underestimate because
06:22we're now 10 years in and like we're
06:25just you know just getting a lot today
06:26yeah we're like one percent of the way
06:30you know uh I think it was it was a
06:32combination of basically saying
06:34well hey like if we're going to commit
06:36ourselves to this we could pick a bigger
06:37scope we could pick a bigger ambition we
06:40wanted to pick a problem that would
06:41really suit the technology well
06:43um and we just had this sense that
06:45Logistics was letting people down that
06:46it could be so much better and more
06:50well that kind of answers one of my
06:51questions which is like how did you
06:53decide to work on blood and
06:56um medical supplies versus let's say
07:01well it wasn't it wasn't us but we
07:04didn't I mean I mean to a certain degree
07:06yes you know we we had a sense
07:09you know doing something like this
07:11for the first time ever you know
07:13everything's hard it's hard to make the
07:16unit Economics work it's hard to get
07:18regulatory permission it's hard to get
07:20the technology to do the thing and so we
07:23did have a sense that you know if we
07:25were going to start and launch this in
07:27the real world fast we would need to
07:28pick a use case that was incredibly
07:30important and clear to customers and
07:34for that reason it made a lot of sense
07:37to focus on Healthcare Logistics I mean
07:39zipline you know Healthcare Logistics
07:40today is still our bread and butter it's
07:42the majority of the deliveries that we
07:43do globally focusing on Health Care
07:46really enabled us to it really helped us
07:49on each of those fronts it was something
07:52that was super clear to our customers it
07:54was uh easier to make the unit Economics
07:57work on day one because there's more
07:58willingness to pay for a delivery that
08:00is literally going to save a life and it
08:02was easier to get regulatory permission
08:04because we would go hand in hand with
08:06the minister of Health to the aviation
08:08Authority and say you know hey here's
08:11the challenge hear the lives that are on
08:13the line we think it's worth taking a
08:15little bit of risk from a technology or
08:17regulatory perspective in order to save
08:19as many lives but I would emphasize it
08:22we were not smart enough to know like
08:24what we should focus on you know we
08:26I remember a conversation with the
08:28minister of health of Rwanda in 2016
08:31where we went in saying oh we'll deliver
08:34to every hospital and health facility in
08:36the country we'll deliver every medical
08:37product it'll be this full national
08:39scale Logistics system and she basically
08:42looked at us and said shut up just do
08:46uh and it was a good direction yeah and
08:50she was you know she was like I don't
08:51believe you this sounds crazy but she
08:54was willing to give us a shot so the
08:56government ended up signing a contract
08:57with us to deliver blood to 21 hospitals
08:59and as she was explaining it to us she
09:01explained you know blood is total
09:03Logistics nightmare but it's really
09:05important for Family Health so you know
09:0750 of blood transfusions are going
09:09toward moms with postpone hemorrhaging
09:1030 are going toward kids uh it's a when
09:14you need it you really really need it
09:16and it's hard because there are all
09:18these different uh components you have
09:20platelets plasma crowd precipitates
09:22packed red blood cells for a lot of
09:24those you have different types a b a b
09:27and O positive and negative Rh factor
09:30and then each of those components has
09:31different shelf lives and storage
09:33requirements so plasma Frozen lasts a
09:36year uh packed red blood cells
09:38refrigerated last 30 days platelets room
09:41temperature have to be constantly
09:42agitated and only last six days so it's
09:45a really really hard Logistics problem
09:46for a Health Care system
09:48and uh we we said yes you know we we
09:51said great we'd love you know we're
09:54happy to do blood we thought that was
09:55going to be easy it turned out it was
09:57not easy it was incredibly hard and the
09:59company almost died just trying to serve
10:00those 21 hospitals during the first year
10:03um and then you know over the last six
10:04years it's it's expanded dramatically
10:06I'm gonna talk about a little bit more
10:09yeah before we um get into that maybe
10:11you could just help our listeners
10:13visualize the problem of how a delivery
10:16actually works can you just walk through
10:17if it's you know the Ministry of Health
10:19or one of those hospitals like uh how
10:21one drone delivery happens today
10:24yeah the you know the idea is really
10:27simple it really is that any doctor
10:29nurse or you know Healthcare technician
10:31can push a button on a phone and summon
10:35you know we talk about what we do as
10:37teleportation it really means they can
10:38basically teleport a product directly to
10:41their GPS coordinates
10:43that's that's the vision it should it
10:45should you know feel like magical uh
10:48teleportation it should be nearly
10:50instantaneous and it needs to be
10:52reliable and that also means we need to
10:54be able to work in any weather 24 7. uh
10:57so what's practically happening behind
11:00the scenes to make that work is that uh
11:02you know any of our any of our customer
11:04hospitals so today we serve 3000
11:06hospitals and health facilities across
11:09um anybody at those facilities can press
11:11a button on a phone and place an order
11:13that order is then transmitted to our
11:16fulfillment center so each of our
11:17distribution centers zipline now
11:18operates uh I think by the end of
11:21January we'll be operating over 20
11:22global distribution centers each
11:24Distribution Center has a fulfillment
11:26area where we stock all of these
11:29different Medical Products blood was the
11:31first thing that we stocked uh so we get
11:33the order uh we will confirm it we'll
11:36pack the products whatever that is
11:38that's needed into a box under that box
11:40gets handed to flight operations flight
11:42operations then uh will pre-flight and
11:45aircraft load the box into the aircraft
11:51don't have runways where we operate
