Learn. Share. Evolve…

0:00:02 What is the journey over the next decade?
0:00:06 It’s about the speed at which humans can change their workflows.
0:00:08 Why doesn’t the breakthrough that we just saw get released?
0:00:11 Why doesn’t that permeate every corporation within six months?
0:00:16 It’s so strange to me how many disruptions are happening all at the same time.
0:00:17 Your R&D is changing.
0:00:18 Yeah.
0:00:20 Every part of the stack is changing.
0:00:20 Like everything.
0:00:23 We’re not in like a fear of AI world.
0:00:28 We’re in a, we know this is going to happen and it needs to happen to us faster than it
0:00:31 happens to our competitors, which is a totally different dynamic than we saw with cloud.
0:00:36 What do you think is the best metric for anybody interested in tracking this stuff as far as
0:00:37 like how fast it’s going?
0:00:37 Is it GDP?
0:00:39 Is it margin?
0:00:40 Is it top line?
0:00:42 Is it headcount growth?
0:00:43 Is it all the above?
0:00:47 It’s basically fully assumed that AI is going to take over the enterprise.
0:00:50 How does AI actually change the enterprise?
0:00:54 Not just in theory, but in how software is built, sold, and used?
0:01:00 In today’s episode, A16Z general partner Martin Casado sits down with Aaron Levy, co-founder
0:01:05 and CEO of Box, to explore what it means to be an AI-first company from product strategy
0:01:06 to internal workflows.
0:01:11 They talk about why incumbents may be better positioned than expected, how startups can still
0:01:15 break out, the rise of agents and vibe coding, and what happens when the bottleneck isn’t the
0:01:16 tech, but the org chart.
0:01:22 Aaron also shares how Box is using AI internally today and why he thinks the next generation of
0:01:25 employees may spend more time managing agents than writing code.
0:01:26 Let’s get into it.
0:01:33 As a reminder, the content here is for informational purposes only, should not be taken as legal
0:01:37 business, tax, or investment advice, or be used to evaluate any investment or security,
0:01:42 and is not directed at any investors or potential investors in any A16Z fund.
0:01:46 Please note that A16Z and its affiliates may also maintain investments in the companies
0:01:47 discussed in this podcast.
0:01:54 For more details, including a link to our investments, please see A16Z.com forward slash disclosures.
0:02:01 Aaron, thank you very much for joining us.
0:02:01 Thank you.
0:02:03 Everybody here already knows you.
0:02:05 However, I still think you should intro yourself, just for completeness.
0:02:12 Aaron Levy, CEO, co-founder of Box, and at Box, we help enterprises basically take all of their
0:02:18 unstructured data or enterprise content and turn it into valuable information, and AI is
0:02:20 absolutely this incredible accelerant for that problem.
0:02:22 I just learned that we’re investors, didn’t you?
0:02:24 Well, many years ago.
0:02:24 Many years ago.
0:02:26 So no claims post-IPO.
0:02:31 But actually, Ben Horowitz had this early kind of blog post on basically, I think it was
0:02:32 the title of The Fat Startup.
0:02:32 Yeah.
0:02:33 Yeah, yeah, yeah.
0:02:35 In response to enterprises, the lean startup.
0:02:36 Yeah, that’s right.
0:02:40 And let’s just say we very much took that to heart, and we basically like deployed every
0:02:45 single lesson, which was like the name of that game is you get big fast, you scale aggressively,
0:02:48 and that was a very important period in our company’s journey.
0:02:52 So the notional topic of this is AI in the enterprise.
0:02:56 But I think it’s good to be kind of nuanced about this, because it’s less obvious than
0:03:02 people think, and you’ve been talking a lot about AI on X, but also you’re thinking about
0:03:03 it in the terms of your business.
0:03:07 So let me just kind of set up the first question as follows, which is, AI has historically been
0:03:11 this very B2B enterprise thing, like chatbots or whatever, personalization systems.
0:03:16 But what’s unique about Gen AI is a lot of the use cases are actually like a consumer or
0:03:17 prosumer, right?
0:03:23 Think like creativity or developers, and it actually hasn’t made intros as much.
0:03:24 into the enterprise yet.
0:03:25 It’s just starting now.
0:03:27 So maybe just a couple of questions.
0:03:30 First off, A, does that match with your experience?
0:03:34 And then B, how are you thinking about this transition to the enterprise?
0:03:35 Yeah.
0:03:41 I think if you were to probably like do the idiosyncrasies of AI and then reverse engineer
0:03:45 why that was the journey, basically up until, let’s say, pre-chat to be team moment,
0:03:47 AI was extremely hard to use.
0:03:52 It required in many cases having custom models for basically every problem you tried to solve.
0:03:57 And so there was almost no way that a consumer ecosystem could flourish based on that.
0:03:59 It was just not generalizable enough.
0:04:03 There was really few products other than like maybe Siri, Alexa, et cetera, that you’d interact
0:04:05 with that would even have some sense of AI.
0:04:10 And so enterprises were the early adopters of AI systems to bring workflow automation to
0:04:11 their companies.
0:04:16 Then boom, ChatGPT happens and all of a sudden it’s the exact right form factor for mass adoption.
0:04:17 There’s no startup costs.
0:04:19 It costs two seconds to learn the product.
0:04:20 It’s a chat interface.
0:04:24 So it was like perfectly ripe for just taking off in the consumer space.
0:04:29 And then you have also these incredible conditions set up for mass adoption.
0:04:31 You have billions of people on the internet.
0:04:32 It was set up as a free product.
0:04:36 Again, it kind of solved this sort of latent kind of question mark that everybody had, which
0:04:40 is like, when are we going to see AI work and touch our lives?
0:04:45 And so everything was kind of like the perfect conditions to get mass consumer adoption.
0:04:49 On the enterprise side, you have unfortunately kind of the opposite, right?
0:04:53 You have lots of workflows that have been kind of ingrained for decades and decades.
0:04:59 You have lots of legacy IT systems that have data kind of not set up well to be accessed by AI.
0:05:05 You have a sort of shadow IT problem, which is most corporations don’t want, and users just
0:05:10 injecting text into prompts that might contain information that the AI models could learn off of.
0:05:14 So it’s sort of a difficult environment for that same level of virality.
0:05:20 With the exception of a few of these pro-sumer categories, I have talked to large corporation
0:05:24 CIOs that are seeing people just show up with Windsurf and Cursor and Replit.
0:05:28 And so you’re getting actually this sort of shadow IT version that we saw 15 years ago.
0:05:30 DevTools has always had that.
0:05:31 Yeah, 100% fair.
0:05:31 100%.
0:05:32 So DevTools have had that.
0:05:36 But I think that you’re still seeing that now in the chat to BT kind of leakage into organizations.
0:05:37 Right.
0:05:42 I’m sure their pro-sumer inside of a corporation firewall usage is off the charts, even separate
0:05:43 from the people that pay for it.
0:05:43 Totally.
0:05:48 So now the question, though, is what is the journey over the next decade for the real
0:05:54 change management of deployment of AI systems that drive the more like GDP changing productivity
0:05:55 gains?
0:05:58 And that’s something where I do think we have to be prepared for.
0:05:59 This is many years.
0:06:03 It’s about the speed at which humans can change their workflows as opposed to how kind of quickly
0:06:06 the technology can just sort of evolve in advance.
0:06:11 And so we in Silicon Valley and certainly anybody tuning into this sort of imagines like, well,
0:06:13 why doesn’t the breakthrough that we just saw get released?
0:06:16 Why isn’t that permeate every corporation within six months?
0:06:20 And it’s because like people just have meetings and they have budget processes and they have
0:06:24 to go through a governance council and they have to get compliance on board and they have
0:06:28 to figure out like who has the liability when the thing recommends this stock and then the
0:06:30 financial services provider shares that with a client.
0:06:33 Like that takes years and there’s going to be case law that needs to happen.
0:06:36 And we still have lawsuits that are going on about who owns the IP of this stuff.
0:06:38 So that part is going to take years.
0:06:42 What’s interesting, and I think you’ll especially appreciate this on the cloud side is I remember
0:06:47 when we first were scaling up in the enterprise, let’s say 2007, 2008, 2009, let’s say that three
0:06:53 to five year period, post AWS, post kind of cloud starting its journey, basically to a
0:06:58 T, every conversation you’d have with a CIO or a group of CIOs was basically like, yeah,
0:06:59 that’s nice.
0:07:01 Maybe some little corner of our organization could use this.
