Why AI Moats Still Matter (And How They’ve Changed)

0:00:04 The thing that is fundamentally different about this product cycle is that the software itself
0:00:08 can actually do the work. And therefore, the market opportunity for software today is no
0:00:13 longer just IT spend, it’s largely labor. It’s not like all the jobs will go away. I actually
0:00:17 think that’s not going to happen at all. There are a lot of things where if I could hire somebody
0:00:23 for a dollar to do this task, I would 100% do that. I’ve never been able to hire somebody for
0:00:27 a dollar. Now I can hire software for a dollar. While it is important to understand model capabilities
0:00:30 and what’s happening in the frontier, you still need to figure out how to apply that
0:00:35 technology. I think modes matter just as much as they did before. The one change is that in
0:00:40 the supply-demand equation, there’s conceptually more supply of software on the cup because the
0:00:45 barrier to creating this stuff has gone down dramatically. I think AI is an incredible tool
0:00:50 for differentiation. The idea that a voice agent can speak in 50 languages fully compliantly,
0:00:56 24-7, highly differentiated, you know, certainly versus the human. The AI-ness of that capability,
0:01:01 in my opinion, is not a source of defensibility. It is just so consensus. Like cloud was not
0:01:06 consensus. Mobile was not consensus. And that’s why the incumbents kind of screwed up.
0:01:12 Everyone’s saying that AI killed the concept of most. That anyone can vibe code a Zendesk
0:01:16 competitor in their bedroom. That 20 companies are building the exact same thing you are.
0:01:21 So why are software companies potentially more defensible today than any other time in history?
0:01:27 A16Z general partners, David Haber and Alex Rampell are seeing companies charge $20,000 for what used
0:01:32 to be called a feature because that so-called feature now replaces an entire person. They’re
0:01:37 watching startups attack markets that were never worth touching with software like plaintiff law and auto
0:01:43 loan servicing because suddenly the market isn’t IT spent, but labor spent. The counterintuitive reality
0:01:49 is this. The same force creating infinite competition is also creating trillion dollar opportunities in places nobody’s looking.
0:01:56 In today’s episode, we explore the relationship between momentum and moats, why the 19th player always dies,
0:02:01 and how to find the Goldilocks zone where you’re too small for giants to care about, but big enough to build an empire.
0:02:10 We’ve spent a lot of time talking about moats and how moats have evolved and are there still even moats in this new era?
0:02:16 And so why don’t you reflect and share some of the conversations we’ve been having here, some of your perspectives on this broader moat question.
0:02:17 Maybe David, we’ll start with you.
0:02:21 Maybe just to jump right into it with a hot take. I think moats still matter.
0:02:24 And I think a lot of the moats still matter.
0:02:29 Still matter. Exactly. And I think they’re largely the same. I often think about this between
0:02:35 sort of differentiation and defensibility. I think AI is an incredible tool for differentiation, right?
0:02:41 The idea that a voice agent can speak in 50 languages fully compliantly, 24-7, highly differentiated,
0:02:47 certainly versus the human. But the source, the AI-ness of that capability, in my opinion, is not a source of defensibility.
0:02:52 It’s largely differentiation. The defensibility of a software product resides
0:02:56 in my opinion, from owning the end-to-end workflow, from the context in which that it’s applied.
0:03:01 Becoming the system of record, having a network effect, deeply embedding yourself within your customer.
0:03:06 And I think these were the heuristics that were always, you know, things that we would always look for
0:03:11 when evaluating software companies. I think the thing that is fundamentally different about this product cycle
0:03:14 is that the software itself can actually do the work, right?
0:03:19 And therefore, the market opportunity for software today is no longer just IT spend, it’s largely labor.
0:03:27 The challenge often has been that everybody can build something at small scale and a lot of the,
0:03:34 I wouldn’t call them network effects, but some of the defensibility moats only become apparent at large scale.
0:03:39 So like a lot of people talk about, okay, take an example from like long time ago, pre-AI era.
0:03:49 If I am building an anti-fraud company, and I’ve seen lots of people, right, am I going to do a better job than a net new anti-fraud company that’s seen a few people?
0:03:55 And the reason why this would be called a data network effect, although there’s another podcast that Martine and I did a long time ago
0:04:00 debating whether or not data network effects are real, but it’s something that really, it’s almost like gravity.
0:04:08 Gravity actually, like one atom actually has, exerts gravity on you, but you only really see it at like very, very large scale.
0:04:09 Like the Earth, you notice the gravity.
0:04:10 The Sun, you notice the gravity.
0:04:11 Jupiter, you notice the gravity.
0:04:13 You don’t notice it for like that glass.
0:04:21 And it’s the same thing for a lot of these data network effects where at very, very small scale, when you have 20 companies that are all saying, I’m going to stop fraud.
0:04:23 All right, they’re all building the same things.
0:04:24 They all have the same algorithms.
0:04:35 But when you’ve seen 4 billion people and like these people are bad, now you can sell each incremental customer, each customer of your anti-fraud technology, to use this example,
0:04:39 because you’ve seen more customers and you can get actually better results.
0:04:45 But the challenge is that a lot of these moats only really are evident at mega, mega, mega scale.
0:04:47 And the same argument would apply.
0:04:49 It’s like, oh, like I’ve seen four customers.
0:04:50 David’s seen three.
0:04:51 I’ve seen four.
0:04:52 He’s seen three.
0:04:53 Pick my software.
0:04:54 But it’s like, you’ve seen four customers.
0:04:56 That means there are 8 billion customers you haven’t seen.
0:04:58 There are 8 billion customers he hasn’t seen.
0:04:59 What’s the difference?
0:05:02 Whereas at mega scale, it’s like, all right, I’ve seen 4 billion customers.
0:05:04 He’s seen 1 billion customers.
0:05:08 Well, it’s actually kind of easy to see that the results of my product will be better.
0:05:09 But that’s at scale.
0:05:18 And a lot of the question is on the zero to one phase, it’s hard to make the argument that I have better, if it’s fraud, I have better fraud underwriting.
0:05:24 If it’s AI do the work, like I’ve done more phone calls to a particular type of customer and therefore I do a better job.
0:05:27 It’s hard to make that argument at subscale.
0:05:33 And this is often the challenge is that it’s kind of self-evident that if you become the biggest company in the world, then you have a moat.
0:05:37 But how do you get to the scale where you actually could show?
0:05:45 You can’t get to that scale if you have 9 million ankle biters and you are yourself an ankle biter of just, we are trying to get to scale.
0:05:48 And nobody can because it’s so easy to actually produce software.
0:05:52 And that’s the double-edged sword of AI is that it’s very, very easy to produce software.
0:05:58 Everybody can go do something that is a very obvious idea because it’s obvious everybody’s going to go build it.
0:06:02 But can you get to the type of scale where you actually could show a moat?
0:06:08 And that has gotten arguably harder because you have a larger end count of potential competitors.
0:06:11 But if you get to mega scale, then you could show the moat.
0:06:13 And that’s kind of the zero to one versus one to N.
0:06:21 Maybe talk about what’s different about defensibility for even the bigger players today in the AI era than it was in, let’s say, the Web2 era.
0:06:25 Are the companies today more defensible, less defensible, or how should we think about sort of the strength?
0:06:31 I think the less defensible part, this is why a lot of enterprise software has gotten beaten up in the public markets.
0:06:32 It’s kind of two reasons.
0:06:39 Number one is that if you’re doing per-seat pricing, like, how do you come up with a pricing model that people feel is fair?
0:06:41 And a lot of it is just psychology.
0:06:48 And for whatever reason, for the last 20 years, it’s like per-seat per month with, you’ve heard my joke, the tall, grande-venti model of, like, software charging.
0:06:50 It’s like somehow that felt fair.
0:06:53 And whether that is fair or not, I don’t know.
0:06:56 But, like, people are like, oh, yeah, it’s $85 a seat per month.
0:06:57 Yeah, okay, that sounds reasonable.
0:07:01 Whereas if you proposed that pricing 40 years ago, you would have been laughed out of town.
0:07:03 So this just became the norm.
0:07:10 And the reason why, as I was saying, public software companies have been beating up a little bit is like, uh-oh, maybe you won’t sell as many seats.
0:07:15 Is Adobe going to sell as many seats if now you don’t have to hire as many graphics designers?
0:07:19 Or is Zendesk going to sell as many seats if the software does answer all the queries?
0:07:20 Like, the answer is no.
0:07:23 It doesn’t mean that the companies are toast.
0:07:27 They might actually quintuple their revenue because now they charge per outcomes as opposed to charging per seats.
0:07:28 But that’s kind of part one.
0:07:31 Part two is, wait a minute.
0:07:34 Now everybody can vibe code up a Zendesk competitor.
0:07:37 So maybe companies will just stop buying software.
0:07:39 This one I don’t think we’ve seen at all.
0:07:42 But I think there is, like, these two-sided, these two risks.
0:07:46 But to answer your question, does defensibility change?
0:07:50 Well, if you now are able to code your own software, like, why am I paying?
0:07:52 Like, your margin is my opportunity.
0:07:54 Well, look at the margin of software companies.
0:07:57 Like, Salesforce has an 80% gross margin.
0:07:58 Like, they should have a 1% gross margin.
0:08:01 Or nobody should use Salesforce anymore.
0:08:05 That would be the pro case of moats really starting to disintegrate.
0:08:07 But I don’t think we’ve seen that happen at all.
0:08:12 Because it turns out people, on the one hand, two things are actually happening.
0:08:15 One is that this is kind of like Clay Christensen theory.
0:08:17 It’s like the incumbents overshoot the market.
0:08:21 So the amount of features in Salesforce or Zendesk or NetSuite,
0:08:24 it way exceeds the feature set that you need,
0:08:26 that any individual customer needs.
0:08:29 Because it’s meant to encompass, it’s like all of these weird edge cases.
0:08:32 And you kind of see this if you use Microsoft Word.
0:08:34 When was the last time you wrote a book?
0:08:35 When?
0:08:36 Never, right?
0:08:38 I haven’t written a book.
0:08:39 It has all of these things.
0:08:40 They probably have 50 software engineers.
0:08:43 But if you do write a book, guess what?
0:08:44 Microsoft Word has all these features
0:08:48 just for book authors to, like, make a table of contents or something.
0:08:49 It’s like, I don’t use that.
0:08:52 So they keep bundling more stuff in there.
0:08:53 So they overshoot the market.
0:08:56 And theoretically, it’s going to make it easier for somebody.
0:08:58 But kind of going back to where I started with this topic,
0:09:00 like, it turns out that this concept of,
0:09:02 I’m just going to vibe code Microsoft Word,
0:09:04 it’s like there are these edge cases that you just don’t know about.
0:09:06 So it’s actually, you know,
0:09:08 why don’t you grow your own food or weld your own aluminum
0:09:10 or build your own house.
0:09:13 It’s just, it’s kind of easier to use this concept of comparative advantage
0:09:16 and just say, I’m going to buy something off the shelf.
0:09:19 So anyway, so I think moats matter just as much as they did before.
0:09:23 The one change is that in the supply-demand equation,
0:09:26 there’s conceptually more supply of software on the come.
0:09:30 Because the barrier to creating this stuff has gone down dramatically.
