Box CEO on AI Agents & Why Enterprise Can't Keep Up | a16z

Steven Sinofsky, board partner at a16z, Aaron Levie, CEO of Box, and Martin Casado, general partner at a16z, discuss the reality of AI inside enterprises. They cover the gap between Silicon Valley and the rest of the world, why most AI initiatives fail in large organizations, and how agents, infrastructure, and workflows are evolving beyond the hype. Timestamps: 00:00 - Trailer 01:05 - Introductions & The Silicon Valley vs Enterprise Gap 04:30 - Why Enterprise AI Efforts Keep Failing 09:16 - The Architectural Shift: Treating AI as a User, Not Software 14:38 - The Integration Wall Agents Can't Climb 20:12 - Should Agents Be Treated Like Humans? 24:40 - Salesforce Goes Headless & What It Means for SaaS 39:16 - Scale, Entropy & Why AI Coding Creates as Many Problems as It Solves 47:53 - Will AI Kill Jobs or Create More of Them? Resources: Follow Aaron Levie on X: https://twitter.com/levie Follow Steve Sinofsky on X: https://twitter.com/stevesi Follow Martin Casado on X: https://twitter.com/martin_casado Stay Updated: If you enjoyed this episode, be sure to like, subscribe, and share with your friends! Find a16z on X: https://twitter.com/a16z Find a16z on LinkedIn: https://www.linkedin.com/company/a16z Listen to the a16z Podcast on Spotify: https://open.spotify.com/show/5bC65RDvs3oxnLyqqvkUYX Listen to the a16z Podcast on Apple Podcasts: https://podcasts.apple.com/us/podcast/a16z-podcast/id842818711 Follow our host: https://x.com/eriktorenberg Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see http://a16z.com/disclosures.

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Speaker 00

  The world is going to need platforms that support billions of transactions per second, billions of transactions per second, which no payment rail or platform does today.

• 0:11Speaker 01

  Where we think things will go is just there will be a huge amount of agentic commerce. And again, we're seeing a little bit of it today. We think there'll be a torrent of it. And that is what unites stablecoins and AI, because we think you're going to need blockchains and better blockchains.

• 0:27Speaker 00

  Up until now, the economics of software have been conceived of as fixed costs and then infinitely monetize or monetize as much as possible. That has the winner take all dynamics. But once there are inference costs and custom creation involved, it really shifts.

• 0:43Speaker 01

  Finally, one executive who said, oh yeah, we started, you know, augmenting our customer service with AI so people are more productive, but we're just going to go back to doing it the old fashioned way.

• 0:53Erik Torenberg?

  Stripe processed more than a trillion dollars in payments last year. It grew 34%. And according to Patrick Collison, 2026 Q1 may be looked back on as the first quarter of the singularity. That's not a marketing line. It's what the data is showing. The 2025 cohort of businesses on Stripe is larger and performing better on a per business basis than any prior cohort, and the trend is accelerating. In this conversation recorded live on TDPN, John and Patrick Collison walk through what they're actually seeing in the real economy, why agentic commerce will require blockchains capable of billions of transactions per second, and a reframe on software itself, from mass-produced product

• 1:34Erik Torenberg?

  to something bespoke, cooked fresh at the moment of use, like pizza. This conversation originally aired on TDPN, hosted by John Coogan and Georgie Hayes.

• 1:47Speaker 03

  We have John and Patrick Collison. Yo, geez. From Stripe. How are you guys doing? What's going on?

• 1:53Speaker 04

  Greetings.

• 1:54Speaker 03

  Welcome to the show. Thank you so much. This is, uh, this is huge. Uh, I went through YC. You guys were massively influential in my career and, uh, it's a joy to speak to you today on such a big day, but I'd love for you to kick it off with the actual news. What happened? Why are we talking today?