0:07:03 We are never going to go fully to the cloud.
0:07:05 They had their arms wrapped around their servers.
0:07:06 I remember.
0:07:06 Yeah.
0:07:09 And basically they did not want to give up the infrastructure.
0:07:11 There was too many questions, too many compliance issues.
0:07:15 There was just existential job questions of, well, what happens when this, you know, gets
0:07:16 delivered as a service?
0:07:17 Here’s a super interesting.
0:07:21 Let’s say we’re now two and a half years into the ChatGPT moment.
0:07:24 That same group of CIO conversations, none of that.
0:07:29 It is basically assumed, it’s basically fully assumed that AI is going to take over the enterprise.
0:07:36 Like the CEO, the CEO, the CIO, the CDO, every job, every org leader is basically like, we know this is going to happen.
0:07:39 This is not like a, oh, we’re trying to kind of push it off.
0:07:42 It is purely a sequence of events.
0:07:43 Who do I deploy?
0:07:44 How do I deploy it?
0:07:45 How do I drive the change management?
0:07:46 Is the model ready?
0:07:55 So what’s really interesting is I think the level of buy-in you have now in the enterprise is like five times greater than we had in the early days of cloud.
0:07:57 And you can even see it.
0:08:04 Like to me, the classic litmus test was, if you remember like 15 years ago, I think Jamie Dimon was probably most famous for saying like, we’re never going to go to the cloud.
0:08:04 Yes.
0:08:07 So like they basically said, J.P. Morgan will never go to the cloud.
0:08:07 Yeah.
0:08:19 Today, that equivalent commentary, whether I don’t have a perfect Jamie Dimon quote, but David Solomon at Goldman Sachs has given this anecdote of they can write now an SEC filing or an S1 for an IPO in like a few minutes.
0:08:21 That used to take a number of analysts a few days.
0:08:28 And so the fact that like those are the anecdotes already coming out of the biggest banks means that we’re not in like a fear of AI world.
0:08:36 We’re in a, we know this is going to happen and it needs to happen to us faster than it happens to our competitors, which is a totally different dynamic than we saw with cloud.
0:08:42 So do you think this has implications for companies today that are building products that are pre-AI products?
0:08:49 So for example, with the cloud wave, you basically had a bunch of cloud native companies that ended up taking over.
0:08:54 So for example, Snowflake is a great example of this, which is like the ones that decided not to go all in and were hybrid.
0:08:57 Like hybrid kind of became known as like means it won’t work.
0:08:59 You know, anything called hybrid hasn’t worked.
0:09:09 So do you think because the buyer and the enterprise is more ready that like companies that are pre-AI have more of an opportunity?
0:09:13 Or do you think that you’re going to see the same thing with a lot of like AI native companies do well?
0:09:16 I’m going to basically give you the non-answer of I think both.
0:09:28 And one benefit that the cloud cohort has or the SaaS kind of posts like us all understanding and agreeing on what SaaS would look like, what we all have is whether we adhered to this perfectly or not is a question.
0:09:31 But we basically all tried to build API first platforms.
0:09:31 Yeah.
0:09:34 Or at least like API kind of like equal platform.
0:09:36 So we have the UI and we have the API.
0:09:42 And if you think about it, like AI and AI agents are like the perfect consumers of an API, right?
0:09:46 And so they basically become these super users within your system on your APIs.
0:10:01 So if I had to just say, okay, I want to deploy agents to go and automate my ServiceNow workflows, I think I’m better off just deploying the ServiceNow agent to go do that than do an entire reinvention of my ITSM system to solve that use case.
0:10:02 And you can just go down the list.
0:10:11 Like Workday, if I want an AI agent to do some kind of HR-related task, I think I’m better off to just do that within Workday than I am building an entire new system.
0:10:14 So you have a bunch of different factors versus the pre-cloud days.
0:10:17 Like pre-cloud to post-cloud was an entire rewriting of your software.
0:10:19 You had to go from single-tenant to multi-tenant.
0:10:21 The scaling of the systems were totally different.
0:10:25 Even the functionality and application logic was different because like it should be real-time.
0:10:26 It should be collaborative.
0:10:30 It shouldn’t be as sort of async and batches as the on-prem systems were.
0:10:35 And so in a cloud world, it is a reinvention of the user experience and what you’re doing in the system.
0:10:36 And we should definitely get to that.
0:10:40 Well, I just want to make sure I tease this out because it’s actually a very interesting point.
0:10:47 So your claim is to go from pre-cloud to post-cloud, like that ripped through the entire stack all the way down to like the infrastructure, for example, like tenancy.
0:10:49 Like you have to rewrite everything.
0:10:55 And then what you’re saying about AI is more of a consumption layer thing, which is you just treat the existing systems as they are.
0:10:56 And then the AI becomes the consumption layer.
0:11:02 Do you think this is like a 1.5 step and like the 2.0 step kind of rips through the entire stack?
0:11:04 Okay, so let’s bookmark that one for one second.
0:11:09 But like if you do pure Clay Christensen sort of approach, you know, sustaining innovation, disruptive innovation.
0:11:14 Disruptive innovation is this thing that looks like so much harder, so different, so less profitable.
0:11:17 Sustaining is like, actually, no, I’d like to build that because it’s incremental.
0:11:19 It’s better for our business overall.
0:11:22 The on-prem guys had a disruptive innovation.
0:11:26 Everything about the business model of SaaS looked different, harder, stranger.
0:11:28 I don’t have the talent.
0:11:32 I’m running a service delivery operation as opposed to I ship you a CD-ROM with my code.
0:11:35 Everything about the finances and pricing model.
0:11:35 Yes, everything.
0:11:36 Everything, the business model, everything.
0:11:41 AI, again, with the bookmark being like the really big disruption that you could contemplate,
0:11:45 right now with AI, everything kind of looks like a sustaining innovation if you’re an incumbent,
0:11:49 which is like, instead of a user pressing the buttons in the application,
0:11:53 let’s have an agent run through the API and operate as if they were that user.
0:11:57 And so all of a sudden, for a lot of SaaS providers, this looks like a TAM expansion
0:12:02 because now, for the first time ever, I can actually deploy my software for use cases
0:12:06 where the customer didn’t have users on the other end before to do those things.
0:12:07 So I think you have a lot of TAM expansion.
0:12:09 Now, the good news for startups.
0:12:11 With one caveat, which maybe we’ve bookmarked and we’re going to get to,
0:12:12 but let me just say the one caveat.
0:12:17 The one caveat is you now have a component that has a very different COGS model
0:12:19 if you’re a software provider.
0:12:19 Yes.
0:12:22 And so like now, it’s almost like when we went from like on-prem to cloud,
0:12:24 we went from perpetual to recurring.
0:12:30 And it feels like with AI, you kind of have to go from recurring to usage-based just because.
0:12:30 Yeah.
0:12:33 So business model will shift for some of the use cases
0:12:37 because even if you look at the cursors, replets, windsrifts of the world,
0:12:40 there does seem to be this baseline seat price.
0:12:43 And then your consumption usage thing is sort of this add-on.
0:12:43 This overage sheet.
0:12:47 And so SaaS providers are kind of well-structured to be able to have that kind of dynamic.
0:12:48 Yeah.
0:12:51 If it was 100% usage and the user seat goes away,
0:12:53 I do agree that you have this, then you have a little bit of a business model crisis.
0:12:57 Oh, so you think, but right now, it’s not clear that that’s going to go all the way over.
0:13:00 Well, until the human literally is not a seat on the system,
0:13:04 I think you don’t remove the end user license as a component.
0:13:04 Okay.
0:13:06 But again, that could be like the much bigger disruption.
0:13:11 Now, just to fully lay out the market dynamics, I think SaaS incumbents,
0:13:14 especially you have a couple other idiosyncrasies right now versus the on-prem days.
0:13:17 Another idiosyncrasy is I would say like on the margin,
0:13:20 you tend to have founders still leading the SaaS companies.
0:13:20 100%.
0:13:22 And so we didn’t really have that in the on-prem world.
0:13:28 And so like Siebel already had three CEOs later and PeopleSoft already had multiple CEOs later.
0:13:30 So it was a different leadership structure in these organizations.
0:13:33 A lot of times you still have the founder around, they’re poking around,
0:13:33 they’re really into AI.
0:13:37 So there can be a more natural pivot of the company from the leadership standpoint.
0:13:39 So a bunch of different factors.
0:13:43 Now, to the benefit of startups, which is why I can hold both of these in my head,
0:13:48 which is I’m very bullish on the SaaS incumbent being the natural place for that AI agent
0:13:49 relative to that category.