11:53might not be obvious so that means that
11:54we we launch using a catapult system we
11:57land using something that sort of looks
11:58like the landing system for an aircraft
12:00carrier you sort of have to see it to
12:04um but we'll put links in the show notes
12:05yeah but the you know the the simple
12:08vision is aircraft launches flies itself
12:11autonomously to one of these hospitals
12:13or Primary Care Facilities descends to
12:15about 30 feet in ever lands and never
12:18stops uh we drop the package using
12:20really simple paper parachute that
12:22enables us to deliver to a uh what we
12:25call their mailbox which is the size you
12:27sort of just think of it as like a
12:28imaginary rectangle on the ground that's
12:29the size of a couple parking spaces and
12:32then the aircraft comes comes home lands
12:34we can swap batteries and we can have
12:37the aircraft back in the air about two
12:38minutes later so our distribution
12:40centers were designed to do about 150
12:42flights a day for the first three years
12:44that was a total pipe dream everybody
12:45thought that that number was completely
12:46impossible we have distribution centers
12:48doing north of 300 deliveries a day
12:51so over the last six years it's gone
12:54from being this sort of Niche thing
12:55where figuring out the technology to
12:57then scaling getting to the point where
12:59you know we deliver a majority of a lot
13:02of these critical medical products in
13:03all these different countries where we
13:04operate and then today where it's the
13:07largest commercial autonomous system on
13:11so this is just unbelievable uh it's a
13:14first of its kind systems company it's
13:17um I'm sure I'm gonna miss components
13:19but you make a fixed-wing long range
13:21low-cost autonomous drone you're doing
13:22Hardware software sensors autonomy
13:24weather forecasting external UI for
13:27doctors internal apps Integrations with
13:29Hospital systems and traffic control I'm
13:31sure a thousand other things not to
13:32mention all of the massive operational
13:34capabilities on the ground you just
13:36described with distribution and
13:38fulfillment centers you mentioned the
13:40company almost died trying to serve the
13:42first 21 hospitals how big was the team
13:44that made the first delivery who did you
13:46have to make that happen
13:48when we launched in Rwanda in 2016 I
13:52think we're about 20 people and that's
13:55yeah and for sure that was a near-death
13:58experience and it was actually a
13:59near-death experience for exactly the
14:01sort of the reason you were just
14:02pointing out we didn't fully appreciate
14:04we needed to be how full stack of a
14:06company we needed to be we had sort of
14:08fooled ourselves into thinking well
14:09we've had this awesome vehicle it's
14:11going to work great like we can just
14:12launch and it turns out that you know
14:15the vehicle is only like 15 of the
14:17complexity of the solution here this is
14:18by the way a problem
14:21or a realization that I think most
14:24Robotics and autonomy companies have not
14:26yet had like you don't really realize it
14:29until you start trying to serve
14:30customers and suddenly have this
14:32really rude awakening that like the
14:35customer does not care how you know cool
14:38the robot is or you know how how even
14:41how capable it is like for our customers
14:43the only thing that matters is does this
14:45product go from A to B
14:48Ultra fast reliably 24 7. fast enough to
14:53and it turns out that all of the
14:56different layers of technology that have
14:58to be built in the background sort of
14:59running behind the curtain to make that
15:01overall customer experience work
15:04the Drone is 15 of the complexity
15:06there's so much complexity and a lot of
15:09things you mentioned you know Air
15:10Traffic Control software that we have to
15:12provide to the regulator computer vision
15:14based pre-flight checks data logging
15:16very unsexy but turns out it's a very
15:19hard problem you know each aircraft is
15:21generating a gigabyte of data that you
15:24you have to upload that data and use it
15:27to make the vehicles more reliable over
15:28time that's a hard problem in and of
15:29itself detect and avoid an autonomy
15:32solutions that enable the vehicles to
15:33communicate with each other and avoid
15:35each other in in an airspace and avoid
15:37other airplanes uh so there's a huge
15:40amount of complexity
15:42I think people just intuitively
15:45get very focused on you know the
15:47physical thing that looks cool which is
15:48yeah the autonomous airplane
15:50but uh the reality is that's a small
15:53part of what was actually required by
15:56uh so how did you decompose such a a
15:59hairy problem was there like you
16:01obviously discovered more of it as you
16:03got into it but was there an overarching
16:07uh you know there was no plan we were we
16:13severely naive when we launched in 2016
16:16there is no doubt about it I mean we
16:21we had done some testing in Half Moon
16:23Bay California and we thought we were
16:25ready to operate at national scale in
16:26Rwanda in any weather it sounds kind of
16:29silly when you describe it in retrospect
16:31but you know we we were so confident
16:33when we launched and for the first nine
16:36months we only served One hospital not
16:3821. you know we were wise enough to say
16:41well we're going to onboard the first
16:42hospital make sure it's working reliably
16:44for that hospital before we try to do
16:46we thought that was going to take two
16:47weeks it ended up taking nine months uh
16:50we were killing ourselves pulling
16:52constant all-nighters trying to fix all
16:55of these problems that suddenly became
16:56apparent when we started trying to
16:59deliver reliably to this one hospital
17:01and I don't think this was a philosophy
17:04like I think we were you know it's not
17:06like we were wise enough to know how
17:08hard this was going to be maybe if we
17:09knew how hard it was going to be we
17:11wouldn't have done it to begin with but
17:12uh I do think you might have thought you
17:14needed uh I don't know 200 more money
17:16more time more people but we didn't have
