--:-- Host (Martin Casado): What is the journey over the next decade? It's about the speed at which humans can change their workflows. Why doesn't the breakthrough that we just saw get released? Why doesn't that permeate every corporation within six months?
--:-- Guest (Aaron Levie): It's so strange to me how many disruptions are happening all at the same time. Your R&D is changing.
--:-- Host (Martin Casado): Yeah.
--:-- Guest (Aaron Levie): Every part of the stack is changing. Like everything.
--:-- Host (Martin Casado): We're not in a fear of AI world. 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.
--:-- What do you think is the best metric for anybody interested in tracking this stuff as far as how fast it's going? Is it GDP? Is it margin? Is it top line? Is it headcount growth? Is it all of the above?
--:-- [Intro music fades in and out]
--:-- Narrator: It's basically fully assumed that AI is going to take over the enterprise. How does AI actually change the enterprise? Not just in theory, but in how software is built, sold, and used? In today's episode, a16z General Partner Martin Casado sits down with Aaron Levie, co-founder and CEO of Box, to explore what it means to be an AI-first company, from product strategy to internal workflows. They talk about why incumbents may be better positioned than expected, how startups can still break out, the rise of agents and vibe coding, and what happens when the bottleneck isn't the tech, but the org chart. Aaron also shares how Box is using AI internally today and why he thinks the next generation of employees may spend more time managing agents than writing code. Let's get into it.
--:-- As a reminder, 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. Please note that a16z and its affiliates may also maintain investments in the companies discussed in this podcast. For more details, including a link to our investments, please see a16z.com/disclosures.
--:-- Host (Martin Casado): Aaron, thank you very much for joining us.
--:-- Guest (Aaron Levie): Thank you.
--:-- Host (Martin Casado): Everybody here already knows you. However, I still think you should intro yourself, just for completeness.
--:-- Guest (Aaron Levie): Okay. Aaron Levie, CEO, co-founder of Box. At Box, we help enterprises take all of their unstructured data or enterprise content and turn it into valuable information. AI is absolutely this incredible accelerant for that problem.
--:-- Host (Martin Casado): I just learned that we're investors, didn't you?
--:-- Guest (Aaron Levie): Well, many years ago, many years ago. So no claims post-IPO. But actually, Ben Horowitz had this early blog post on—I think it was the title, "The Fat Startup."
--:-- Host (Martin Casado): Yeah, in response to "The Lean Startup."
--:-- Guest (Aaron Levie): Yeah, and let's just say we very much took that to heart. We deployed every single lesson, which was: the name of the game is you get big fast, you scale aggressively. That was a very important period in our company's journey.
--:-- Host (Martin Casado): The notional topic of this is AI and the enterprise. But I think it's good to be nuanced about this because it's less obvious than people think. You've been talking a lot about AI on X, but also you're thinking about it in terms of your business.
--:-- Let me set up the first question as follows. AI has historically been this very B2B enterprise thing—chatbots, personalization systems. But what's unique about Gen AI is that a lot of the use cases are actually consumer or prosumer. Think creativity or developers. It actually hasn't made inroads into the enterprise yet. It's just starting now. So first off, does that match with your experience? And second, how are you thinking about this transition to the enterprise?
--:-- Guest (Aaron Levie): If you reverse engineer why that was the journey, up until the ChatGPT moment, AI was extremely hard to use. It required custom models for every problem you tried to solve. There was almost no way a consumer ecosystem could flourish based on that—it was just not generalizable enough. There were really few products other than maybe Siri or Alexa that you'd interact with that would have any sense of AI. So enterprises were the early adopters of AI systems to bring workflow automation to their companies.
--:-- Then, boom, ChatGPT happens. All of a sudden, it's the exact right form factor for mass adoption. There's no startup costs. It costs two seconds to learn the product. It's a chat interface. It was perfectly ripe for taking off in the consumer space. You also had incredible conditions for mass adoption: billions of people on the internet, it was set up as a free product, and it solved this latent question mark everybody had, which is, "When are we going to see AI work and touch our lives?" Everything was the perfect conditions for mass consumer adoption.
--:-- On the enterprise side, you have the opposite. You have lots of workflows ingrained for decades. You have legacy IT systems with data not set up well to be accessed by AI. You have a shadow IT problem—most corporations don't want end users injecting text into prompts that might contain information the AI models could learn from. So it's a difficult environment for that same level of virality.
--:-- With the exception of a few pro-sumer categories, I've talked to large corporation CIOs who are seeing people just show up with Windsurf, Cursor, and Replit. So you're getting this shadow IT version that we saw 15 years ago.
--:-- Host (Martin Casado): DevTools has always had that.
--:-- Guest (Aaron Levie): 100% fair. DevTools have had that. But I think you're still seeing that now with ChatGPT leakage into organizations. I'm sure their prosumer usage inside a corporation firewall is off the charts, even separate from the people who pay for it.
--:-- Host (Martin Casado): Totally.
--:-- Guest (Aaron Levie): Now, the question is: what is the journey over the next decade for the real change management of deployment of AI systems that drive the GDP-changing productivity gains? We have to be prepared for this to take many years. It's about the speed at which humans can change their workflows, as opposed to how quickly the technology can evolve.
--:-- In Silicon Valley, anybody tuning into this imagines, "Why doesn't the breakthrough that we just saw get released? Why doesn't it permeate every corporation within six months?" It's because people have meetings, they have budget processes, they have to go through a governance council, they have to get compliance on board, they have to figure out who has the liability when the thing recommends a stock and the financial services provider shares that with a client. That takes years. There's going to be case law that needs to happen. We still have lawsuits about who owns the IP of this stuff. That part is going to take years.
