How AI is Transforming Labor Markets

0:00:07 Now you have software agents that are effectively doing what for 65 years have been human work.
0:00:10 Is that going to increase software revenue 2x?
0:00:12 It could potentially increase it 10x.
0:00:14 It’s not even on the same kind of playing field.
0:00:16 All the data is here in the cloud,
0:00:18 all the compute is in the cloud,
0:00:19 and now you just mix them together.
0:00:22 Many of these incumbents aren’t going to evolve.
0:00:24 Every time there’s a new technology shift,
0:00:28 we have to challenge every investment thesis that we thought was not going to work,
0:00:31 and so many of them now are going to work.
0:00:32 Moats still matter,
0:00:36 and a lot of the moats in software today are the same that they’ve always been.
0:00:42 It’s both defense and offense for these companies to figure out what the hell to do.
0:00:46 There is a lot of talk and investment in AI,
0:00:50 but that opportunity is often framed in software terms,
0:00:52 the budgets, the market caps,
0:00:54 and the existing companies of the last era.
0:00:57 But might this era be fundamentally different?
0:00:59 Consider that for centuries,
0:01:02 the biggest school of science was Alchemy to turn lead into gold,
0:01:06 and today AI enables a far more powerful transmutation,
0:01:08 turning software into labor.
0:01:11 Capital no longer just buys engineers or hardware,
0:01:14 it buys code that replaces or augments labor,
0:01:17 unlocking completely new markets.
0:01:20 And that’s exactly what we discussed in today’s episode,
0:01:23 together with three A16Z general partners,
0:01:26 Alex Rampell, Angela Strange, and David Haber.
0:01:29 Together, we traced through the evolution of previous cloud eras
0:01:32 and what those tell us about what’s to come.
0:01:35 Perhaps most importantly, what’s really new here?
0:01:38 Namely that the $300 billion enterprise software market
0:01:43 is just a fraction of the multi-trillion-dollar white-collar labor market.
0:01:45 These niche markets that weren’t that interesting
0:01:48 are now potentially very interesting.
0:01:50 We’ll also cover where we are in the adoption curve,
0:01:52 wedges and defensibility for startups,
0:01:55 and how pricing is being upended.
0:01:56 If they don’t do this right,
0:01:59 they could lose all of their revenue, or most of it.
0:02:01 If they do it really well, they could 10x their revenue.
0:02:05 Finally, if you’re building this space, we’d love to hear from you.
0:02:08 Just reach out to podpitches@a16z.com
0:02:10 and we’ll route you to the right place.
0:02:12 All right, let’s get started.
0:02:17 As a reminder, the content here is for informational purposes only.
0:02:21 Should not be taken as legal, business, tax, or investment advice,
0:02:23 or be used to evaluate any investment or security,
0:02:26 and is not directed at any investors or potential investors
0:02:27 in any A16Z fund.
0:02:30 Please note that A16Z and its affiliates
0:02:33 may also maintain investments in the companies discussed in this podcast.
0:02:36 For more details, including a link to our investments,
0:02:39 please see a16z.com/disclosures.
0:02:48 Alex, you wrote an article recently, “Input coffee, output code.”
0:02:51 But this idea of turning capital into labor,
0:02:53 hasn’t this always been true?
0:02:54 What’s new here?
0:02:57 Well, it certainly has been true for a long time.
0:03:00 If you watch some old movie about the Romans,
0:03:03 you’d have all these Roman slave laborers or Roman soldiers
0:03:05 rowing in unison on a boat.
0:03:07 And then, of course, you got the seamship,
0:03:10 and then you don’t need these 50 people rowing anymore.
0:03:13 So, clearly, there has been this long historical arc
0:03:18 of technology is, in some cases, augmenting labor.
0:03:21 But it was always the brawn and not the brains.
0:03:23 I have a bunch of people that are so enclosed,
0:03:25 and now I have the loom.
0:03:28 But everything that was what we would now call white collar,
0:03:29 that hasn’t happened before.
0:03:33 And what I talked about was the three or four different eras
0:03:35 of software, just story information.
0:03:38 So for the longest time, if I wanted to keep track of who
0:03:41 is on my airplane from the Wright brothers onward,
0:03:43 I would have a filing cabinet with, it’s like, oh,
0:03:47 here’s Pan Am Flight 192, and here’s who’s on it.
0:03:50 And I’m writing down their name, and then they might call or send
0:03:52 a telegram back in the old days of,
0:03:54 I don’t want to be on that plane anymore than I erase it,
0:03:55 and then I refile it.
0:03:58 And one of the first examples of digitization
0:04:02 of the filing cabinet was something called Saber.
0:04:05 And this was developed by American Airlines,
0:04:08 I think in 1959 or 1960, in concert with IBM.
0:04:10 And this was one of the first examples
0:04:12 of taking this filing cabinet that really
0:04:15 kept track of who’s on Pan Am Flight or American Airlines
0:04:18 in this case, and putting it in a database.
0:04:21 So now, instead of having filing cabinets
0:04:22 with lots of erasers and whiteout,
0:04:25 you replace the filing cabinet with a computer.
0:04:29 And this, in turn, begot travel agents and travel agencies,
0:04:32 because they had a thin client, a little computer,
0:04:35 a terminal that I remember like booking airline tickets
0:04:36 with my mom in the 1980s.
0:04:38 You’d go into the travel agent, and they
0:04:40 had like a green screen computer that
0:04:43 connected to a mainframe in Texas, which is where Saber was.
0:04:47 So the first realm of software was really called from 1960
0:04:51 onwards when computers became a thing of take a filing cabinet.
0:04:54 So it could be the HR filing cabinet.
0:04:57 It could be the medical filing cabinet.
0:04:59 It could be the financial filing cabinet,
0:05:01 and put that in software.
0:05:02 And what was that?
0:05:05 It’s a database with a front end to actually go enter things.
0:05:08 So Quicken famously did this in the 1980s
0:05:09 for financial statements.
0:05:11 There was a company called PeopleSoft that famously
0:05:12 did this.
0:05:15 It was the first HR filing cabinet put as software.
0:05:18 Saber obviously did this for airline tickets.
0:05:20 Email, I would even argue, did it for mail.
0:05:22 It’s like you have files of mail, and now you just
0:05:25 have a filing cabinet of people on your computer.
0:05:27 But the actions that were done on the software
0:05:28 were really the same.
0:05:33 So imagine an HR department that has 50 people working in HR.
0:05:35 They might have one person that’s in charge
0:05:37 of the filing cabinets, and the HR person says, hey,
0:05:39 get me David’s file, because I want
0:05:41 to talk to him about something, which is very scary.
0:05:42 Don’t worry.
0:05:43 But get me David’s file.
0:05:45 So the filing cabinet person–
0:05:47 the gopher– this is what the filing cabinet person would
0:05:48 often be called.
0:05:49 Go for something.
0:05:49 That’s what it means.
0:05:51 There’s the gopher that obviously Bill Murray tries
0:05:53 to kill in Caddyshack, but then there’s
0:05:56 the gopher of the person who goes for something.
0:05:59 If you are a gopher at Creative Artists Agency,
0:06:01 you are going and getting files.
0:06:03 So that person went away.
0:06:04 The filing cabinet went away.
0:06:07 It was more efficient from a space perspective.
0:06:10 But the 50 HR people are still 50 HR people today.