17:19any of those things you know we had no
17:21credibility nobody was going to invest
17:23money into the company
17:25um yeah you just don't have those things
17:27and you don't know what you don't know
17:28and I guess in this case you know our
17:31night it was lucky that we were so naive
17:34I think the reality is the thing that
17:37really did differentiate zipline during
17:38that year of extreme pain and near death
17:42and and constant all-nighters was
17:45you know we were never the biggest
17:47company in this space we were not the
17:49best funded company for sure I mean you
17:51know there was a big e-commerce company
17:52in Seattle investing billions of dollars
17:54trying to build similar technology
17:56I but we and we never even necessarily
17:59thought ever since ourselves as the
18:01smartest team but we were always by far
18:03the most practical like that was in our
18:05DNA we knew we wanted to get if we were
18:07going to work on this we were going to
18:08launch it into the real world and ask
18:10customers to pay us for the service
18:13with no excuses and the reality is
18:16especially with something that's
18:17complicated when you take the first
18:20version of your product and put it into
18:21a customer's hands and ask them to pay
18:22money it is a deeply humbling experience
18:25you instantly realize why the thing that
18:28you built in love sucks
18:30and is failing you know and that's like
18:35it's a deeply unpleasant upsetting
18:38um and I would say that is how it felt
18:40for you know the whole first year of
18:42trying to operate and falling on our
18:46making every possible mistake but
18:48interestingly all the things that we
18:50would have thought were going to be the
18:51problems ended up being irrelevant and
18:53all the things that really ended up
18:54being major problems or kind of screwing
18:57Us in terms of reliability or weather or
18:59pre-flight checks we weren't even
19:01thinking about them that's so
19:03interesting like what's uh what's an
19:05example of something you thought was
19:06going to be like fatally difficult and
19:11um I think that we invested way too much
19:15money in the early days on things like
19:18you know flight computer reliability and
19:20redundancy we designed this really
19:24uh sexy like you know two by two system
19:28where you know there's two flight
19:29computers running in lockstep and then
19:31and then you know another redundant
19:33flight computer so it's like if anything
19:34failed on the flight computer you know
19:36we felt confident that the vehicle could
19:38still get itself home and I mean to this
19:41day I'm not sure that we have ever
19:43actually had to use that architecture
19:45take care like it's just not the thing
19:47that ended up failing there are a lot of
19:49other things on the aircraft that end up
19:50so we ended up investing all this time
19:52building this you know sexy architecture
19:54on the flight compute side that went
19:55that wound up um we were off by orders
19:58of magnitude in terms of the things that
20:00were really going to hurt us what about
20:04oh there's so many things so many things
20:06I mean yeah where do you start but we
20:09could talk for an hour about you know
20:10the hard learnings from from that side
20:12but just to give a few examples I mean
20:14one data logging wound up being way
20:16harder than when we were anticipating uh
20:19and you know that sounds like trivial I
20:22mean it sounds trivial but it's
20:24definitely not it turns out it's hard
20:27um you know the maintenance of the
20:29vehicle wound up you know we had 43
20:31different kinds of Fasteners in the
20:33first design of the aircraft today's
20:35aircraft has two total kinds of
20:38Fasteners in the airplane uh you just
20:40realize when you're trying to keep 43
20:41different kinds of Fasteners in stock
20:43and then you cannot fly an aircraft if
20:46you run out of stock of one of those
20:47it's just you realize like okay we're
20:49this is a huge problem
20:51uh you know acquiring pre-flight checks
20:54and uh acquiring GPS pre-launch fast
20:58enough so that you you know you get an
21:00order you put a package into an aircraft
21:02and then that aircraft automatically has
21:04GPS lock and can launch turns out to be
21:06really hard and the entire vehicle has
21:08to be designed around optimizing for
21:12that problem and we did not optimize the
21:14first aircraft around that problem
21:15because we didn't know it was a thing
21:16reliability of the rotors and the
21:19Staters inside of the electric motor
21:22wound up being a major problem we saw
21:26some of the components inside the motors
21:28because we Airship the first uh 20
21:31aircraft to Rwanda I mean these are like
21:33really you know in the weeds
21:34detail-oriented things that we had no
21:36idea were going to hurt us and wound up
21:40life-threatening to the company
21:43um for us to learn these lessons and you
21:45just can't imagine it until you run into
21:46it you cannot imagine until you run into
21:48it and and that's the thing like we you
21:51know when we talk to robotics or
21:53Hardware companies today and a lot of
21:54them have raised large amounts of money
21:57I mean billions of dollars tens of
21:59billions of dollars in some cases and
22:01have yet to serve customers
22:03you just have this sense of like
22:06doesn't matter how much money you've
22:08raised it is going to be a rude
22:09awakening when you actually start
22:11charging customers for your product
22:12because that is when you actually learn
22:15all of the things that you need to fix
22:18and all the things that are going to be
22:19super important to build
22:21and so I do think the thing although is
22:24incredibly painful the thing that really
22:26enabled us to survive that period was
22:28that we we were very practical and
22:31unfancy and we got things into customers
22:33hands quickly and we learned by doing
22:34and we assumed that we were idiots and
22:37it turns out that that assumption was
22:39100 correct and even to this day
22:43well I that's how we talk about it
22:46and you know even to this day when I
22:48talked to new zipline team members I