--:-- What's interesting—and I think you'll especially appreciate this on the cloud side—I remember when we were first scaling up in the enterprise, let's say 2007, 2008, 2009, that three-to-five-year period post-AWS. To a T, every conversation with a CIO was, "Yeah, that's nice. Maybe some little corner of our organization could use this. We are never going to go fully to the cloud." They had their arms wrapped around their servers.
--:-- Host (Martin Casado): I remember.
--:-- Guest (Aaron Levie): They did not want to give up the infrastructure. Too many questions, too many compliance issues. Existential job questions of, "What happens when this gets delivered as a service?"
--:-- Here's the super interesting part. We're now two and a half years into the ChatGPT moment. That same group of CIO conversations? None of that. It is basically fully assumed that AI is going to take over the enterprise. The CEO, the CIO, the CDO, every org leader is basically saying, "We know this is going to happen. This is not us trying to push it off. It is purely a sequence of events. Who do I deploy? How do I deploy it? How do I drive the change management? Is the model ready?"
--:-- The level of buy-in you have now in the enterprise is five times greater than we had in the early days of cloud. The classic litmus test: 15 years ago, Jamie Dimon was famous for saying, "We're never going to go to the cloud." JP Morgan would never go to the cloud. Today, David Solomon at Goldman Sachs has given anecdotes about how they can write an SEC filing or an S1 for an IPO in a few minutes that used to take a number of analysts a few days. The fact that those anecdotes are already coming out of the biggest banks means we're not in a fear of AI world. We know this is going to happen, and it needs to happen to us faster than it happens to our competitors. That is a totally different dynamic than we saw with cloud.
--:-- Host (Martin Casado): Do you think this has implications for companies today that are building products that are pre-AI products? With the cloud wave, you had a bunch of cloud-native companies that ended up taking over. Snowflake is a great example—the ones that decided not to go all in and were hybrid. Hybrid became known as "it won't work." Do you think because the buyer in the enterprise is more ready, companies that are pre-AI have more of an opportunity? Or do you think you're going to see the same thing with a lot of AI-native companies doing well?
--:-- Guest (Aaron Levie): I'm going to give you the non-answer: I think both.
--:-- One benefit the cloud cohort has—or the SaaS cohort, post all of us understanding and agreeing on what SaaS would look like—is that we all tried to build API-first platforms. We have the UI and the API. AI and AI agents are the perfect consumers of an API. They become super users within your system on your APIs.
--:-- If I want to deploy agents to automate my ServiceNow workflows, I think I'm better off deploying the ServiceNow agent to do that than doing an entire reinvention of my ITSM system. You can go down the list. Workday—if I want an AI agent to do some HR-related task, I'm better off doing that within Workday than building an entire new system.
--:-- Pre-cloud to post-cloud was an entire rewriting of your software. You had to go from single-tenant to multi-tenant. The scaling of the systems was totally different. The functionality and application logic was different because it should be real-time, collaborative, not async and batched like on-prem systems. In a cloud world, it is a reinvention of the user experience. We should definitely get to that.
--:-- Host (Martin Casado): I want to make sure I tease this out because it's a very interesting point. Your claim is that to go from pre-cloud to post-cloud, you ripped through the entire stack all the way down to the infrastructure—tenancy, you had to rewrite everything. What you're saying about AI is more of a consumption layer thing. You treat the existing systems as they are, and the AI becomes the consumption layer. Do you think this is a 1.5 step, and the 2.0 step kind of rips through the entire stack?
--:-- Guest (Aaron Levie): Let's bookmark that one for a second. If you do a pure Clay Christensen approach—sustaining innovation versus disruptive innovation—disruptive innovation looks so much harder, so different, so less profitable. Sustaining is, "Actually no, I'd like to build that because it's incremental, it's better for our business overall."
--:-- The on-prem guys had a disruptive innovation. Everything about the business model of SaaS looked different, harder, stranger. "I don't have the talent. I'm running a service delivery operation as opposed to shipping a CD-ROM with my code." Everything about the finances and pricing model.
--:-- Host (Martin Casado): Yes, everything. The business model, everything.
--:-- Guest (Aaron Levie): With AI—again, with the bookmark being the really big disruption you could contemplate—right now, everything kind of looks like a sustaining innovation if you're an incumbent. Instead of a user pressing buttons in the application, let's have an agent run through the API and operate as if they were that user. For a lot of SaaS providers, this looks like a TAM expansion. For the first time ever, I can deploy my software for use cases where the customer didn't have users on the other end to do those things. So you have a lot of TAM expansion.
--:-- The good news for startups comes with one caveat, which we've bookmarked and will get to. The caveat is you now have a component with a very different COGS model if you're a software provider. It's almost like when we went from on-prem to cloud, we went from perpetual to recurring. With AI, you kind of have to go from recurring to usage-based.
--:-- Host (Martin Casado): Yeah.
--:-- Guest (Aaron Levie): The business model will shift for some use cases. Even if you look at the Cursors, Replits, and Windsurfs of the world, there is a baseline seat price, and then consumption usage is an add-on. SaaS providers are well-structured to have that kind of dynamic. If it was 100% usage and the user seat goes away, then you have a business model crisis.
--:-- Host (Martin Casado): So you think right now it's not clear that's going to go all the way over?
--:-- Guest (Aaron Levie): Until the human literally is not a seat on the system, you don't remove the end user license as a component. But that could be the much bigger disruption.
--:-- To fully lay out the market dynamics, SaaS incumbents have a couple of other idiosyncrasies versus the on-prem days. On the margin, you tend to have founders still leading the SaaS companies. We didn't have that in the on-prem world. Siebel had three CEOs later. PeopleSoft had multiple CEOs later. It was a different leadership structure. A lot of times you still have the founder around, they're poking around, they're really into AI, so there can be a more natural pivot of the company from a leadership standpoint.