0:13:53 I just think we have this incredible expansion of categories for the first time
0:13:55 that we haven’t seen in probably 15 years.
0:14:02 So the SaaS 1.Wave actually expanded the software universe where we had these new categories of
0:14:04 software that we didn’t expect before.
0:14:08 Like nobody would have predicted the confluence and the snowflakes in the pre-on-prem days.
0:14:11 We didn’t have all of these different cuts of how do you work with data?
0:14:12 How do you do this workflow?
0:14:16 Like lines of business didn’t have 15 different applications they got to use.
0:14:17 Post-SaaS, they did.
0:14:23 So for startups in the AI world, the equivalent of that is I think there’s a lot of categories now
0:14:29 where there is no actually software incumbent in that category where AI agents all of a sudden
0:14:31 let you go build software for that category.
0:14:33 Legal, healthcare, education, and so on.
0:14:36 So that’s definitely true on the consumer side, right?
0:14:39 If you look at the top use cases of open AI, it’s almost like the top of the pyramid of needs,
0:14:40 right?
0:14:42 It’s like creativity and fulfillment, et cetera.
0:14:46 And I think like number five is like professional coding, but everything above that is one of these.
0:14:48 So on the consumer, that’s very clear.
0:14:50 Is that clear on the enterprise side?
0:14:50 I absolutely think so.
0:14:57 I think if we did a snapshot 10 years ago of the size of the contract management market or
0:15:00 the legal document market, it’s like sub 2 billion.
0:15:01 I’m making up the numbers.
0:15:02 It could be plus or minus a billion.
0:15:08 Would you agree that in five years from now, the AI agent related spend on legal services
0:15:12 should be in the many, many billions to double digit billions?
0:15:13 Absolutely.
0:15:13 Okay.
0:15:13 No question.
0:15:18 So all of a sudden there’s like not these natural incumbents that were like, oh, we captured all
0:15:19 that market.
0:15:23 AI agents all of a sudden expands the size of the software related spend in that space.
0:15:28 So I can underwrite that for healthcare, legal, consulting services.
0:15:31 I think there’s entire areas of financial services.
0:15:33 Like we always think, oh, finance has been wired up for so many years.
0:15:36 No, banking, consumer banking has been wired up.
0:15:37 Trading has been wired up.
0:15:39 Investment banking never went digital.
0:15:41 Wealth management never went digital.
0:15:46 Like these were not categories where you ever had like major software platforms to help these
0:15:47 entire categories of the economy.
0:15:51 And the reason it was because the work was unstructured.
0:15:56 It’s very ad hoc, very dynamic, lots of unstructured data as opposed to stuff that goes into databases.
0:15:58 All of that is now ripe for AI.
0:16:02 And that will then largely be ripe for many startups because there won’t be a natural incumbent
0:16:03 in those spaces.
0:16:08 I mean, it’s so strange to me how many disruptions are happening all at the same time with AI,
0:16:08 right?
0:16:11 I mean, if you think about like everything you said, which is basically vertical SaaS or vertical
0:16:13 use cases, which a lot of that is actually human budget, right?
0:16:13 Yep.
0:16:14 That’s being disrupted.
0:16:17 There’s a bunch of new use cases that we never really thought about before, which is
0:16:21 like the creativity and I mean, who would have thought that 2D image would be some massive
0:16:21 market?
0:16:22 Yes.
0:16:23 But it’s a massive market, right?
0:16:27 It turns out, you know, I’ve been a programmer for 30 years, right?
0:16:30 And in that time, like software would disrupt other things.
0:16:30 Yeah.
0:16:33 Like we disrupt all of these things, but we never got disrupted.
0:16:34 We’re safe.
0:16:35 We’re screwing you guys.
0:16:38 But clearly now software is being disrupted, right?
0:16:40 For the first time like I’ve ever seen in 30 years.
0:16:46 And so do you think this level of disruption is something that existing companies will not?
0:16:48 Like maybe a more fine point.
0:16:49 You are a business leader right now.
0:16:50 You have to think about product.
0:16:52 You have to think about your organization.
0:16:53 Does it require you to have to think about too much?
0:16:55 Like how do you structure your company as well?
0:16:59 How do you structure your product or do you think this is actually all pretty manageable?
0:17:01 I think it’s your R&D product.
0:17:04 Like literally like I’m putting myself in your shoes, right?
0:17:04 Yeah, yeah, yeah.
0:17:05 Which is like your CEO.
0:17:06 Like your R&D is changing.
0:17:07 Yeah.
0:17:08 You’re like-
0:17:09 Every part of the stack is changing.
0:17:09 Like everything.
0:17:10 Yeah.
0:17:23 I think the reason that I’m probably frankly more distracted by what we’re building that I don’t have enough time to stress out about the actual organizational side because I’m stressed out enough about just literally like the actual pure like delivery of the product.
0:17:27 I think if I had a little bit more time, I’d get more stressed out about all the other change.
0:17:30 We are very much leaning into the idea of being AI first.
0:17:31 We have a twofer on this.
0:17:36 Like one, by being as AI first as possible, we’ll see the use cases that our product should go solve for customers.
0:17:37 So like check that box.
0:17:41 And then second is I’m just a believer of the efficiency productivity gains.
0:17:41 Yeah, yeah, sure.
0:17:44 And I do think it does change basically everything about work.
0:17:46 And there’s lots of these interesting examples of what it means.
0:17:51 So in the future, does the individual contributor basically become a manager of agents?
0:17:52 Yeah, yeah, yeah.
0:17:53 So that’s a totally different job.
0:17:54 Right.
0:17:54 Right?
0:18:03 Like my recent kind of go-to is just thinking about it as a lot of the productivity of your organization was rate limited by literally like how fast can somebody use a computer to do something?
0:18:03 Yeah.
0:18:06 To type an email, to write code, to generate a marketing asset.
0:18:06 Yeah.
0:18:10 When that’s no longer a limiter, how do these jobs begin to change?
0:18:18 And it’s like, okay, your job is now orchestration, integration of work, planning, task management, reviewing, auditing, and that will radically change work.
0:18:33 Interestingly, this probably behooves us to not over-rotate on transforming yet internally for any given company simply because the technology is changing so fast that like you probably wouldn’t want to snap the line right now, run your whole business on this technology.
0:18:35 Because in two years from now, it’s going to happen.
0:18:37 Because in two years from now, it’s going to be so much better.
0:18:45 And so I think progressively figuring out which workflows have high impact upside, getting it rolled out in a decentralized way so people can experiment.
0:18:47 Like I think you want to do a few of those kind of things first.
0:18:50 I mean, I can’t imagine a listener not knowing what Box does.
0:18:57 But just for completing this, maybe can you just talk to us very quickly about what Box does and how you’re thinking about how that dovetails with AI?
0:19:02 Yeah, so we started the company with a really simple premise, make it easy to access and share your files from anywhere.
0:19:06 And we pivoted about two years into the journey to focus on the enterprise market.
0:19:10 And the whole idea was enterprises are awash with all this unstructured data.
0:19:17 So corporate documents, research files, marketing assets, M&A documents, contracts, invoices, all of this.
0:19:22 And as companies move to the cloud and as they move to mobile, they need a way to access that information.
0:19:25 They need a way to collaborate securely on it.
0:19:28 They want to be able to integrate that data across different systems.
0:19:30 So we built a platform to help companies do that.
0:19:34 We have about 120,000 customers, about 65 or so percent of the Fortune 500.
0:19:40 And so what’s incredible right now is we’ve had this ongoing problem since the creation of the company,
0:19:46 which is with structured data, the stuff that goes into your database, you can query it, you can synthesize it,
0:19:49 you can calculate it, you can analyze it, your unstructured data, the stuff that we manage,
0:19:54 you create it, you share it, you look at it, and then you basically kind of get forgotten about.
0:19:57 Like it goes into some folder and you almost never see it again.
0:20:01 And maybe you kind of find it once every five years for some task you’re doing, but that’s about it.
0:20:06 And so most companies are sitting on most of their data being unstructured
0:20:11 and getting the least amount of value from it relative to their other structured data.
0:20:13 AI is basically the unlock.
0:20:17 So AI lets you finally say, okay, we can ask this data questions.
0:20:22 We can structure it so we can look at a contract, pull out the 10 most important fields.
0:20:25 Once we have all that data, we can analyze that information.
0:20:26 We can get insights from it.