22:49mean zip lines now almost a thousand
22:51people um but you know when we talk to
22:53new zipline team members at distribution
22:54centers or joining the engineering team
22:56we're basic we basically we say assume
22:58they were idiots because we're proven
22:59correct more often than not uh you know
23:02your customer will tell you what really
23:03matters like we don't know anything and
23:06so even even six years in seven years
23:08into like commercial operations when
23:10we're launching new products or new
23:13we always learn hard lessons when we
23:15actually get the product into people's
23:18okay Keller for you know fast forward
23:20six or seven years a lot of this stuff
23:22actually works pretty well now at
23:24zipline assume I'm an idiot and I'm
23:26going to ask you a very basic question
23:28we have this largely technical audience
23:30but even for a lay software engineer
23:32it's not obvious what the difference is
23:33between like Ai and autopilot which
23:36every commercial flight uses some degree
23:38of already I recently took like a small
23:41plane to Serious Vision jet that only
23:43requires one pilot because it's got this
23:44like cool automatic emergency landing
23:47system where are we in terms of aircraft
23:50and drone autonomy and that might be
23:52like a stupid question asked because
23:53those might be very very different but
23:55you know what are the parts of flying
23:57that existing autonomy systems can't
23:59handle where where you guys had to do a
24:02yeah the uh the advantage of uh flying
24:08in the air is there are not very many
24:09things to hit so actually I think this
24:12might be counter-intuitive but
24:14you know the self-driving problem in the
24:16air is way easier than the self-driving
24:18problem on roads like people I might
24:21have this sense so you know airplanes
24:22even harder than cars actually
24:24definitely not like thousand times
24:27uh and the reality is much like uh the
24:31self-driving problem on the ground you
24:32know certainly there are all kinds of
24:34cool use cases there's a lot of cool
24:35research going on about using you know
24:39neural Nets and uh you know doing
24:41Advanced State estimation and motion
24:45planning and it's called slam and
24:46Robotics so you know building control
24:49algorithms State machines
24:52um and then building computer vision
24:53systems they can help you know or or
24:55light our systems that can help like
24:57model and estimate the environment
24:59around you and recognize you know when a
25:01kid runs in front of the car or when you
25:04know another airplane is flying too
25:06close to you in the air like these kinds
25:08of autonomy problems are still very hard
25:12um and are not solved at like human
25:14safety level in either direction but the
25:17good thing about the air is that the
25:20chances are so vanishingly small of
25:22having an air-to-air I think the air is
25:24a really big place the sky is a big
25:26place you know so the reality is you
25:29know we were able to operate at
25:31multinational scale without having to
25:36like Cutting Edge computer vision
25:38problems uh and I think that wound up
25:42being a good choice you know we thought
25:44like the company that kind of inspired
25:46us on this front was Tesla and we saw
25:48like Tesla and then Google we were
25:51looking at right and Tesla had kind of
25:53come up with had said we're going to
25:55uh what I think they thought at the time
25:58was a dead simple product just like put
25:59a battery in a car and then like sell
26:01the car right don't try to design an
26:03autonomous car from scratch and then
26:05there was like the waymo approach of
26:07we're gonna build a fully autonomous car
26:10it's a total Moon shot and we're not
26:11gonna like sell a single thing to a
26:13customer until it's working and so
26:16felt 100 the right path was like the
26:19Tesla path launched something that's
26:21Ultra simple that we felt confident we
26:23could get into the real world quickly
26:24that we could make money on and learn
26:27from and then start to integrate
26:29autonomy thereafter and I think in
26:33retrospect that was definitely the right
26:34choice there are other companies in you
26:37know the drone delivery industry that I
26:38think have taken more of a moonshot
26:40they've spent billions of dollars and
26:43have not served a single customer yet so
26:46the advantage is now you know zipline
26:49has been able to get to this scale where
26:51our ability to collect data our ability
26:53to know what the actual problems are our
26:55ability to know what our customers want
26:57and are willing to pay for
26:59um is much higher and we're now
27:01integrating you know new autonomy
27:03systems into the overall
27:06service network every you know every
27:08quarter we just announced this new
27:10detect and avoid technology and kind of
27:11our zipline's full autonomy stack last
27:13year we are going through a
27:16certification process with the FAA for
27:18that autonomy stack as we as we speak
27:19can we talk about that a little bit um
27:22see you mentioned it before it's a it's
27:24an acoustic based detect and avoid
27:26system which is a new idea
27:28um I think a lot of people who might
27:29come from the the software side or you
27:32know other possible autonomy might be
27:33like radar lidar et cetera for context
27:36for our listeners problematically in the
27:38US a significant number of aircraft
27:40don't have to carry transponders which
27:42are like little radios that broadcast
27:43your presence and location other
27:44aircraft and Keller am I correct that
27:47drones don't have any right-of-way
27:49you know in airspace that's exactly
27:51right and so it's their job to avoid all
27:53these other aircraft but you don't know
27:55where they are so person looking around
27:57for them can you walk us through just
27:59like how you're solving this issue and
28:02like I think people can infer you're
28:03basically listening for aircraft that's
28:06it is it's a hundred percent wild uh so
28:10detect and avoid is kind of what you
28:13would call the problem it's you know
28:15detect and avoid has been a problem uh
28:17for autonomous flight in the US for 20