--:-- To the benefit of startups—which is why I can hold both of these in my head—I'm very bullish on the SaaS incumbent being the natural place for that AI agent relative to that category. But we have this incredible expansion of categories for the first time in probably 15 years. The SaaS 1.0 wave expanded the software universe with new categories nobody would have predicted in the pre-on-prem days. Nobody would have predicted Confluence and Snowflake. Lines of business didn't have 15 different applications to use. Post-SaaS, they did.
--:-- For startups in the AI world, the equivalent is that there are a lot of categories now where there is no software incumbent. AI agents let you go build software for that category. Legal, healthcare, education, and so on.
--:-- Host (Martin Casado): That's definitely true on the consumer side. The top use cases of OpenAI are almost like the top of the pyramid of needs—creativity, fulfillment. Number five is professional coding, but everything above that is one of these. On the consumer side, that's very clear. Is that clear on the enterprise side?
--:-- Guest (Aaron Levie): I absolutely think so. If we did a snapshot 10 years ago of the size of the contract management market or the legal document market, it's sub-$2 billion. I'm making up the numbers. Would you agree that in five years, the AI agent-related spend on legal services should be in the many, many billions to double-digit billions?
--:-- Host (Martin Casado): No question.
--:-- Guest (Aaron Levie): There are no natural incumbents that captured all that market. AI agents expand the size of the software-related spend in that space. I can underwrite that for healthcare, legal, consulting services. There are entire areas of financial services. We always think finance has been wired up for years. No. Consumer banking has been wired up. Trading has been wired up. Investment banking never went digital. Wealth management never went digital. These were not categories where you ever had major software platforms to help these entire categories of the economy. The reason was the work was unstructured—very ad hoc, very dynamic, lots of unstructured data as opposed to stuff that goes into databases. All of that is now ripe for AI. That will largely be ripe for many startups because there won't be a natural incumbent in those spaces.
--:-- Host (Martin Casado): It's so strange to me how many disruptions are happening all at the same time with AI. Everything you said—vertical SaaS or vertical use cases, a lot of which is actually human budget—that's being disrupted. There are a bunch of new use cases we never really thought about before, like creativity. Who would have thought 2D image generation would be some massive market? But it is.
--:-- I've been a programmer for 30 years. In that time, software would disrupt other things. We disrupted all of these things, but we never got disrupted. Clearly, now software is being disrupted for the first time I've ever seen in 30 years. Do you think this level of disruption is something existing companies will not—maybe a more fine point. You are a business leader right now. You have to think about product. You have to think about your organization. Does it require you to think about too much? How do you structure your company? How do you structure your product? Or do you think this is actually all pretty manageable?
--:-- Guest (Aaron Levie): I think your R&D is changing.
--:-- Host (Martin Casado): Yeah.
--:-- Guest (Aaron Levie): Every part of the stack is changing. Everything.
--:-- I'm putting myself in your shoes as a CEO. Your R&D is changing. Every part of the stack is changing. I think the reason I'm probably more distracted by what we're building—I don't have enough time to stress out about the actual organizational side because I'm stressed out enough about the pure delivery of the product. If I had a little bit more time, I'd get more stressed out about all the other change.
--:-- We are very much leaning into the idea of being AI-first. We have a twofer on this. One, by being as AI-first as possible, we'll see the use cases our product should go solve for customers. Check that box. Two, I'm a believer in the efficiency productivity gains. I do think it changes everything about work. There are lots of interesting examples of what it means.
--:-- In the future, does the individual contributor basically become a manager of agents? That's a totally different job. My recent go-to is thinking about it as: a lot of the productivity of your organization was rate-limited by how fast somebody can use a computer to do something—type an email, write code, generate a marketing asset. When that's no longer a limiter, how do these jobs begin to change? Your job becomes orchestration, integration of work, planning, task management, reviewing, auditing. That will radically change work.
--:-- Interestingly, this probably behooves us to not over-rotate on transforming internally for any given company simply because the technology is changing so fast. You wouldn't want to snap the line right now and run your whole business on this technology, because in two years it's going to be so much better. Progressively figuring out which workflows have high-impact upside, getting it rolled out in a decentralized way so people can experiment—I think you want to do a few of those things first.
--:-- Host (Martin Casado): I can't imagine a listener not knowing what Box does, but just for completeness, can you talk to us very quickly about what Box does and how you're thinking about how that dovetails with AI?
--:-- Guest (Aaron Levie): We started the company with a really simple premise: make it easy to access and share your files from anywhere. We pivoted about two years into the journey to focus on the enterprise market. The whole idea was enterprises are awash with unstructured data—corporate documents, research files, marketing assets, M&A documents, contracts, invoices. As companies move to the cloud and mobile, they need a way to access that information, collaborate securely on it, and integrate that data across different systems. We built a platform to help companies do that. We have about 120,000 customers, about 65% of the Fortune 500.
--:-- What's incredible right now is we've had an ongoing problem since the creation of the company. With structured data—the stuff that goes into your database—you can query it, synthesize it, calculate it, analyze it. Your unstructured data—the stuff we manage—you create it, share it, look at it, and then it gets forgotten about. It goes into some folder and you almost never see it again. Maybe you find it once every five years for some task. Most companies are sitting on most of their data being unstructured and getting the least amount of value from it relative to their structured data.
--:-- AI is the unlock. AI lets you finally ask this data questions. We can structure it—look at a contract, pull out the 10 most important fields. Once we have that data, we can analyze it, get insights from it, and start to do workflow automation that was never possible with unstructured data. 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. The computer previously was not able to know what's in the contract. For us, there's a huge unlock of what you can finally do with your information and your content. We're building an AI platform to handle the plumbing and user experience to make your content AI-ready effectively.
--:-- Host (Martin Casado): I don't want to be too bullshitty and provocative, but I have to ask this.
--:-- Guest (Aaron Levie): Please.