0:09:33 I think the flip side to that, too, is that while there will be more software,
0:09:36 and again, the kind of marginal cost of producing software
0:09:39 is declining asymptotically towards zero,
0:09:41 the way that these companies are getting more deeply entrenched
0:09:43 within their customers has differed.
0:09:45 Because, again, the software is doing the work,
0:09:48 and therefore, in many cases, it’s actually replacing labor.
0:09:51 And so if you’ve transitioned a team out
0:09:53 that has now become your software,
0:09:55 like, you’re now much more dependent
0:09:57 on that product to run your business.
0:09:59 And again, is it more difficult to replace that software
0:10:02 with another piece of software or to rehire that team?
0:10:03 I think it’s an open question.
0:10:05 But again, the software is doing more of the work,
0:10:07 and therefore, I think, getting more deeply embedded
0:10:08 within their customers.
0:10:10 One part of it is just, like, the Goldilocks zone of pricing.
0:10:13 So I wrote some tweet or whatever it’s called,
0:10:15 X thread about this a long time ago.
0:10:16 I call it the janitorial services problem.
0:10:19 Because if I went to you, you’re the CEO of a giant company
0:10:21 where you write your books in the future.
0:10:23 So you have a 300,000-person company.
0:10:24 I find you as Eric.
0:10:26 I can get your toilets 9% cleaner
0:10:29 and save you 1% on your toiletry spend.
0:10:31 or your janitorial services spend.
0:10:32 Not only do you not care,
0:10:34 you don’t even care enough.
0:10:36 You won’t even exercise the mental energy
0:10:39 to find the person in the company who does care, right?
0:10:41 And that means that your janitorial services spend
0:10:42 will never change.
0:10:44 And the problem is it’s hard to get in.
0:10:46 The good news is it’s hard to get out.
0:10:47 Whereas for something,
0:10:51 it’s like 90% of my profits go to, like, you.
0:10:55 I’m now 90% of your profits as the CEO of GE.
0:10:56 They’re going to me.
0:10:59 Your number one priority is, like,
0:11:00 getting the hell off of me, right?
0:11:02 And, like, doing RFPs left and right.
0:11:05 So part of it is also just, like, how relevant this is.
0:11:07 And there are some companies that operate
0:11:09 in this Goldilocks zone of irrelevance,
0:11:10 like these janitorial services,
0:11:13 where even if you have 9 million competitors,
0:11:15 like, they’re just not going to go anywhere,
0:11:17 which is why, like, a lot of the strategy
0:11:20 that we talk about internally is Greenfield, right?
0:11:24 It’s like those companies are, they’re stuck for good.
0:11:28 Is there a high rate of new company creation
0:11:31 that will not use the crappy old janitorial services company,
0:11:34 but will actually resonate?
0:11:36 Like, your pitch of, like, I will get your toilets cleaner
0:11:38 and I will charge you less money.
0:11:39 That really resonates.
0:11:41 But that’s not going to resonate to the people
0:11:43 that are using the old-fashioned stuff.
0:11:47 What are examples of company or space in the Goldilocks zone?
0:11:49 And what was an example of companies or space
0:11:50 in the Greenfield zones?
0:11:53 Well, like, payroll companies, right?
0:11:54 Like, ADP and Paychex.
0:11:56 I mean, these are companies that are collectively worth
0:11:57 hundreds of billions of dollars.
0:11:59 Very, very profitable.
0:12:01 And how does, like, you could do your own payroll.
0:12:03 Actually, it’s kind of a good metaphor for software in general.
0:12:06 Like, why is it that you have to, like,
0:12:07 why can’t I just pay you?
0:12:08 You’re my employee.
0:12:09 Why can’t I just, like, cut you a check?
0:12:11 Well, because I have to withhold taxes.
0:12:13 Well, how much tax do I have to withhold?
0:12:14 Well, it depends, right?
0:12:17 And there’s this, like, super complicated lookup table.
0:12:18 It’s like, well, you live in this county,
0:12:20 but you spend this many days in New York
0:12:21 and this, that, and the other thing.
0:12:24 Oh, and you owe, like, child support
0:12:26 and the IRS is garnishing your wages.
0:12:28 Like, all of these things that are very complicated.
0:12:31 So it turns out it’s just cheaper to go to ADP
0:12:33 and ADP just charges you, like, I don’t know,
0:12:35 like 50 bucks a month per person
0:12:36 that you might be paying 100,000.
0:12:39 It’s a paltry sum compared to the overall amount of payroll.
0:12:43 So nobody really switches their payroll companies.
0:12:44 Like, that would be an example of one.
0:12:48 On the other side, I had a lot of companies
0:12:50 coming out of 2022 where the market
0:12:52 really went through a downturn
0:12:54 and they’re like, wait a minute, I’m spending four,
0:12:55 I had 1,000 employees.
0:12:57 I downsized to 200 employees.
0:13:01 I had 1,000 licenses for Salesforce, right?
0:13:04 What’s 1,000 times $100 a month times 12?
0:13:06 That’s $1.2 million a year.
0:13:08 Wow, like, that’s a lot of money
0:13:09 because I only have 200 employees
0:13:11 and I only have six months of cash.
0:13:12 Like, I got to save that.
0:13:14 And they didn’t do that for their payroll spent.
0:13:18 So you see it, like, a lot of companies
0:13:21 do want to rationalize their overall software cost,
0:13:23 especially for these things
0:13:24 where they recognize in aggregate,
0:13:27 like, most people aren’t actually using the seats.
0:13:32 So I’d say, like, you know, Salesforce type stuff,
0:13:34 you know, some of the creative tools,
0:13:37 like if you, like, Adobe is very expensive
0:13:40 and you might just do, like, a wall-to-wall license
0:13:41 saying, why not?
0:13:43 But then you look at, if you’re like,
0:13:44 how do I save $5 million?
0:13:47 Nobody’s using this, well, it’s $5 billion.
0:13:50 Whereas for things where inextricably
0:13:54 the delivery and the payment are linked, right?
0:13:55 Which is very, very different
0:13:57 than pricing for software.
0:13:59 Like, payroll, like, obviously,
0:14:00 I’m not going to pay for payroll services
0:14:02 unless you are employed here.
0:14:05 Whereas I might, like, we have 600 people
0:14:06 that work at our firm.
0:14:08 I think we have 600 licenses
0:14:10 from Microsoft Office 365.
0:14:11 Like, we probably,
0:14:13 I bet there are a lot of people here
0:14:15 who have not opened Microsoft Excel in a year.
0:14:17 So why are we paying for that?
0:14:18 And that would be the idea
0:14:20 of kind of rationalizing software spend.
0:14:22 So it kind of depends,
0:14:24 but I think per-seed pricing,
0:14:25 where it’s like, it’s just easier
0:14:27 to pay for the entire thing wall-to-wall,
0:14:29 you know, in your entire organization,
0:14:31 those are often the first to go
0:14:32 versus things that are, again,
0:14:34 inextricably linked to the actual usage.
0:14:37 Yeah. So you mentioned earlier
0:14:38 that we’ve seen, you know,
0:14:39 basically, you mentioned
0:14:40 there was this concern
0:14:42 that maybe instead of Zendesk,
0:14:43 people will, you know,
0:14:44 their companies will, you know,
0:14:46 there’ll be a Vibe-coded version of it,
0:14:47 but we’ve seen none of that so far.
0:14:49 Is your mental model is,
0:14:51 we’ll see it in examples
0:14:53 where the cost is significantly high
0:14:55 or in which there’s sort of
0:14:57 greenfield opportunities,
0:14:58 or what is sort of your mental model
0:14:59 for the types of software
0:14:59 that we’ll replace?
0:15:01 Yeah, I mean, I think the greenfield one
0:15:02 is always true,
0:15:04 but when you look at greenfield opportunities,
0:15:05 you need two things to be true.
0:15:06 You need the entrepreneur
0:15:07 to be very, very patient
0:15:09 and say, I’m not going to try
0:15:11 to sell to everybody who’s,
0:15:13 if I’m starting a net new payroll company,
0:15:15 I’m not going to try to sell to GE
0:15:18 because I recognize that they are,
0:15:19 they are hostages to ADP
0:15:20 and that’s never going to change.
0:15:23 So one is that patience of entrepreneur
0:15:25 and the other one is
0:15:26 you just need a high enough rate
0:15:27 of new company creation
0:15:29 to really make it work,
0:15:32 which is why I like to pick on one space
0:15:34 of electronic health records
0:15:35 or electronic medical records.
0:15:37 How many new hospital systems
0:15:38 are created every day?
0:15:39 I mean, it rounds to zero.
0:15:42 So if I’m trying to go build
0:15:43 a new EHR system
0:15:44 to go compete with Epic or Cerner,
0:15:46 I can do that.
0:15:48 There are a lot of edge cases there,
0:15:48 but it’s like,
0:15:49 and I might have patience
0:15:50 as an entrepreneur,
0:15:51 but wait a minute,
0:15:53 like I need to sell $5 million deals
0:15:54 to big hospital systems.
0:15:56 Every single hospital on earth
0:15:57 is currently using an EHR system.
0:15:58 It’s going to be really,
0:15:59 really hard to make that work.
0:16:01 So I think both of those
0:16:01 need to be true.
0:16:03 like the right type of entrepreneur
0:16:04 who’s willing to be patient
0:16:05 because it’s often
0:16:06 a very lonely game
0:16:07 of it’s like,
0:16:08 I built this great product.
0:16:08 Wait a minute,
0:16:09 I don’t have any customers yet
0:16:11 and you want to see high traction
0:16:11 because you’re seeing
0:16:12 in the rest of the market
0:16:13 like some companies
0:16:14 are just going like this
0:16:15 and my company’s not
0:16:17 and I’m in Silicon Valley
0:16:18 and I need to recruit
0:16:18 the best people.
0:16:19 It’s like they want to work
0:16:19 at the company
0:16:21 that has the graph like this,
0:16:21 but you need this
0:16:23 greenfield requires patience.
0:16:26 So we’re talking about
0:16:27 how moats still matter
0:16:29 and in many ways
0:16:30 they look pretty similar.
0:16:32 Let’s steel man the other side
0:16:32 for a second.
0:16:34 Where are we even having
0:16:35 this conversation
0:16:36 where some people say,
0:16:37 hey, you know,
0:16:38 brand is the,
0:16:40 is the motor shipping velocity
0:16:41 or because this era is different.
0:16:43 What’s the steel man
0:16:44 of their argument?
0:16:46 Look, I think this market
0:16:48 is noisier than ever, right?
0:16:49 And so I think finding ways
0:16:50 to sort of, you know,
0:16:51 stand out from the crowd
0:16:53 probably matters more today
0:16:54 than it has, you know,
0:16:55 in the past I would argue.
0:16:56 I think the other thing
0:16:58 is that the underlying technology
0:16:59 is changing so quickly
0:17:00 and so, you know,
0:17:01 as a founder,
0:17:02 you want to be living
0:17:03 on the frontier
0:17:04 and understanding
0:17:05 kind of what model capabilities
0:17:05 look like
0:17:06 because it can dramatically
0:17:08 change the efficacy
0:17:09 or the, you know,
0:17:10 the capability
0:17:11 of your underlying product.
0:17:13 And so I think, you know,
0:17:15 one of the things
0:17:16 that’s changed,
0:17:16 I think that’s been
0:17:17 really interesting
0:17:18 in this sort of,
0:17:19 you know, current wave
0:17:20 of especially vertical applications
0:17:21 that we’ve seen
0:17:22 is the type of founder.
