00:01hi this is Chris Dixon this is the a 16z
00:03podcast i'm here with benedict Evans and
00:06Balaji Srinivasan so we've talked about
00:09this concept a lot internally we call it
00:11full stack startups I wrote a blog post
00:13about it recently I guess just to get
00:16out of the way because people say what
00:17isn't just just another fancy word for
00:19vertical integration let me define what
00:22I mean by these which are basically
00:24there's sort of a new kind of trend in
00:27startups to instant it to have a kind of
00:31a core insight that that's around
00:32technology but then to build a company
00:35that does multiple activities beyond
00:38just that technology not just sort of
00:40you know so instead of building software
00:42and then going out and selling that
00:43software building a complete service
00:45around it that includes things that we
00:48would traditionally not call technology
00:50functions in order to go to market so
00:52I'll give some examples so you know uber
00:55I think is a good example where prior to
00:58BRR there were actually companies out
00:59there that built software for that they
01:01tried to license to taxi companies that
01:04would help you know do automated
01:06dispatching and things for a variety of
01:08reasons those companies were not
01:10technology forward-thinking you were
01:12selling them something that they was
01:13going to replace part of their staff
01:14maybe even the people buying it they
01:16didn't catch on and then you have
01:19someone like Cooper come along who who
01:21does it sort of in this full-stack way
01:22they actually you know pay the drivers
01:24and do all these things we don't think
01:25of it's technology companies you know
01:27Tesla building a full car versus a lot
01:29of companies before they tried to just
01:31build electric batteries and sell them
01:33you know you take Netflix they they're
01:36now you know both funding content movie
01:39development they're they have a website
01:41where you can discover it and then they
01:43also you know and you can watch it and
01:45they're sort of providing the whole
01:46stack there whereas prior to them there
01:48were lots of companies that tried to do
01:49things like streaming services that they
01:52license to you know Hollywood
01:53distributors or something or
01:55recommendation services for Hollywood's
01:57websites or something so Netflix going
01:59to kind of full stack another example
02:00might be BuzzFeed BuzzFeed actually
02:02began as a company that was trying to
02:05provide services for media properties to
02:07be more viral and then figured out that
02:10actually it would be better to apply it
02:11to their own media property and build
02:14so just a variety we recently announced
02:16our investment in altschool which is an
02:18attempt to kind of instead of taking
02:20technology and selling it to other
02:21schools actually build a new school
02:24using that technology so I call these
02:28full stack startups I blogged about it a
02:30lot people say isn't that vertical
02:31integration I think vertical integration
02:33is a term I think you could argue this
02:35maybe it's a boring point the reason one
02:37I think vertical integration is a term
02:38that maybe is overloaded and most
02:41people's minds refers to kind of non
02:43technology companies that might go and
02:45acquire their suppliers or buyers so you
02:47know Standard Oil or something like this
02:49I think it's a you know I think my mind
02:52that's what it implies to people it's
02:54good to have a new term number two I
02:56think some of these functions that these
02:57full stack companies are doing aren't
02:59really sort of strictly the buyer
03:01supplier it's sort of more almost like
03:03horizontal integration in some cases so
03:06I thought a new term was warranted yeah
03:07maybe it's not but regardless I think
03:10it's a new phenomenon yeah so briefly on
03:12that point just for a second so I think
03:14there's at least two reasons where I
03:16think the term full stack is interesting
03:17so one is that it is sort of in keeping
03:20with just how people think about
03:22software right you've got a full stack
03:23developer who can do the front end the
03:25back end everything in between and the
03:28interesting thing about software
03:30companies in particular is that they
03:31have the option to build rather than by
03:34many more things than previous kinds of
03:36companies is that to say many more
03:37things are within their core competency
03:38as software eats the world acquiring a
03:41new skill is learning a new API or being
03:44able to develop a new thing is learning
03:45a new API frequently and so they have
03:47the option to do more of the stack
03:50that's that's kind of point number one
03:51and put number two is that also means