• 2:11Speaker 01

  Uh, we had two announcements today. One is we're launching a tender offer for, uh, employees and that and kind of the valuation, everything tended to get a bunch of the headlines. Uh, the thing that was honestly, uh, more work was we released our annual letter where every year we, uh, sum up all the trends, uh, that we're seeing, uh, on Stripe and, uh, Stripe is growing a lot. We grew at 34% last year because the businesses on Stripe, uh, are growing a lot. And there's just, as you guys know, there's a lot happening in tech right now. This is why we need TVBN. This is why we need a non-stop stream of everything that's going on because there is so much happening. So.

• 2:44Speaker 04

  Yeah, we'll move, we'll move to 24 hours.

• 2:47Speaker 03

  Eventually, eventually. Uh, I mean, but, uh, I feel like there is a ton of AI noise and stories and drama and we are, you know, never running out of stuff to talk about, but what are you actually seeing in the data? Because there's always this disconnect between the market and the real economy. Like people are still shopping in retail stores occasionally. Uh, what do you, where is AI actually moving the needle?

• 3:12Speaker 01

  Well, the, um, generally speaking, I would say from the Stripe data, it looks like the economy is in pretty good shape. And there's been, um, to say the, to say the least, uh, there's been, uh, some degree of volatility in markets, uh, over the last two years and, you know, all sorts of different events and deep-seak moments and what have you. But if you look at the actual real economy time series, if you look at what's actually happening substantively over the last two years, things, I mean, it's, it's always hard to prognosticate the future, but over the last two years, things really seem to be in good shape. The thing that's really catching our attention.

• 3:44Speaker 04

  One second, cause I'm just curious. Have you, have you guys tried to think about, uh, maybe the businesses are doing well on Stripe because they're, you know, uh, uh, kind of like forward looking, extremely tapped in, you know, working on the right things. And if you look at a bunch of legacy providers, you would see that actually there are a bunch of businesses out there that are slowing down, that maybe are feeling effective just overall, uh, consumer spending. Like, have you tried to kind of like break that out or understand that dynamic?

• 4:15Speaker 00

  It's obviously hard to measure because we don't have that data. We only have our data, but I think there is some of that composition effect. And we see it, I guess, both in Stripe's data compared to, say, public earnings from others. Like clearly the respective populations are performing somewhat differently. But I guess we also see it qualitatively in the conversations we're having with customers, where what tends to happen, say, for some incumbent is they built some business, they installed some system long before Stripe even existed. Maybe there's some sense that, well, if that broke, then don't fix it. But then they decide, hey, we're going to do something new. And when they're doing something new, then they want to use the best infrastructure

• 4:59Speaker 00

  that will enable them to move the fastest and launch in the most countries and support stable coins and do things with AI and whatever. And then they tend to launch that on Stripe. And so there is this qualitative sense that once a company decides to do something innovative, new, retool, what have you, they're more likely to come straight.

• 5:15Speaker 03

  Are you seeing overlap between stablecoin activity and AI activity? There's been sort of a new narrative around agents will use stablecoins, but I feel like agents can use legacy payment rails just fine. And then also you can do really cool things with stablecoins that are not really AI-native necessarily. And so I'm wondering how much overlap there is there.

• 5:38Speaker 00

  I would distinguish between how things work today and how things will work in the future. In terms of how things work today, agents absolutely can. You know, a lot of people build with Stripe. You know, you can have a one-time-use credit card that your agent can go out and spend. But if you look at what's happening, there's lots of, you know, agents having to solve captures to, you know, be able to kind of do stuff on the wider web. Clearly, the web is not built for agents. And as a result, they have to get creative to actually do any real-world tasks. And that's true in the kind of economic activity as well. Where we think things will go is just there will be a huge amount of agentic commerce. And again, we're seeing a little bit of it today. We think there'll be a torrent of it.

• 6:19Speaker 00

  And that is what unites stablecoins and AI, because we think you're going to need blockchain.

• 6:25Speaker 01

  And better blockchains. Honestly, I mean, this is what this was our thinking behind incubating Tempo. Because we're going to need really high throughput blockchains for the for the for the agents.