0:06:13 So round one of software was really take the filing cabinet
0:06:15 and put it not as physical files,
0:06:17 but as a database with the front end.
0:06:21 Round two, which started arguably 1998, 1999,
0:06:23 and this is what Salesforce did.
0:06:26 The idea of a customer relationship management product
0:06:28 has been around for a long time.
0:06:30 The Rolodex is an actual physical thing
0:06:31 where you put every business card,
0:06:33 organize them alphabetically, and you’d
0:06:35 find the person that you want to contact that way.
0:06:39 Salesforce put that not in software, but in the cloud.
0:06:42 So what QuickBooks had done for a long time
0:06:44 or something called Great Plane Software
0:06:48 had done for a long time, Netswe did in the cloud.
0:06:50 They put financial statements in the cloud.
0:06:53 Zendesk put email support in the cloud.
0:06:56 So it was still software, but instead of having a giant
0:06:58 mainframe in your office somewhere,
0:06:59 you now had it in the cloud.
0:07:00 But again, going back to the HR example,
0:07:04 so PeopleSoft first did this with on-premise,
0:07:08 mainframe computer HR filing cabinet now as software.
0:07:11 And then Workday came along, did the exact same thing,
0:07:12 and now it’s in the cloud.
0:07:13 It’s just much, much easier.
0:07:15 You don’t have to have a dedicated IT team
0:07:18 to go worry about your server exploding in flames
0:07:20 if your office building burns down or something.
0:07:22 So it’s more secure.
0:07:25 So it was software 1.0, then became software 2.0,
0:07:27 which is in the cloud.
0:07:31 And that played out from call it 1998 to maybe 2010.
0:07:34 And then that grew a little bit with the insertion
0:07:36 of financial services because the way that I like to think
0:07:39 about this is how many restaurants need and will pay
0:07:42 tens of thousands of dollars a year for software.
0:07:45 Like Pan Am needed this in 1960,
0:07:48 but does a restaurant with one location,
0:07:51 are they going to spend $100,000 on a server
0:07:52 and pay for software?
0:07:55 No in 1960, no in 2000.
0:07:58 But when the idea of bundling and payment processing
0:08:00 became a thing in other financial services,
0:08:02 now the market became big enough
0:08:05 for restaurant software to exist.
0:08:07 And this is where Toast came from.
0:08:10 Toast is a $15 billion company that does this
0:08:12 or Service Titan, again,
0:08:14 what’s the software market in 1965
0:08:16 for HVAC contractor zero?
0:08:17 What’s the cloud software market
0:08:19 for HVAC contractor zero?
0:08:20 But once you bundle on these other things,
0:08:21 it becomes big enough.
0:08:23 But the point that I’m getting to
0:08:27 is that the same 50 HR people that worked in 1960
0:08:31 are the same 50 HR people in 2024.
0:08:34 The same email support team in 2024
0:08:37 was the phone support bank in 1985,
0:08:42 was the letter writing typewriter support team in 1965.
0:08:44 What’s exciting about AI
0:08:46 is that it’s taking this filing cabinet
0:08:48 and now allowing actions on the filing cabinet.
0:08:50 And that’s what I think is really revolutionary
0:08:52 because you can actually ask
0:08:56 the software filing cabinet application like Workday,
0:08:59 hey, I want to add a dependent
0:09:01 and now Workday will do all the work involved with that
0:09:03 and Workday can charge a premium for that.
0:09:06 So why is it input, coffee and output code?
0:09:10 The whole idea is that you now have software engineers
0:09:12 that can build products
0:09:15 on top of these cloud-based filing cabinets
0:09:18 that now do the job that the end user
0:09:23 of that software product did in 1960, 1970, 1990, 2000,
0:09:28 2010, 2023, 2024, but 2024, 2025, onward.
0:09:30 Now you have software agents
0:09:34 that are effectively doing what for 65 years
0:09:35 have been human work.
0:09:37 Yeah, and that sounds really important.
0:09:38 So I want to underscore that.
0:09:39 You kind of went through the eras.
0:09:43 You have the non-software era, then you have software,
0:09:44 then you moved into cloud,
0:09:46 then you have this financial services enabled,
0:09:49 cloud era, and now we’re in this new era.
0:09:50 Can you talk about how,
0:09:52 as we chart through these different eras,
0:09:56 the scale maybe in this new era is fundamentally different?
0:09:57 Yeah, it’s completely different
0:10:00 because it’s really comparing wages to software
0:10:03 and to take an example from a completely different field
0:10:05 that also has no software market,
0:10:09 there are about 4.7 million registered nurses in the US.
0:10:12 Average wage for a nurse is a little over $120,000 a year.
0:10:15 So it’s a high-paying profession.
0:10:18 And that means that the annual nurse,
0:10:20 not software market, but wage market
0:10:23 is over $600 billion a year, which is a lot.
0:10:26 And the worldwide software market is under $600 billion.
0:10:27 Not just America.
0:10:29 By the way, that nurse figure is just for the US.
0:10:32 Of course there are nurses in the UK and France and Angola,
0:10:34 like every country on earth, right?
0:10:35 Probably Antarctica has nurses, I’m sure.
0:10:36 Almost certainly.
0:10:37 Almost certainly.
0:10:38 They’re probably even higher paid down there.
0:10:42 So the labor market is enormous,
0:10:45 but what is the dedicated software market for nurses,
0:10:47 like probably zero,
0:10:49 because nobody took the time to develop software.
0:10:50 And you could try developing software,
0:10:52 but it’s just, there was no budget.
0:10:53 Economics didn’t make sense.
0:10:54 Exactly.
0:10:56 But if every hospital in the US
0:10:58 currently has a nursing shortage,
0:11:00 where maybe you’re in Minneapolis,
0:11:03 where there’s a giant Somalian expat community,
0:11:06 and you need nurses that speak that language,
0:11:08 how do you find them?
0:11:09 And it takes three to four years
0:11:10 to actually get trained as a nurse.
0:11:12 Now you have a software product
0:11:15 that can deliver not everything that a nurse does.
0:11:16 Of course, it can’t be a phlebotomist,
0:11:18 it can’t perform CPR,
0:11:21 but it can call you the night before your colonoscopy
0:11:23 and say, “Don’t eat food.”
0:11:25 And it can say that in 45 languages,
0:11:26 and it can actually have a conversation with you.
0:11:28 So that’s a labor example,
0:11:31 or going back to financial services land.
0:11:32 NetSuite is used by,
0:11:35 I think it’s 70% of companies that go public
0:11:36 are on NetSuite.
0:11:37 It’s some very, very large–
0:11:39 I’m sure I’ve heard a podcast ad with that exactly.
0:11:41 Yeah, maybe it’s higher.
0:11:42 Something on that order of magnitude.
0:11:43 If I’m wrong by 10 points,
0:11:44 it doesn’t belittle the point.
0:11:46 But the key thing is,
0:11:48 you always have people that are paying you late.
0:11:49 Like you’ll look at your accounts receivable,
0:11:50 and you’ll see, wow,
0:11:52 a bunch of customers owe me millions of dollars.
0:11:53 And that’s what you’ll see
0:11:55 when you look at your financial statements.
0:11:56 And then again, this is where humans
0:11:59 take the operation on that information.