03:53that they would grow into these other
03:55layers by building more than acquisition
03:57and more than the earlier things so
03:58that's kind of I think one of the
04:00reasons this is interesting the third is
04:01that after kind of a decade in which
04:05every layer of the stack the computing
04:11stack was sort of split off and so you
04:12have CloudFlare and you have AWS and you
04:15have all of these services which kind of
04:17are based on a fair to be or maturity
04:21of for example x86 and arm architecture
04:23and so and so for there's some degree of
04:24maturity of those individual pieces so
04:26you can you can in theory focus on a
04:28very thin slice and just work on that
04:30but when you're doing a real-world
04:31applications almost the opposite right
04:33so you actually want to be very
04:34integrative and and I think extremely
04:37importantly when you're going into a new
04:38industry where software has not actually
04:40been present for a while often it is
04:42because there's a few different parts
04:43that need to get or a few different
04:44layers that need to be replace at the
04:46same time now with that said one thing
04:48that I said it's interesting yet so it's
04:49you're saying it's sort of the so twenty
04:52years ago if you wanted to build a you
04:54know Amazon had to go build their
04:55Williams let's say whatever random
04:57startup had to go build their own Cola
04:59or their old datacenter and do the whole
05:01thing and then as each of the tools
05:04matured you got any good api's you
05:07didn't you could build a thinner slice
05:08and now we're sort of seeing the same
05:10pattern repeat in the exact world Adams
05:12that's exactly right and I think that
05:13then again in like 10 or 15 years you'll
05:16be able to build a logistics company on
05:18top of ruber or lift for logistics or
05:21something like that because those things
05:22will also be produced api's but I think
05:24when you first move into a new vertical
05:26there's so much stuff that is broke and
05:28there's reasons that software is not yet
05:30even that vertical and you often need to
05:32go after a few different components at
05:33the same time so with that said one of
05:35the things that I noticed in some like
05:37the comments all said this like you know
05:39long Twitter series is it a mark a P
05:42Market tweet storm 4141 tweets but you
05:47know the reason I like actually those
05:48like multi-part you know tweets is that
05:51blog yeah well so I would blog except
05:55first you can tweet like in the car and
05:57so you can tweet partially you know you
05:59can you know tweet in the house you can
06:00tweet like a mouse like a dr. Seuss kind
06:03of thing but second you can also just
06:04cut it off at any time okay so that's
06:06that's like it's you know blah gift edit
06:08it more but Haman they're all drafts of
06:11like MOOC lectures but regarding the
06:14some of the responses to the chair pose
06:17I saw a lot of people who are saying oh
06:19you know the : oh yes I want to build a
06:21uranium mine now you were telling me yes
06:23we can go in fund it
06:24I'm not against like extremely ambitious
06:26initial things but I want to clarify
06:28that doing a full stack startup does not
06:30mean like a takeover for example yes
06:34they they did the the full stack but
06:36they didn't buy a hundred thousand cars
06:38they you know they figured out okay here
06:41is an ultimate ambition maybe they
06:43weren't even thinking about a hundred
06:44thousand cars at the beginning they
06:45picked the minimum number of layers to
06:47replace at the beginning they had the
06:48ambition of doing the whole thing
06:49ultimately I think that's really what
06:51distinguishes it you still want to have
06:53a Minimum Viable Product it just maybe
06:55it's not quite as minimum but you you
06:58want to have a plan for games there's
06:59other layers and in the interim you use
07:01the stopgaps use the things that are
07:03commercial off-the-shelf uses things and
07:05some of these things like that like
07:06Tesla 2 versus a 1-2-3 batteries or
07:09something like the Tesla you just have
07:10to remember a product is a really nice
07:12car like right not it's pretty ambitious
07:14so but even then you can stage it you
07:16can say hey hey look we're gonna show
07:17the engine works ok that's good and they
07:18start at the high end for like the niche
07:20yeah a few people or something exactly
07:22presumably they're gonna have a very
07:23mainstream for examples of cars some
07:26point like you know even starting with
07:27I'm sure you know I didn't see Tesla's
07:29you know roadmap and so and so forth but
07:31I'm sure that there was staging where
07:33they said okay first let's prove out
07:34superchargers in California right and