• 6:34Speaker 03

  Can you take us through some of the the historical technologies that led to growth in just internet payments? I'm thinking about like mobile, social commerce, one click checkout, Apple Pay. Like, there's so many things when I think about the agentic commerce boom that's coming like it could be hooking a better version of Siri up and, you know, ChatGPT rolling this out very aggressively, but also, you know, smart speakers, smart lamps, like your watch. Like, there's so many different pieces to unblock and unhobble the the actual agents as they go about their day.

• 7:12Speaker 00

  Well, can I answer a slightly different question without even come back to that? A point I just, sorry, this is a

• 7:20Speaker 04

  We'll tell you the questions, you tell us your answers.

• 7:24Speaker 00

  So you you know how brothers are, but then so I just want to lose one point for the prior question about, you know, what we're seeing in the economy because I feel like, I mean, this is very arbitrary, obviously, but I feel like there's at least a reasonable chance that 2026 Q1 will be looked back upon as the first quarter of the singularity. Maybe in three years, in hindsight, that'll look completely delusional. I don't know. But what we're seeing, I mean, there's this kind of the macroscopic picture, the Stripe user base and things overall looking pretty good and so forth. And the tumult's not quite showing up. But when we look at the cohorts and then when we look at the businesses that signed up in 2023 and their

• 8:10Speaker 00

  progression and trajectory over the subsequent months, the businesses that signed up in 2024, and then the business signed up in 2025, there's been a phased transition in 2025 where there are both more of them and on a per business basis, they are on average doing better. Which is really striking because you might think, okay, well, there's this cavalcade of new lightweight Code it applications or something, but you know, there's not really a lot of substance there. We're actually seeing both numbers move together. There are many more businesses getting started and the average median business is in fact performing better.

• 8:48Speaker 00

  We're only a couple of weeks into 2026, but it looks tentatively like 2026 may plausibly be an acceleration even over that significant leap of 2025. So I don't know, I mean, there's, we've, we've had all sorts of dramatic AI inventions and innovations over the last couple of years. There's a bit of a question of, well, how and when and how should we think about how it'll translate to the economy? I would say looking at real purchasing behavior on Stripe, 2025, end of 25, beginning of 26 is when I feel like we're really starting to see it.

• 9:28Speaker 04

  That's super interesting data. One, because we were, there was some survey that came out yesterday, or maybe it was late last week that said, they asked like, they asked a bunch of executives, are you getting any value out of AI? And 80% of them said no, but clearly if you look, when you look at.

• 9:47Speaker 00

  Come on, that's hogwash. Like, find me one executive who wants a refund on their tokens. Find me one executive who said, oh yeah, we started, you know, augmenting our customer service with AI. So people are more productive, but we're just going to go back to doing it the old fashioned way. Or like, we're spinning our code by hand and you know, we don't need any of this automated loom, you know, technology. It just like.

• 10:12Speaker 04

  Yeah, I'm not, I'm not saying, I'm not, I could, I could pick out a bunch of reasons. No, no, I'm not saying I agree with it.

• 10:17Speaker 03

  Techno pessimist.

• 10:18Speaker 04

  No, I could pick out a bunch of reasons why it would be wrong. One, one reason it might be wrong is they, is they're not in the weeds actually using the tools. And so they just think like.

• 10:25Speaker 03

  You might not even be aware that they're using the tools because it's buried under.

• 10:28Speaker 04

  And they're not, they're not feeling, they're not feeling the acceleration because they're not, they're not.

• 10:32Speaker 03

  Yeah.

• 10:33Speaker 04

  I wanted to ask how you guys think about incubations like Tempo. When you look at, when I look at Atlas and what Jeff and the team have done there. You think even in your, I don't know, kind of like the most wild projection that you had early with Atlas. Like, hey, maybe someday a quarter of the, of the C corps in the United States could be, you know, built on this platform. Anybody would have said that was insane and yet here we are.