28:20years because as you said you know a
28:23unmanned aircraft does not have any
28:25right-of-way the assumption is you have
28:27to be the one avoiding everybody else
28:29and by the way the detect and avoid of
28:31like a human pilot sitting in a cockpit
28:34is also extremely poor I don't want to
28:37freak anybody out who you know flies in
28:39aircraft a lot but the reality is that
28:42um if you're in an unmanned vehicle you
28:44have to solve the problem you and and
28:46you're right you know transponders
28:49um are you know technically required but
28:51compliance is very low especially in um
28:53GA you know general aviation and so you
28:56can't rely on transponders you have to
28:58be able to know what you know you have
29:00to be able to see something when it
29:01enters we call it the you know the two
29:04mile hockey puck it's a you know a two
29:06mile hockey puck around around your
29:08vehicle you need to know if something
29:10enters that hockey puck and then you
29:12need to be able to actively avoid it why
29:14is that called a hockey puck
29:16um because it's like a circle around you
29:18but then it's a uh you know a set amount
29:21of space above you and Below you so
29:23imagine the shape of a hockey puck yep
29:26um yeah and a lot of companies have been
29:28trying to solve this problem for the
29:30last decade it's a really hard problem
29:33there's just no really good sensing
29:34solution you can imagine lidar radar
29:37computer vision you know cameras each of
29:40those I mean lidar and radar very heavy
29:42very power consumptive they're expensive
29:45yeah very expensive and like the you
29:48know the the fee the field of view is
29:50not very I mean radar you might have
29:51like 30 degree you know field of view
29:54and so then how many like radar systems
29:57you're gonna have on your vehicle and
29:58you know we're talking about a fit an
30:00aircraft that weighs 50 pounds in total
30:01and you care a lot about payload right I
30:04mean we're optimizing we spend huge
30:06amounts of time optimizing for ounces I
30:08mean in our case grams you know we but
30:12spend a lot of time basically every gram
30:15matters and so you have no weight to
30:16spare as an electric aircraft like the
30:18battery is heavy you want to deliver a
30:20lot of payload there's no you care a lot
30:23about weight and power consumption and
30:26uh and so radar lidar like getting those
30:29to provide a really good 360 view that's
30:33it's doesn't work very well and then you
30:35can say well cameras they're lighter and
30:36cheaper yes but they don't work in
30:38clouds so basically you know immediately
30:41you can rule out cameras as a sole
30:44solution because you need to work in
30:45clouds like you know clouds are a common
30:47thing so you know about four years ago
30:50an engineer on the team as we were
30:53basically realizing we got to go solve
30:54this problem like there's no technology
30:56available off the shelf that's going to
30:57solve this problem and zipline is at a
30:59scale now that is you know a hundred
31:00times the next closest player and so
31:02it's like not not like anybody's going
31:04to build this for us like we had to
31:05build it for ourselves and a member of
31:07the team basically said
31:09well I ever you know had this really
31:11stupid idea which was like what if we
31:13just listened what if we could just
31:15listen for for other aircraft and then
31:18triangulate them you know in the way
31:20that you can imagine like our ears
31:22echolocate we're not that good at it but
31:24yeah we literally echolocate yeah like
31:28yeah well a dolphins actually
31:30um active acoustic Acoustics yeah yeah
31:34Sonar is a little different
31:36um but uh yeah so you're you know can we
31:39passively listen and we thought and we
31:40basically heard that and we're like yeah
31:42that for sure will work so let's just
31:44quickly disprove it via you know some
31:46engineering prototypes and we then did
31:49some experiments and we failed at
31:51disproving the idea in those early
31:52experiments and so we said okay well why
31:54don't we put like a two-person team on
31:56this uh because we're pretty sure like
31:57the first prototype will show us why it
31:59won't work and then first prototype also
32:01failed at showing us why it wouldn't
32:02work and it kind of became like the plan
32:05of record for us and over the last four
32:06years you know we invested a lot like we
32:08had a large like mechanical engineering
32:10signal processing uh machine learning
32:14team working together to figure out how
32:16to build a microphone array make it work
32:19when it's traveling through the air at
32:2180 miles an hour which is not an easy
32:23thing in and of itself like if you
32:24imagine you know putting this microphone
32:27like and having to travel through the
32:29air and yeah you can imagine what that
32:32um so you're really hard mechanical
32:34engineering problems you know microphone
32:36design problems there are hard signal
32:38processing problems there are hard
32:40machine learning problems speaking of
32:42that did you did you collect your own
32:44data for this because I don't like you
32:46know there's not like good location
32:49trajectory audio at 80 miles an hour
32:53pairs just sitting out there on the
32:54internet for you yep 100 and so we had
32:57to collect you know we had to collect
32:59our own data by flying a lot of
33:00microphones on airplanes but we also
33:02collected a lot of data by designing
33:04this little system that we had at our
33:06office where we would literally
33:07literally just had a microphone array
33:09sitting on the ground 24 7 recording
33:11airplanes flying overhead uh so
33:14yeah we had to build the data set
33:16ourselves but I mean the amazing thing
33:18is that four years in you know we rolled
33:20the hardware into full production uh and
33:23then you know we turned it on in Shadow
33:25mode across a lot of the countries where
33:27we operate and we've now gotten active
33:28permission from regulators and are
33:31actually flying using it where it can
33:33now control the aircraft if it needs to
33:35to deconflict from another airplane uh