07:37let's do it and even Bill net because
07:40they you know they started the first car
07:42they didn't make half of the components
07:43since I've been a steady load yes okay
07:46you know they start with low well let's
07:47do a roadster and we'll buy in half a
07:49bit and then you kind of step slowly
07:52through and you know there's an end
07:53point somewhere when Ellen Ellen masks
07:55mind that you know he's making as many
07:57cars as Toyota yes you know how do you
07:59get there I mean there's a sort of the
08:01census isn't really unique technology
08:03and there's always been kind of asset
08:04heavy and asset like models in here you
08:06know obviously but doesn't own the cars
08:08um and it's interesting to look at
08:10nobody cars all there now but they are
08:12thinking about leasing but it's
08:13interesting looking at air B&B and
08:16comparing that with say a big hotel
08:17chain like you know a merit or Hilton
08:19because they don't know in hotels you
08:22know they've got you know dozens or
08:23hundreds of franchisees who've gone out
08:25or individual real estate developers
08:27who've gone out and raise the money and
08:28so there's always that kind of tension
08:30of well how far into that you actually
08:32need to go if you look at Apple we would
08:35think of Apple as being the ultimate
08:36integrated company but they don't own
08:38any of the factories I mean they you
08:39know they own a few things but you know
08:42most of the places where iPhones are
08:43made they don't know and they put
08:45capital in all their I think well
08:48now where do you where do you put the
08:49wetter you draw the ownership line
08:51weight you'd put the inventory where is
08:52it hell's how what can you do you know
08:55back to Apple you know apples the
08:56world's biggest SAP installation so that
08:58that degree of disaggregation of
09:02manufacturing that Apple is a internal
09:04function of software yeah yeah and
09:06you're right so the Apple is clearly the
09:08canonical full-stack technology company
09:11I think the way that the principle they
09:14use to decide which parts of the chain
09:16they have to own in which parts they can
09:18outsource is just ultimate product what
09:21ultimate product experience we need to
09:22deliver and which of the you know which
09:25of the layers topologies point are
09:27mature enough that I can use their quote
09:29API so maybe the China factories they
09:32decide the API is mature enough right
09:34but you know for the clearly for the the
09:37main core processor that wasn't yeah
09:39they pulled that it's kind of its kind
09:41of a bundling and unbundling think I
09:42don't mean the thing that you see in
09:44media and telecoms is clearly the
09:46unbundling of things that were
09:48previously bundled that's enabled by the
09:49internet and enabled by smartphones and
09:51so on so you have um you know magazines
09:53and newspapers so on one hand or you
09:56know data and SMS and voice on the other
09:57getting pulled apart into the component
09:59pieces one of the things I think that's
10:00under appreciate about full sex arts in
10:03particular is so why do you need them to
10:05move into a new vertical often not
10:07always but often Airbnb for example is
10:10moving into new vertical without doing
10:11the whole thing but they're doing
10:13probably more of it then other web
10:15competitors have gone to for example
10:17they sent photographers to people's
10:19houses and they photograph them and
10:22there were a lot more invasive about
10:23that than preview was an enormous
10:26component of making the listings look
10:27attractive than soda and so forth and
10:28now they're getting even more invasive
10:29or maybe invasive is not the right term
10:31let's say they're being more considered
10:34of the full experience by thinking about
10:35it as a hospitality experience they're
10:37giving guidelines for their hosts and
10:38they're saying you know here's the
10:39cleaning staff will provide cleaning
10:41services and so on and so forth so you
10:43know there's a I think the logic of
10:45providing a good customer experience
10:46often means that full stack and when the
10:48reasons for that is that if the more
10:49layers you can operate you can propagate
10:51feedback for one layer to another so
10:53I'll give two examples I think of a lot
10:55one is so development of like the
10:57MagSafe you know cable as I understand
11:00because Apple controlled the Genius Bar
11:03they have people who had come in and
11:05they would be able to log all the
11:07defects that happened with Apple
11:08computers right and they found that a
11:10relatively small number of defects but
11:12with a very high dollar value or caused
11:15by people yanking the computer off the
11:17desk and so they propagated this from