0:12:01 I have teams that are in collections
0:12:03 that will go call you and remind you to pay,
0:12:05 or I’m gonna cut off the product.
0:12:08 That’s an operation that can now be done within NetSuite.
0:12:09 They haven’t done this yet,
0:12:10 but I’m sure they will.
0:12:13 And versus charging for the filing cabinet.
0:12:14 Now they can say, well,
0:12:17 we know that you wanted a higher five collections people.
0:12:18 We know that you pay those collections people
0:12:21 $80,000 a year with benefits and everything else.
0:12:23 We know that it takes a year to train them.
0:12:25 Now our software product can do that
0:12:29 not for $80,000 a year, but for $2,000 a year.
0:12:30 And that’s incredible.
0:12:32 So the question is actually,
0:12:34 how will the customer think about this?
0:12:37 Because once they start paying NetSuite 10 times
0:12:38 more than they paid NetSuite last year,
0:12:41 they’re like, wow, my software budget has ballooned.
0:12:43 I gotta cut my software spend.
0:12:46 Or will they say, wow, I’m saving so much money
0:12:48 because as opposed to these five job openings,
0:12:50 I’m waiting to pay $400,000 a year
0:12:52 for these five collections people.
0:12:55 Now I can pay $10,000 a year to NetSuite,
0:12:57 and I’m paying more for software, but less for labor.
0:12:59 And that part, it’s very, very new,
0:13:01 but a lot of the hyper growth that we’re seeing
0:13:03 in this category of company
0:13:06 is because they are really moving into the labor market
0:13:07 and less the software market.
0:13:09 And that nurse example is a prime example of that.
0:13:11 – Definitely, it’ll be very interesting
0:13:12 to see the appetite.
0:13:15 If you think back to the app store early days,
0:13:19 when people were so reticent to pay 99 cents for an app
0:13:21 for arbitrary reasons,
0:13:23 but just because it was what they were used to paying.
0:13:26 So as we round out these areas of clouds,
0:13:29 you mentioned PeopleSoft or Quicken or Zendesk,
0:13:31 and all these companies that are capturing data,
0:13:33 how are we uniquely set up now
0:13:35 because of the previous areas?
0:13:38 – Yeah, so I would argue that if we just went straight
0:13:41 to AI in 1960, this just wouldn’t have worked
0:13:43 because you still needed human input
0:13:45 to go collect the customer information.
0:13:49 We have everything built out where right now
0:13:51 this isn’t an API,
0:13:53 but if I wanna go answer a customer question,
0:13:55 the question’s already on the internet,
0:13:56 it’s already in a database.
0:13:58 Like all of these things have set this up
0:13:59 to be the ultimate platform.
0:14:01 So this is why we love investing
0:14:03 in what I would call systems of record.
0:14:06 And a system of record is just something
0:14:08 that has every single piece of minutia
0:14:09 that runs a business.
0:14:11 And it could be the strangest business
0:14:12 that you can imagine.
0:14:14 If I’m running a laundromat,
0:14:17 there is actually dedicated laundromat management software.
0:14:19 All of these systems of record that have popped up
0:14:21 for all sorts of different businesses
0:14:22 and all sorts of consumer use cases.
0:14:26 Like the fact that now the mainstream form of communication
0:14:30 for many adults and children is texting and email.
0:14:33 Like it’s not voice, it’s all in the cloud.
0:14:35 Now you can perform these operations.
0:14:39 So I would almost say there’s been a 60 year period
0:14:43 of digitization of physical things,
0:14:44 putting them in the cloud.
0:14:45 Why is the cloud part important?
0:14:49 Well, again, if this were the mainframe era of the 1970s,
0:14:53 how would you get the AI that’s in some Google server somewhere?
0:14:56 How would it actually have access to all the information
0:14:59 that’s in some server basement in Indiana?
0:15:01 Like that’s really hard to do.
0:15:03 So all the data’s here in the cloud,
0:15:04 all the compute is in the cloud
0:15:05 and now you just mix them together.
0:15:08 So the fact that systems of record
0:15:10 for so many different types of businesses
0:15:12 and so many different kinds of consumer use cases
0:15:15 are now widespread and hundreds of billions of dollars
0:15:18 of company market cap has been created
0:15:21 from these systems of record either horizontal or vertical.
0:15:23 And a vertical one would be something like a toast
0:15:25 that vertically runs a restaurant.
0:15:27 And a horizontal one would be like a Zendesk
0:15:30 that just does customer support software in the cloud
0:15:32 for every different type of company.
0:15:34 – Building on that, if you take the toast example,
0:15:35 start with the cloud wave,
0:15:37 move to the financial services wave.
0:15:38 It was initially a hypothesis
0:15:40 that these vertical SaaS companies
0:15:42 would make a lot more from financial services.
0:15:46 Fast forward to today, 80% of toast revenue
0:15:48 is payments, insurance,
0:15:50 all sorts of financial services versus software.
0:15:52 And so if you’re operating toast
0:15:54 or one of my favorite examples is MindBody
0:15:57 which runs fitness studio software.
0:15:59 It does scheduling for employees, it’s a CRM.
0:16:02 They also make a lot of money from financial services,
0:16:04 but they still need a lot of people
0:16:05 beyond the yoga instructor.
0:16:07 You’ve got your financial back office,
0:16:09 you’ve got people answering the phone
0:16:10 to answer very basic questions.
0:16:14 Like all of that can start to be done with AI.
0:16:16 And I think the most bullish version of that
0:16:19 is all right, you then don’t need to hire
0:16:21 people to do the tasks that are not human facing
0:16:23 that AI can do better.
0:16:25 Is that going to increase software revenue 2x?
0:16:27 It could potentially increase it 10x
0:16:28 depending on how the customer views it
0:16:30 and how much they’re willing to let their software budget
0:16:32 bleed into their labor budget.
0:16:33 – I think part of the challenge
0:16:34 or the potential for disruption
0:16:35 is that the pricing model
0:16:37 may need to change pretty significantly, right?
0:16:40 You talk a bit about this in your piece, Alex,
0:16:43 where Zendesk today is charging on a per seat basis.
0:16:46 But if you’re actually eating away at some of the labor,
0:16:48 you can charge per the output of work.
0:16:49 And how does that create the potential
0:16:51 both for increased ACV,
0:16:53 but again, creates potential disruption
0:16:55 for a lot of these larger incumbent players
0:16:57 given their existing pricing structure.
0:16:59 – Can we actually talk about that example from Zendesk?
0:17:02 What is the difference between the software component
0:17:03 and the human component?
0:17:06 – Well, it shows how stark the difference is.
0:17:10 So most companies like Salesforce charges per seat.
0:17:13 And so Zendesk, at the time that I wrote my piece,
0:17:14 they might have changed their pricing a little bit,
0:17:17 but it was $115 per seat per month.
0:17:20 So imagine that you have 1,000 people
0:17:22 that work in an email-based support center
0:17:24 and they use Zendesk.
0:17:27 And Zendesk then profits as you hire more people
0:17:29 because 1,000 is better than 100.
0:17:34 So $115,000 a month is about $1.4 million a year
0:17:36 in spend on software for Zendesk.
0:17:39 And Zendesk, I think, has about $2 billion in revenue,
0:17:40 something in that order of magnitude
0:17:43 of annual recurring revenue from all these seats
0:17:45 that are paying every single month.
0:17:48 And of course, they want their customers to grow seats.