• 11:03Speaker 00

  Gosh, I am not sure what to say really, except we just try to pay a lot of attention to the... I mean, as you guys know, there's a lot of pain points that go into starting a company, and we just try to take them seriously. And then, you know, it's the line, so much of such of these things is just a long obedience in the same direction. Like, Atlas is now this great overnight success, but we launched Atlas, I think, in 2014, maybe 2015. And so, you know, 10 years of compounding, and yeah, now it's at some pretty meaningful scale.

• 11:46Speaker 00

  And, you know, like, I think Tempo will probably end the same shape where we think it's... I mean, again, to this AI discussion and us sounding a bit unworn and untethered, like, I think the world is going to need platforms that support billions of transactions per second, billions of transactions per second, which no payment rail or platform does today. But even in a success case, it's not going to be an overnight thing. It's going to be, you know, five, six, seven years. And then maybe we'll have conversations about how, you know, Tempo suddenly became an overnight success or something, but done.

• 12:20Speaker 01

  I think Patrick's a bit the fish in water who can't, you know, who doesn't know things are wet. My framework would be, you can't get too MBA-brained about new products. You can't have your spreadsheet that's like, oh, the TAM is this, and just like reason about things in natural terms.

• 12:39Speaker 04

  Yeah, you should never say, "We want 1% of global GDP."

• 12:43Speaker 01

  No, exactly.

• 12:46Speaker 04

  You guys never, wait, you guys never pitched that?

• 12:49Speaker 01

  Well, companies... We actually never thought about Stripe in GDP terms until one day we realized, oh, hang on.

• 12:55Speaker 04

  Oh, wow. That's such an important lesson, because so many founders, how many pitch decks have you seen over the last decade? We're like, sorry. They're like, "Yeah, we just need 1%." And it's kind of, it's a meme.

• 13:07Speaker 01

  You can go back in the Wayback Machine and find the early Stripe websites, but we were very focused on payment.

• 13:11Speaker 01

  For developers and making that experience good. But where I'm going is I think you have to reason in product specifics. And so, again, I think any MBA would have told you that the adjacency of incorporation makes no sense. It's not related to what's our right to win. It was all these things people say, whereas you actually go talk to founders. They're like, guys, it's like this is the single biggest issue I ran into starting my company. And similarly with Tempo, and just as we think about incubations, we're trying to solve a real problem here. Where we talked in the letter about Bridge having operational issues, not because of Bridge,

• 13:50Speaker 01

  but because of blockchain congestion, where you have coins that are or blockchains both used for kind of meme coin trading and also serious real world payments. And so we just want low latency, high throughput payments, and we're going to need much higher throughput for the agents. But anyway, I think you have to reason in very specific product terms.

• 14:12Speaker 03

  What specific products are you excited about in the unhobbling of agentic commerce?

• 14:20Speaker 01

  We laid out in the letter, basically, these levels of agentic commerce. Because I think like everything in AI, people want to sell a hypey story. And so they talk about how the machines will buy everything without even consulting you. And people aren't actually, that seems far off. They're not that excited about that. You can start from just the basics of why are we filling out forms like that? You were talking about the progression of commerce. Why can't I just send something to a link to ChatGPT and have it buy us? Or why can't I search outside of just doing a basic keyword search or something like that? And so a lot of the work Stripe is doing is building the infrastructure.

• 15:01Speaker 01

  We're working with all the big retailers that you would expect, the Etsy's and Shopify's and Best Buys and Walmarts and folks like this to make. Product catalogs viable within the AI apps. And there's basically a ton of boring API and protocol and infrastructure work, which we love, that's our business. But people just want to be able to do shopping, do discovery, do purchases within the AI apps.

• 15:27Speaker 00

  And then just more kind of abstractly, we've been in this kind of this specific agenda commerce thing. And then there was just the general question of how software will change because of agents. And I've been thinking about it a bit, maybe software becomes a bit like pizza. That is to say, you know, you, software historically has been created.

• 15:53Speaker 01

  Not like pizza, some would say.

