Apr 8, 2026 · 58m · a16z
Box CEO on the AI Adoption Gap | The a16z Show
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of The a16z Show, Box CEO Aaron Levie joins Steven Sinofsky and Martin Casado to analyze the enterprise AI adoption gap, the transition toward API-first software architectures for agents, corporate security risks, and the evolving economics of compute budgeting.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The host as informed peer 0.1 Guest teaching 3.1 Guest disagreement 3.4 The host pushing back 0.0
speaking balance: gold is the host, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 19:33 Levie rejects treating enterprise agents as regular human employees
Aaron Levie strongly opposes Martin Casado's premise that agents can be managed with standard HR and RBAC frameworks, explicitly asserting that liability, context window vulnerabilities, and prompt injection make agents fundamentally different from human workers.
Hardest push from the host ▶ 47:28 Host invites panel to keep speakingThe podcast host remained completely passive and uncritical throughout the episode, offering no direct pushback or premise refusal; her sole audible interjection was telling the panel she was happy to let them continue speaking.
Biggest teaching moment ▶ 42:42 Sinofsky reframes AI market size using PC and Cloud adoption cyclesSteven Sinofsky educates the panel on historical technology transitions, demonstrating how current Wall Street models miss the true market potential by analyzing AI spend through fixed zero-sum frameworks rather than elastic consumption shifts seen in early PC and Cloud eras.
The host holds their own ▶ 47:28 Host remains passive facilitatorThe host did not demonstrate personal technical expertise or challenge any panelist's assertions during the conversation, acting purely as a background facilitator while the three guests debated directly.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The host as informed peer | Guest teaching | Guest disagreement | The host pushing back | Why |
|---|---|---|---|---|---|---|
| Designing Software Interfaces for AI Agents | 0 | 2 | 3 | 0 | Aaron Levie contends that software must be fundamentally redesigned for agents via APIs and CLIs. Steven Sinofsky moderates this theoretical view by noting that algorithmic thinking and flowchart creation are notoriously difficult for average enterprise workers. | |
| AI Leverage and the Anthropic Growth Marketer | 0 | 3 | 4 | 0 | Levie cites a viral post about an Anthropic growth marketer automating work, suggesting this is the future of all jobs. Sinofsky directly counters that this is an extreme outlier, urging the panel to look at typical PC marketing roles instead. | |
| The Spreadsheet Analogy and Evolving Workflows | 0 | 4 | 1 | 0 | Sinofsky shares a personal story about his cousin using interns before learning spreadsheets to illustrate how technology shifts work abstraction layers upward. Levie concurs that domain expertise ultimately remains the anchor. | |
| Computer Use versus Autonomous Code Generation | 0 | 2 | 4 | 0 | Martin Casado explicitly takes the opposing side to Levie, arguing AI is moving toward computer use rather than autonomous code generation. Levie defends on-the-fly code generation as an essential capability for un-preplanned document workflows. | |
| Personal Productivity versus Enterprise Agent Oversight | 0 | 3 | 3 | 0 | Casado details his experience building personal agents, which Levie notes differs greatly from global enterprise supply chain complexity. Sinofsky recalls CFOs and CIOs reacting strongly against user-created ad-hoc system integrations. | |
| Managing Agent Identity, Permissions, and Governance | 0 | 2 | 3 | 0 | Levie describes the power and governance challenges of Box CLI when thousands of agents hit shared enterprise repositories. Casado argues agents should simply be given separate identities and RBAC permissions like human employees. | |
| Enterprise Liability, Prompt Injection, and Security Risks | 0 | 4 | 6 | 0 | Levie forcefully rejects Casado's proposal to treat agents like regular humans, pointing out that employers bear total liability for agents and cannot guarantee context window security against prompt injection. | |
| The Open Source Analogy and the Enterprise AI Adoption Gap | 0 | 4 | 3 | 0 | Sinofsky draws a historical parallel to enterprise open-source adoption, predicting that security concerns will temporarily lock down enterprise adoption while startups move ahead without such assets to protect. | |
| SaaS Business Models and the Data Access Dilemma | 0 | 3 | 2 | 0 | Sinofsky explains the data access dilemma facing legacy SaaS platforms like SAP and Workday, noting that vibe coding to complex ERP systems is absurd due to embedded business logic in UIs and middle tiers. | |
| Building Software for Agents and API-First Architecture | 0 | 3 | 5 | 0 | Casado challenges the popular concept of marketing to agents or building specialized agent interfaces, claiming agents care primarily about underlying backend semantics and cost parameters. | |
| System Layers versus Direct Prompting | 0 | 3 | 2 | 0 | Sinofsky warns that giving agents unbridled latitude risks creating macro-driven shadow IT systems of record inside corporations, repeating historical web-era security vulnerabilities. | |
| First-Principles AI Services and Knowledge Work Disruption | 0 | 3 | 3 | 0 | Casado contrasts direct prompting to machine code with the historical persistence of system architectural layers. Levie suggests first-principles services startups will disrupt incumbents by operating without traditional information boundaries. | |
| Expanding Internet Business Models and Historical Parallels | 0 | 5 | 4 | 0 | Levie envisions agent micropayments for data access, but Sinofsky strongly reframes the debate, arguing Wall Street and financial analysts are off by orders of magnitude on market sizing just as they were with PC MIPS and cloud compute. | |