11:19the retail level right the like highest
11:21level of the business all the way down
11:23to the very lowest level of the business
11:24how you solid-state physics right
11:26develop a new device right to then sell
11:29and propagate it all the way back up to
11:31this layer that sensed it right as
11:32another example so when running counsel
11:35the fact that we controlled the entire
11:37stack all the way from mixing reagents
11:39and procuring reagents and doing that
11:41ourselves up through robotics and
11:43sequencing and so on and so forth up
11:44through the National salesforce and like
11:46clinical integration and so on meant
11:48that you know for example when doctors
11:49asked us for a turnaround time
11:51improvement right because people needed
11:53you know rapidly if it's like an IVF you
11:55know cycle or something like that we
11:57could go and we could you know take
11:59certain reactions and put them into
12:01triplicate rather than cereal and so we
12:03could optimize things at the region
12:04level top to fix things that were at the
12:06patient level and so like controlling
12:07all the layers like that I think is very
12:09helpful and in a new business the one
12:14challenge in doing it is you need
12:15managers of these different layers that
12:18are very competent and you know
12:20initially as an executive you need to
12:22actually be able to articulate that you
12:24need a robotics expert when nobody in
12:26the industry is hired the robotics
12:27expert can quite this way before right
12:29so I think like I think of recruiting is
12:31one of the most difficult challenges
12:32associated with building a full-stack
12:34startup because you need to have
12:35expertise you need to you know what you
12:36lack expertise well that's the other
12:38engine thing knows if you can do it it's
12:39very very hard to compete against
12:41because you've got you've got to do sort
12:43of this like it's this and function of
12:46like this and this and this like you
12:49know you want to go head to head with
12:50you know Apple like I guess you can take
12:53the radically different approach like
12:54Google which is you try to break the
12:55stack up in different ways with Android
12:58but in general like you got to be good
13:00at branding you got to be good at
13:01semiconductors you got to be good at
13:02operating systems you got to be good at
13:03apps like there's a lot of things and it
13:05really is a you know goes against the
13:09whole kind of business school theory of
13:10like have a single core competency and
13:13and Google is arguably pretty full-stack
13:14themselves I mean they build their own
13:15data centers to build their own hardware
13:16and chips and routers and all this
13:18gemstone I was gonna say I'm looking
13:19specifically at startups isn't part of
13:21this a function of the collapse of in
13:22cost of actually getting something out
13:24of the door so you know you could look
13:26at it at Airbnb or uber or you know any
13:28of these companies and say well 10 years
13:30ago it would have taken you 100 million
13:33dollars just to get the app out of the
13:34door yeah today you can get that out of
13:36the door 5 and so actually the amount of
13:39capital that you can deploy to spread
13:41out in all directions is completely
13:43different yep and and in parallel with
13:45that do you know just the size of the
13:47addressable market now means it is
13:49actually feasible to do something I
13:51Caribbean bit you could not have done
13:52Airbnb 15 years ago you couldn't have
13:54done well it also it affects like our
13:58you know I think we people like us are
14:01willing to because the markets are so
14:03huge you know right these hundred
14:06million dollar checks knowing these can
14:07be you know many tens of billions of
14:10dollars in value that's actually worth
14:12if you try to do air B&B 10 years ago
14:14you'd have got to you've got an
14:16innovative you know tiny addressable
14:18market and it would have cost you 10
14:19times as much to get to the product that
14:21you see today so you just couldn't have
14:22done it yeah so the next question is and
14:24when where do we think the next wave of
14:26like what industries what will the next
14:27wave a full stock startups be I'll just
14:29give you my bias view which is for my
14:31bias which is so I think the best lens
14:34to look at this through is the Carlota
14:35Perez lens of basically she divides
14:39every technology revolution into two
14:40phases the installation phase which
14:42always corresponds with the financial
14:43bubble there's always a crash and
14:45there's like a long period of the
14:46deployment phase where you're kind of
14:47been taking the fruits of that
14:49technology and and deploying it
14:52throughout the economy
14:52so the automobile in the automobile