0:17:51 Now, if you assume that each seat,
0:17:52 how much is that person paid?
0:17:54 How much does their healthcare cost?
0:17:57 How much does their stipend for commuting
0:17:59 and yoga benefit and all these other things
0:18:00 that the company throws in?
0:18:04 Maybe it’s $50,000 a year per person.
0:18:06 So what’s 50,000 times 1,000?
0:18:08 Again, 1,000 seats, that’s $50 million.
0:18:10 So you’re spending $50 million on people.
0:18:12 And by the way, it’s very hard
0:18:13 to hire and train these people.
0:18:16 So I think one misnomer is like,
0:18:18 oh, AI is terrible, it’s gonna end all employment
0:18:20 and it’s gonna be chaos.
0:18:21 I always like to point out that
0:18:23 when the United States was founded,
0:18:26 97% of people in the US were farmers.
0:18:28 And most of them were put out of jobs
0:18:29 by things like the tractor,
0:18:31 but that’s actually fine because they moved on
0:18:33 to other kind of productive parts of labor.
0:18:34 And by the way, the average life expectancy
0:18:36 was a little over 30 back then.
0:18:37 So I feel like these have done–
0:18:38 I’ll take what we have.
0:18:38 Things have gone better.
0:18:41 Like I like penicillin and all the other benefits
0:18:42 that we’ve got and fertilizer
0:18:44 and all these things that have made our lives better.
0:18:46 So you have $50 million a year for people
0:18:49 and then you have $1.4 million a year for software.
0:18:50 Which one’s bigger?
0:18:54 And this is the concern is that on the intermediate basis,
0:18:55 and Zendesk is actually very lucky
0:18:57 because they were a public company
0:18:58 and they got taken private.
0:19:01 So two big private equity firms bought it
0:19:03 and they’re actually working on this right now
0:19:05 because they’re like, uh-oh,
0:19:07 if we make AI really good,
0:19:10 then the customer that has 1,000 seats
0:19:12 might cut down to 10 seats
0:19:16 because there are two forms of AI tools.
0:19:17 There’s more than two,
0:19:19 but the common example that we talk about
0:19:22 is there’s autopilot and then there’s copilot.
0:19:25 So copilot is a productivity enhancer.
0:19:27 So I’m trying to figure out
0:19:29 how do I answer Angela’s query?
0:19:30 I just got hired yesterday.
0:19:31 I don’t even know where the bathroom is.
0:19:32 What do I do?
0:19:33 And then it’s like, hey,
0:19:35 we think this is the right answer.
0:19:38 So it makes Angela so much more productive in her job
0:19:39 and that’s great.
0:19:42 Autopilot is like Angela quit yesterday
0:19:44 and we need somebody else to go answer emails
0:19:46 because it’s Black Friday.
0:19:46 What do we do?
0:19:47 What do we do?
0:19:51 Okay, now we just throw the tool at the customer directly
0:19:53 and have them answer the questions.
0:19:55 And that’s the big danger.
0:19:57 Copilot is actually a danger for revenue as well
0:20:00 because why do I have 1,000 support reps
0:20:02 because they can only answer 10 questions a day?
0:20:04 And I get 10,000 queries a day.
0:20:05 So it’s just basic math.
0:20:07 Now with Copilot,
0:20:09 each one of my reps can answer a hundred questions a day.
0:20:11 I only need a hundred reps.
0:20:13 So now Zendesk lost 90% of its revenue.
0:20:14 And then by the way,
0:20:17 if autopilot becomes a thing
0:20:18 and it actually works very, very well,
0:20:20 then I need nobody and therefore I sell no seats
0:20:21 if I’m Zendesk.
0:20:25 So it’s both defense and offense for these companies
0:20:27 to figure out what the hell to do.
0:20:29 Because if they play offense,
0:20:31 they could maybe 10x the revenue
0:20:32 because of the example.
0:20:34 Like it’s $50 million for people or 1.4 million for,
0:20:37 I’d rather get 50 million if I’m Zendesk than 1.4 million.
0:20:38 But they’re probably not going to be able
0:20:40 to take all 50 million, right?
0:20:42 Because ultimately like the cost
0:20:44 of delivering these services is very low.
0:20:46 The moat is not that high.
0:20:47 It’s higher for companies
0:20:49 that have the system of record, I would argue.
0:20:51 Because all of your past correspondence
0:20:54 with all of your customers is in Zendesk.
0:20:55 Where if I’m Salesforce,
0:20:56 like every single communication
0:20:58 that I’ve ever had with any of my customers,
0:21:01 my pipeline, everything, it’s in Salesforce.
0:21:02 It’s hard to yank that stuff out
0:21:03 because everybody’s using it.
0:21:05 And it’s not like this binary thing
0:21:07 where tomorrow we’re all autopilot.
0:21:09 We’re going to see a lot of these co-pilot tools.
0:21:11 They will run on the systems of record.
0:21:14 But even as they’re running on the systems of record,
0:21:15 now I need less seats.
0:21:17 That’s why Salesforce,
0:21:19 it’s a $200 billion plus public company.
0:21:20 If they don’t do this right,
0:21:23 they could lose all of their revenue or most of it.
0:21:25 If they do it really well, they can 10x the revenue.
0:21:26 And it’s like, where is it going to go?
0:21:28 – As early as HBCs, we’re very excited about this.
0:21:29 – Yes, yes.
0:21:31 – This is the question, right?
0:21:32 – You know more about finding
0:21:33 the next entrepreneurs versus the end of the year.
0:21:35 – Well, it’s an interesting moment in time
0:21:37 for enterprising young founders to reinvent the model
0:21:39 and charge radically different
0:21:41 because many of these incumbents aren’t going to evolve.
0:21:43 They’re not going to change their perceived pricing
0:21:44 and they risk disruption.
0:21:46 – And is that the wedge, right?
0:21:47 If you’re a startup and you’re trying to figure out
0:21:48 how do I enter the market,
0:21:51 especially when you have companies like Salesforce
0:21:53 which have the system of record,
0:21:55 they have all the data, they have all the customers.
0:21:57 Is that the wedge where you say, I’m going to undercut
0:21:59 and I’m going to charge a 10th of the price,
0:22:01 even though I still have great margins
0:22:04 because I’m entering the labor part of the equation?
0:22:05 – I think so, yeah.
0:22:06 I wrote a piece recently that I called
0:22:07 the messy inbox problem,
0:22:10 which is my way of describing a wedge strategy
0:22:12 that we’re seeing across lots of different industries.
0:22:15 And the idea is basically that there’s a class of founders
0:22:16 that are building software products
0:22:18 to solve a lot of the,
0:22:21 what was historically judgment intensive work
0:22:23 in lots of different industries.
0:22:25 There is some sort of human administrator’s job.
0:22:28 It is to basically extract information
0:22:30 from a wave of unstructured information,
0:22:33 whether it’s emails, faxes, transcribing phone calls,
0:22:35 and then put that information
0:22:37 into one of these downstream systems of record.
0:22:39 It could be an EMR, it could be an ERP,
0:22:41 it could be a CRM system.
0:22:44 And historically that work lived upstream
0:22:45 of any of that software, right?
0:22:47 Because it was the human’s job and software couldn’t do that.