0:20:31 And then you can start to do things like workflow automation that was never possible with your unstructured data.
0:20:36 So if I want to move a contract through an automatic process, I can’t do it if I don’t know what’s in the contract.
0:20:40 And the computer previously was not able to know what’s in the contract.
0:20:46 So for us, there’s just a huge unlock of now what you can finally do with your information and your content.
0:20:50 So we’re building an AI platform to handle all of the kind of plumbing user experience
0:20:53 to make then your content AI ready effectively.
0:20:57 I don’t want to be like too bullshitty and provocative, but I have to ask this.
0:20:58 Please.
0:21:00 I’ve been in enterprise software for a very long time.
0:21:05 A lot of the business model is predicated on the fact that building software is hard and takes a long time.
0:21:05 Yeah.
0:21:08 To what extent do you worry about that not being the truth going forward?
0:21:13 Do you think we enter like this time of bespoke software being upon us?
0:21:17 I’m bearish on the extreme version of the essence of that.
0:21:22 So the extreme version of that, if you imagine the polls of this, like the extreme on one poll,
0:21:24 basically all software is prepackaged.
0:21:26 It’s the Ford Model T.
0:21:28 It’s going to work only in one way.
0:21:29 Everybody uses the same thing.
0:21:30 Okay.
0:21:30 That’s not going to happen.
0:21:31 We get that.
0:21:34 The other extreme is like everything is just like homebrew.
0:21:37 You wake up in the morning, you utter something, you get your software for the day.
0:21:38 You get your software for that thing.
0:21:40 And then the next day you do it again and you change it.
0:21:40 Okay.
0:21:45 The downsides of that model of why basically I think it doesn’t work is I think if you
0:21:49 ask like the world population, you probably find that 90 plus percent just don’t care enough.
0:21:54 They just don’t care about the tabs on their software and the modules on their dashboard.
0:21:57 Like it’s like they want someone else to just be like, this is what you should look at in
0:21:58 the morning.
0:21:58 Yeah.
0:22:01 They don’t want to have to even prompt the AI to tell them what to look at.
0:22:01 Yeah.
0:22:05 So given that that’s basically guaranteed to be where 90% of the world, no matter how you
0:22:06 cut anything.
0:22:06 Yeah, that’s a great point.
0:22:11 That means that basically 90% of our software should largely be like, okay, you log into the
0:22:13 HR system and it just looks like an HR system.
0:22:19 And in fact, there’s another interesting dynamic, which is like over many years, our software
0:22:24 and our actual way that we operate companies, there’s this flywheel relationship between them.
0:22:29 And so the way we run our HR department is like not so different than the way Workday wants
0:22:31 us to run our HR department.
0:22:31 Yeah.
0:22:35 And it’s fine because that’s not the area that we’re going to have a lot of upside innovating
0:22:35 on.
0:22:39 And like the way that we do our ticket management from customer tickets is like the way that
0:22:41 Zendesk decided to do ticket management.
0:22:44 And that’s fine because that’s not the core IP of the company.
0:22:47 In a way, it solves an operational problem for you.
0:22:47 Yes.
0:22:48 You don’t have to figure it out.
0:22:48 Right.
0:22:50 And people miss that about software.
0:22:54 I don’t want to have to think about the workflow of an HR payroll process.
0:22:56 I just want the software to do that.
0:22:59 And so that’s what people are buying.
0:23:01 And so nobody wants to customize those things.
0:23:05 Now, again, given that we’re going to be in this world of many different outcomes playing
0:23:11 out, the reason I’m still bullish on Replit and Vibe coding is for a different category,
0:23:15 which is like I’m the IT person and I just have this crazy queue of tasks.
0:23:17 And then someone’s like, can you build a website for this thing?
0:23:21 Can you like code up some inventory random plugin for this product?
0:23:24 It’s like that now becomes 10 times easier.
0:23:27 So the new prototyping, scripting, the long tail of stuff that we want to do.
0:23:31 And that long tail is so long and people never get to any of those things in that long tail.
0:23:36 And so I could underwrite a 10x growth of the amount of custom software that gets written
0:23:41 and the fact that these core systems don’t go away because there’s just actually going
0:23:43 to be way more software in the world that gets created.
0:23:44 Let me pressure test this.
0:23:47 So like, okay, so I can imagine why it would be hard to rebuild Box because what you do
0:23:48 is actually hard.
0:23:49 This is core infrastructure.
0:23:50 You store data like that’s really important.
0:23:52 And so I don’t think you just Vibe code that away.
0:23:56 But from my perspective, a lot of SaaS apps just look like CRUD.
0:24:00 To me, CRUD is, I don’t know what the acronym stands for, but it’s basically you’re reading
0:24:02 and writing data from like a backend.
0:24:08 And so do you think that there is a world where the consumption layer evolves to just using AI
0:24:09 and this class of companies go away?
0:24:13 Or do you actually think, if I heard what you just said, that the durability of these companies
0:24:16 is that it basically teaches you what the workflow is?
0:24:17 Well, I’m still going to say the latter.
0:24:20 Now, I don’t know if you need to bleep it out, but if you want to share a couple examples
0:24:25 of who you put in the not hard CRUD layer, then we can parse that.
0:24:26 But up to you.
0:24:27 The not hard CRUD layer?
0:24:28 Yeah.
0:24:34 I mean, I would say most vertical SaaS companies I see, the technology is trivial.
0:24:35 Yeah.
0:24:36 But the understanding of the domain is not.
0:24:36 No, no, no.
0:24:37 This is what you said before.
0:24:38 This is what I want to present.
0:24:38 That’s the thing.
0:24:40 That’s actually a great insight.
0:24:44 I’ve always underestimated vertical SaaS relative to the outcome.
0:24:44 Yeah.
0:24:47 And 20 years into doing enterprise software, I’m just like, no longer going to underestimate
0:24:48 vertical SaaS.
0:24:49 It’s not about the technology.
0:24:52 It’s the fact that somebody else has figured out the business model that works.
0:24:57 Like they have 10 people from the pharma industry that is like sitting next to the engineer
0:25:00 being like, this is how you should do the clinical trial workflow.
0:25:02 And that becomes so much of the IP.
0:25:07 Now, that translates fine to agents, but I still would then bet on that vertical player
0:25:13 doing that as opposed to somebody prompts their way into ChattoBT to build a FDA compliance
0:25:13 agent.
0:25:19 I would still largely bet on complianceagent.ai to do that over the pure horizontal system
0:25:21 that has no particular domain kind of expertise for that.
0:25:26 And then I think the other thing, I still think that there’s a relationship between some
0:25:31 amount of GUI and the agent and the APIs, because again, like you don’t want to every day
0:25:34 of your life, go to a blank empty screen and say, what’s our revenue today?
0:25:37 You just want a dashboard at some point and it just shows you the revenue.
0:25:37 That’s right, of course.
0:25:39 It’s almost like cast queries in a way.
0:25:40 Like somebody has made the decision.
0:25:43 Yes, this is like a known way to solve this problem in the enterprise.
0:25:48 And so I think that’s why the theory of the full abstraction away from the interface and
0:25:49 it’s all an API call.
0:25:50 I don’t think that happens.
0:25:55 And so ironically, probably what will happen is in a couple of years from now, we will see
0:25:59 agents like rebuild entire webpages and dashboards.
0:26:01 And then we’re going to find ourselves like, wait, why are we having an age?
0:26:06 Why do I have to spend tokens to create a thing that is a config on a dashboard?
0:26:10 And we’ll just be back to where we started for some amount of software, which will mean
0:26:13 that basically like these things are going to live together.
0:26:14 Cool.
0:26:16 Let’s move from software to decision process.
0:26:21 So I won’t say the name of the company, but I just spoke with a very, very legit company,
0:26:22 household name.
0:26:23 It’s a private company, though.
0:26:24 It’s not a public company.
0:26:30 We’re at the board level for every decision they ask the AI for like basically more information
0:26:31 for the decision.
0:26:32 Okay.
0:26:36 And this has actually been great from like discussion fodder to be provocative.
0:26:42 And it also shows how like fundamentally unoriginal the board members are.
0:26:45 Like this founder was telling me, it’s literally better than half of my board members.
0:26:46 Right.
0:26:51 And so like, how much have you thought about bringing AIs in to like help with decision
0:26:52 process?
0:26:52 Yeah.
0:26:57 And by the way, I think the board is like low hanging fruit because boards tend to not have
0:26:58 a lot of context to the business.
0:27:00 And so the incidents are probably less anyways.