14:55revolution the first phase happen in
14:57Detroit it was around building better
14:59the second phase happened over the next
15:0150 years which is about building the
15:03highway system and suburbs and trucking
15:04systems and logistics and fast food and
15:07big-box retail and all these other
15:08things that happen as a result and
15:10happen by the way all over the place in
15:12different industries and to me but I
15:14think of full sex Arps is kind of the
15:16what we've learned to be one of the
15:18optimal deployment phase strategies yeah
15:21now that we're going into healthcare
15:23healthcare is a very very complicated
15:26three where you could argue a lot of
15:28parts of it have misaligned incentives
15:31have people who are averse to technology
15:33in many cases that you know one of the
15:36trickiest parts is you it's one thing to
15:38have a breakthrough you invent a new
15:39there are there exist expert systems
15:41that can diagnose illness better than
15:43any doctor and yet they're not never
15:44used why because what he can do sell to
15:46doctors to replace themselves like right
15:48there's all sorts of technologies just
15:49aren't out there now and I think the
15:52full-stack approach is the answer yeah I
15:54think in particular with the healthcare
15:56right it's a industry where you have
15:57third-party payment and fourth party
15:59pricing and fifth party regulation and
16:01the sense that pricing is set by CPT and
16:03regulation is by a number of bodies and
16:05you know payment is by insurers and so
16:08you have like when you're talking to the
16:10doctor there's actually like three other
16:11people maybe you know four in that in
16:13that room with you effectively right
16:15doctors know the price you don't know
16:16the price nobody knows the price of
16:18anything and so going full stack looks
16:20something like Kaiser for example
16:21actually does start to get aligned
16:23incentives because they aren't just
16:24paying for it they're also doing the
16:26treatment and so they start to actually
16:27you know close the loop on those I would
16:29say that three industries in particular
16:31that are very good candidates for you
16:34know full-stack startups are finance
16:36education health care and these three
16:39have a few things in common first their
16:41information heavy businesses which are
16:43good for a software approach second they
16:45have regulations which can be partially
16:50automated in terms of compliance via
16:52software internally third there are very
16:55large markets fourth we're seeing a lot
16:57of entrepreneurs in them and fifth is
16:59they don't require that much in the way
17:00of physical components they're still
17:02actually pretty friendly to software I
17:03mean old school is actually building
17:04schools but it's not quite at the same I
17:06would also say that those three
17:08industries are the ones where the the
17:11the striking how little they have
17:14in the you know the greatest invention
17:16in the history of of information
17:19delivery the internet you would have
17:21expected it to have chained to have had
17:23a major impact on those information
17:25heavy businesses yet it hasn't right
17:27like think about the way health care is
17:28really not that different or finance
17:30then I mean yeah they have online
17:31brokers and these but the fundamental
17:32structure has not changed in the way for
17:34example in media is just dramatically
17:36yeah right and so I think that to me the
17:39disparity between those two things
17:40their information intensive and the fact
17:42that so little has changed right means
17:43they are there by far the right targets
17:45now that said they're very difficult I
17:46think you refer to them as boss mode if
17:48boss level yeah the video game lays is
17:50yeah we're you know this is the big big
17:53scary dragon that you have to you know
17:55fight in the last level yeah I think so
17:57I mean like I think um so what would be
18:00like I think are relatively so finally
18:02education healthcare I think are quite
18:03difficult thing that's like about the
18:05level of sector that we can take on
18:08right now I would say that what I think
18:10of as probably infeasible right now is
18:12like nuclear power plant right like
18:15energy and transportation like aviation
18:17like at the level like a Boeing or
18:19something like that very large physical
18:21businesses are still not I think good
18:24candidates for this right now I mean
18:25there's obviously exceptions you've got
18:26your you know Musk's and you know you've
18:29got people who are doing interesting
18:30things with drones and so on and so
18:31forth but generally speaking like I
18:33think that right sweet spot is like
18:35finance education health care things in