0:22:50 Now we’re seeing companies sort of wedge in
0:22:54 and replace again, that messy inbox problem with software
0:22:55 and slowly begin to eat away
0:22:57 at all the kind of downstream workflows.
0:22:59 And over time, I think that the thesis is that
0:23:03 while that initial wedge is highly differentiated
0:23:06 against the human, it really is the opportunity
0:23:06 to eat away at everything else
0:23:10 and become the kind of new AI native system of record.
0:23:12 We have a company as an example called Tenor
0:23:14 that is doing this in a healthcare context.
0:23:16 So the problem that they’re solving specifically
0:23:17 is around patient referrals.
0:23:20 So you go to your general practitioner,
0:23:21 they’re referring you to a specialist,
0:23:23 could be a dermatologist or an imaging center.
0:23:27 Today they’re often faxing your medical records.
0:23:30 And it is somebody’s job to go physically to the fax machine,
0:23:34 re-enter that information into the EMR system.
0:23:36 Tenor has trained a model against,
0:23:38 I think four million healthcare specific documents,
0:23:40 and now can basically extract all that information
0:23:42 about the patient programmatically
0:23:45 and have effectively begin to solve
0:23:47 this patient intake problem.
0:23:50 And they’re able to reduce now about 90% of the admin costs
0:23:51 of that patient intake
0:23:54 before the patient’s actually sitting in the clinician.
0:23:57 So they again, wedged in with the messy inbox problem.
0:23:58 And over time they’re now eating away
0:24:01 at things like scheduling and eligibility and benefits.
0:24:03 And over time we’ll see if they become
0:24:05 the kind of core AI native system of record.
0:24:06 – And something you’re pointing to there
0:24:08 is also the defensibility of it all, right?
0:24:11 So you can get the wedge maybe through pricing,
0:24:13 but how do you actually protect your customers?
0:24:15 I think one place people jump to is okay,
0:24:17 you need your own models
0:24:18 or you need some sort of proprietary data.
0:24:22 Is that the defensibility of this future
0:24:23 or how do you think about that?
0:24:24 – It’s sort of this distinction
0:24:26 between differentiation and defensibility.
0:24:30 I think AI is an incredible catalyst for differentiation, right?
0:24:32 Solving the messy inbox problem with software
0:24:34 is a thousand times better than the human doing.
0:24:36 It’s not even on the same kind of playing field.
0:24:38 Super differentiated way to wedge in,
0:24:40 again, kind of own the downstream workflow.
0:24:43 Is that wedge product alone defensible?
0:24:45 I would argue no, right?
0:24:47 Today it feels like magic
0:24:49 to the providers that they’re working with,
0:24:50 but I think that capability
0:24:52 is gonna become commoditized over time.
0:24:53 They may have an advantage
0:24:55 because they’ve trained a model for now,
0:24:56 but I think that is ephemeral.
0:24:59 I think the defensibility comes from, again,
0:25:01 owning all of the downstream workflows,
0:25:02 deeply integrating themselves
0:25:04 into every other system that they have,
0:25:09 effectively owning that sort of core end-to-end workflow.
0:25:12 And I guess the hot take would be that moats still matter
0:25:14 and a lot of the moats in software today
0:25:16 are the same that they’ve always been.
0:25:18 So becoming a system of record,
0:25:20 having a network effect, becoming a platform,
0:25:22 having virality baked into your product,
0:25:25 deeply embedding yourself into the existing system
0:25:26 so it’s hard to rip out.
0:25:28 These were all the heuristics
0:25:30 that we would always have looked for in software,
0:25:32 and they’re still true today.
0:25:33 I agree with all of that.
0:25:35 The other way to think about this
0:25:39 is why did software start with airlines?
0:25:41 Well, a lot of people traveled.
0:25:43 Airplane tickets were very expensive,
0:25:45 and this was a pittance for them,
0:25:46 and it made so much sense
0:25:49 versus having throngs of filing cabinets and gophers,
0:25:53 and it makes sense to go pay hundreds of thousands of dollars
0:25:57 in 1960s money to go buy some giant IBM mainframe.
0:26:00 And this is why I brought up that financial services example.
0:26:03 Software for restaurants did not make sense.
0:26:05 It just wasn’t a problem to be solved,
0:26:07 and the market wasn’t big enough.
0:26:09 And you made the market big enough once to Angela’s point,
0:26:12 you threw in payment processing and insurance,
0:26:14 all these other things that they were paying for anyway.
0:26:16 So you can either try to come up with a wedge,
0:26:17 as David mentioned,
0:26:20 and then figure out how you expand your wedge
0:26:22 and make it defensible and become the system of record.
0:26:23 The other thing that you do
0:26:26 is you just find things like restaurants in 1980
0:26:28 that have no software and needed no software,
0:26:30 wouldn’t pay for software,
0:26:32 but their labor budgets are enormous.
0:26:34 And the example that I would give here,
0:26:36 what is the incumbent software product
0:26:41 for compliance officers at banks and financial institutions?
0:26:46 Excel, Word, Microsoft Edge browser looking for bad things.
0:26:47 So what does a compliance officer do?
0:26:49 I mean, it’s now in the news a lot
0:26:50 because of this debanking thing.
0:26:53 And I found this on the Bureau of Labor Statistics,
0:26:54 the fourth fastest growing job in America
0:26:56 is compliance officer.
0:26:59 There isn’t an incumbent software product that they use.
0:27:02 Every bank and financial services company on earth,
0:27:04 they’re all hiring for compliance officers.
0:27:06 It takes a long time to train them.
0:27:07 And what if account openings goes down?
0:27:08 I don’t need as many.
0:27:10 What if account openings go up? I need more.
0:27:11 It takes in some cases like a month
0:27:14 to open a business bank account at a bank
0:27:17 because the compliance officers are backlogged.
0:27:19 What if you deliver that via software?
0:27:21 And there is no software, there is no incumbent
0:27:22 that can now add this AI module.
0:27:24 And the same way that NetSuite was an incumbent
0:27:28 for accountants and financial officers at companies
0:27:32 where they can add a module for AI work of collecting money.
0:27:34 So you find these other areas
0:27:36 where there really isn’t an incumbent
0:27:39 or the incumbent is like Microsoft Excel.
0:27:41 And sometimes these are just so bizarre.
0:27:44 Like I wouldn’t think about compliance officer
0:27:45 until I saw this random report
0:27:47 from the Bureau of Labor Statistics
0:27:49 that showed manicurist is number one.
0:27:51 And that’s a hard one to have AI do.
0:27:53 And then number four is compliance officer.
0:27:54 And then you talk to banks like,
0:27:56 what software do you use? And it’s Excel.
0:27:59 Again, giant labor budget,
0:28:01 not enough people that are using software.
0:28:03 You can not have to worry about the incumbent.
0:28:05 And it still is a wedge,
0:28:07 but you could probably turn it into a system of record.
0:28:08 And while we’re into venture-backed companies
0:28:11 or even non-ventured-backed companies built in this space,
0:28:13 it’s just like, well, I can’t charge that much money.
0:28:14 It’s just not a big market in the same way
0:28:17 that like there was no restaurant software market in 1980.
0:28:19 It’s the exact same reason,
0:28:20 but there’s enough budget there
0:28:23 to go fill this like very, very pressing need.
0:28:25 – I think one of the most interesting parts of our job
0:28:27 is every time there’s a new technology shift,
0:28:29 we have to challenge every investment thesis
0:28:31 that we thought was not gonna work.