33:37and you know that system it actively
33:40listens it can identify the make and the
33:43model of an aircraft is that accurate
33:45because it turns out these you know
33:48these neural Nets if you train them on a
33:50lot of different aircraft and what they
33:51sound like it's really good at telling
33:54um what it's hearing and where the thing
33:56is in the hockey puck and the beauty of
33:58it of course is that microphones are
34:00extremely cheap they weigh almost
34:03um and it is naturally a 360 solution
34:05because that's how Sound Works so super
34:07counter-intuitive and honestly I think
34:09most people would have just rejected
34:10that idea out of hand because it sounds
34:12dumb but it turns out that it solves
34:14problem and that's what's going to
34:16enable zipline to fly full scale Beyond
34:21um in in anything yeah and in class
34:24let's zoom out and talk a little bit
34:26about zipline as a business so
34:27counter-intuitively to perhaps many
34:29people like me and you that kind of live
34:31in the Bay Area and it's a hotbed for
34:33Innovation the government of Rwanda was
34:36her first customer like why do you guys
34:38start in Africa we have Healthcare
34:39deserts and access issues in the US too
34:42well uh when we knew we wanted to start
34:45in healthcare we thought that it
34:46probably made sense to start with a
34:48public health care System rather than a
34:49private Health Care system because you
34:51know advantage of a Public Health Care
34:52system is you can go and work directly
34:54with the country and then you serve all
34:55hospitals and Primary Care Facilities
34:56whereas in the U.S we have so many
34:59different Health Systems it's really
35:00complicated you have to work one by one
35:02and so and we had a sense that we
35:06probably needed work to work with a
35:07country that was small Innovative and
35:10agile in the same way that we were and
35:13so that led us to focusing on some of
35:15these really high performing Public
35:17Health Systems in Africa we felt like
35:20the problems were super clear
35:22we had a government that was as
35:24Innovative and entrepreneurial as we
35:27and they were willing to move quickly
35:30and make regulatory exceptions also in a
35:32way that the FAA just can't do so all
35:36three of those things wound up creating
35:38a a strong partnership between us and
35:41the government of Rwanda in 2016. that's
35:43a partnership that you know we have been
35:46grateful for I mean they put up with a
35:49lot like we were we had no idea what we
35:51were doing in the first year you know as
35:53I mentioned we were trying to serve 21
35:54we served One hospital for nine months
35:55they were very patient with us they knew
35:57we were trying to do something for the
35:59first time in the world ever and that it
36:00wasn't going to necessarily be a smooth
36:03um but you know looking back I mean just
36:06a couple weeks ago we announced this new
36:09national scale partnership that we
36:11signed with the government it's a 61
36:12million dollar partnership zipline is
36:14now delivering all medical products to
36:16every hospital and health facility in
36:18the country and we are the largest
36:20Logistics Network in the country we
36:23not only deliver all Healthcare products
36:25but and we began delivering medical
36:27products to people directly to patients
36:29homes then we began delivering a lot of
36:31other products you know we're now we now
36:33deliver a wide variety of agriculture
36:34products that increases productivity of
36:36farmers decreases childhood malnutrition
36:39brain stunting cycles of poverty
36:42um we're working with them around
36:43providing an e-commerce solution you
36:46know it's becoming full national scale
36:48infrastructure and by the way you know
36:51as part of that Rwanda uh made I think
36:53what's the largest investment into any
36:55private company the country has made
36:56into zipline so is now a shareholder in
36:59the company so you know seven years in
37:01it's been it's been an amazing
37:03partnership we're really really proud
37:05that we've been able to kind of learn
37:08you know now so many other countries are
37:10following suit I mean zipline launched
37:11in Ghana today we serve I think 1800
37:14hospitals and health facilities in Ghana
37:16uh late last year we launched in Nigeria
37:18in January we're launching in code like
37:21this month we're launching in Cote
37:22d'Ivoire in Kenya and we began operating
37:24in the U.S we now have three
37:25distribution centers in the U.S so yeah
37:27started small but it you know I do think
37:29Ziplines survival depended on finding a
37:32partner that would move equally fast and
37:38so what is the biggest challenge to you
37:41know getting zipline everywhere for more
37:44more Goods is it make the system
37:46magnitude cheaper is it just engagement
37:53I think the meta challenge is that
37:55Hardware is hard you know and especially
37:58when you're a full stack company where
38:00you design the vehicle you manufacture
38:02the vehicle and then you operate the
38:04entire vehicle and service for customers
38:07there's a lot that goes into that
38:09zipline has to be really good at supply
38:11chain we have to be really good at
38:13supplier industrialization engineering
38:14we have to be really good at
38:16manufacturing we have to be really good
38:17at Logistics we have to be really good
38:18at all the different kinds of
38:20engineering right whether it's
38:21electrical engineering firmware
38:22engineering software engineering
38:23mechanical engineering error
38:25aerodynamics Aero Acoustics and then we
38:28also have to be really good at
38:29operations like we have to know how to
38:30operate fulfillment centers we have to
38:32know how to get regulatory approval in
38:34the countries where we operate we have
38:35to know how to do flight operations we
38:36have to do maintenance so I think the
38:38challenge is that this kind of
38:40infrastructure is complicated and it
38:42it's not like a software company where
38:44you can just scale 100x in a year
38:47you know it's sort of similar to Tesla
38:48where Tesla has been in this crazy
38:51um massively Supply constrained for five