0:17:24 You know, I think founders today
0:17:25 are often younger
0:17:26 and more technical
0:17:27 than we’ve seen
0:17:28 in prior generations.
0:17:30 You know,
0:17:31 and so they’re less often native
0:17:33 to the particular industry
0:17:34 but they’re fluent
0:17:35 in the tool set, right?
0:17:36 And I think that’s really important
0:17:37 because, you know,
0:17:38 to the same point,
0:17:39 you’ve got to stay
0:17:40 on the frontier
0:17:41 and understand what’s coming.
0:17:42 At the same time,
0:17:43 you know,
0:17:44 I wrote this piece
0:17:44 that I call
0:17:45 Context is King.
0:17:46 You know,
0:17:46 while it is important
0:17:47 to understand,
0:17:47 you know,
0:17:48 model capabilities
0:17:49 and what’s happening
0:17:50 in the frontier,
0:17:52 you still need to figure out
0:17:53 how to apply that technology.
0:17:55 And so while
0:17:56 the founders themselves
0:17:57 are maybe less native
0:17:58 to the particular industry,
0:17:59 they’re still hiring
0:18:00 for Context,
0:18:00 you know,
0:18:01 very early
0:18:02 in a company’s life cycle.
0:18:04 A good example of this
0:18:05 that I sit on the board of
0:18:06 is a company called EVE.
0:18:06 You know,
0:18:07 the two founders of EVE
0:18:08 were the earliest employees
0:18:09 at Rubric,
0:18:09 which is, you know,
0:18:10 now a public
0:18:11 infrastructure company.
0:18:12 You know,
0:18:14 they built a legal AI company
0:18:15 in the plaintiff law space.
0:18:16 Neither of them
0:18:17 had any particular background
0:18:18 in employment law
0:18:20 or personal injury,
0:18:22 but they deeply understood,
0:18:22 you know,
0:18:23 how to apply,
0:18:24 you know,
0:18:25 document extraction technology
0:18:26 and sort of,
0:18:27 you know,
0:18:28 voice and LMs more broadly
0:18:29 to this very particular
0:18:30 work,
0:18:30 you know,
0:18:31 workflow.
0:18:33 And they’ve hired
0:18:34 plaintiff attorneys
0:18:34 actually on staff.
0:18:36 So anytime a new model
0:18:37 is released,
0:18:38 you know,
0:18:38 they’re understanding,
0:18:39 you know,
0:18:40 from people in industry
0:18:41 the impact that it’s having
0:18:42 on drafting,
0:18:43 on,
0:18:43 you know,
0:18:44 their ability to,
0:18:44 you know,
0:18:46 to reason through a case,
0:18:46 you know,
0:18:47 or a matter.
0:18:48 And so again,
0:18:49 it’s sort of this tension
0:18:50 of like,
0:18:51 you know,
0:18:51 building the brand,
0:18:52 having momentum,
0:18:53 you know,
0:18:54 understanding what’s happening
0:18:54 in the frontier,
0:18:55 and yet,
0:18:56 you know,
0:18:56 figuring out ways
0:18:57 to apply that technology
0:18:59 in the context,
0:18:59 you know,
0:19:01 of your specific customer.
0:19:01 Because again,
0:19:02 I deeply believe
0:19:03 that that is where
0:19:04 a lot of the sources
0:19:06 of defensibility reside.
0:19:06 you know,
0:19:07 I’d love to find
0:19:08 other examples
0:19:08 of businesses
0:19:11 is where the technology
0:19:11 like reinforces
0:19:12 their business model.
0:19:14 It doesn’t compete
0:19:14 with it.
0:19:14 Meaning,
0:19:15 in lots of areas
0:19:16 of legal,
0:19:18 if you make your employee
0:19:19 50 times more efficient,
0:19:19 you’re eroding
0:19:20 your billable hour.
0:19:21 In their business,
0:19:22 they operate
0:19:23 on a contingency basis,
0:19:24 meaning,
0:19:25 you know,
0:19:25 they only get paid
0:19:26 if they win.
0:19:28 So there’s no sort
0:19:29 of limit to the amount
0:19:29 of AI that they want
0:19:30 to adopt.
0:19:31 And if you can become
0:19:32 5x more efficient,
0:19:33 you can take on
0:19:34 5x more clients.
0:19:36 Anyway,
0:19:36 these are sort
0:19:37 of characteristics
0:19:37 that I think,
0:19:39 you know,
0:19:40 I’d love to find more of
0:19:40 and hopefully
0:19:41 that can be
0:19:42 kind of a bat signal too.
0:19:43 I think the other
0:19:44 steel man is
0:19:45 if you believe
0:19:47 that brand matters,
0:19:48 which it almost
0:19:50 tautologically does,
0:19:52 because what do I buy?
0:19:53 I buy the thing
0:19:54 that I’ve heard of,
0:19:55 right?
0:19:55 So there’s an advantage
0:19:56 there.
0:19:57 And if you believe
0:19:58 that for a lot
0:19:58 of companies
0:19:59 and products,
0:20:01 somehow having scale
0:20:02 is effective,
0:20:02 right?
0:20:03 So not a network effect,
0:20:04 but a scale effect.
0:20:04 So if I’m
0:20:05 Honey Nut Cheerios
0:20:06 and I know
0:20:07 that people
0:20:07 are going to buy
0:20:08 lots of my Cheerios,
0:20:10 I can build a big factory
0:20:11 and not, you know,
0:20:12 hand crank out
0:20:13 each Cheerio.
0:20:14 I’m going to have
0:20:15 these compounding
0:20:16 advantages just in terms
0:20:18 of economies of scale,
0:20:18 right?
0:20:19 Like Amazon,
0:20:20 does that really
0:20:20 have a network effect?
0:20:21 No, it’s like
0:20:22 it’s kind of nice
0:20:23 that everything
0:20:24 that I buy
0:20:25 will show up
0:20:25 the next day
0:20:26 or in two days
0:20:26 and how can they
0:20:27 do that at low cost
0:20:29 because so many people
0:20:29 are buying things.
0:20:30 So there are some
0:20:31 things that have scale
0:20:32 and those things
0:20:33 also benefit
0:20:34 from brand.
0:20:35 So if you can move
0:20:36 the fastest,
0:20:37 right?
0:20:37 So if you can
0:20:38 agglomerate capital
0:20:38 and labor,
0:20:39 so it’s like
0:20:40 I raised the most money,
0:20:41 it’s a very,
0:20:42 very generic idea,
0:20:44 but somehow,
0:20:45 like most other things
0:20:46 on planet Earth,
0:20:47 if it’s the biggest
0:20:48 and like really,
0:20:48 really big
0:20:50 kind of gravitational scale,
0:20:50 then it’s just
0:20:51 going to work better.
0:20:53 So can I get there
0:20:53 the most quickly?
0:20:55 But there are 20 companies
0:20:56 that are doing
0:20:57 the exact same thing.
0:20:58 And at that point,
0:20:58 I wouldn’t say
0:20:59 that momentum
0:21:00 is a moat per se,
0:21:01 but momentum
0:21:02 has the highest chance
0:21:03 of getting you
0:21:04 to gravitational scale
0:21:05 where you do have a moat.
0:21:07 And if you don’t do that,
0:21:07 by contrast,
0:21:08 you’re just going
0:21:09 to get eaten alive
0:21:10 because you can’t
0:21:12 hand crank out the Cheerios.
0:21:12 You have to get
0:21:13 to the scale
0:21:14 where you’re able
0:21:15 to build a factory
0:21:16 and with the,
0:21:17 you have the biggest factory,
0:21:18 you can crank out
0:21:19 the most things
0:21:20 at the lowest cost.
0:21:21 So what is the trajectory?
0:21:22 What is the slope
0:21:23 of you versus
0:21:24 all of your competition?
0:21:25 And if you have
0:21:26 not a good slope,
0:21:28 you’re just not
0:21:29 going to win that game.
0:21:29 No.
0:21:32 One of the questions
0:21:32 for defensibility
0:21:33 in Web2 companies
0:21:34 was, hey,
0:21:35 would Google,
0:21:35 you know,
0:21:37 will they someday
0:21:37 build this
0:21:38 or Facebook
0:21:39 or name your incumbent?
0:21:42 In the AI era,
0:21:43 it will open AI
0:21:44 or will some other,
0:21:44 you know,
0:21:45 major company.
0:21:46 How should company,
0:21:47 how should we think
0:21:48 about that framework
0:21:48 in the AI era?
0:21:52 You know,
0:21:52 I mean,
0:21:53 it’s funny.
0:21:54 I feel like 18 months ago,
0:21:55 this, you know,
0:21:57 GPT wrapper
0:21:57 was on everybody’s lips
0:21:58 and I think it was
0:21:59 largely used
0:22:00 as a pejorative.
0:22:00 You know,
0:22:01 it was like,
0:22:02 and I think,
0:22:02 you know,
0:22:03 to some degree,
0:22:04 I think there are
0:22:04 some spaces
0:22:05 where like the model capability
0:22:07 and the application capability,
0:22:08 if they’re very overlapping,
0:22:09 I think you’re in a risky spot.
0:22:10 You know,
0:22:12 but the reality is
0:22:13 that there’s so many,
0:22:15 I think one of the remarkable
0:22:16 things that’s happened
0:22:17 is there’s so many markets
0:22:18 that were never particularly
0:22:19 interesting to sell software into
0:22:21 that are now radically
0:22:22 interesting spaces
0:22:23 to build companies in.
0:22:23 Again,
0:22:24 in large part because,
0:22:25 you know,
0:22:26 the market is now labor,
0:22:27 not just IT spend.
0:22:29 Plaintiff law,
0:22:29 being an example,
0:22:30 you know,
0:22:31 you know,
0:22:32 Alex says that we have
0:22:32 a company called
0:22:34 Salient in applying
0:22:35 voice agents
0:22:36 to auto loan servicing.
0:22:38 five,
0:22:38 six years ago,
0:22:38 would it be back
0:22:39 to software companies
0:22:40 selling to,
0:22:40 you know,
0:22:41 non-bank auto lenders?
0:22:42 Probably not.
0:22:43 The company’s doing
0:22:44 incredibly well,
0:22:45 again,
0:22:46 in large part because,
0:22:47 you know,
0:22:47 the capability
0:22:49 of being able to,
0:22:49 you know,
0:22:51 speak in 50 languages,
0:22:51 you know,
0:22:52 fully compliantly,
0:22:52 you know,
0:22:54 with customers in 50 states
0:22:55 working 24-7,
0:22:57 you know,
0:22:58 it’s just so differentiated,
0:22:59 you know,
0:23:00 versus the individual
0:23:00 and they’re finding
0:23:01 that their ability
0:23:02 to collect
0:23:03 is meaningfully higher,
0:23:04 you know,
0:23:04 than that labor,
0:23:06 that the kind of
0:23:07 cost-benefit trade-off
0:23:07 is,
0:23:08 so dramatic,
0:23:08 the company is getting
0:23:09 a lot of,
0:23:09 you know,
0:23:10 revenue from those customers
0:23:11 who may not have had,
0:23:13 you know,
0:23:14 millions of dollars
0:23:15 of IT budget historically
0:23:16 and are now very willing
0:23:17 to pay for a product
0:23:17 like that,
0:23:18 you know,
0:23:18 given the impact
0:23:19 on the business.