0:28:33 And so many of them now are going to work.
0:28:35 If we come back to financial services,
0:28:39 there’s a lot of pretty terrible systems of record
0:28:42 where smart people have tried to get them ripped and replaced,
0:28:43 and it was just not gonna happen.
0:28:45 And my new conclusion with AI
0:28:47 is not that they would never do it,
0:28:49 it’s that the replacements were 2x better,
0:28:50 they weren’t 10x better.
0:28:53 And so if we come back to compliance, lots in Excel,
0:28:54 but for instance, you’ve probably read
0:28:57 the $4 billion fine by TD in transaction monitoring, right?
0:29:00 And so they said in an old transaction monitoring system,
0:29:01 MacTomyce is one of them.
0:29:02 They should probably get one
0:29:04 that throws up fewer false alerts,
0:29:06 but they were trying to clear a backlog
0:29:09 of several tens of thousands alerts.
0:29:11 They can’t hire enough compliance people.
0:29:13 So now an interesting wedge in
0:29:16 is we’ll provide you with all of these agents.
0:29:18 And oh, by the way, we also have
0:29:19 a much better transaction monitoring system
0:29:21 that is actually gonna fix the problem.
0:29:24 So this labor plus software bundle
0:29:25 also helps the sales process
0:29:26 and then helps the defensibility
0:29:28 ’cause you’re really solving the major problem,
0:29:32 which is better software and also I can’t hire the people.
0:29:34 – Yeah, and I think that’s such an important point
0:29:36 because you all have pointed out different areas
0:29:39 where quite frankly, the labor is not there.
0:29:42 And so as we’re talking about software disrupting labor,
0:29:43 the natural question is, okay,
0:29:45 so what happens to all these jobs?
0:29:47 But maybe we can talk about both that and the flip side
0:29:49 of what new jobs are created
0:29:51 because the previous arcs we saw,
0:29:55 like product managers, UX designers, social media managers,
0:29:58 those were all remnants of the previous era.
0:30:01 How do we think this will shape up?
0:30:03 – It’s always hard to prophecy these things
0:30:05 because in 1789 or something,
0:30:07 it would be hard to say what will these farmers be doing?
0:30:08 Post tractor.
0:30:09 – Sitting in a room talking?
0:30:12 – Exactly, talking with these electronic microphones
0:30:13 and this like amazing fire.
0:30:15 – Look at the fire in the sky here
0:30:16 that we have in the basement.
0:30:17 – Our ancestors would be proud.
0:30:19 – Exactly, what was a nurse, right?
0:30:22 It’s like medicine was bloodletting leeches and prayers.
0:30:24 So it’s obviously changed a lot.
0:30:27 So on the one thing that I think AI cannot do,
0:30:29 and in fact, if you have an AI sales rep,
0:30:32 like Salesforce, why do I need seats for salespeople
0:30:34 if AI is doing selling?
0:30:35 But AI cannot build a relationship
0:30:37 with somebody over golf.
0:30:41 So I think the in-person things that only humans can do,
0:30:44 that skill set might go up in value tremendously.
0:30:46 At the far extreme, I talked to somebody
0:30:47 who believes that in the distant future,
0:30:49 there will only be two jobs.
0:30:51 You either tell a computer what to do
0:30:53 or you were told by a computer what to do.
0:30:55 And there’s a whole set of things
0:30:57 where people could be a lot more productive
0:30:59 in whatever job they’re doing
0:31:01 when you have this little coach by your side
0:31:02 saying do this, do that.
0:31:03 But I think the human connection thing
0:31:04 is almost the most important
0:31:06 because if I think about every other era
0:31:08 of a new communications tool,
0:31:10 imagine having the first telephone,
0:31:13 Alexander Graham Bell invents the telephone,
0:31:14 nobody has a telephone.
0:31:15 Like how do you scale this network?
0:31:17 You get one and the only person that calls you
0:31:18 is your mother saying,
0:31:20 “Why don’t you call me more often or something?”
0:31:23 So then the first telemarketer shows up
0:31:26 and then starts taking advantage of the fact
0:31:28 that you have a phone and they can,
0:31:30 rather than go horse and bug you to your house,
0:31:31 they can try selling you something
0:31:33 over this like old fashioned telephone.
0:31:35 That was a very advantageous place
0:31:37 for that first telemarketer to be.
0:31:40 Or the history of Sears Roebuck is really fascinating
0:31:42 because even though they had the Sears Tower,
0:31:44 they had these giant stores,
0:31:47 it really was the first giant mail order catalog.
0:31:49 So they were the ones that kind of figured out
0:31:51 how to use the US Postal Service.
0:31:52 Faxes came out,
0:31:54 people started sending unsolicited faxes.
0:31:56 So the reason why I bring all this up
0:31:58 is you can imagine a world where like AI
0:32:00 is selling you everything, pushing you everything.
0:32:02 And then right now it’s a novelty and it works really well.
0:32:04 But once it becomes so mainstream
0:32:06 that everybody’s doing it,
0:32:07 it’s like this yogi-bearer expression,
0:32:09 like it’s so crowded, nobody goes here anymore,
0:32:10 you can imagine like the need
0:32:13 for actual human connectivity to go up dramatically.
0:32:16 It’s followed this pattern of once it gets so crowded,
0:32:18 somebody that’s doing something different,
0:32:19 in this case the old fashioned way,
0:32:20 it might be more valuable.
0:32:22 – One of the ways we think about it is
0:32:25 all of us have some percentage of our job
0:32:28 that’s wrote tasks that could be automated with AI.
0:32:29 And I think we strongly believe
0:32:31 that at least every white collar job
0:32:32 is gonna have a co-pilot.
0:32:34 Some might be fully gigantic
0:32:36 back to our L1 compliance reviewers.
0:32:40 And so if you imagine all of us not doing any menial tasks
0:32:42 or anyone focusing on the human connection,
0:32:43 the most creative parts,
0:32:45 and having all of our days to spend on that,
0:32:47 like what might be enabled,
0:32:49 then that’s a pretty exciting way to think about it.
0:32:51 – Definitely, and as you think about the companies
0:32:52 that can be created in this wave,
0:32:55 I’m so curious because it does feel fundamentally different
0:32:58 as you three are assessing companies,
0:33:00 are there a new layer of metrics
0:33:01 that you pay attention to,
0:33:02 again, using a previous wave,
0:33:04 maybe we got social media
0:33:06 and all of a sudden we’re thinking of daily active users.
0:33:08 That was a key metric that people started
0:33:09 to pay attention to.
0:33:10 Are there new metrics?
0:33:12 Are they the same metrics that matter?
0:33:14 Is it too early to tell?
0:33:16 – I think it’s actually the exact same metrics.
0:33:17 It’s not like, ooh, it’s AI,
0:33:19 so therefore future profits don’t matter.
0:33:20 It’s the present value of future profits.
0:33:22 And that really comes down to like,
0:33:23 how many customers do you have?
0:33:25 Do you retain those customers?
0:33:27 And then how much gross profit do you make per customer?
0:33:29 And then how much overhead do you have?
0:33:31 And I don’t think any of that changes.
0:33:34 So, I mean, the reason why social networks were interesting
0:33:38 is we knew that the customers retained, right?
0:33:39 But will people pay for it?