--:-- Host (Martin Casado): I've been in enterprise software for a very long time. A lot of the business model is predicated on the fact that building software is hard and takes a long time. To what extent do you worry about that not being the truth going forward? Do you think we enter this time of bespoke software being upon us?
--:-- Guest (Aaron Levie): I'm bearish on the extreme version of that. The extreme version—on one pole, all software is prepackaged. It's the Ford Model T. It works only one way. Everybody uses the same thing. That's not going to happen. The other extreme is everything is homebrew. You wake up in the morning, you utter something, you get your software for the day. You give your software for that thing, and the next day you do it again and change it.
--:-- The downsides of that model: if you ask the world population, 90% plus just don't care enough. They don't care about the tabs on their software and the modules on their dashboard. They want someone else to say, "This is what you should look at in the morning." They don't want to have to prompt the AI to tell them what to look at. Given that's guaranteed to be where 90% of the world is, 90% of our software should largely be: you log into the HR system and it just looks like an HR system.
--:-- There's another interesting dynamic. Over many years, our software and the way we operate companies have a flywheel relationship. The way we run our HR department is not so different from the way Workday wants us to run our HR department. That's fine because that's not the area where we're going to have a lot of upside innovating. The way we do ticket management from customer tickets is the way Zendesk decided to do ticket management. That's fine because that's not the core IP of the company.
--:-- Host (Martin Casado): In a way, it solves an operational problem for you.
--:-- Guest (Aaron Levie): Yes. You don't have to figure it out. People miss that about software. I don't want to have to think about the workflow of an HR payroll process. I just want the software to do that. That's what people are buying. Nobody wants to customize those things.
--:-- Given that we're going to be in this world of many different outcomes playing out, the reason I'm still bullish on Replit and vibe coding is for a different category. I'm the IT person, and I have this crazy queue of tasks. Someone says, "Can you build a website for this thing? Can you code up some inventory random plugin for this product?" That now becomes 10 times easier. The new prototyping, scripting, the long tail of stuff we want to do. That long tail is so long, and people never get to any of those things. I could underwrite a 10x growth of the amount of custom software that gets written, and the fact that these core systems don't go away because there's going to be way more software in the world that gets created.
--:-- Host (Martin Casado): Let me pressure test this. I can imagine why it would be hard to rebuild Box, because what you do is actually hard. This is core infrastructure. You store data. That's really important. I don't think you just vibe code that away. But from my perspective, a lot of SaaS apps just look like CRUD. I don't know what the acronym stands for, but it's basically reading and writing data from a backend. Do you think there is a world where the consumption layer evolves to just using AI and this class of companies goes away? Or do you think, if I heard what you just said, that the durability of these companies is that it basically teaches you what the workflow is?
--:-- Guest (Aaron Levie): I'm still going to say the latter. Now, I don't know if you need to bleep it out, but if you want to share a couple examples of who you put in the "not hard CRUD layer," then we can parse that. But up to you.
--:-- Host (Martin Casado): The "not hard CRUD layer"?
--:-- Host (Martin Casado): Yeah. I mean, I would say most vertical SaaS companies I see, the technology is trivial. Yeah, but the understanding of the domain is not.
--:-- Guest (Aaron Levie): No, no, this is what you said before. This is what I want to present. That's actually a great insight.
--:-- Host (Martin Casado): I've always underestimated vertical SaaS relative to the outcome. And 20 years into doing enterprise software, I'm just like, no longer going to underestimate vertical SaaS. It's not about the technology. It's the fact that somebody else has figured out the business model that works. Like they have 10 people from the pharma industry that are sitting next to the engineer being like, "This is how you should do the clinical trial workflow." And that becomes so much of the IP. Now that translates fine to agents, but I still would then bet on that vertical player doing that as opposed to somebody prompting their way into ChatGPT to build an FDA compliance agent. I would still largely bet on complianceagent.ai to do that over the pure horizontal system that has no particular domain expertise.
--:-- And then I think the other thing: I still think that there's a relationship between some amount of GUI and the agent and the APIs. Because you don't want to every day of your life go to a blank empty screen and say, "What's our revenue today?" You just want a dashboard at some point that just shows you the revenue.
--:-- Guest (Aaron Levie): That's right, of course.
--:-- Host (Martin Casado): It's almost like cached queries in a way. Somebody has made the decision. Yes, this is a known way to solve this problem in the enterprise. And so I think that's why the theory of the full abstraction away from the interface where it's all an API call — I don't think that happens. So ironically, probably what will happen is in a couple of years from now, we will see agents rebuild entire webpages and dashboards. And then we're going to find ourselves like, "Wait, why are we having an agent? Why do I have to spend tokens to create a thing that is a config on a dashboard?" And we'll just be back to where we started for some amount of software, which will mean that basically these things are going to live together.
--:-- Guest (Aaron Levie): Cool. Let's move from software to decision process.
--:-- Host (Martin Casado): So I won't say the name of the company, but I just spoke with a very, very legit company, household name. It's a private company, though. It's not a public company. We're at the board level. For every decision, they ask the AI for basically more information for the decision. And this has actually been great discussion fodder to be provocative. And it also shows how fundamentally unoriginal the board members are. This founder was telling me, "It's literally better than half of my board members." So how much have you thought about bringing AIs in to help with decision process? And by the way, I think the board is low-hanging fruit because boards tend to not have a lot of context to the business, so the incidents are probably less anyways. But is this something that you've thought about?
--:-- Guest (Aaron Levie): The board one is an interesting one, so maybe we can unravel that one. But I already use it for, let's say, our earnings calls where we'll do a draft of the initial earnings script. And then, because BoxAI deals with unstructured data, I just load up the earnings script and I'll use a better model and say, "Give me 10 points that analysts are going to ask about this. And how would I improve the script?" And it just spits out a bunch of things.