0:23:19 And the way
0:23:20 that we used to talk
0:23:21 about this a long time ago
0:23:22 is,
0:23:23 and this almost had
0:23:24 a pejorative slant to it,
0:23:25 but it’s like,
0:23:26 are you building a feature,
0:23:27 a product,
0:23:27 or a company?
0:23:29 And what’s the difference
0:23:30 between the three?
0:23:31 Well,
0:23:32 a feature is like
0:23:33 there’s an existing product
0:23:34 and you tweak that product
0:23:36 to make it marginally better.
0:23:38 a product is,
0:23:38 you know,
0:23:39 not that.
0:23:40 It’s like some,
0:23:40 hopefully,
0:23:41 system of record
0:23:42 or something
0:23:42 that keeps track
0:23:43 of something.
0:23:45 And then a company
0:23:47 is probably the most
0:23:48 defensible of those three
0:23:50 where you have a product
0:23:51 and, you know,
0:23:52 maybe you own a platform.
0:23:53 Like,
0:23:54 the platforms tend to be
0:23:55 the most valuable companies.
0:23:55 But,
0:23:56 you know,
0:23:57 a feature is like
0:23:58 I built a Chrome plug-in.
0:24:00 And that doesn’t mean,
0:24:00 and there were,
0:24:00 by the way,
0:24:01 there were a lot of Chrome plug-ins.
0:24:03 Like Honey was a Chrome plug-in
0:24:03 that got bought by
0:24:05 for $4 billion.
0:24:05 Like,
0:24:06 I wish I had done that.
0:24:06 Right?
0:24:08 That’s a good feature.
0:24:09 But that was a feature,
0:24:09 you know,
0:24:11 a product would be like,
0:24:11 oh,
0:24:12 I built my own browser.
0:24:13 And a company is like,
0:24:13 all right,
0:24:13 well,
0:24:14 like my own browser company
0:24:15 actually makes money.
0:24:15 Like,
0:24:16 you don’t actually have a company,
0:24:18 even if you have 10 products,
0:24:19 if you don’t have
0:24:19 a sustainable path
0:24:20 to have that company
0:24:21 be around in 10 or 20 years.
0:24:23 And I think
0:24:23 kind of another way
0:24:24 of thinking about
0:24:25 what David just said
0:24:27 is that now
0:24:28 the features,
0:24:29 like,
0:24:29 you know,
0:24:31 the feature was the most pejorative
0:24:32 and seemingly small
0:24:33 of all of those three,
0:24:34 almost obviously.
0:24:35 Some of the features
0:24:37 can be incredibly profitable
0:24:38 because it’s like,
0:24:39 wait a minute,
0:24:39 like this,
0:24:41 it feels like a feature
0:24:43 because it could get added
0:24:44 to Salesforce,
0:24:44 right?
0:24:45 Or it could get added
0:24:46 to one of these other things.
0:24:48 But the amount of money
0:24:48 that I can charge
0:24:49 for my feature
0:24:51 is like orders
0:24:52 of magnitude more
0:24:53 because it’s like,
0:24:53 hey,
0:24:54 I’m going to be
0:24:55 the front office receptionist
0:24:57 for your,
0:24:57 you know,
0:24:58 orthodontic clinic.
0:24:59 Like,
0:24:59 that’s my job.
0:25:00 Like,
0:25:01 that’s the feature.
0:25:02 And it sits on top
0:25:03 of whatever software
0:25:04 you currently use,
0:25:06 but the feature
0:25:06 I can now charge
0:25:08 $20,000 a year for
0:25:09 because it is doing
0:25:10 the job of labor.
0:25:10 But,
0:25:11 uh-oh,
0:25:12 will the existing product
0:25:13 that my feature
0:25:14 is riding on top of,
0:25:15 will they go build
0:25:16 those pieces
0:25:17 of functionality?
0:25:18 And or,
0:25:19 will another company
0:25:20 show up
0:25:21 that just says,
0:25:21 hey,
0:25:21 we’re going to sell
0:25:22 the green field
0:25:23 with the new product
0:25:24 that kind of has
0:25:25 this feature set embedded?
0:25:26 And,
0:25:26 you know,
0:25:27 feature product company,
0:25:28 it still is out there,
0:25:29 but,
0:25:30 um,
0:25:31 I’ve just never seen
0:25:32 a world where the features,
0:25:32 if you will,
0:25:34 can get to revenue scale
0:25:35 as quickly.
0:25:36 And by the way,
0:25:37 you kind of often have
0:25:38 to start with the feature
0:25:41 because a customer isn’t,
0:25:41 like,
0:25:42 think of it from
0:25:42 the customer’s perspective,
0:25:43 the customer being
0:25:44 the business buyer
0:25:44 of software.
0:25:45 It’s like,
0:25:45 I know,
0:25:46 I want to be locked
0:25:47 into a piece of shit
0:25:48 software company
0:25:49 for 20 years.
0:25:50 That’s what I’m looking
0:25:50 for as a buyer.
0:25:51 No,
0:25:51 it’s like,
0:25:51 ooh,
0:25:52 I have a problem
0:25:53 to solve.
0:25:54 My problem is,
0:25:55 I can’t hire a front office
0:25:56 receptionist for my
0:25:57 orthodontic clinic.
0:25:57 Or,
0:25:58 I can’t call people
0:25:59 in Mandarin
0:26:00 or Cantonese
0:26:01 to go,
0:26:01 like,
0:26:02 repay their auto loans.
0:26:02 Like,
0:26:03 what do I do?
0:26:04 Oh,
0:26:05 something shows up
0:26:05 and it offers
0:26:06 that functionality,
0:26:06 boom,
0:26:07 I’m a buyer.
0:26:09 And then that functionality
0:26:09 has to,
0:26:10 that feature has to
0:26:11 backfill product,
0:26:12 backfill company
0:26:13 as quickly as possible.
0:26:14 So that’s still true today
0:26:15 as it was 10 or 20
0:26:16 or 30 years ago.
0:26:18 But the difference,
0:26:18 again,
0:26:19 is that the feature,
0:26:21 the revenue for the feature
0:26:22 is just so high
0:26:23 and the demand for it
0:26:24 is so high
0:26:24 because,
0:26:24 again,
0:26:25 in many cases,
0:26:26 you’re just responding
0:26:27 to help one of that
0:26:27 effectively.
0:26:29 and so I think the effect
0:26:30 of that is there’s been
0:26:31 sort of like a Cambrian
0:26:32 explosion of interesting
0:26:33 markets to go after.
0:26:34 You know,
0:26:35 I think it’s unrealistic
0:26:35 to believe that like
0:26:36 OpenAI is going to go
0:26:37 build,
0:26:37 you know,
0:26:39 the front office
0:26:40 assistant for the,
0:26:41 you know,
0:26:41 the dental clinic
0:26:42 like as their core,
0:26:43 you know,
0:26:44 kind of business.
0:26:44 They’re not going to do that
0:26:45 across every single market.
0:26:47 I think the other dynamic
0:26:47 is that for many
0:26:48 of these companies,
0:26:49 part of the product value
0:26:50 is actually orchestrating
0:26:51 the work across lots
0:26:52 of different model companies.
0:26:54 And so I think having one,
0:26:55 you know,
0:26:55 you know,
0:26:56 foundation model business,
0:26:57 you know,
0:26:59 going kind of up the stack,
0:27:00 I think limits the actual
0:27:01 impact of the actual,
0:27:02 of the application,
0:27:02 you know,
0:27:03 potentially as well.
0:27:04 Well,
0:27:05 I think that,
0:27:05 you know,
0:27:06 if you kind of think about
0:27:07 this versus other
0:27:08 platform companies,
0:27:11 so Facebook was the
0:27:12 preeminent platform company
0:27:13 of Web 2.0.
0:27:14 So call it from,
0:27:15 whenever they opened up
0:27:16 Facebook platform,
0:27:17 which I think was like 2007,
0:27:20 people built their businesses
0:27:21 on top of Facebook.
0:27:23 Facebook would never do
0:27:24 those particular things.
0:27:25 Like,
0:27:26 so Facebook is never
0:27:27 going to show up and say,
0:27:27 hey,
0:27:27 you know what,
0:27:28 we should build a farming game.
0:27:29 Like,
0:27:29 they were like,
0:27:30 no,
0:27:30 we’re going to have a platform
0:27:31 that allows companies
0:27:32 like Zynga to build
0:27:33 these farming games.
0:27:34 But what the platform
0:27:35 normally does,
0:27:37 if they don’t actually
0:27:38 go compete with the
0:27:39 underlying products,
0:27:40 is they say,
0:27:42 I’m going to tax it,
0:27:43 but I’m going to tax it
0:27:44 in ways that are
0:27:45 kind of at my fancy.
0:27:46 So this week,
0:27:47 it’s 10% taxes,
0:27:48 that’s my promise.
0:27:48 Oh,
0:27:48 wait,
0:27:49 I changed my mind,
0:27:50 now it’s going to be
0:27:51 40% taxes.
0:27:51 So that’s why
0:27:52 it’s always dangerous
0:27:53 to build on somebody
0:27:54 else’s platform.
0:27:55 So I think the two things
0:27:56 to look at are,
0:27:57 number one,
0:27:58 is will the platform owner
0:27:59 compete with what I’m doing?
0:28:01 And that’s also another
0:28:02 Goldilocks zone question,
0:28:03 right?
0:28:04 Because why is it,
0:28:06 I published this graph
0:28:08 of VisiCalc versus Lotus 1-2-3
0:28:09 versus Excel?
0:28:11 So VisiCalc invented the spreadsheet
0:28:12 in 1979,
0:28:14 had 100% of the market
0:28:14 because they were the only
0:28:15 player in town.
0:28:17 Lotus built a better version
0:28:17 of that.
0:28:19 Lotus got to like,
0:28:19 I think,
0:28:20 70% market share by 1985,
0:28:22 which was when Microsoft
0:28:25 released Excel for a Mac.
0:28:27 And then by 2000,
0:28:29 Microsoft had 96% market share.
0:28:30 And why is it?
0:28:31 Because they owned Windows,
0:28:32 like the platform owner
0:28:33 normally wins.
0:28:34 But that’s because it was
0:28:35 just such a huge,
0:28:36 like why do I buy a computer
0:28:37 in 1997?
0:28:38 Because I want to use
0:28:38 a spreadsheet.
0:28:40 Like it was just so
0:28:41 intrinsically linked.
0:28:41 Like that was one of the
0:28:43 main use cases for computers
0:28:44 and business use, right?
0:28:45 It’s like using spreadsheets.
0:28:47 So that was like a violator
0:28:48 of Goldilocks zone.
0:28:49 Whereas other things
0:28:50 where it’s like all you have
0:28:51 to worry about from the
0:28:52 platform owner is that
0:28:53 they’re going to tax you,
0:28:54 but they might tax you
0:28:54 in very,
0:28:56 very bizarre ways.
0:28:57 But part of what David
0:28:58 was saying in terms of like
0:28:59 there are multiple model
0:29:00 companies,
0:29:00 which is great.
0:29:01 like the problem with
0:29:02 Windows was that it was
0:29:04 like 95% of the market.