0:27:02 But is this something that you’ve thought about?
0:27:04 The board one is an interesting one.
0:27:05 So maybe we can unravel that one.
0:27:10 But like I already use it for, let’s say, our earnings calls where we’ll do a draft of the
0:27:11 initial earnings script.
0:27:16 And then, I mean, again, because BoxAI deals with unstructured data, I just load up the
0:27:21 earnings script and I’ll use a better model and say, give me 10 points that analysts are
0:27:22 going to ask about this.
0:27:23 And like, how would I improve the script?
0:27:25 And it just spits out a bunch of things.
0:27:25 And it’s…
0:27:26 And how good is it at predicting?
0:27:27 Oh, extremely good.
0:27:28 Oh, 100%.
0:27:29 But the thing is, that’s not surprising.
0:27:33 Like it has access to every public earnings call in history.
0:27:34 Yeah, yeah.
0:27:38 And like at the end of the day, analysts can only ask you like, tailwinds, headwinds, who’s
0:27:38 buying what?
0:27:41 It’s not because the analysts are smart or not smart.
0:27:44 It’s just like, those are the things that like you would try and deduce from an earnings
0:27:45 call, buying a stock.
0:27:47 And you wouldn’t have thought of these questions beforehand?
0:27:48 Or is it just like…
0:27:49 I think you’re doing…
0:27:50 On the margin, on the margin.
0:27:51 No, no, sorry.
0:27:56 So what I’m using is then the specific parts of the document that is missing the answers
0:27:57 to those questions.
0:27:59 So I can actually inject the answers into that.
0:28:03 Because like you’re typing out a thing and like, I forgot to give two case studies in
0:28:04 this section or whatever.
0:28:07 It’s a quick way to just do some analysis on something.
0:28:12 But yeah, I mean, so Bezos famously had this memo-oriented, essay-oriented kind of meeting
0:28:12 structure.
0:28:13 And we never did that.
0:28:16 But I was always fascinated by the companies that could do it.
0:28:19 And actually, we’re entering a world where probably you could just pull that off, right?
0:28:23 So imagine if, whether it’s a board meeting or product meeting, you just do a quick, deep
0:28:24 research essay on the topic.
0:28:29 Like, obviously, every meeting, every strategy meeting in history would be better off if
0:28:32 you probably had that as a starting asset to get everybody informed.
0:28:36 I think the argument against that would be, the reason Bezos said it is because it forced
0:28:39 people to think clearly about what they’re doing and writing it down.
0:28:42 So the exercise meant that people walking in the meeting had more context.
0:28:46 This would almost argue that they would have less context because something else did the
0:28:46 thinking.
0:28:47 Well, two things.
0:28:51 It was to make sure that the person doing the thing had the clarity to write it, for sure.
0:28:54 But it was also still to inform everybody else that didn’t do that work.
0:28:57 And so it certainly would have helped everybody else in the room.
0:28:59 And I’m not 100%.
0:29:03 I mean, we should do a full longitudinal analysis of like the people that wrote the essay.
0:29:04 Did they actually have the better products?
0:29:06 Or like, I mean, there’s some Amazon products I don’t like.
0:29:09 And so they obviously wrote an essay also for those.
0:29:12 So I don’t know the hit rate ultimately on the essay specifically as much as the idea of
0:29:14 like write down a strategy, think it through.
0:29:18 And so why not have an agent do 90% of the heavy lifting?
0:29:24 So a lot of my workflows are like, if I have a topic where like maybe the direct change
0:29:29 of my workflow on this front is the kind of thing that three years ago, I might sort
0:29:32 of lob over to the chief of staff and say, hey, can you like go research like the pricing
0:29:35 strategy of this ecosystem or something?
0:29:37 That’s just a deep research query now.
0:29:39 And then I’ll wake up and it all look at this thing.
0:29:44 But what that does is because now I’m not having to calculate that person’s time, their
0:29:45 tasks, their trade-offs.
0:29:51 I just do it for the most random things, which means like I’m expanding and exploring
0:29:54 way more spaces mentally than I would have before.
0:29:55 And these are the kind of parts.
0:29:59 And this is equally why I’m like actually more optimistic on the jobs front, because what we
0:30:04 do too many times with an AI is we like look at today’s way of working and we’re just like
0:30:06 AI will come in and take 30% of that.
0:30:06 And it’s like, no, no, no.
0:30:08 We’ll just do totally different things with AI.
0:30:12 I wouldn’t have researched that thing before when it was people required to research it
0:30:15 because that would have been an inane task to send to somebody.
0:30:15 Yeah, yeah, yeah.
0:30:17 One thing.
0:30:22 So when we run the numbers and by run the numbers, I mean look through how AI companies are doing,
0:30:23 where does the value accrue?
0:30:26 There’s basically one takeaway.
0:30:30 And that is like these markets are very large and growing very fast.
0:30:32 And value is kind of accruing at every layer.
0:30:35 Everything from like literally chips up to apps.
0:30:40 And so like the only real sin is zero-sum thinking to be like, oh, like the models are
0:30:43 not going to be defensible or whatever your zero-sum thinking is that just hasn’t proven
0:30:44 out.
0:30:47 Now, this has still largely been a consumer phenomenon.
0:30:48 So what I’ve been thinking about, and I don’t have an answer.
0:30:53 I’d love to hear your thought is, is when it comes to enterprise budgets, like you can’t
0:30:55 just create budget out of thin air.
0:30:57 So like you actually do have a limited resource.
0:31:02 And so as budgets get reallocated, to what extent do you think this is like zero-sum,
0:31:06 like the old budget gets robbed versus like budget accretive?
0:31:07 Or like, how do you think about that?
0:31:10 Because again, like where we’ve come from, that has not been an issue.
0:31:12 I think in the enterprise, it probably will be.
0:31:14 So it does have to come from somewhere.
0:31:14 It’s fully logical.
0:31:15 A couple of things.
0:31:16 Yeah.
0:31:20 A large number for startups can also be a very small number for a large corporation.
0:31:21 Yeah.
0:31:23 So you have that dynamic playing out.
0:31:28 I’ll make up random stats, but you could probably take a meaningful engineering team and
0:31:32 probably for the price of five of those engineers or 10 of those engineers, you could
0:31:34 probably pay for cursor licenses for the entire engineering team.
0:31:38 But this would argue that it’s actually coming out of headcount.
0:31:41 So here’s where the asterisk is.
0:31:43 There’s an infinite set of ways.
0:31:46 This is why like you can never take a point in time snapshot on these kinds of things.
0:31:46 Yeah, totally.
0:31:48 There’s an infinite set of ways that this actually plays out.
0:31:49 Yeah.
0:31:56 Next year’s planning process, maybe in a perfectly like parallel universe, the salary increase
0:31:59 that year would have been 3.5% for employees.
0:32:02 And this coming year, it’s 3% because we’re going to take 0.5% and we’re going to deploy
0:32:04 AI for the company.
0:32:09 Or maybe next year, we would have added 50 engineers, but we’re going to add 25 and then pay for AI.
0:32:10 But guess what?
0:32:14 The year after, we’re going to have engineering productivity gains.
0:32:16 So it increases because it’s still a competitive environment.
0:32:21 We then now add engineers the year later because we’re getting higher productivity gains.
0:32:21 Yeah.
0:32:25 I think that most companies of any reasonable scale post 100 employees, let’s say, have
0:32:31 enough sort of dynamism in the financial model within a one to two year period where it doesn’t
0:32:33 look like what the economists would think it looks like.
0:32:34 Can I just spit this back?
0:32:37 Because I think this is actually a very good point that’s buried in there.
0:32:42 I just want to make sure I’m following along, which is the software license cost to a startup
0:32:46 relative to like a large people organization is relatively small.
0:32:50 It’s just a couple of headcount, which if you just look like normal performance management,
0:32:55 normal attrition, normal like variability, and even like hiring timelines is kind of in
0:32:56 the noise.
0:32:59 And so you already have an annual budgeting cycle to fix that up.
0:33:03 And so like basically within the noise, even of just like headcount planning, all of this
0:33:05 could work out without some massive disruption.
0:33:06 And I think that’s such a cool point.
0:33:10 And there could be an upper limit of this point, but let’s say the going rate in Silicon Valley
0:33:15 of a new engineer coming out of college, let’s just say it’s somewhere between 125 and 200.
0:33:16 Okay.
0:33:17 I’m just making up.
0:33:17 Okay.
0:33:17 Yeah.