38:54years because it's like you can only
38:55build factories so fast but I think the
38:58flip side of that is that although
39:00Hardware is hard I do think that those
39:01companies wind up being some of the most
39:03defensible companies on earth like they
39:05have very powerful competitive modes
39:08um because you know really hard to
39:10compete with a hardware company at
39:13massive scale it's the reason it'd be
39:15hard to build a new kind of smartphone
39:17and compete with Apple
39:19um so certain things about building the
39:21company are really hard but I think the
39:24I do think a lot of the most important
39:26companies for Humanity that are going to
39:28be built over the next 20 years are
39:29going to be hardware and infrastructure
39:30companies they're more Capital intensive
39:32they're harder to build but they also
39:34have way more momentum and inertia and I
39:36think powerful competitive advantages
39:39and they're worth they're worth building
39:41right right as an investor I'm excited
39:43to back them yes I mean I think so like
39:46I I although I feel like there are a lot
39:48of investors who I you know I'm speaking
39:51to a venture capitalist but I do think
39:53that over the last 20 years
39:55I think a lot of America and investors
39:59like I feel like we lost our vision for
40:01the future you know I grew up
40:04watching Star Wars Star Wars and Star
40:06Trek I was a big Trekkie huge nerd me
40:08too like the vision that we were
40:12of Robotics and you know Automation and
40:14nuclear fusion and genetic engineering
40:17and life extension you know and some of
40:20that's working Interstellar space travel
40:22I think you're right yeah absolutely I
40:24think it is but I think in general you
40:26know the tech industry in the US uh was
40:29a lot more focused at least for a you
40:32know a decade or two on smaller ideas of
40:36you know incremental SAS uh products and
40:39things like that and it was impossible
40:41to raise money for it was very difficult
40:43to raise money for zipline for the first
40:45six years right like I I mean I think
40:47it's only recently that investing in
40:49robotics and genetic engineering has
40:51become a thing so I think in general the
40:54exciting vision of the future is not one
40:57where you know we're designing like the
40:59metaverse and you know nfts and we're
41:02all like living in a digital world well
41:04actually like depending on the crumbling
41:06infrastructure that our grandparents
41:09like that's crazy why is the US not you
41:11know why are we not building
41:12more highways more tunnels you know more
41:15new new kinds of aircraft new kinds of
41:18spaceships like there's so much to to
41:20build and I think that we lost our
41:23imagination and appetite for building
41:25things in the real world a little bit at
41:27least and I I but I agree with you I
41:29think that that's changing and I think
41:30that that's really exciting and that is
41:32the thing that you know it inspires us
41:34every day it's like let's go build
41:36National or multinational scale
41:38infrastructure that can save millions of
41:40lives and make the world a more equal
41:41place like what's cooler than that
41:43well you're an inspirational
41:45entrepreneur for I think the Next
41:47Generation Keller um and I I hope to see
41:50many more companies that you know have
41:52those goals maybe one last question for
41:54you about the business 2022 was a big
41:57year you uh you know expanded geography
41:59as you described you got that F of a a
42:02certification you signed a deal with
42:04Walmart I believe you made over 200 000
42:06deliveries which is more the last five
42:08years combined you hired uh Deepak Ahuja
42:11former Tesla CFO among other amazing
42:14talents like what can you tell us about
42:17yeah I mean I think I was just talking
42:19to the team about this I mean 2023
42:20zipline is definitely in the middle of a
42:22very intense growth period
42:25um and I don't know I guess our timing
42:26stinks because we're doing it during a a
42:28large you know macroeconomic recession
42:30but actually the reality is you know the
42:33the the forces that are causing the
42:35recession also are sort of accelerating
42:37zipline because inflation like the
42:39increasing cost of Labor and gasoline
42:41are the things that sort of cause
42:43customers to adopt our business partly
42:45for that reason 2023 is going to be an
42:48intense year of growth we're going to
42:49increase uh you know total flight volume
42:52by 400 percent in 2023 based just on
42:56contracts that are already signed
42:58um so that means we'll do about two
43:00times as many autonomous deliveries in
43:032023 as we've done in the history of the
43:04company since its founding
43:07so you know there that's a lot of work
43:10um handling that kind of scale is hard
43:11things break basically every single day
43:13when you're growing that fast we're also
43:16you know dramatically expanding you
43:18mentioned the partnership with Walmart
43:19we've signed a lot of other Partnerships
43:21with big players in the US Intermountain
43:23Healthcare Avant Healthcare Multi-Care
43:26Michigan medicine these are all big
43:29Hospital systems that are now relying on
43:31zipline to take over to basically build
43:33autonomous instant delivery directly to
43:35homes we really think of teleportation
43:37as just the other half of telepresence
43:39so if a health system can if you can
43:41pull out a phone and talk remotely to a
43:43doctor and that doctor can diagnose you
43:45the health system should then be able to
43:46say awesome we know what you need it's
43:49going to be on your doorstep in five
43:51that that's expanding the service in the
43:54US is is kind of a big priority but as I
43:57mentioned I mean we're building four new
43:58we have four new distribution centers
44:00coming online just in January
44:02um to a Nigeria one in Kenya one in Cote
44:04d'Ivoire it's a huge amount of work we
44:08um well there's a really big new thing
44:11that we'll be announcing in a couple
44:12months that I you know can't can't talk
44:14too much about but yeah something that
44:16we've been building for the last three