0:33:40 Will it make money?
0:33:42 There was this open question.
0:33:46 And therefore there was this, if you will, alpha of,
0:33:48 oh wow, we call this the smile curve, which is very rare.
0:33:51 Obviously 100% of people use the product on day zero
0:33:52 because day zero is when they installed it.
0:33:55 But then people stop using it on day one, day two, day three.
0:33:57 And normally most products,
0:33:58 they just have exponential decay.
0:34:02 So by day 200 of the 100 people that downloaded on day zero,
0:34:04 zero people use it at the end.
0:34:06 What’s interesting is things like Uber or Facebook
0:34:09 where again, 100% of people use it on day zero.
0:34:10 Then it drops off on day one, day two,
0:34:12 and then it picks back up.
0:34:16 And then it plateaus at maybe 50, 70, 90%
0:34:17 of the original starting bunch.
0:34:18 And that’s so rare.
0:34:22 But then the question was, will Facebook ever make money?
0:34:24 Oh, they won’t make money because it’s free.
0:34:26 But they figured out that advertising was very valuable.
0:34:29 I think the vast majority of things that we’re seeing
0:34:31 right now just monetize via subscription.
0:34:33 So it’s actually very clear how they make money.
0:34:36 And the DAU thing is actually just as useful today
0:34:39 as it was before, but the money part is almost automatic.
0:34:42 Like the thing that was unique about the internet era
0:34:45 was it’s like, oh, get big and then monetize later.
0:34:47 And we’re not seeing as many of those,
0:34:50 but I don’t think any of the fundamental isms
0:34:52 of evaluating a business have really changed.
0:34:54 The only thing that’s more dangerous is
0:34:56 since AI can now write software,
0:34:59 it’s just so much easier to spin these things up.
0:35:02 Whereas to build something and scale it out,
0:35:04 the reason why Friendster failed,
0:35:05 like Friendster should have been
0:35:07 the social networking winner,
0:35:08 but their servers couldn’t stay up.
0:35:10 And MySpace should have been the winner,
0:35:12 but they couldn’t hire good engineers.
0:35:13 There are all these different reasons
0:35:16 that are not relevant today
0:35:18 because the technology stack is so different.
0:35:20 But again, it’s present value of future profits
0:35:22 that’s unchanged.
0:35:24 – I think one thing, and this is not a business metric,
0:35:28 but that’s changed is just potential market size.
0:35:29 And we talked about how it scales
0:35:31 in the large vertical SaaS markets.
0:35:32 So there’s something in the U.S. called
0:35:34 the North American Industry Classification System,
0:35:35 the NAICS codes.
0:35:37 And there’s 600 of them and they’ll classify industries.
0:35:39 How many companies are in there?
0:35:40 What’s their labor budget?
0:35:42 And there’s a whole host of industries.
0:35:44 Whereas if you looked at them before, you say,
0:35:47 well, there’s 1,000 potential buyers.
0:35:50 Maybe they’ll pay $1,000 a month for my software service.
0:35:52 That’s $120 million market.
0:35:54 That’s really not that interesting
0:35:56 if I’m gonna build a venture-backed business.
0:35:58 Now, if you think you can layer an AI,
0:36:00 replace some of the labor budgets,
0:36:03 those markets get dramatically bigger.
0:36:04 And so I think the different pockets
0:36:06 of where software can be built,
0:36:08 these niche markets that weren’t that interesting
0:36:10 are now potentially very interesting.
0:36:12 – I think the other dimension that we’re seeing pitches for
0:36:15 is are you selling software into the incumbent industry
0:36:17 or are you building the full-stack version?
0:36:19 Alex wrote a bit about this and our brigands at the gate,
0:36:22 which is sort of like the evolution of private equity
0:36:24 kind of been in AI context.
0:36:26 And so you think about an area like professional services
0:36:27 who are illegal, for example,
0:36:28 the challenge that a lot of these law firms have
0:36:31 is that they’re charging on a per hour basis, right?
0:36:33 So if AI can do what used to take three hours
0:36:36 in three seconds, where does the revenue go?
0:36:38 And so we’re seeing some people pitch
0:36:40 the full-stack kind of AI native law firm,
0:36:42 which might have a totally different cost structure
0:36:44 to Karath or one of these big firms.
0:36:47 Or there are other areas within professional services
0:36:50 that are much more aligned to benefit from that efficiency.
0:36:53 So for example, we have a company that is solving
0:36:55 a lot of the workflow challenges in plaintiff law.
0:36:58 They operate in both employment and personal injury,
0:37:01 where in that model, unlike on a per hour basis,
0:37:03 they’re charging on a contingency model.
0:37:05 Meaning they don’t get paid unless there’s an outcome
0:37:06 or a sediment in the case.
0:37:09 And in personal injury, as an example,
0:37:11 for every hundred leads that these lawyers get,
0:37:13 they take one case.
0:37:15 And there’s a ton of, again, that messy inbox problem
0:37:19 of sifting through medical records or employment documents
0:37:22 and having to essentially quantify the value of each case
0:37:23 that they’ll take on because any case that they take
0:37:25 is an investment of their labor.
0:37:27 And so what this company is doing is essentially
0:37:30 programmatically helping solve that sort of intake challenge,
0:37:34 that messy inbox problem, to help automatically qualify
0:37:35 the value of those cases.
0:37:38 And then works essentially as a co-pilot for the lawyer
0:37:42 to draft a medical chronology, to create a demand letter,
0:37:44 to file a complaint, and basically walks through
0:37:48 the entire sort of pre-litigation and litigation process
0:37:50 that allows that lawyer to take on three X or four X
0:37:51 the number of cases.
0:37:53 But again, the value that software is delivering
0:37:55 to the practice is reducing labor costs.
0:37:57 So one way to do it is you just have fewer lawyers
0:37:58 and the same amount of revenue.
0:38:01 Or in this case, I think what’s gonna happen is
0:38:03 it’s gonna significantly grow these practices.
0:38:06 And in this case, they’re actually passing the cost
0:38:08 of that software to the end client
0:38:09 in the form of a technology expense,
0:38:11 which they often had done historically.
0:38:14 And so the value that the software is delivering
0:38:17 to each of these firms is super aligned to the impact
0:38:18 that it’s having on the business.
0:38:20 And so I think the more clients you can take on,
0:38:22 the more people that can pay for the software.
0:38:24 And so on a per-firm basis,
0:38:26 there’s a significant kind of revenue expansion opportunity.
0:38:28 I think that’s an interesting tension
0:38:29 that you’ll see across industries
0:38:33 where does the AI help by reducing cost?
0:38:34 Is it better to build a full stack version
0:38:36 or sell the software in?
0:38:38 And I think there’ll be successes in both dimensions,
0:38:40 but that’s something we are seeing more of.
0:38:41 – Yeah, and one follow-up there,
0:38:44 as you’re talking about the cost being passed along
0:38:46 to the end user or buyer,
0:38:49 is that just net deflationary
0:38:51 as this permeates across the system?
0:38:52 I know it would take time,
0:38:54 but eventually if you see more competition
0:38:58 and more people creating these AI-based labor products,
0:39:00 and then people are competing on price,
0:39:01 and all of a sudden to take on a new case,
0:39:04 it’s no longer $5,000, it’s 500,
0:39:07 is that something that you guys are thinking about?