0:33:22 Let’s say your most aggressive cursor usage or something is like a thousand bucks a year,
0:33:22 2000 bucks a year.
0:33:25 So you’re at like 1% salary maybe.
0:33:26 Here’s the question.
0:33:28 Again, do this like crazy apples to apples thing.
0:33:33 If you went and recruited from Stanford right now and you said, okay, you Stanford grad have a
0:33:33 choice.
0:33:41 You can work at this company and get paid 125K with no AI, or you can get paid 123K with
0:33:42 full access to AI.
0:33:43 Which one are you going to do?
0:33:45 They would do the 123 all day long.
0:33:48 But even that, yeah, I mean like your argument is, which makes a lot of sense to me.
0:33:50 It’s kind of on the margin when it comes to like 20.
0:33:55 But just as a way of exploring like why these things are not the high order bit of the cost
0:33:55 increase on budgets.
0:33:56 Oh, I love that.
0:33:57 That’s great.
0:34:02 And I did one kind of late night sort of like modeling once, but I’m afraid to say all the
0:34:03 numbers here because I think they’re just going to be so wrong.
0:34:07 But I think it’s something on the order like five or six trillion in knowledge worker headcount
0:34:08 spend in the U.S.
0:34:08 Yeah.
0:34:11 Everybody says for developers, let’s say 40 million.
0:34:12 Let’s just say it’s 30 million.
0:34:12 Yeah.
0:34:14 Let’s say that the average is 100K or you’re at three trillion.
0:34:16 Man, these are just massive, massive numbers.
0:34:17 So it’s many trillion.
0:34:17 Yeah.
0:34:19 So you have many trillions of dollars.
0:34:23 If you take a couple percent of that or five percent of that, you’re already doubling the
0:34:24 entire sort of U.S.
0:34:25 enterprise software spend.
0:34:26 Yeah.
0:34:28 So you can just make it work within.
0:34:31 And this is why I don’t think that people will not make cuts because they have to pay for
0:34:31 AI.
0:34:33 They might make cuts for other reasons.
0:34:33 Sure.
0:34:37 But even in those cases, I think you’ll often have it be for myopic reasons temporarily.
0:34:38 Yeah.
0:34:42 And there’s enough flexibility to basically consume this and then actually like recap on
0:34:43 the productivity game.
0:34:43 Yeah.
0:34:43 I think that’s great.
0:34:49 I try and parse everything you say through the lens of like, where are you landing on AI
0:34:50 coding?
0:34:53 And you seem to have a very pragmatic view of where things actually are at.
0:34:54 Yeah.
0:34:55 Where are you landing right now?
0:34:57 Well, it’s been an evolving.
0:35:02 So I would say in the entire AI thing, the biggest surprise to me is how effective it is
0:35:02 at code.
0:35:08 And so my sense is, so I’m just going to say a couple of, I think, facts, and then we can
0:35:10 kind of back out what this means in aggregate.
0:35:16 Because I think one fact is, the reality is, I do think that AI helps better developers more
0:35:17 than not better developers.
0:35:22 And the reason is you just have to deal with and be able to like know what to ask for and
0:35:22 know how to deal with the outcome.
0:35:23 So I think that’s one.
0:35:27 Someone said it, I thought, beautifully, which was, this is a very good developer.
0:35:28 And this was on X.
0:35:29 I forgot who it was, but I thought it capsulated.
0:35:35 He’s like, you know, 90% of what I know, the value of it has gone to zero, but 10% has
0:35:37 tripled more than 10X or whatever it is.
0:35:38 They’ve got 100X.
0:35:40 And I think that’s exactly right.
0:35:46 I do think that for a lot of rote use cases, the AI can do it and it doesn’t need to be
0:35:47 double checked, right?
0:35:50 So there’s a lot, to your point, like things like prototyping, things like scripting.
0:35:55 And so I do think if you look at usage of like open AI, if you actually look at code
0:35:59 usage, it’s like the primary use is actually professional developers, which means it’s part
0:36:00 of a developer workflow.
0:36:07 And then probably the most controversial stance I have is, and this is probably like sunk cost
0:36:10 fallacy because I’ve been a programmer for, I mean, like my PhD is in computer science,
0:36:12 like, you know, so maybe this is sunk cost fallacy.
0:36:17 But I just don’t see a world where you get rid of formal programming languages just because
0:36:20 they arose out of natural languages for a reason.
0:36:24 Like we started with English and then we made programming languages so that we could formally
0:36:25 describe stuff.
0:36:27 And so it’d be kind of a regression to go back.
0:36:30 So I still think we’ll use languages and maybe they’ll change, maybe more like a scripting
0:36:35 language, but I think like the existing tool set will evolve, but it’ll still be a professional
0:36:35 developer.
0:36:38 Like I think we’ll still have developers, still have developer tools.
0:36:38 So that’s kind of where I am.
0:36:39 I would love to hear where you’re at.
0:36:42 If I’m fully, I’m on the exact same page.
0:36:46 The fun thing to me is how coding is just at the tip of the kind of iceberg.
0:36:51 It’s the best thing to first sort of experience agentic automation, but I think you’ll see
0:36:52 this in basically every other space.
0:36:59 But what’s so fun is just in a one-year shift, let’s say, of like the nature of the relationship
0:37:00 with the AI.
0:37:04 So if you think about the GitHub co-pilot moment was like, oh, this thing is incredible.
0:37:07 It’s going to type ahead and predict what I’m typing.
0:37:12 And then you’re basically using it to work 20 or 30% faster and which parts of it do you
0:37:13 take on or not.
0:37:19 And now the relationship is like totally different within, again, a year or two period where you’re
0:37:23 using cursor, Windsor for whatever, and the agent is generating this chunk of an output
0:37:25 and then you’re just reviewing it.
0:37:30 But what’s incredible is like none of your expertise is any less valuable in that review.
0:37:34 In fact, it’s probably even more important than ever before because in some cases, like
0:37:39 it’s just going to be like wrong 3% of the time and then you review it, but then you’re
0:37:41 literally doing 3x the amount of output.
0:37:47 And the nature of how that changes both programming, but just like, why not have that for basically
0:37:51 everything is sort of this new way that both software should work and then actually we will
0:37:56 work is like, you know, the big joke a year after Chat to BT is like, okay, this thing generates
0:37:59 a legal case and it’s like wrong 10% of the time.
0:38:00 And it’s like, well, actually, hold on.
0:38:04 If you think about what the new paradigm of work looks like, and it’s like such a weird inversion
0:38:07 of it used to be the AI was fixing your errors.
0:38:09 That’s what we thought the AI was going to be.
0:38:10 And it’s just like a total flip.
0:38:12 It’s like the human’s job is to fix the AI errors.
0:38:14 And that’s the new way that we are going to work.
0:38:15 Right.
0:38:16 So this begs a very obvious question.
0:38:17 I’m going to work up to the question.
0:38:22 So there is a great paper in NSDI from an MIT team, which basically says you can optimize
0:38:24 a running system with agents.
0:38:29 And the way they did it is they basically have a teacher agent and then like more junior
0:38:32 agents and then the more junior agents would go try a bunch of stuff.
0:38:37 And of course, they had much more knowledge of the literature than any single human being.
0:38:39 So they try all of different things to try it.
0:38:41 And then the one at the senior agent would say, oh, this is good.
0:38:42 This isn’t good.
0:38:44 And then once it optimized the system, they would do it.
0:38:48 And then, you know, like the human being is then kind of helping the teacher agent decide
0:38:52 what are the parameters, what is good, what is not good and provide high level direction.
0:38:52 Right.
0:38:58 And so you’re already starting to see cases where human beings are running multiple agents
0:39:02 and even that already is starting to have some kind of bifurcation, which one way to think
0:39:08 about it is in any R&D organization, of course, people start as like ICs, but then they very
0:39:11 quickly get interns and go into management.
0:39:12 And so maybe we’re just skipping that step.
0:39:17 So the obvious question is what happens to entry level engineers?
0:39:22 Like does this change how people get introduced to computer science, for example?
0:39:26 The cool thing is probably more people will even now get introduced to computer science
0:39:27 because you’ll be able to…
0:39:28 Anybody can learn.
0:39:29 Anybody can learn it.
0:39:33 And, you know, it’s been 25 years for me, but like in the early days of programming basic
0:39:37 applications or putting other websites, it was just extremely frustrating that you would
0:39:40 spend days and days being like, why does that thing not work?
0:39:40 Yeah, yeah, yeah.