0:29:06 Like 95% of your customers
0:29:07 used Windows.
0:29:08 So if I’m going to go
0:29:09 build a competing spreadsheet,
0:29:10 I’m just toast because
0:29:11 the platform owner is just
0:29:12 going to drown me.
0:29:13 Now,
0:29:15 there are five model
0:29:15 companies,
0:29:17 or, you know,
0:29:17 more,
0:29:18 like when you include
0:29:19 all the Chinese models
0:29:19 and whatnot,
0:29:19 open source.
0:29:21 Like I don’t have to
0:29:23 worry about that,
0:29:24 but I do have to worry
0:29:25 about them saying,
0:29:25 wow,
0:29:26 this is so relevant.
0:29:28 Like why is it that
0:29:30 OpenAI got a public
0:29:30 company’s CEO
0:29:32 to quit her job
0:29:33 and just to become
0:29:35 the CEO of applications
0:29:35 at OpenAI?
0:29:37 Maybe because they have
0:29:37 a huge application
0:29:38 opportunity.
0:29:40 But this is the nice thing
0:29:41 is that a lot of these
0:29:42 things are so obscure,
0:29:44 but they’re still big.
0:29:46 But I don’t think
0:29:46 OpenAI is going
0:29:47 to go do them
0:29:48 because it’s like,
0:29:49 are they going to do
0:29:50 like dental care
0:29:50 management?
0:29:51 Like they could,
0:29:53 but if they’ve done that,
0:29:54 then I would be
0:29:55 short OpenAI
0:29:56 because it’s like
0:29:56 they’ve run out
0:29:57 a lot of good stuff
0:29:57 to do.
0:29:58 That’s something
0:29:59 that they should do
0:29:59 in 2029.
0:30:00 And this is,
0:30:01 I think I told you
0:30:02 this story before.
0:30:05 This changed my outlook
0:30:06 on life when I pitched
0:30:07 this guy Dan Rose
0:30:08 at Facebook
0:30:08 who was running
0:30:09 business development
0:30:09 there.
0:30:10 I’m like,
0:30:11 this is a huge
0:30:12 opportunity.
0:30:12 You should use us
0:30:13 for payments.
0:30:13 We’re going to do
0:30:14 this.
0:30:14 We can make so much
0:30:15 money for Facebook.
0:30:17 And he was so patient
0:30:17 and nice.
0:30:18 And I love this guy.
0:30:19 I’m on a board
0:30:19 with him to this day.
0:30:20 He was like,
0:30:20 Alex,
0:30:21 that’s such a great idea.
0:30:21 I was like,
0:30:22 all right,
0:30:22 I got the deal.
0:30:22 Yes,
0:30:23 he said it’s a great idea
0:30:24 but we’re not going
0:30:25 to do it
0:30:26 because you’re pitching
0:30:26 me a gold,
0:30:27 like we have gold bricks
0:30:28 all around us.
0:30:29 And he was right.
0:30:29 I mean,
0:30:31 like Facebook in 2010,
0:30:31 I mean,
0:30:32 how much money,
0:30:33 Facebook has grown
0:30:33 their revenue,
0:30:35 they have more profit
0:30:36 every quarter today
0:30:37 than they had revenue
0:30:38 per year in 2010.
0:30:39 It’s just such
0:30:40 an incredible company.
0:30:41 And he’s like,
0:30:42 you’re pitching me
0:30:42 a gold brick
0:30:43 that’s like 100 feet away
0:30:44 and it’s real.
0:30:45 Like,
0:30:46 I love that gold brick
0:30:46 but we have like
0:30:47 hundreds of gold bricks
0:30:48 where I just have to
0:30:48 like stoop down
0:30:49 at my feet
0:30:49 and pick them up
0:30:50 so I’m just not going
0:30:51 to do that one
0:30:53 as big companies think.
0:30:55 But the nice thing
0:30:56 is that these are gold bricks,
0:30:57 these gold bricks
0:30:57 are bigger than
0:30:58 they’ve ever been
0:30:59 because you have software
0:31:00 that can do
0:31:01 the job of labor.
0:31:04 Which is on that note,
0:31:06 if you were running
0:31:06 OpenAI
0:31:07 and you were thinking
0:31:08 about which gold bricks
0:31:08 or how do you even,
0:31:09 what mental model
0:31:10 to think about
0:31:11 sort of what are the things
0:31:11 that you should be doing
0:31:12 first versus things
0:31:13 that hey,
0:31:14 maybe let other people do,
0:31:15 how would you be thinking
0:31:15 about that question?
0:31:17 I mean,
0:31:17 I think a lot of it
0:31:18 is where,
0:31:19 well,
0:31:19 it’s two things.
0:31:20 Number one is
0:31:21 we want to be
0:31:21 the back end
0:31:22 for everybody.
0:31:23 like the platform,
0:31:24 I think it’s two things.
0:31:25 Number one is
0:31:26 can we be the platform
0:31:27 for pretty much
0:31:27 everybody who’s
0:31:28 building anything?
0:31:30 So we’re not going
0:31:31 to go into these
0:31:31 obscure spaces
0:31:33 like orthodontic care,
0:31:34 at least not until
0:31:35 2045.
0:31:37 So let’s make sure
0:31:38 that every single
0:31:39 developer is using us.
0:31:41 And this is part
0:31:42 of why Microsoft
0:31:43 crushed Apple
0:31:44 in the 1980s
0:31:45 because Apple
0:31:46 made it really hard
0:31:47 to develop software.
0:31:48 And what’s actually
0:31:49 kind of interesting
0:31:50 is that both Apple
0:31:50 and Microsoft
0:31:52 had,
0:31:52 like Microsoft
0:31:53 started off
0:31:53 as a compiler
0:31:54 company.
0:31:54 Like their very,
0:31:55 very first products,
0:31:56 they were not
0:31:56 Microsoft Office,
0:31:57 it was not DOS,
0:31:58 they built a
0:31:59 basic interpreter
0:32:00 for the programming
0:32:01 language basic
0:32:02 and they had
0:32:02 a big business.
0:32:03 Their biggest
0:32:04 competitor was
0:32:04 Borland,
0:32:06 which only made
0:32:07 compilers.
0:32:08 And like the early
0:32:08 rallying cry,
0:32:09 if you talk to any
0:32:09 early Microsoft
0:32:10 employee,
0:32:10 was beat Philippe.
0:32:11 Philippe Kahn
0:32:12 was the CEO
0:32:12 of Borland.
0:32:14 So Microsoft
0:32:14 was focused on
0:32:15 that,
0:32:16 made a lot of
0:32:16 money on that,
0:32:17 and Apple was
0:32:17 like,
0:32:18 we should make
0:32:18 money on that
0:32:19 too.
0:32:19 And they had a
0:32:20 product,
0:32:20 it was called
0:32:21 MPW,
0:32:22 Macintosh
0:32:22 Programmer’s
0:32:22 Workshop.
0:32:23 I remember I
0:32:24 used to use it
0:32:25 in the 1980s.
0:32:26 And it was
0:32:27 like $2,000,
0:32:28 I think,
0:32:28 in 1980s money
0:32:30 to buy this,
0:32:31 you know,
0:32:31 IDE,
0:32:32 or,
0:32:32 you know,
0:32:33 programming thing.
0:32:35 And it’s like,
0:32:36 how do you
0:32:37 afford that?
0:32:37 So like,
0:32:38 but it was like,
0:32:39 we have to make
0:32:39 money on that.
0:32:40 Microsoft’s making
0:32:40 money on this.
0:32:41 And then,
0:32:41 lo and behold,
0:32:42 there were like
0:32:43 10,000 times more,
0:32:43 you know,
0:32:45 DOS and Windows
0:32:45 software products
0:32:46 than there were
0:32:46 Macintosh software
0:32:47 products.
0:32:47 And of course,
0:32:48 Apple corrected
0:32:48 that mistake
0:32:49 when the iPhone
0:32:50 came out,
0:32:51 when they did
0:32:52 now like Xcode,
0:32:52 which is the way
0:32:53 that you build
0:32:54 products for Mac
0:32:55 products,
0:32:56 or Macintosh
0:32:57 and iPhone,
0:32:58 iOS,
0:32:58 it’s free.
0:32:59 So like,
0:33:00 they kind of
0:33:01 corrected that mistake.
0:33:02 But I’d say two
0:33:03 things to answer
0:33:03 your question.
0:33:04 number one is,
0:33:04 can we be the
0:33:05 biggest consumer
0:33:06 brand in the world?
0:33:07 So ChatGPT has
0:33:07 800 million
0:33:08 weekly active users,
0:33:09 like get that
0:33:09 to 5 billion,
0:33:10 right?
0:33:11 Like even if
0:33:12 Gemini 3 came out
0:33:13 today,
0:33:13 it might be
0:33:14 five times better,
0:33:15 but are people
0:33:16 that are using
0:33:17 ChatGPT just
0:33:18 as consumers,
0:33:19 are they going
0:33:20 to switch?
0:33:20 Like maybe,
0:33:21 but it’s unlikely
0:33:22 just because they
0:33:23 kind of make that
0:33:24 their default,
0:33:25 and then be the
0:33:26 back end for
0:33:26 everybody who’s
0:33:27 building anything.
0:33:28 And that way,
0:33:29 it’s like kind
0:33:30 of all the
0:33:30 gold bricks
0:33:30 kind of come
0:33:31 to you.
0:33:33 I think the
0:33:34 other thing
0:33:35 that we should
0:33:35 anticipate,
0:33:35 we’re already
0:33:36 beginning to see
0:33:37 from some of
0:33:37 these big model
0:33:37 companies,
0:33:38 are like,
0:33:38 what are the
0:33:39 big horizontal
0:33:40 applications that
0:33:41 they can likely
0:33:42 sell to every
0:33:43 large enterprise?
0:33:43 And I think,
0:33:44 you know,
0:33:45 you saw today
0:33:46 with Google’s
0:33:47 anti-gravity launch,
0:33:48 like the IDE
0:33:48 is going to be
0:33:49 one of those
0:33:49 things.
0:33:49 I think like that,
0:33:50 you know,
0:33:50 if there’s like
0:33:51 product market fit
0:33:52 for LMs,
0:33:52 like,
0:33:52 you know,
0:33:53 coding is definitely,
0:33:55 one of the
0:33:55 top categories.
0:33:57 So I think,
0:33:57 you know,
0:33:58 thinking about
0:33:58 what are the
0:33:58 big horizontal
0:33:59 kind of
0:34:00 applications in
0:34:00 the enterprise?
0:34:01 I think there’s
0:34:01 also,
0:34:02 to some degree,
0:34:03 and,
0:34:03 you know,
0:34:04 I think this has
0:34:05 been earlier
0:34:05 to sort of
0:34:06 play out,
0:34:06 it’s sort of
0:34:07 the Palantir
0:34:07 opportunity.
0:34:09 I think we’re
0:34:09 still very early
0:34:10 in sort of
0:34:11 the proliferation
0:34:11 of this technology
0:34:12 into large
0:34:12 enterprise.
0:34:14 At the same
0:34:14 time,
0:34:15 you know,
0:34:16 unlike prior
0:34:17 product cycles,
0:34:17 you know,
0:34:18 like the cloud,
0:34:19 if I’m the
0:34:20 CEO of a large
0:34:20 public company
0:34:21 and I’m asking
0:34:21 myself,
0:34:22 do I need to
0:34:22 be in the
0:34:22 cloud?