0:39:10 Or how you said that in the previous era,
0:39:12 the open question was, can we make money?
0:39:13 Can we monetize?
0:39:16 Is that an open question that this overtime
0:39:20 just becomes deflationary and firms can’t charge as much?
0:39:22 – I don’t know, this legal example might be unique
0:39:23 in the sense that the clients
0:39:25 are also waiting for a settlement themselves.
0:39:28 And so the cost of the software is essentially coming out
0:39:31 of whatever the lawyer is able to help win for the client.
0:39:34 And so they’re not really feeling the cost of the technology
0:39:36 as much as maybe other industries.
0:39:38 – I think technology, if it’s done right,
0:39:40 is always deflationary because you get productivity gains.
0:39:45 So I think for sure, I think the defensibility point
0:39:47 is the most relevant one that we struggle with a lot,
0:39:51 which is, wow, it’s so easy to build one of these things
0:39:53 because the number one use case,
0:39:54 it’s almost like a recursive thing.
0:39:59 It’s like, which profession is using AI tools the most?
0:40:02 It’s probably the tech people that actually build tools.
0:40:05 And that’s where things like cursor have gotten so popular.
0:40:08 Companies like Stack Overflow are suffering so much
0:40:11 because I don’t need to go to Stack Overflow anymore
0:40:14 because I just get the answer from Claude
0:40:15 or I get the code sample from cursor.
0:40:18 It’s just so much easier to do this stuff.
0:40:19 And so it is fundamentally deflationary,
0:40:21 but what if there are 50 companies
0:40:23 that end up doing the exact same thing?
0:40:24 That’s the hard part.
0:40:26 But I think I can’t see a scenario
0:40:29 where prices are more expensive than humans
0:40:31 or where prices don’t just keep going down.
0:40:32 – And significantly.
0:40:33 – Significantly.
0:40:35 And that’s the history of technology in a nutshell.
0:40:38 I mean, again, it’s like the 100 megabyte hard drive in 1960
0:40:41 that weighed like tons, literally tons,
0:40:43 was probably a million dollars or something crazy.
0:40:45 And now it’s just comical.
0:40:47 Like these little, I bought some for Cyber Monday.
0:40:50 I bought all these like little micro SSDs
0:40:52 for a terabyte was like $10.
0:40:53 It’s just incredible.
0:40:56 So I think that is an inexorable process
0:40:57 for technology cost in general.
0:40:59 And then you also get new use cases
0:41:01 where I’ve written about this separately,
0:41:02 which is the everything to the right
0:41:04 of the supply demand curve.
0:41:06 This is a really interesting use case
0:41:08 where it’s like there just wasn’t demand
0:41:10 where there was supply.
0:41:12 There’s a lot of supply to do something
0:41:15 for $2,000 an hour if I wanna, you know,
0:41:18 file a trademark with the leading trademark attorney.
0:41:21 So the trademark market or the patent market,
0:41:24 like that might be very small because it costs too much.
0:41:27 But now if it only costs $5, wow, maybe everybody does it.
0:41:30 Or like translation is one that I find fascinating
0:41:32 where it just doesn’t make sense.
0:41:33 If you’re a small company,
0:41:36 so you go translate your introductory video
0:41:39 into like 45,000 different languages that have ever existed.
0:41:42 Like why would I translate it into ancient Greek?
0:41:43 But you know what, why not?
0:41:44 It’s free, right?
0:41:46 So you have all these other things
0:41:48 that just expand the market
0:41:51 because the cost has dropped so precipitously.
0:41:52 Closing things out.
0:41:54 Where do you guys want to see more builders
0:41:55 applying themselves?
0:41:56 You’re obviously seeing a lot of companies,
0:41:57 a lot of people excited.
0:41:59 Also the incumbents are clearly excited
0:42:01 about getting in on this wave.
0:42:03 Is there an area that you’d like to see
0:42:05 more attention being put toward?
0:42:07 My view is obscure is good.
0:42:09 We love it when somebody walks in,
0:42:13 has had a decade or more of like obscurity.
0:42:14 They serve some weird job
0:42:16 or they were like something that nobody’s ever heard of,
0:42:18 the farming industry, the mining industry,
0:42:20 the whatever industry.
0:42:22 And then they actually have an insight
0:42:23 that somebody else doesn’t
0:42:25 and they actually know the potential of AI.
0:42:28 Because the other thing is that it’s very important to know
0:42:30 that the technology is not ready for autopilot
0:42:32 for a lot of these things.
0:42:34 It’s just like the use cases are too complicated.
0:42:35 The integrating the different pipes
0:42:36 is too complicated.
0:42:37 Overshooting early.
0:42:39 Like there are gonna be a lot of failures
0:42:40 because there inevitably are
0:42:42 in every technology revolution,
0:42:44 not because the idea is bad,
0:42:48 but it’s not good enough to be a hundred times better.
0:42:49 I think it’s like obscure
0:42:50 and then finding the ones we’re like,
0:42:51 at least for right now,
0:42:53 but actually the technology is good enough
0:42:55 for the obscure use case at hand.
0:42:57 Yeah, I’d say also across,
0:42:58 there’s many industries this,
0:43:00 but financial services and insurance
0:43:02 have a host of old systems,
0:43:05 like 30 plus year old systems of record
0:43:08 that now can be made 10x better incorporating labor,
0:43:09 redoing them in a workflow.
0:43:11 And so deep knowledge of those areas,
0:43:13 like we have transaction monitoring in Sardine,
0:43:16 we’ve got a mortgage loan origination system in Vesta,
0:43:18 we’ve got servicing, we’ve got a couple in insurance.
0:43:21 And so entrepreneurs that really understand those space
0:43:23 and can bring AI thinking there,
0:43:25 I think is a big opportunity.
0:43:27 I think we’ll continue to see lots of entrepreneurs
0:43:29 wedge in with the messy inbox problem
0:43:30 across lots of niche vertical industries.
0:43:33 We’re still on the lookout for horizontal software,
0:43:36 AI native versions selling into the sales teams,
0:43:39 marketing, product management, analytics, CFOs.
0:43:40 In those categories,
0:43:43 you often do have a large incumbent software competitor.
0:43:44 And so that’s sort of the tension.
0:43:45 You have to understand the market structure
0:43:47 and how likely it is for that incumbent
0:43:49 to change their pricing model
0:43:51 and build more AI native features.
0:43:53 But I think there will be generationally
0:43:56 defining companies built in an AI native way
0:43:57 in horizontal software as well.
0:43:59 – Very exciting, thanks guys.
0:44:00 – Thank you.
0:44:01 – Thanks moderator.
0:44:03 You would have been a good farmer too, I’m sure.
0:44:05 (laughing)
0:44:07 All right, that is all for today.
0:44:10 If you did make it this far, first of all, thank you.
0:44:12 We put a lot of thought into each of these episodes,
0:44:14 whether it’s guests, the calendar touchers,
0:44:16 the cycles with our amazing editor Tommy
0:44:18 until the music is just right.
0:44:20 So if you like what we’ve put together,
0:44:24 consider dropping us a line at ratethespodcast.com/a16z.
0:44:27 And let us know what your favorite episode is.
0:44:30 It’ll make my day and I’m sure Tommy’s too.
0:44:31 We’ll catch you on the flip side.
0:44:35 (gentle music)
0:44:37 (gentle music)