0:39:43 And like I have very few resources of figuring out why the thing didn’t work.
0:39:47 It would have been a hundred times easier if I could have had an agent write the thing.
0:39:48 I would have learned 10 times faster.
0:39:49 Yeah, yeah.
0:39:53 Honestly, what you did is you was like, well, not 25 years ago, but 10 years ago, you go
0:39:53 to Stack Overflow.
0:39:55 And so it was like the slow version.
0:40:00 Yeah, but so think about how many people missed the window pre-Stack Overflow that got sort
0:40:03 of pushed out of the ecosystem because they’re just like, this is too frustrating.
0:40:07 And so you’re going to have a way bigger funnel at the top of people now learning programming
0:40:08 and computer science.
0:40:11 I think a similar percentage of people will fall out.
0:40:14 So it’s not like, again, you’re going to get a 10x increase in programmers because you
0:40:17 still have to enjoy it and you have to like solving problems and whatnot.
0:40:21 It’s going to change the nature of the incoming class of engineers that you hire.
0:40:25 They will literally not be able to code without AI assisting them.
0:40:30 And it’s not 100% obvious that’s a bad thing because assuming you have internet and the site
0:40:32 stays up, like we should have access to the agents.
0:40:37 So I think it’s mostly just like we have to adapt how we think about the role of an engineer
0:40:40 and what these tools are giving us in terms of the productivity gains.
0:40:45 Actually, I meet with a lot of larger, not tech-oriented companies as customers.
0:40:50 And generally, the thing I’m recommending is hire a bunch of these people because they’re
0:40:55 going to flip your company on its head of how much faster the organization can run.
0:40:56 So I do understand.
0:40:59 I want to be sympathetic to the job market for anybody coming out of college because I don’t
0:41:00 think it’s easy right now.
0:41:02 And it probably hasn’t been easy in a number of years.
0:41:10 But if you are graduating, the thing I would be selling any corporation some way or another
0:41:15 is that if you are AI native right now coming out of college, the amount you can teach a company
0:41:16 is unbelievable.
0:41:20 And then conversely, if you’re a company, you should be actually like prioritizing this talent
0:41:26 that is just like, why does it take you guys two weeks to research a market to enter?
0:41:29 I can do that in deep research and get an answer to you in 30 minutes.
0:41:32 They will be able to show companies way faster ways of working.
0:41:37 Do you think there’s any stumbling into problems this way, which is like you kind of adopt too quickly.
0:41:42 It’s like you get into a morass you can’t get out of or you think at this point it’s pretty clear
0:41:44 this stuff can be practically consumed.
0:41:46 What would the morass be that you’d get into?
0:41:50 You hire a bunch of vibe cutters and then they create something that nobody can maintain
0:41:51 and it’s really slow.
0:41:51 Oh, yeah, yeah, totally.
0:41:54 Which, by the way, I will say I have seen this.
0:41:54 Yeah, yeah, yeah.
0:41:56 Okay, you could easily overdo this whole thing.
0:42:01 So I think as with anything, like deploying these strategies in moderation
0:42:05 while we’re all collectively still getting the technology to work better and better
0:42:08 is super important and understanding the consequences of these systems.
0:42:12 So, yes, this is not like a moment to just have your whole company vibe code.
0:42:15 I will say one of my favorite things that I’m witnessing in the whole coding thing,
0:42:18 I don’t know, the point of this talk is kind of the AI and the enterprise generally,
0:42:20 but like the coding thing is just so salient,
0:42:23 is that a lot of these OG programmers that I’ve known for a long time
0:42:26 that are off-creating companies or CEOs of public companies like yourself
0:42:27 are all back to programming.
0:42:27 Yeah.
0:42:30 And then you talk to them, you know, many of them, like I code, you know,
0:42:33 most nights with Cursor just because it’s really enjoyable.
0:42:37 And the reason I didn’t code before is because I just couldn’t keep up with the fucking frameworks.
0:42:39 I’m like, dude, I don’t know how to install the fucking thing.
0:42:41 And what is this Python environment stuff?
0:42:45 And like, you’re literally learning bad design choices that somebody else just made up.
0:42:47 Like, they’re not fundamental to the laws of the universe.
0:42:49 They don’t make you any smarter.
0:42:51 It’s just like waste of brain space.
0:42:55 And so in some way, the AI just gets rid of this kind of crufty stuff
0:42:57 that you probably shouldn’t be wasting brain space on anyway.
0:43:01 The amount of frustration I have when I look through, let’s say, our product roadmap.
0:43:01 Yeah.
0:43:04 Let’s say pre-AI, although this still obviously happens
0:43:05 because we haven’t fully transformed everything about how we work.
0:43:08 But pre-AI, when you would see things like,
0:43:12 we have to upgrade the Python library in this particular product.
0:43:15 And it’s like three engineer, two quarters.
0:43:16 No, exactly.
0:43:20 And like, at the end of that project, zero customers will notice that we did something.
0:43:24 We resolved some fringe vulnerability that is not going to even happen.
0:43:26 But you have to do it because there’s some compliance thing
0:43:30 where you have to make sure you’re on the latest version, which is super important.
0:43:32 But like, the thing is never going to happen.
0:43:37 And all of a sudden, like, you are wasting hundreds of thousands of dollars of engineering time.
0:43:41 And the fact that like, that’s now like a codex task is just unbelievable.
0:43:45 And the amount of just now things that you can actually relieve your team to go and work on is incredible.
0:43:51 And the other big, like, boon for the economy, and this is again where the economists just totally miss this stuff,
0:43:56 is think about every small business on the planet, of which there’s millions, tens of millions, whatever,
0:44:01 that for the first time ever in history, they have access to resources that are somewhat approximate
0:44:03 to the resources of a large company.
0:44:06 Like, they can do any marketing campaign.
0:44:08 Did you see the NBA finals video from Kalshi?
0:44:09 No.
0:44:10 The VO3 video?
0:44:11 Oh, yeah, yeah, yeah, yeah, yeah.
0:44:15 Like, you can now put together an otherwise million dollar marketing video.
0:44:18 For a couple hundred bucks of tokens.
0:44:21 And that being applied to every domain in every service area,
0:44:24 I can run a campaign that translates into every language.
0:44:29 I can have this long tail of bugs that I never got around to automatically get solved.
0:44:34 I can have the analysis of a top tier consulting firm done for my particular business.
0:44:43 So for the people or companies that are resourceful and are creative and imaginative, the access to resources right now is just truly unprecedented.
0:44:48 What do you think is the best metric for anybody interested in tracking this stuff as far as, like, how fast it’s going?
0:44:49 Is it GDP?
0:44:51 Is it margin?
0:44:52 Is it top line?
0:44:53 Is it headcount growth?
0:44:54 Is it all the above?
0:44:55 Like, how do you measure it?
0:44:56 Yeah.
0:45:02 I mean, for us, so internally first, and then maybe we’ll spitball some macro solutions to this.
0:45:08 Internally, we’ve actually explicitly taken the stance that we want to use AI to increase the capacity and capability of the company.
0:45:10 So just do more.
0:45:12 For anything that you track, just make sure it happens fast.
0:45:13 Just do more.
0:45:14 Just, like, do more or do faster.
0:45:16 In a given time period, yeah.
0:45:21 And so that somewhat relieves the pressure from people that, like, this is about cost cutting.
0:45:21 Yeah, yeah, yeah.
0:45:23 It’s just like, no, no, just, like, do more right now.
0:45:24 Let’s figure out what works.
0:45:25 Some things won’t work.
0:45:26 We want experimentation.
0:45:28 So just use AI to do more.
0:45:29 Okay, so that’s us.
0:45:35 The way you should measure that then in a couple years from now is either the growth rate of the company should be faster.
0:45:36 Yeah.
0:45:45 Or the amount of things that we’re collectively doing should be more, and the only reason that wouldn’t show up in growth rate is that every other company also does more, and so that gets competed away.
0:45:45 Yeah.
0:45:50 Which is, like, also a very viable outcome, is this is just the new standard of running a business.
0:45:51 But there’s no shift in the equilibrium.
0:45:52 Right.
0:45:52 There’s no shift in the equilibrium.
0:45:54 You just have to do it.
0:46:00 And then the ultimate product of all of that is some other kind of metric of satisfaction of, like, our products get better.
0:46:02 It could be, like, consumer price index or something.
0:46:04 Yeah, but, like, did the iPhone show up in GDP?
0:46:07 I don’t know, but my life is better with the iPhone pre than without the iPhone.