--:-- Host (Martin Casado): How good is it at predicting?
--:-- Guest (Aaron Levie): Oh, extremely good. 100%. But the thing is, that's not surprising. It has access to every public earnings call in history. And at the end of the day, analysts can only ask you about tailwinds, headwinds, who's buying what. It's not because analysts are smart or not smart, it's just those are the things you would try and deduce from an earnings call when buying a stock.
--:-- Host (Martin Casado): And you wouldn't have thought of these questions beforehand, or is it just...
--:-- Guest (Aaron Levie): On the margin. No, no, sorry. What I'm using it for is the specific parts of the document that are missing the answers to those questions, so I can actually inject the answers into that. Because you're typing out a thing and you forgot to give two case studies in this section or whatever. It's a quick way to just do some analysis on something.
--:-- So Bezos famously had this memo-oriented, essay-oriented meeting structure, and we never did that, but I was always fascinated by the companies that could do it. And actually, we're entering a world where probably you could just pull that off. Imagine if, whether it's a board meeting or product meeting, you just do a quick deep research essay on the topic. Every meeting, every strategy meeting in history would be better off if you probably had that as a starting asset to get everybody informed.
--:-- Host (Martin Casado): I think the argument against that would be, the reason Bezos said it is because it forced people to think clearly about what they're doing and writing it down. So the exercise meant that people walking in the meeting had more context. This would almost argue that they would have less context because something else did the thinking.
--:-- Guest (Aaron Levie): Well, two things. It was to make sure that the person doing the thing had the clarity to write it, for sure. But it was also still to inform everybody else that didn't do that work. And so it certainly would have helped everybody else in the room. And I'm not 100% sure. I mean, we should do a full longitudinal analysis of the people that wrote the essay. Did they actually have the better products? There's some Amazon products I don't like, and they obviously wrote an essay for those too. So I don't know the hit rate ultimately on the essay specifically as much as the idea of writing down a strategy and thinking it through. So why not have an agent do 90% of the heavy lifting?
--:-- A lot of my workflows now are like: if I have a topic where three years ago I might have lobbed it over to the chief of staff and said, "Hey, can you go research the pricing strategy of this ecosystem or something?" That's just a deep research query now. And then I'll wake up and look at this thing. But what that does is, because now I'm not having to calculate that person's time, their tasks, their trade-offs, I just do it for the most random things. That means I'm expanding and exploring way more spaces mentally than I would have before.
--:-- And these are the kind of parts where I'm actually more optimistic on the jobs front. Because what we do too many times within AI is we look at today's way of working and we just say, "AI will come in and take 30% of that." And it's like, no, no, no. We'll just do totally different things with AI. I wouldn't have researched that thing before when it required people to research it, because that would have been an inane task to send to somebody.
--:-- Host (Martin Casado): Yeah. One thing. So when we run the numbers — and by "run the numbers," I mean look through how AI companies are doing, where does the value accrue — there's basically one takeaway. And that is that these markets are very large and growing very fast. And value is kind of accruing at every layer, from chips up to apps. So the only real sin is zero-sum thinking. To be like, "Oh, the models are not going to be defensible" or whatever your zero-sum thinking is — that just hasn't proven out.
--:-- Now, this has still largely been a consumer phenomenon. So what I've been thinking about, and I don't have an answer, I'd love to hear your thought, is: when it comes to enterprise budgets, you can't just create budget out of thin air. You actually do have a limited resource. So as budgets get reallocated, to what extent do you think this is zero-sum — the old budget gets robbed — versus budget accretive? Because where we've come from, that has not been an issue. I think in the enterprise, it probably will be. So it does have to come from somewhere.
--:-- Guest (Aaron Levie): It's fully logical. A couple of things. A large number for startups can also be a very small number for a large corporation. So you have that dynamic playing out. I'll make up random stats, but you could probably take a meaningful engineering team, and for the price of five or ten of those engineers, you could probably pay for Cursor licenses for the entire engineering team. But this would argue that it's actually coming out of headcount. So here's where the asterisk is.
--:-- There's an infinite set of ways this actually plays out in next year's planning process. Maybe in a perfectly parallel universe, the salary increase that year would have been 3.5% for employees. And this coming year, it's 3% because we're going to take 0.5% and deploy AI for the company. Or maybe next year we would have added 50 engineers, but we're going to add 25 and then pay for AI. But guess what? The year after, we're going to have engineering productivity gains. So it increases because it's still a competitive environment. We then add engineers the year later because we're getting higher productivity gains.
--:-- I think that most companies of any reasonable scale, post-100 employees, have enough dynamism in the financial model within a one to two year period where it doesn't look like what the economists would think it looks like.
--:-- Host (Martin Casado): Can I just spit this back? I think this is actually a very good point that's buried in there. I just want to make sure I'm following along. The software license cost to a startup relative to a large people organization is relatively small. It's just a couple of headcount, which if you look at normal performance management, normal attrition, normal variability, and even hiring timelines, it's kind of in the noise. And you already have an annual budgeting cycle to fix that up. So basically, within the noise of even just headcount planning, all of this could work out without some massive disruption.
--:-- Guest (Aaron Levie): Totally. And there could be an upper limit to this point, but let's say the going rate in Silicon Valley of a new engineer coming out of college is somewhere between 125 and 200. Let's say your most aggressive Cursor usage is like a thousand bucks a year, two thousand bucks a year. So you're at like 1% of salary maybe. Here's the question. Do this crazy apples-to-apples thing. If you went and recruited from Stanford right now and you said, "Okay, Stanford grad, you have a choice. You can work at this company and get paid 125K with no AI, or you can get paid $123,000 with full access to AI." Which one are you going to do? They would do the $123,000 all day long.