0:34:23 It was
0:34:23 sort of
0:34:24 an esoteric
0:34:25 idea.
0:34:25 You know,
0:34:26 today,
0:34:27 I can plug a
0:34:27 prompt into
0:34:28 any one of
0:34:28 these models
0:34:29 and intuitively
0:34:30 understand the
0:34:30 impact that it
0:34:31 could have on
0:34:31 my business,
0:34:31 right?
0:34:32 The efficiency
0:34:33 gains in my
0:34:34 customer support
0:34:34 organization,
0:34:35 in my engineering
0:34:36 organization,
0:34:37 in all of my
0:34:37 back office
0:34:38 functions,
0:34:39 at the same
0:34:39 time,
0:34:40 many of them
0:34:41 don’t know
0:34:41 where to
0:34:41 start.
0:34:42 And so I
0:34:42 think you
0:34:43 will see
0:34:43 sort of
0:34:43 this
0:34:44 consultative
0:34:45 sort of
0:34:45 forward
0:34:45 deployed
0:34:47 Palantir-esque
0:34:47 sort of
0:34:48 sale into
0:34:49 very large
0:34:49 enterprise from
0:34:50 some of
0:34:50 these big
0:34:51 model
0:34:51 companies.
0:34:52 Again,
0:34:52 I think
0:34:53 we’re early
0:34:53 in that,
0:34:53 but you’ve
0:34:54 heard
0:34:55 inklings of
0:34:56 this with
0:34:57 Anthropic
0:34:58 talking about
0:34:58 wanting to
0:34:58 build into
0:34:59 financial services
0:35:00 in other
0:35:00 markets.
0:35:00 So,
0:35:02 you know,
0:35:02 I agree.
0:35:02 I think the
0:35:02 biggest
0:35:03 opportunities
0:35:03 are the
0:35:03 one that
0:35:04 Alex is
0:35:04 describing,
0:35:05 but I
0:35:05 think you
0:35:05 will see
0:35:05 them
0:35:06 selectively
0:35:07 try to
0:35:07 build
0:35:08 applications
0:35:08 that
0:35:09 cut across
0:35:09 every one
0:35:09 of those.
0:35:10 And then
0:35:10 they’ll
0:35:10 probably
0:35:11 choose
0:35:12 a few
0:35:12 sort of
0:35:13 like
0:35:13 lighthouse
0:35:13 customers
0:35:14 to
0:35:14 build
0:35:16 largely
0:35:16 bespoke
0:35:17 kind of
0:35:17 custom
0:35:18 integrations
0:35:19 into these
0:35:19 bigger
0:35:19 enterprises.
0:35:20 But where
0:35:21 the ACBs
0:35:22 just really
0:35:23 make sense.
0:35:25 In Web2,
0:35:26 there was a lot
0:35:27 of winner
0:35:27 take most.
0:35:28 You were
0:35:28 talking about
0:35:29 one of the
0:35:29 benefits in
0:35:30 AI is that
0:35:30 there’s multiple
0:35:30 winners.
0:35:31 To what
0:35:32 extent is
0:35:33 consolidation
0:35:34 inevitable?
0:35:35 Or how
0:35:35 do you
0:35:35 think
0:35:36 sort of
0:35:36 this
0:35:37 plays out?
0:35:38 Well,
0:35:39 I think
0:35:40 if you
0:35:40 have 20
0:35:40 companies
0:35:41 that are
0:35:41 all doing
0:35:41 the same
0:35:42 thing,
0:35:43 what has
0:35:44 historically
0:35:44 happened
0:35:46 is that
0:35:46 it’s a
0:35:47 bad market
0:35:47 if there
0:35:47 are 20
0:35:48 companies
0:35:48 doing it.
0:35:49 But then,
0:35:50 I don’t
0:35:50 know,
0:35:50 the bottom
0:35:51 15 just
0:35:51 go bankrupt.
0:35:53 And then
0:35:53 maybe there’s
0:35:54 some consolidation
0:35:55 where number
0:35:56 one buys
0:35:56 number two,
0:35:57 number two
0:35:57 buys number
0:35:58 three,
0:35:59 and assuming
0:35:59 that we have
0:36:00 a functional
0:36:00 FTC and
0:36:01 whatnot,
0:36:01 it’s like
0:36:02 all of this
0:36:02 is approved
0:36:02 because it’s
0:36:03 not like
0:36:03 you’re taking
0:36:04 this is like
0:36:05 orthodontic
0:36:05 clinic
0:36:06 answering
0:36:06 software
0:36:06 or something.
0:36:08 And then
0:36:09 what was
0:36:09 a bad
0:36:10 market
0:36:10 becomes
0:36:11 a good
0:36:11 market.
0:36:13 And this
0:36:13 kind of
0:36:14 goes back
0:36:14 to why
0:36:14 momentum
0:36:15 is important
0:36:16 because if
0:36:16 you have
0:36:17 20 companies
0:36:17 that are
0:36:17 all at
0:36:18 the exact
0:36:18 same
0:36:19 scale,
0:36:20 then it’s
0:36:21 actually great
0:36:21 for the
0:36:21 customer,
0:36:22 which is
0:36:22 like the
0:36:23 prices go
0:36:24 to zero
0:36:25 or they
0:36:25 converge
0:36:25 on the
0:36:26 price of
0:36:26 electricity,
0:36:27 whereas
0:36:28 if you,
0:36:29 this is not
0:36:29 saying you
0:36:30 want to
0:36:30 go build
0:36:31 a monopoly
0:36:31 in
0:36:32 orthodontic
0:36:32 answering
0:36:33 software
0:36:33 or something,
0:36:34 but rather
0:36:35 you can
0:36:36 charge more
0:36:36 if you get
0:36:37 to a certain
0:36:37 scale because
0:36:38 whatever the
0:36:39 quality of the
0:36:39 product that
0:36:40 you’re delivering
0:36:40 at the end of
0:36:41 the day is
0:36:42 just higher.
0:36:43 And you have
0:36:44 to get to the
0:36:44 critical scale
0:36:45 to get there.
0:36:46 And sometimes
0:36:46 you just need
0:36:47 these markets
0:36:47 to work
0:36:48 themselves out.
0:36:48 I mean,
0:36:49 like when I
0:36:49 was running
0:36:50 my company
0:36:50 TrialPay,
0:36:51 we had,
0:36:52 I don’t know,
0:36:53 20 competitors
0:36:54 and it was
0:36:54 tough because
0:36:55 it’s like,
0:36:55 you know,
0:36:57 everybody would
0:36:57 be pricing
0:36:58 their product
0:36:59 at a loss.
0:36:59 You know,
0:37:00 this loss
0:37:01 leader only works
0:37:01 if you end
0:37:02 up leading with,
0:37:03 like you have
0:37:03 to make money
0:37:04 at the end
0:37:04 and nobody
0:37:05 really had a
0:37:06 plan for that
0:37:06 because the
0:37:07 venture capital
0:37:07 dollars were
0:37:08 really subsidizing
0:37:09 everything and
0:37:10 that does not
0:37:10 get a good
0:37:11 market.
0:37:11 What does
0:37:12 become a good
0:37:12 market at the
0:37:12 end and
0:37:13 sometimes this
0:37:13 is what,
0:37:14 you know,
0:37:15 Vista,
0:37:15 the private
0:37:16 equity firm
0:37:16 would do,
0:37:16 is like we’re
0:37:17 going to buy
0:37:18 one as our
0:37:18 anchor,
0:37:19 we’re going
0:37:20 to go low
0:37:21 ball and put
0:37:21 the other five
0:37:22 out of their
0:37:23 misery and now
0:37:23 we end up with
0:37:24 actually a pretty
0:37:25 good product at
0:37:25 the end or a
0:37:26 pretty good business
0:37:26 at the end,
0:37:27 pretty good company
0:37:27 at the end.
0:37:28 So I think
0:37:29 that will
0:37:29 probably play
0:37:29 out the same
0:37:30 way here
0:37:30 because you
0:37:31 just can’t
0:37:31 have a market
0:37:33 where you
0:37:34 have everybody
0:37:34 lost leading
0:37:37 and nobody’s
0:37:37 big enough to
0:37:38 get any kind
0:37:38 of scale
0:37:39 effects.
0:37:40 is there going
0:37:41 to be a world
0:37:42 where the
0:37:43 19th player
0:37:43 survives?
0:37:44 I mean,
0:37:46 Jack Welch would
0:37:47 always say you
0:37:47 have to be number
0:37:48 one or number
0:37:49 two and there’s
0:37:50 no value to
0:37:50 being number
0:37:51 three through
0:37:51 100.
0:37:51 I don’t think
0:37:52 that’s changed.
0:37:52 Right.
0:37:53 Even in the
0:37:54 model provider
0:37:55 example.
0:37:56 I’m also curious
0:37:57 if prices go
0:37:57 down.
0:37:58 Yeah, I don’t
0:37:59 see how,
0:38:00 like, there
0:38:00 actually are,
0:38:01 I mean, people
0:38:02 know XAI,
0:38:03 Anthropic,
0:38:04 OpenAI,
0:38:05 Gemini, like,
0:38:05 they know, or
0:38:07 Quinn, they
0:38:08 know the big
0:38:09 ones, but there
0:38:09 are actually,
0:38:10 there’s a long
0:38:11 tale of things
0:38:11 that people
0:38:12 haven’t heard
0:38:14 of where it’s
0:38:15 like they’ve
0:38:15 raised lots of
0:38:16 money.
0:38:16 It’s just like
0:38:17 not, it works
0:38:19 fine, but how
0:38:19 can you, like,
0:38:20 the model company
0:38:20 is the most
0:38:21 cutthroat because
0:38:22 like, unless
0:38:23 you’re state, if
0:38:24 you’re state-of-the-art
0:38:24 minus, minus,
0:38:26 minus, and you’re
0:38:26 trying to earn a
0:38:27 living, it’s just
0:38:28 like, that’s just
0:38:29 not going to work.
0:38:30 So that game is
0:38:30 super cutthroat.
0:38:32 I think the one
0:38:33 area where that
0:38:35 may have diverged,
0:38:35 and Martine talks
0:38:36 about this a lot,
0:38:37 it’s like, you
0:38:38 know, when markets
0:38:38 are growing so
0:38:38 quickly, you
0:38:39 end up having
0:38:40 specialization.
0:38:41 And so I think in
0:38:41 other kind of
0:38:43 modalities, you
0:38:43 know, in some of
0:38:44 the creative tools
0:38:45 or, you know,
0:38:45 people have
0:38:46 specialized to
0:38:47 like serve, you
0:38:47 know, the up
0:38:48 market, you
0:38:49 know, like I’m
0:38:49 producing, you
0:38:50 know, movies,
0:38:51 okay, I want to
0:38:51 create sort of
0:38:52 like social, you
0:38:52 know, quality
0:38:53 content.
0:38:54 Like these are
0:38:55 different, you
0:38:55 know, markets
0:38:56 that the models
0:38:57 can kind of
0:38:57 specialize in.