0:46:08 I’m pretty sure it did.
0:46:09 Okay, yeah, fine.
0:46:15 So, but, like, it would ultimately then show up in, like, new cures to diseases, better health care.
0:46:20 I don’t know that the dollars would move around all that differently as much as just, like, life expectancy should go up.
0:46:22 Like, cost of housing should go down.
0:46:30 Like, weird metrics that productivity gains will then drive that the economists wouldn’t naturally associate to, like, enterprise software and AI.
0:46:36 By the way, this is where I am, which is, like, clearly there’s a disruption because marginal costs are going down on a bunch of stuff.
0:46:36 Yeah.
0:46:38 Like, writing code and language reasoning and whatever.
0:46:43 And, like, some companies will take advantage of that, but I don’t think, like, the fundamental equilibrium changes.
0:46:46 I think, to your point, I think we just do more tech, products get better faster.
0:46:47 Yeah.
0:46:51 We saw problems that we haven’t before, but, like, it’s not asymmetric that we’ve seen in other companies.
0:46:59 The way I kind of think about it is, you know, if we go back to, like, 1985 and we just looked at how everybody works, I think we would just be totally stunned by how slow everything is.
0:47:05 And just, like, how long did it take you to research a thing or analyze a market or create a campaign or whatever?
0:47:05 Yeah.
0:47:10 But, like, it just has now been baked into our human productivity that we just do all those things really fast now.
0:47:10 Yeah.
0:47:16 And so, in 10 years from now, when we all have AI agents running around, we will just look back to today and be, like, how did we function?
0:47:22 Like, you spent two weeks to decide, like, the message for the marketing campaign?
0:47:24 Like, how is that possible?
0:47:27 Like, what we do now is we run 50 experiments with AI agents.
0:47:28 They all come back with versions.
0:47:32 We look at them all together, and then we make a decision in an hour, and we move on.
0:47:34 That’s obviously, like, how work works.
0:47:36 And, like, that’s what we will be saying 10 years from now.
0:47:38 Do you think we’ll ever saturate the consumer?
0:47:44 I caveat this by saying this comes up every one of these inflection points, and so I just wanted to ask it again for the umpteenth time.
0:47:46 I think I’ll say yes, just because at some point, maybe.
0:47:50 But, like, my list of purely consumer demands has not gone down.
0:47:55 Like, healthcare is, like, a totally unmet need that I have.
0:48:01 I do not like to go to doctors or dentists or anybody because of just how hard it is to get scheduled.
0:48:03 I mean, buying a fucking car, man.
0:48:05 Like, there are so many things that, like, just need to be sold.
0:48:06 The cost of housing.
0:48:07 We clearly don’t have enough houses.
0:48:10 Like, where will AI, you know, drive that?
0:48:12 Okay, so, you know, maybe robotics would be then the play there.
0:48:17 But, like, I don’t think we’re anywhere close to consumer satisfaction or satisfying all needs of consumers.
0:48:25 Well, actually, I meant more, will things change so fast that, like, it saturates the ability to adopt new things?
0:48:27 I do think that that is certainly possible.
0:48:34 I think I track sort of, let’s say, my parents as a decent kind of proxy or even just, like, college friends that aren’t particularly in tech.
0:48:34 Yeah.
0:48:38 And they’re still, like, in their ChatGPT phase of adoption.
0:48:39 And they haven’t moved on from that.
0:48:40 They haven’t made a VO video yet.
0:48:45 They’re just, like, using ChatGPT to ask questions about life experiences they have.
0:48:50 And so, maybe, ironically, like, one of the problems was ChatGPT was so good.
0:48:58 If you, like, you know, imagine what people thought AI should be able to do for them, it already met, like, 80% of, like, where they would have sort of projected it.
0:48:58 Yeah, yeah.
0:49:01 But when we know, actually, no, it can probably still do 10 to 20x more.
0:49:02 Yeah.
0:49:05 But their needs are going to be satisfied for some time on those core use cases.
0:49:06 Yeah.
0:49:10 So, I think for, like, the most basic consumer query type things.
0:49:19 But then this is the opportunity for startups, which is, like, now AI will show up in sort of ways that maybe the person isn’t even, like, in the market for an AI thing.
0:49:21 They just want a better version of that category.
0:49:24 I was going to say, this could just, like, simply be another market constraint.
0:49:27 As soon as it saturates, you just make the product better given, like, the existing.
0:49:35 Then, if I could just get better healthcare, but I don’t need to think about that as an AI problem or not an AI problem, but AI will be behind the scenes delivering that.
0:49:35 Yeah.
0:49:37 Then I don’t think you’re saturated anytime soon.
0:49:39 Yeah, it’s just the consumption capacity just becomes another market constraint.
0:49:41 But there’s a ton of other ways that you can improve things.
0:49:42 100%.
0:49:42 That’s great.
0:49:42 Good.
0:49:44 I love that you’re so optimistic about it.
0:49:47 I am, I think, 98th percentile optimistic.
0:49:47 Same.
0:49:48 Good.
0:49:48 All right.
0:49:54 So, I think we’ve had a fairly pragmatic conversation about the current impacts and the near-term impacts.
0:50:00 If you do a longer view, can you dare to guess what things look like in five to ten years?
0:50:04 So, Sam Altman and Jack Altman had a podcast recently.
0:50:05 Yeah, yeah, yeah.
0:50:12 And I’m going to paraphrase probably in some wrong way, but they were going back and forth about how, like, we just got what we would have predicted as AGI five years ago.
0:50:14 And it’s just like, we use it.
0:50:16 And it’s, like, it’s now just built in.
0:50:17 The most anticlimactic.
0:50:18 Yeah, it’s anticlimactic.
0:50:19 Anti-anticlimactic.
0:50:24 And I think that’s my instinct for a lot of this is five years, ten years, whatever your number is.
0:50:34 And this is why I’m so optimistic on just society and jobs and all this stuff is I don’t think it’s the Terminator kind of crazy outcome scenario of we automate away everything.
0:50:48 I think the human capacity for wanting to solve new problems, for creating new products, for serving customers in new ways, for delivering better healthcare, to try and do scientific discovery, like, all of this stuff is just built in us.
0:50:48 Yeah.
0:50:49 And it will continue.
0:50:55 And AI is this kind of up-leveling of the tools that we use to do all those things.
0:50:59 And so I think the way we work will be totally different in five years or ten years.
0:50:59 Totally.
0:51:05 But you’re already seeing enough of probably what it will look like that I think it’s an extrapolation of that.
0:51:15 It’s when you want the marketing campaign done, you have a set of agents that go and create the assets and choose the markets and figure out the ad plan.
0:51:19 And then you have a few people review it and debate it and say, okay, let’s go in this direction instead.
0:51:21 And then you deploy it and you’re on to the next thing.
0:51:25 And so each company, their units of output grow.
0:51:29 As a result of that growth, we’re all still in competitive spaces.
0:51:33 So some of it gets competed out and others will keep growing faster than they would have before.
0:51:35 So they’ll hire more people and you’ll have new types of jobs.
0:51:38 Like we’ll have jobs for people just to manage agents.
0:51:40 And like you’ll have operations teams.
0:51:43 You know, Adam D’Angelo had this cool role that just kind of got announced.
0:51:44 Yeah, that was really cool.
0:51:49 Yeah, the role is to work with Adam at Quora and figure out which workflows can be automated with AI.
0:51:51 I think you’ll have a lot of those kind of functions.
0:51:56 But I think one of the exciting things about at least being in Silicon Valley or anybody kind of tuning in, being in this ecosystem,
0:52:00 is like we’re seeing the change happen faster here.
0:52:05 And it’s going to be five or ten years of this rolling out to the rest of the economy.
0:52:11 And so I think we will spend the next five years making the technology actually deliver on the things that we’re all collectively talking about
0:52:18 to make it more and more robust and the accuracy goes up and the costs go down and the workflows it can tie into are better.
0:52:20 And we will be working on that for quite some time.
0:52:29 And you think ultimately this leads to the biggest peace dividend of better products for users, better user experience?
0:52:31 Yeah, I think the software gets better.
0:52:33 Our healthcare gets better.
0:52:34 The life sciences discoveries increase.
0:52:36 I think it’s all a society net positive.
0:52:37 I love it.
0:52:42 Thanks for listening to the A16Z podcast.
0:52:48 If you enjoyed the episode, let us know by leaving a review at ratethispodcast.com slash A16Z.
0:52:50 We’ve got more great conversations coming your way.
0:52:51 See you next time.