44:17years and that I think is going to be
44:18quite transformational to the logistics
44:23um we'll look out we'll look out for
44:25um yeah I hope you you won't mind me
44:27seeing so but there are a lot of ways in
44:29which a zipline as a company is kind of
44:30unlikely right as you said everything is
44:33hard for those Founders or so many of
44:37our listeners are founders who may want
44:39to tackle similarly audacious problems
44:41instead of um perhaps something trivial
44:43and incremental what what advice would
44:46you offer them about a secret that has
44:47made zipline work today I heard one
44:50thing from you which is B ruthlessly
44:53yeah I think that zipline was 100
44:57improbable there's no doubt about it and
45:00when we started building the company you
45:01know I told the team I remember we were
45:04at a Christmas party where we all fit
45:06around one table at a little Hole in the
45:07Wall Chinese Restaurant and
45:10I think someone on the team asked like
45:11well what do we think are the chances of
45:12success and I said I think the chances
45:14of success are about one percent
45:16and everybody was kind of upset but then
45:18I was like but guys it's it but you know
45:20it's one percent of of a totally
45:23world-changing I mean if we actually
45:25succeeded at automating Logistics and
45:28making it serve all people equally and
45:30providing universal access to health
45:32care to every single person on Earth and
45:34transitioning all of Last Mile Logistics
45:36to zero emission I mean
45:39that's that would be so world changing
45:42I think it takes a special kind of
45:44person to want to work on
45:47a project that has a one percent chance
45:49of success and zipline was thoughtful
45:51about making sure that we hired people
45:53who were okay doing something that was
45:55risky and where failure was a very real
46:02um but that doesn't mean we shouldn't
46:03work on those things and also you know a
46:06lot of those ideas that are most
46:08challenging most ambitious most scary
46:10there's less competition
46:12yeah zipline uh today I mean
46:16yeah we don't we look around it's like
46:18no other companies are operating at
46:20anywhere near the scale right and so
46:22it's I think you know choosing these
46:24less ambitious more incremental problems
46:27um you end up being in Perfect Price
46:29competition with a lot of other folks
46:31working on the exact same thing whereas
46:33you know setting your
46:35sites a little farther ahead and saying
46:38hey we think the world is going to be
46:39this way in five years or 10 years and
46:41you know we'll work in unfancy Scrappy
46:44practical ways for a decade I do think
46:47you know that's is I mean you can build
46:49really important companies that way and
46:50I think that there are definitely
46:52investors out there who are willing to
46:55fund those kinds of ideas because I
46:56think those kinds of ideas are often
46:57become the biggest companies so I guess
47:00that's my advice it's like I don't know
47:02that it's for everybody but I also think
47:05I mean I guess I think the biggest thing
47:07that I would say is technology has a bad
47:09rap like we can look around and see the
47:11biggest problems that are affecting
47:14and I think unfortunately the way the
47:16world kind of thinks about it is like oh
47:18you know technology is going to serve
47:19the richest people on the coast of the
47:22United States and then all the problems
47:24that really matter for Humanity like
47:26childhood malnutrition and brain
47:27stunting you know reproductive health
47:34you know energy and sustainability like
47:38all of these problems actually in many
47:40cases get left up to like government or
47:42non-profits which is really bad because
47:45government and nonprofits are are bad at
47:47solving problems and
47:49are incapable of deploying technology
47:52sorry to be I mean that's not no I
47:56vast majority of the time yeah okay so
47:59I'm like you know being a little bit
48:01provocative but uh I think that the
48:03reality is like all of these problems
48:06that impact six billion humans plus and
48:09that like are crying out to be solved
48:12for Humanity at global scale like they
48:15require engineering they require
48:17technology they require sustainable
48:20business models so that we can actually
48:22scale a solution once it's built and I
48:26you know having the smartest engineers
48:30and entrepreneurs graduating from the
48:32best colleges and then going to work for
48:34you know a large search engine to help
48:36them sell like point one percent more
48:38ads like it's not actually a great
48:40application of Humanity's Talent one of
48:43the most important things that we have
48:45you know tried to prove is that you can
48:48build a multi-billion dollar company
48:50focusing on important problems for
48:52Humanity and you you know and there's an
48:54opportunity for like the smartest most
48:57driven engineers and entrepreneurs and
48:59operations specialists in the world to
49:02work on problems that really matter and
49:04and you know will make the world a
49:06better place if they can be solved and
49:08by the way showing that that's not like
49:10that's not about philanthropy the LA you
49:12know the thing that drives me crazy a
49:13lot of people look at zipline and
49:14they're like oh it's so generous of you
49:16it's so philanthropic that you operate
49:17in these countries and we're always like
49:18that is exactly the wrong way to
49:21understand what we do you know we make
49:22money in all of these countries it is
49:24gross margin profitable at almost all of
49:27our distribution centers and the whole
49:31like sustainable unit economics building
49:34a profitable business is how we actually
49:36scale solutions that can work and and
49:38and solve problems for like six billion
49:40people so we believe very strongly and
49:42like entrepreneurship and you know as an
49:45innovation yeah as an engine to solve
49:48these problems and to make the world a
49:56it's all we can do today I was so
49:58looking forward to this interview thank
50:00you for joining us on the podcast thank
50:02you so much for having me