--:-- Host (Martin Casado): But even that, yeah. Your argument, which makes a lot of sense to me, is it's kind of on the margin when it comes to $12,000.
--:-- Guest (Aaron Levie): Oh, I love that. That's great. And I did one kind of late-night modeling once, but I'm afraid to say all the numbers here because I think they're just going to be so wrong. But I think it's something on the order of five or six trillion in knowledge worker headcount spend in the U.S.
--:-- Host (Martin Casado): Everybody says for developers, let's say 40 million, let's just say it's 30 million. Let's say that the average is 100K, or you're at 3 trillion. These are just massive, massive numbers.
--:-- Guest (Aaron Levie): So it's many trillions. If you take a couple percent of that or 5% of that, you're already doubling the entire U.S. enterprise software spend. So you can just make it work within that. And this is why I don't think that people will make cuts because they have to pay for AI. They might make cuts for other reasons. But even in those cases, I think you'll often have it be for myopic reasons temporarily. And there's enough flexibility to basically consume this and then actually recoup on the productivity game.
--:-- Host (Martin Casado): I think that's great. I try and parse everything you say through the lens of, where are you landing on AI coding? And you seem to have a very pragmatic view of where things actually are at. Where are you landing right now?
--:-- Guest (Aaron Levie): Well, it's been an evolution. In the entire AI thing, the biggest surprise to me is how effective it is at code. So I'm just going to say a couple of facts, and then we can kind of back out what this means in aggregate. One fact is, I do think that AI helps better developers more than not-better developers. The reason is you have to be able to know what to ask for and know how to deal with the outcome.
--:-- Someone said it beautifully on X — I forgot who it was, but I thought it encapsulated it perfectly. He said, "90% of what I know, the value of it has gone to zero, but 10% has tripled, or more than 10x, or 100x." I think that's exactly right. I do think that for a lot of rote use cases, the AI can do it and it doesn't need to be double-checked. So things like prototyping, things like scripting. If you look at usage of OpenAI, if you actually look at code usage, the primary use is actually professional developers, which means it's part of a developer workflow.
--:-- And then probably the most controversial stance I have is — and this is probably sunk cost fallacy because I've been a programmer for, my PhD is in computer science — but I just don't see a world where you get rid of formal programming languages just because they arose out of natural languages for a reason. We started with English and then we made programming languages so that we could formally describe stuff. It would be a regression to go back. So I still think we'll use languages, and maybe they'll change, maybe more like a scripting language. But I think the existing tool set will evolve, but it'll still be a professional developer. We'll still have developers, still have developer tools. So that's kind of where I am. I would love to hear where you're at.
--:-- Guest (Aaron Levie): I'm fully on the exact same page. The fun thing to me is how coding is just at the tip of the iceberg. It's the best thing to first experience agentic automation, but I think you'll see this in basically every other space. What's so fun is just in a one-year shift in the nature of the relationship with the AI.
--:-- If you think about the GitHub Copilot moment, it was like, "Oh, this thing is incredible. It's going to type ahead and predict what I'm typing." And you're basically using it to work 20 or 30% faster. And now the relationship is totally different within a year or two period where you're using Cursor, Windsurf, or whatever, and the agent is generating this chunk of output and you're just reviewing it. But what's incredible is that none of your expertise is any less valuable in that review. In fact, it's probably even more important than ever before because in some cases it's just going to be wrong 3% of the time and then you review it, but you're literally doing 3x the amount of output.
--:-- And the nature of how that changes both programming, and why not have that for basically everything, is this new way that both software should work and actually we will work. The big joke a year after ChatGPT was, "Okay, this thing generates a legal case and it's wrong 10% of the time." And it's like, well, actually, hold on. If you think about what the new paradigm of work looks like, it's such a weird inversion. It used to be that the AI was fixing your errors. That's what we thought the AI was going to be. And it's just a total flip. The human's job is to fix the AI errors. And that's the new way that we are going to work.
--:-- Host (Martin Casado): Right. So this begs a very obvious question. I'm going to work up to it. There's a great paper in NSDI from an MIT team which basically says you can optimize a running system with agents. The way they did it is they have a teacher agent and then more junior agents. The junior agents would go try a bunch of stuff. And of course, they had much more knowledge of the literature than any single human being. So they try all different things. And then the senior agent would say, "This is good. This isn't good." And once it optimized the system, they would do it. And the human being is then helping the teacher agent decide what the parameters are, what is good, what is not good, and provide high-level direction.
--:-- So you're already starting to see cases where human beings are running multiple agents, and even that already is starting to have some kind of bifurcation. One way to think about it is in any R&D organization, people start as ICs, but then they very quickly get interns and go into management. So maybe we're just skipping that step. The obvious question is, what happens to entry-level engineers? Does this change how people get introduced to computer science?
--:-- Guest (Aaron Levie): The cool thing is probably more people will even now get introduced to computer science because you'll be able to—
--:-- Host (Martin Casado): Anybody can learn it.
--:-- Guest (Aaron Levie): Anybody can learn it. It's been 25 years for me, but in the early days of programming basic applications or putting up other websites, it was just extremely frustrating that you would spend days and days being like, "Why does that thing not work?" And I had very few resources for figuring out why the thing didn't work. It would have been a hundred times easier if I could have had an agent write the thing. I would have learned 10 times faster.
--:-- Host (Martin Casado): Honestly, what you did was — well, not 25 years ago, but 10 years ago — you go to Stack Overflow. And so it was like the slow version.
--:-- Guest (Aaron Levie): Yeah, but think about how many people missed the window pre-Stack Overflow that got pushed out of the ecosystem because they were just like, "This is too frustrating." So you're going to have a way bigger funnel at the top of people now learning programming and computer science. I think a similar percentage of people will fall out. So it's not like you're going to get a 10x increase in programmers because you still have to enjoy it and you have to like solving problems and whatnot.