0:38:58 Time will tell,
0:38:59 you know, how
0:39:01 sort of, you
0:39:01 know, defensible
0:39:02 those become over
0:39:04 time, but maybe
0:39:04 that’s the
0:39:05 optimistic take
0:39:05 that like, you
0:39:07 know, early on
0:39:07 and everything
0:39:08 looks, you
0:39:08 know, overlapping
0:39:09 and competitive,
0:39:10 but we’re still
0:39:11 so, you know, the
0:39:12 market is growing
0:39:12 that everything can
0:39:13 kind of expand and
0:39:14 people can kind of
0:39:15 specialize over time.
0:39:15 Earlier when you
0:39:16 were talking about
0:39:16 the feature versus
0:39:17 product, didn’t
0:39:18 Steve Jobs once
0:39:19 tell Drew Houston
0:39:20 that Dropbox was
0:39:21 just a feature?
0:39:23 Yeah, that’s why it’s
0:39:24 always been this
0:39:25 pejorative thing, but
0:39:26 that’s kind of the
0:39:26 point that I was
0:39:27 getting to is that
0:39:28 nobody wants to
0:39:28 like, oh, I need
0:39:29 this company.
0:39:30 No, it’s like, I
0:39:30 need this feature.
0:39:32 Every now and then
0:39:32 you see a product
0:39:33 that is not a
0:39:34 feature because it’s
0:39:35 just like so far out
0:39:35 of left field.
0:39:36 Like nobody was
0:39:38 anticipating ChatGPT
0:39:38 dominating their
0:39:40 daily workflow in
0:39:42 2022 in October.
0:39:43 But then once it
0:39:44 came out, it was
0:39:45 this like, holy
0:39:46 crap, this is
0:39:46 incredible.
0:39:47 And that’s not a
0:39:48 feature.
0:39:48 You could argue it’s
0:39:49 a feature on top of
0:39:50 your iPhone, but no,
0:39:51 the iPhone is the
0:39:52 delivery mechanism.
0:39:53 That’s a product.
0:39:55 And they’ve
0:39:55 obviously turned that
0:39:56 into a company.
0:39:58 Whereas other things,
0:39:59 it kind of is like,
0:39:59 you know, why is
0:40:00 there antivirus
0:40:01 software?
0:40:02 That almost doesn’t
0:40:02 make any sense.
0:40:03 Like, shouldn’t the
0:40:04 operating system stop
0:40:04 you from getting
0:40:04 viruses?
0:40:06 Like, why do you
0:40:07 need a third-party
0:40:08 tool to do
0:40:09 synchronization between
0:40:09 devices?
0:40:10 But it turns out,
0:40:11 like, the reason why
0:40:13 Dropbox has survived
0:40:14 and thrived since
0:40:15 Steve Jobs made that
0:40:16 comment is, like, it’s
0:40:17 really hard to do
0:40:17 well.
0:40:19 And there’s a lot of
0:40:20 other things.
0:40:21 like, once you’ve
0:40:21 built that feature,
0:40:23 you can backfill with
0:40:24 all sorts of other
0:40:25 product, which is what
0:40:26 Dropbox has done a
0:40:27 pretty good job of.
0:40:28 But it is hard because
0:40:30 this is the danger of
0:40:30 building on somebody
0:40:32 else’s platform, is
0:40:33 that, you know, I’m
0:40:34 going to build this
0:40:35 thing that they should
0:40:36 have had, right, if
0:40:37 they had the
0:40:37 foresight.
0:40:39 And if it doesn’t
0:40:40 operate in the
0:40:41 Goldilocks zone, right,
0:40:42 it’s like, wow, this
0:40:43 is so, this will, like,
0:40:45 triple Apple’s profits.
0:40:46 Let’s just say that
0:40:47 Dropbox would have
0:40:48 tripled Apple’s profits.
0:40:49 Would they have
0:40:49 dropped everything?
0:40:50 Would they have
0:40:52 focused on building that
0:40:53 versus the iPad or
0:40:54 something?
0:40:54 Whatever, like, Steve’s
0:40:56 last gizmo was, like,
0:40:56 sure.
0:40:58 But if it’s kind of in
0:40:58 this, like, Goldilocks
0:41:00 zone of irrelevance, like,
0:41:01 janitorial services, it’s
0:41:01 like, yeah, they should
0:41:02 do that.
0:41:03 But, you know,
0:41:05 platform owners get
0:41:05 lazy.
0:41:07 This is why, like, you
0:41:08 know, half the things on
0:41:09 my iPhone don’t really
0:41:10 work if they’re built by
0:41:10 Apple.
0:41:12 Try, like, any parent
0:41:13 that’s listening to this,
0:41:14 if they’ve tried screen
0:41:15 time, it’s just, like, an
0:41:16 embarrassment upon
0:41:16 humanity.
0:41:18 And because they don’t
0:41:19 have to go sell as a,
0:41:20 it’s like, they don’t
0:41:21 have to compete on
0:41:21 feature.
0:41:22 They compete on the
0:41:23 fact, they don’t even
0:41:23 compete.
0:41:24 They just, like, they’re
0:41:25 the platform.
0:41:26 They roll it out, it’s
0:41:28 going to be bad, and
0:41:28 that does create an
0:41:29 opportunity for somebody
0:41:30 to come up with a
0:41:31 feature and actually
0:41:32 out-compete the,
0:41:33 the platform.
0:41:35 But, like, you have to
0:41:36 be careful because it’s
0:41:37 like, obviously, the
0:41:38 platform owner is going
0:41:39 to go compete with you.
0:41:39 And that’s why often
0:41:40 what I find very
0:41:40 compelling about
0:41:42 entrepreneurs, when
0:41:43 they know this, like,
0:41:45 they’ve studied how is
0:41:46 it that from every
0:41:47 single platform shift
0:41:48 from, like, you know,
0:41:48 we were talking about
0:41:50 AC versus DC current,
0:41:51 like, there have always
0:41:52 been these battles for,
0:41:53 like, who’s going to be
0:41:54 the underlying, you
0:41:55 know, layer.
0:41:57 The best entrepreneurs
0:41:58 have studied this, and
0:41:59 they have a plan.
0:42:00 They’re like, I know
0:42:00 I have a feature.
0:42:01 Like, Drew knew this.
0:42:02 He’s like, I know
0:42:03 that, like, there’s
0:42:04 this stupid comment on
0:42:05 Hacker News.
0:42:05 It’s like, oh, this is
0:42:06 just, like, our sync with
0:42:07 this, that, and the other
0:42:07 thing.
0:42:08 It’s like, yeah, of course
0:42:09 Drew knows that, but he
0:42:10 built this into a $10
0:42:12 billion company because,
0:42:13 like, he had a plan.
0:42:14 And the best entrepreneurs,
0:42:15 they often, like, okay, I
0:42:17 know it’s not this
0:42:17 naivete.
0:42:18 I was like, oh, I’m going
0:42:18 to build this.
0:42:19 There’s no way that
0:42:19 they’re going to build it
0:42:20 because they’re too
0:42:20 dumb and stupid.
0:42:21 It’s like, no, they’re
0:42:21 not.
0:42:23 Like, these companies, if
0:42:23 they get their act
0:42:24 together, they will
0:42:25 marshal a lot of
0:42:26 resources to go compete
0:42:26 with you.
0:42:27 It might take them
0:42:27 five years, but they
0:42:29 will 100% do it.
0:42:30 You have to backfill your
0:42:32 feature with a product.
0:42:34 And you have to have a
0:42:35 moat for that product as
0:42:36 opposed to, like, oh,
0:42:37 yeah, like, the big
0:42:38 company will never figure
0:42:38 this out.
0:42:39 It’s like, that’s not
0:42:39 true.
0:42:41 I think what’s also
0:42:42 unique, I wrote this
0:42:43 piece a while ago called
0:42:44 The Messy Inbox Problem.
0:42:46 And it was sort of a
0:42:46 wedge strategy that we’ve
0:42:48 been observing across lots
0:42:49 of different industries.
0:42:50 And it’s just this idea
0:42:52 that you hook into a
0:42:52 bunch of your different
0:42:53 unstructured data sources.
0:42:54 Could be email, could be
0:42:55 fax, could be phone.
0:42:57 You know, Tenor, as an
0:42:58 example, has trained a
0:42:59 model to be able to
0:43:00 extract all the relevant
0:43:01 patient information from
0:43:03 those data sources to
0:43:04 plug it downstream into
0:43:05 some system of record, in
0:43:06 their case, in EHR.
0:43:08 But this exists in a CRM,
0:43:10 an ERP, what have you.
0:43:11 And I think that that
0:43:12 wedge for that feature is
0:43:13 interesting in large part
0:43:14 because it lives up
0:43:15 funnel from software.
0:43:16 Right?
0:43:18 You’re replacing the kind
0:43:19 of human-level judgment
0:43:20 of the individual, like
0:43:22 often that, you know, the
0:43:22 secretary is sort of like
0:43:23 collecting the physical
0:43:24 facts and then plugging it
0:43:25 into the EHR.
0:43:26 And so now a bunch of
0:43:27 AI companies can kind
0:43:29 of, you know, wedge in
0:43:30 and then eat away at all
0:43:32 the downstream workflows that
0:43:33 might have been their
0:43:34 point solution software
0:43:34 companies.
0:43:36 And so, you know, Tenor is
0:43:37 no longer just doing, you
0:43:38 know, the messy inbox.
0:43:39 They’re now doing
0:43:40 scheduling and prior, you
0:43:41 know, prior auth and
0:43:42 eligibility and benefits.
0:43:45 And they’ve used that wedge to
0:43:46 try to become, you know, kind
0:43:48 of the end-to-end platform.
0:43:49 Eventually, maybe they become
0:43:50 the system of record.
0:43:52 But again, because you can
0:43:53 kind of replace the human
0:43:55 labor now with software, I
0:43:56 think it’s creating
0:43:58 opportunities for these, you
0:43:58 know, features to actually
0:44:00 become products and, you
0:44:01 know, in their case, I think
0:44:02 become, you know, whole
0:44:02 companies.
0:44:04 Well, I think this is the
0:44:05 thing that in my mind is
0:44:06 very dramatically different
0:44:07 than every other platform
0:44:10 shift, is that the, it is
0:44:11 just so consensus.
0:44:13 Like, cloud was not
0:44:13 consensus.
0:44:15 Mobile was not consensus.
0:44:16 And that’s why the
0:44:17 incumbents kind of screwed
0:44:18 up.
0:44:19 Where it’s like, and then
0:44:20 sometimes it was just like
0:44:23 completely, I’ll use the
0:44:23 Silicon Valley term,
0:44:25 orthogonal to their, to
0:44:26 their business model.
0:44:28 Because it’s like, I sell
0:44:29 $5 million a year products
0:44:30 and wait a minute, I’m
0:44:31 going to charge $100,000 a
0:44:31 month?
0:44:32 Like, that’s just hard.
0:44:33 Like, how do I pay my
0:44:34 salespeople?
0:44:34 How do I make my
0:44:35 quarterly numbers?
0:44:36 So that’s why, like, you
0:44:38 know, Workday beat
0:44:38 PeopleSoft.
0:44:40 Or that’s why, you know,
0:44:41 Salesforce beat Siebel.
0:44:44 So all of these things
0:44:44 played out.
0:44:46 But behind it was this
0:44:47 concept of it’s like, that
0:44:49 new thing, that iPhone is
0:44:49 stupid.