• 15:55Speaker 00

  Months, years beforehand, and then freeze dried and whatever you prepare it at the sort of moment of consumption. But we're actually going to, you know, software should be like pizza in a sense, it should be cooked right then and there at the moment of use. And so it's this quite fundamental shift where you don't want mass produced industrial scale software. You want bespoke custom software made for you that moment. That's, that's very fundamentally different. It's kind of the, you know, the, up until now, the economics of software have been conceived of as fixed cost and then infinitely monetize or monetize as much as possible.

• 16:40Speaker 00

  That has these kind of winner take all dynamics. But once there are inference costs and custom creation involved, it really shifts. It's kind of the non Walrasian software regime and just, I don't know, I don't quite know where it goes, but I think, I think it's going to look very different.

• 16:56Speaker 03

  Last question. Pineapple on pizza, yes or no?

• 17:02Speaker 00

  Ireland was was big into pineapple on pizza. Ireland not a big pineapple growing country, I will concede, but a lot of pineapple in the pizza. Good memories.

• 17:13Speaker 01

  A very large fraction of the banana market, don't forget. So we punch above our weight in fruits that don't grow there.

• 17:19Speaker 03

  There we go.

• 17:19Speaker 04

  There we go. The round is exciting. The the the overall growth of volume is exciting, but we wanted to hit the gong for how many books you guys are selling.

• 17:30Speaker 03

  Oh, yeah.

• 17:30Speaker 04

  Can you give us, can you give us the the numbers there?

• 17:33Speaker 03

  Scale of that operation.

• 17:34Speaker 00

  Stripe Press just, well, actually, we, in the letter, we sold our millionth book. But in fact, since.

• 17:47Speaker 04

  Incredible.

• 17:48Speaker 00

  No, look, books, we've actually now sold our 1.1 millionth book. So we'll go through the next gong at two, but.

• 17:57Speaker 04

  One billion soon.

• 17:58Speaker 00

  We we love books and they're very, they're very AGI proof.

• 18:01Speaker 03

  Oh, yeah. No, we've been a huge fan of so much of the Stripe Press catalog. I haven't read them all, but I'm collecting them one at a time and I'm working through them. And every time one drops, it's always a moment and we love them. So thank you for everything.

• 18:16Speaker 04

  Yeah, great to have you guys on and congratulations to the whole team on.

• 18:19Speaker 00

  Congratulations to you guys. TBPM is an amazing startup and it's super cool to see you guys grow and.

• 18:28Speaker 04

  Built on Stripe.

• 18:30Speaker 00

  Streaming world and our sector needs.

• 18:31Speaker 04

  Incorporated on Stripe, built on Stripe.

• 18:34Speaker 03

  Our first, our first ad deal ever was a live read at a live conference. I think we charged $50 and I put and I sent someone a Stripe link.

• 18:43Speaker 01

  We're talking about the 2025.

• 18:46Speaker 00

  Yeah. Well, we'll have to have you to our, to our, to our internal Stripe show, so we'll.

• 18:52Speaker 03

  Yeah, we'll talk to you soon. Have a great rest of your day. Congratulations.

• 18:54Speaker 00

  Thanks, guys.

• 18:55Speaker 03

  Goodbye.

• 18:58Erik Torenberg?

  Thanks for listening to this episode of the A16Z Podcast. If you liked this episode, be sure to like, comment, subscribe, leave us a rating or review, and share it with your friends and family. For more episodes, go to YouTube, Apple Podcasts, and Spotify. Follow us on X at A16Z and subscribe to our Substack at A16Z.substack.com. Thanks again for listening, and I'll see you in the next episode. This information is for educational purposes only and is not a recommendation to buy, hold, or sell any investment or financial product. This podcast has been produced by a third party and may include paid promotional advertisements, other company references, and individuals unaffiliated with A16Z.

• 19:39Erik Torenberg?

  Such advertisements, companies, and individuals are not endorsed by AH Capital Management LLC, A16Z, or any of its affiliates. Information is from sources deemed reliable on the date of publication, but A16Z does not guarantee its accuracy.