--:-- It's going to change the nature of the incoming class of engineers that you hire. They will literally not be able to code without AI assisting them. And it's not 100% obvious that's a bad thing because assuming you have internet and the site stays up, we should have access to the agents. So I think it's mostly just that we have to adapt how we think about the role of an engineer and what these tools are giving us in terms of productivity gains.
--:-- Actually, I meet with a lot of larger, not tech-oriented companies as customers. And generally, the thing I'm recommending is hire a bunch of these people because they're going to flip your company on its head in terms of how much faster the organization can run. So I do understand — I want to be sympathetic to the job market for anybody coming out of college because I don't think it's easy right now. And it probably hasn't been easy in a number of years. But if you are graduating, the thing I would be selling any corporation is that if you are AI native right now, coming out of college, the amount you can teach a company is unbelievable. And conversely, if you're a company, you should be prioritizing this talent that is just like, "Why does it take you guys two weeks to research a market to enter? I can do that in deep research and get an answer to you in 30 minutes." They will be able to show companies way faster ways of working.
--:-- Host (Martin Casado): Do you think there's any stumbling into problems this way? Like you kind of adopt too quickly. You get into a morass you can't get out of. Or do you think at this point it's pretty clear the stuff can be practically consumed?
--:-- Guest (Aaron Levie): What would the morass be that you'd get into?
--:-- Host (Martin Casado): You hire a bunch of vibe coders and then they create something that nobody can maintain and it's really slow.
--:-- Guest (Aaron Levie): Totally. Which, by the way, I will say I have seen this.
--:-- Host (Martin Casado): Yeah, yeah, yeah. Okay, you could easily overdo this whole thing.
--:-- Guest (Aaron Levie): So I think as with anything, deploying these strategies in moderation while we're all collectively still getting the technology to work better and better is super important, and understanding the consequences of these systems. So yes, this is not a moment to just have your whole company vibe code.
--:-- I will say one of my favorite things that I'm witnessing in the whole coding thing — I don't know, the point of this talk is kind of the AI and the enterprise generally, but the coding thing is just so salient — is that a lot of these OG programmers that I've known for a long time that are off creating companies or CEOs of public companies like yourself are all back to programming. You talk to them, many of them, I code most nights with Cursor just because it's really enjoyable. And the reason I didn't code before is because I just couldn't keep up with the frameworks. I'm like, "Dude, I don't know how to install the thing. And what is this Python environment stuff?" You're literally learning bad design choices that somebody else just made up. They're not fundamental to the laws of the universe. They don't make you any smarter. It's just waste of brain space. And so in some way, the AI just gets rid of this kind of crufty stuff that you probably shouldn't be wasting brain space on anyway.
--:-- The amount of frustration I have when I look through our product roadmap pre-AI — although this still obviously happens because we haven't fully transformed everything about how we work — but pre-AI, when you would see things like "We have to upgrade the Python library in this particular product" and it's three engineers, two quarters. And at the end of that project, zero customers will notice that we did something. We resolved some fringe vulnerability that is not even going to happen, but you have to do it because there's some compliance thing where you have to make sure you're on the latest version. And all of a sudden, you are wasting hundreds of thousands of dollars of engineering time. And the fact that that's now a Cursor task is just unbelievable. The amount of things you can actually relieve your team to go and work on is incredible.
--:-- And the other big boon for the economy — and this is where the economists just totally miss this stuff — is think about every small business on the planet, of which there's millions, tens of millions, whatever, that for the first time ever in history, they have access to resources that are somewhat approximate to the resources of a large company. They can do any marketing campaign. Did you see the NBA finals video from Kalshi?
--:-- Host (Martin Casado): The VO3 video?
--:-- Guest (Aaron Levie): Yeah, yeah, yeah. You can now put together an otherwise million-dollar marketing video for a couple hundred bucks of tokens. And that being applied to every domain in every service area — I can run a campaign that translates into every language. I can have this long tail of bugs that I never got around to automatically get solved. I can have the analysis of a top-tier consulting firm done for my particular business. So for the people or companies that are resourceful and are creative and imaginative, the access to resources right now is just truly unprecedented.
--:-- Host (Martin Casado): What do you think is the best metric for anybody interested in tracking this stuff as far as how fast it's going? Is it GDP? Is it margin? Is it top line? Is it headcount growth? Is it all of the above? How do you measure it?
--:-- Guest (Aaron Levie): For us, internally first, and then maybe we'll spitball some macro solutions. Internally, we've explicitly taken the stance that we want to use AI to increase the capacity and capability of the company. Just do more. For anything that you track, just make sure it happens fast. Just do more or do faster.
--:-- Host (Martin Casado): In a given time period, yeah.
--:-- Guest (Aaron Levie): And that somewhat relieves the pressure from people that think this is about cost cutting. It's just like, no, no, just do more right now. Let's figure out what works. Something won't work. We want experimentation. So just use AI to do more. The way you should measure that then in a couple of years from now is either the growth rate of the company should be faster 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, which is also a very viable outcome — this is just the new standard of running a business.
--:-- Host (Martin Casado): But there's no shift in the equilibrium.
--:-- Guest (Aaron Levie): Right, there's no shift in the equilibrium. You just have to do it. And then the ultimate product of all that is some other kind of metric of satisfaction — our products get better. It could be consumer price index or something.
--:-- Host (Martin Casado): But did the iPhone show up in GDP? I don't know, but my life is better with the iPhone than without the iPhone.
--:-- Guest (Aaron Levie): I'm pretty sure it did.
--:-- Host (Martin Casado): Okay, yeah, fine. But it would ultimately show up in new cures to diseases, better healthcare. I don't know that the dollars would move around all that differently as much as just life expectancy should go up, cost of housing should go down. Weird metrics that productivity gains will then drive that the economists wouldn't naturally associate to enterprise software and AI.