0:44:52 Like, there’s no version of
0:44:53 the famous Steve Ballmer
0:44:54 clip of, like, him saying
0:44:55 this, nobody’s going to buy
0:44:56 an $800 phone with no
0:44:57 keyboard.
0:44:59 There’s no version of that
0:44:59 for AI.
0:45:01 It’s like, how do you find
0:45:02 a big CEO?
0:45:03 Or even a small CEO?
0:45:04 It’s like, nobody will use
0:45:05 that tool that makes you
0:45:06 100 times more productive.
0:45:08 Of course.
0:45:10 And this is why it’s kind
0:45:11 of a bonanza for most of the
0:45:13 incumbents as well, because
0:45:14 anybody who has a system of
0:45:16 record will add a button or a
0:45:18 feature, to use our parlance,
0:45:19 that will make them more
0:45:19 money.
0:45:21 So, like, they’re just kind of
0:45:22 gold bricks everywhere.
0:45:25 And the challenge, though, is
0:45:27 that there isn’t this kind of
0:45:29 white space to occupy in the
0:45:30 same way that there was for
0:45:32 cloud or for mobile or for a
0:45:33 lot of the Web 2.0 things,
0:45:34 where it’s like you just, like,
0:45:36 the incumbents screwed up.
0:45:37 They weren’t paying attention.
0:45:39 They scoff at this new
0:45:39 technology.
0:45:40 Like, nobody’s scoffing at
0:45:41 this new technology.
0:45:42 Like, everybody’s just trying
0:45:43 to embrace it.
0:45:45 But, you know, the opportunity
0:45:47 often exists where a lot of
0:45:48 the areas that just seem too
0:45:50 small, they don’t have an
0:45:51 incumbent at all.
0:45:53 Like, those actually might
0:45:54 turn out to be, like, you
0:45:55 know, trillions of dollars
0:45:56 of value.
0:45:56 And that’s kind of what makes
0:45:57 it much more exciting than,
0:45:59 like, last gen, where it’s
0:46:00 like, oh, I’m just going to
0:46:02 copy everything that was on
0:46:03 prem and make it, you know,
0:46:04 recurring billing cloud.
0:46:06 And I’m going to do that at
0:46:07 a time when, like, the big
0:46:09 guys say that’s stupid and I
0:46:09 don’t get it.
0:46:12 Some argue that, you know,
0:46:13 mobile was ultimately
0:46:14 sustaining in that although
0:46:16 there were, you know, net
0:46:17 new companies and use cases
0:46:18 that were, you know, $100
0:46:20 billion like Uber and Airbnb,
0:46:22 et cetera, that, you know,
0:46:23 the incumbents, you know,
0:46:23 some of them became
0:46:24 trillion dollar companies,
0:46:25 you know, have got it by
0:46:25 mobile.
0:46:27 When we look at the, you
0:46:28 know, business impact of the
0:46:30 AI era, what’s your mental
0:46:30 model for thinking about
0:46:31 sort of the incumbent
0:46:33 startup or kind of net new
0:46:35 company in terms of, you
0:46:36 know, value capture?
0:46:38 I think a lot of it is the
0:46:38 same.
0:46:40 Like, unless you really
0:46:41 screw up the pricing model
0:46:42 or like, you know, you’re
0:46:44 all per seat pricing, it’s
0:46:45 very, very hard to just get
0:46:46 the market to adopt something
0:46:47 that is just violently
0:46:49 different and you’re
0:46:50 operating in the public eye
0:46:51 and your technology team is
0:46:52 bad.
0:46:53 There are a lot of ands that
0:46:53 need to happen.
0:46:55 I have a hard time
0:46:56 believing that incumbents
0:46:57 will really suffer.
0:46:59 I mean, there probably are
0:46:59 some things like, you know,
0:47:01 take like one example of,
0:47:02 and this kind of goes back
0:47:03 to distribution versus
0:47:06 technology, like all of
0:47:06 these business process
0:47:08 outsourcing companies, these
0:47:09 BPOs, they’re the largest
0:47:10 employers on the planet.
0:47:13 So like Tata, Wipro, Infosys.
0:47:15 So if I’m JP Morgan and I say
0:47:18 I need a call center and this
0:47:19 call center needs to have
0:47:20 access to like customer
0:47:21 records and it needs to be
0:47:22 safe and everybody needs to
0:47:24 be trained, like, and I need
0:47:25 to have like a hundred
0:47:26 thousand people that can
0:47:27 answer the phone, you know
0:47:28 who can do that for you?
0:47:29 Infosys, right?
0:47:30 Or Tata.
0:47:33 Tata has already done the
0:47:34 integration with JP Morgan
0:47:35 in this case.
0:47:37 They might just add AI and
0:47:38 now they don’t need a
0:47:39 hundred thousand people and
0:47:40 they maintain that JP
0:47:41 Morgan contract and they
0:47:43 operate in the area of the
0:47:44 Goldilocks zone where it’s
0:47:45 like they’re going to make
0:47:46 like a hundred times more
0:47:46 money.
0:47:47 That’s one case.
0:47:48 That’s the bull case for
0:47:48 Tata.
0:47:50 The bear case is like JP
0:47:51 Morgan’s like, wait a
0:47:53 minute, like we should
0:47:54 partner with the startup to do
0:47:55 this or we should do this
0:47:58 ourselves and now like Tata
0:47:59 loses that relationship
0:48:00 altogether and it could go
0:48:00 either direction.
0:48:02 Like, you know, I think a
0:48:02 lot of these things are
0:48:05 really up for grabs, but I
0:48:06 think the default is that the
0:48:08 incumbents probably will do
0:48:09 well, but you can pick a lot
0:48:10 of these cases.
0:48:11 I mean, this is why you see
0:48:12 the public markets kind of
0:48:14 don’t know what to do, where
0:48:15 there is a case that is
0:48:16 very, very bad for a lot of
0:48:18 software companies, but
0:48:19 there is an alternative
0:48:20 case, which is like if you
0:48:21 operate in the right
0:48:23 Goldilocks zone and you’re,
0:48:24 you know, you have the
0:48:25 right momentum to
0:48:25 actually build these
0:48:26 things and embrace these
0:48:28 new technologies, like you’ll
0:48:29 maintain all of your
0:48:31 customer relationships and
0:48:32 you’re just going to have a
0:48:33 more profitable business.
0:48:34 And it’s not that you’re
0:48:35 going to do this, like the
0:48:36 most compelling thing I think
0:48:37 about AI that almost
0:48:38 everybody gets wrong is
0:48:39 like, oh, it’s going to
0:48:40 destroy all the jobs.
0:48:41 Like our beloved
0:48:43 representative from Silicon
0:48:44 Valley is like trying to
0:48:45 like eliminate AI.
0:48:47 It’s just so crazy that our
0:48:48 elected representative wants
0:48:50 to turn us back to farmers
0:48:51 of tangerines and whatnot
0:48:55 in Silicon Valley, but which
0:48:56 which again, I think is
0:48:59 crazy, but it’s not like
0:49:00 all the jobs will go away.
0:49:01 I actually think that’s not
0:49:02 going to happen at all.
0:49:03 What’s going to happen is
0:49:04 there are a lot of things
0:49:05 where it’s like if I could
0:49:07 hire somebody for a dollar
0:49:09 to do this task, I would
0:49:10 a hundred percent do that.
0:49:12 I cannot hire somebody for
0:49:12 a dollar.
0:49:13 I’ve never been able to
0:49:14 hire somebody for a dollar.
0:49:15 Now I can hire software for
0:49:15 a dollar.
0:49:17 So a lot of these tasks like,
0:49:18 you know, look at how many
0:49:21 people took taxis post
0:49:22 Uber, right?
0:49:24 And it’s like, did you hear
0:49:25 people say like you probably
0:49:26 took an Uber to get here
0:49:26 today, right?
0:49:28 Would you have taken a taxi
0:49:29 20 years ago?
0:49:30 Like no way, right?
0:49:31 Because it’s like, where
0:49:32 would you find the taxi?
0:49:33 How would you arrange the
0:49:33 taxi?
0:49:33 It’s just like way too
0:49:34 complicated.
0:49:35 Whereas once you make it
0:49:37 very, very abundant and
0:49:38 less expensive, like
0:49:39 everybody’s going to use
0:49:39 this.
0:49:41 And I think that’s the
0:49:42 that’s what wrote Kanna
0:49:43 and his ilk are missing,
0:49:45 which is it’s not like, oh,
0:49:46 I’m going to go and say
0:49:47 I’m going to like eliminate
0:49:48 all the jobs.
0:49:49 Like think of it in that
0:49:50 JPMorgan example that I
0:49:50 just mentioned.
0:49:51 It’s like, wouldn’t it be
0:49:53 cool if every single
0:49:54 customer of JPMorgan Chase
0:49:55 could have their own
0:49:56 personal friend that they
0:49:57 could talk to every single
0:49:58 day there that would help
0:49:59 them with every single
0:50:00 element of their financial
0:50:01 life or it’s like I’m
0:50:02 stuck downloading the app.
0:50:03 I can’t figure out how
0:50:04 to get it set up.
0:50:06 Oh, talk to somebody in
0:50:06 real time that will help
0:50:07 you about that.
0:50:08 Why don’t they do that?
0:50:09 It’s just like the cost is
0:50:11 known, it’s high, and
0:50:12 then the value is
0:50:12 probably low.
0:50:14 And as soon as you can
0:50:15 bring the cost down to
0:50:17 zero, now you’re going
0:50:19 to start hiring AI in
0:50:20 all of these different
0:50:21 areas that you just would
0:50:22 never bother hiring a
0:50:23 human for because it’s
0:50:24 just like you can’t train
0:50:24 the human, you can’t find
0:50:25 the human, and the human’s
0:50:25 too expensive.
0:50:27 It’s a good place to wrap.
0:50:29 Guys, thanks for coming to
0:50:29 the podcast.
0:50:30 Most don’t matter.
0:50:30 Yeah.
0:50:34 Thanks for listening to this
0:50:36 episode of the A16Z podcast.
0:50:38 If you liked this episode, be
0:50:39 sure to like, comment,
0:50:41 subscribe, leave us a rating
0:50:42 or review, and share it with
0:50:43 your friends and family.
0:50:45 For more episodes, go to
0:50:47 YouTube, Apple Podcasts, and
0:50:47 Spotify.
0:50:50 Follow us on X at A16Z and
0:50:52 subscribe to our sub stack at
0:50:54 A16Z.substack.com.
0:50:55 Thanks again for listening,
0:50:56 and I’ll see you in the
0:50:56 next episode.
0:50:59 As a reminder, the content
0:51:00 here is for informational
0:51:01 purposes only.
0:51:02 Should not be taken as
0:51:03 legal, business, tax, or
0:51:04 investment advice, or be
0:51:05 used to evaluate any
0:51:07 investment or security, and
0:51:08 is not directed at any
0:51:09 investors or potential
0:51:11 investors in any A16Z fund.
0:51:13 Please note that A16Z and
0:51:14 its affiliates may also
0:51:15 maintain investments in
0:51:16 the companies discussed in
0:51:16 this podcast.
0:51:18 For more details, including a
0:51:19 link to our investments,
0:51:21 please see A16Z.com
0:51:23 forward slash
0:51:23 disclosures.