--:-- Guest (Aaron Levie): By the way, this is where I am. Clearly there's a disruption because marginal costs are going down in a bunch of stuff like writing code and language reasoning. And some companies will take advantage of that, but I don't think the fundamental equilibrium changes. I think to your point, we just do more tech, products get better faster. We solve problems that we haven't before, but it's not asymmetric in the way we've seen in other shifts.
--:-- The way I kind of think about it is, if we go back to 1985 and we just looked at how everybody works, I think we would just be totally stunned by how slow everything is. How long did it take you to research a thing or analyze a market or create a campaign? But it has now been baked into our human productivity that we just do all those things really fast now. 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? You spent two weeks to decide the message for the marketing campaign? How is that possible?" What we do now is we run 50 experiments with AI agents, they all come back with versions, we look at them all together, and then we make a decision in an hour and we move on. That's obviously how work works. And that's what we will be saying 10 years from now.
--:-- Host (Martin Casado): Do you think we'll ever saturate the consumer? 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.
--:-- Guest (Aaron Levie): I think I'll say yes, just because at some point maybe, but my list of purely consumer demands has not gone down. Healthcare is a totally unmet need that I have. I do not like to go to doctors or dentists or anybody because of just how hard it is to get scheduled. I mean, buying a car, man. There's so many things that just need to be solved. The cost of housing. We clearly don't have enough houses. Where will AI drive that? Maybe robotics would be the play there. But I don't think we're anywhere close to consumer satisfaction or satisfying all needs of consumers.
--:-- Host (Martin Casado): Well, actually, I meant more, will things change so fast that it saturates the ability to adopt new things?
--:-- Host (Martin Casado): I do think that that is certainly possible. I think I track my parents as a decent proxy, or even college friends that aren't particularly in tech. And they're still in their ChatGPT phase of adoption. They haven't moved on from that. They haven't made a VO video yet. They're just using ChatGPT to ask questions about life experiences they have.
--:-- Guest (Aaron Levie): So ironically, one of the problems was that ChatGPT was too good. If you imagine what people thought AI should be able to do for them, it already met 80% of where they would have projected it.
--:-- Host (Martin Casado): But we know it can probably still do 10 to 20x more.
--:-- Guest (Aaron Levie): Right, but their needs are going to be satisfied for some time on those core use cases. So for the most basic consumer query type things, that's fine. But this is the opportunity for startups: AI will show up in ways where the person isn't even in the market for an "AI thing." They just want a better version of that category. This becomes simply another market constraint. As soon as it saturates, you just make the product better given the existing constraints.
--:-- Host (Martin Casado): Yeah.
--:-- Guest (Aaron Levie): If I could just get better healthcare, but I don't need to think about that as an AI problem or not—AI will be behind the scenes delivering that—then I don't think you're saturated anytime soon.
--:-- Host (Martin Casado): Yeah. It's just the consumption capacity becomes another market constraint, but there are a ton of other ways to improve things.
--:-- Guest (Aaron Levie): 100%.
--:-- Host (Martin Casado): That's great. I love that you're so optimistic about it.
--:-- Guest (Aaron Levie): I am, I think, 98th percentile optimistic.
--:-- Host (Martin Casado): Good. So I think we've had a fairly pragmatic conversation about the current impacts and the near-term impacts. If you take a longer view, can you dare to guess what things look like in five to 10 years? Sam Altman and Jack Altman had a podcast recently, and I'm going to paraphrase probably incorrectly, but they were going back and forth about how we just got what we would have predicted as AGI five years ago. And it's just like, we use it. It's now just built in.
--:-- Guest (Aaron Levie): The most anticlimactic. Anti-anticlimactic.
--:-- Host (Martin Casado): And I think that's my instinct for a lot of this. Five years, 10 years, whatever your number is. And this is why I'm so optimistic on society and jobs and all this stuff. I don't think it's the Terminator crazy outcome where we automate away everything. 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, for trying to do scientific discovery—all of this stuff is just built into us. And AI is this kind of up-leveling of the tools we use to do all those things.
--:-- Guest (Aaron Levie): I think the way we work will be totally different in five years or ten years. But you're already seeing enough of what it will look like that I think it's an extrapolation of that. When you want the marketing campaign done, you have a set of agents that go in, create the assets, choose the markets, and figure out the ad plan. Then you have a few people review it and debate it and say, "Okay, let's go in this direction instead." Then you deploy it and you're onto the next thing. Each company's units of output grow. As a result of that growth, we're all still in competitive spaces. So some of it gets competed out, and others will keep growing faster than they would have before. They'll hire more people, and you'll have new types of jobs. We'll have jobs for people just to manage agents. You'll have operations teams. Adam D'Angelo had this cool role that just got announced—the role is to work with Adam at Quora and figure out which workflows can be automated with AI. I think you'll have a lot of those kinds of functions.
--:-- One of the exciting things about being in Silicon Valley, or for anybody tuning in who's in this ecosystem, is that we're seeing the change happen faster here. And it's going to be five or 10 years of this rolling out to the rest of the economy. I think we will spend the next five years making the technology actually deliver on the things we're all collectively talking about, making it more and more robust. The accuracy goes up, the costs go down, and the workflows it can tie into are better. We will be working on that for quite some time.
--:-- Host (Martin Casado): And you think ultimately this leads to the biggest piece dividend of better products for users, better user experience?
--:-- Guest (Aaron Levie): Yeah, I think the software gets better. Our healthcare gets better. The life sciences discoveries increase. I think it's all a society net positive.
--:-- Host (Martin Casado): I love it.
--:-- Narrator: Thanks for listening to the a16z podcast. If you enjoyed the episode, let us know by leaving a review at ratethispodcast.com slash a16z. We've got more great conversations coming your way. See you next time.