| Infrastructure Usage Explosions and Pricing Models | 1 | 2 | 1 | 0 | Casado shares portfolio metrics showing infrastructure software consumption going asymptotic due to massive code generation. The host speaks briefly to invite the panel to continue standard conversation. | |
| Managing Engineering Compute Budgets & CFO Accountability | 0 | 3 | 5 | 0 | Levie claims managing engineering compute spend will be the most critical debate for CFOs due to its direct impact on EPS. Sinofsky humorously offers to sacrifice CFOs, dismissing short-term panic over historical technology budgeting cycles. | |
| Edge Computing vs. Cloud Repatriation Debate | 0 | 3 | 6 | 0 | Sinofsky raises local edge compute as a release valve for cloud costs, but Casado and Levie challenge his historical directionality, with Levie playfully mocking Sinofsky's reliance on dated hardware analogies. | |
| Short-Term Capacity Bottlenecks vs. Long-Term Tooling ROI | 0 | 4 | 2 | 0 | Sinofsky concludes that short-term token capacity bottlenecks will resolve over a 10-year horizon, drawing analogies to IBM mainframe MIPS pricing curves where supply growth naturally drove unit costs down. |
Statements from this episode (45)
Levie: Corporate compute budget discussions will explode over next two years
“The engineering compute budget conversation is going to be the most wild one in the next couple of years.”
Sinofsky: Market underestimates AI economic opportunity by an order of magnitude
“The biggest problem right now is everybody is trying to figure out the economics of all of this when they're off by at least an order of magnitude on how big the opportunity is.”
Casado: Adapting software by merely marketing APIs to agents is wrong
“People in the abstract say things like, now you're marketing to agents, you're like an API, you've got a good idea. I actually think that's almost exactly wrong.”
Levie: Box spends equal time on agent and human interfaces
“We spend as much time now thinking about the agent interface to our tool as we do the human interface.”
Levie: AI agents will outnumber human software users by 100x to 1,000x
“If you have a hundred or a thousand times more agents than people, then your software has to be built for agents, and then what is the way that those agents are going to interact with your system? It's going to be through an API or a CLI.”
Sinofsky: Most workers lack algorithmic thinking required to flowchart their work
“That, the way to say it is, algorithmic thinking. Is really, really, really hard for the vast majority of people who have jobs. And so the easiest way to think about it is if you were to go into any person and ask them to create a flow chart for a particular t…”
Sinofsky: Multi-agent networks will soon consolidate into broader domain agents
“Where we are with agents is just at this step where you think you need 50 and the abstraction layer is such that we're dividing up in these really small pieces with one super smart person coordinating them all and pretty soon that, that whole thing is just gon…”
Sinofsky: Autonomous AI action remains costly until non-determinism is solved
“Until the whole, like, non-reproducible, non-random element of this AI stuff goes away, the doing stuff is gonna get very costly.”
Sinofsky: Technical barriers to orchestrating AI agents will quickly evaporate
“Right now, you have to be an absolute, you have to be a rocket scientist and the growth marketing person to create 42 agents and spin them all up and do all of this stuff. But the rocket science part of it just Is going to evaporate in very short order.”
Casado: 2026 will be the year of AI computer use over coding
“And now this year is going to be the year of computer use. So it's almost like they're much more like humans using computers than them generating code”
Levie: Box is building an AI agent that writes code dynamically
“We have an agent that, that we're working on where you, it just makes a determination whether it should use an existing skill, it should be using an existing tool from Box, or it should write code to solve that problem”
Levie: AI agents face no cognitive limits in software usage
“An agent that is going to use tools and APIs and be able to code things doesn't have any of the same constraints that we have”
Sinofsky: Humans are the primary bottleneck to enterprise software adoption
“Humans have been a bottleneck In tapping the past 25 years of software capabilities.”
Casado: Software backends will converge into generic APIs for agents
“It'll probably converge into like some database, like some generic set of APIs, like that they'll connect to, and like that seems to be the direction it's going.”
Sinofsky: CIOs fear AI agent integrations will break enterprise systems
“Their fear is, like, unleashing, not just the agents themselves, but humans to do integration. Because you put, People creating new integrations. And you just say, please break my system of record.”
Levie: Enterprise AI agent integrations will remain read-only for years
“I think we have a read only version of this for a number of years before”
Levie: Autonomous agent coordination will be enterprise executives' next big problem
“This is going to be, like, the new big question that every CFO, CIO, et cetera, is running around trying to, with their hair on fire, trying to figure out.”
Casado: Enterprise AI agents should be treated as separate human identities
“You know, we've actually built in a lot of these permission systems. You have to treat it like a human. As a separate human, and then instead of like building another auth layer”
Levie: AI agents cannot be governed like human employees
“So you can't fully treat them like humans because here's the thing. And with regular humans, you don't get to look at the Slack channel of the person that, that is working with you or working for you. You don't get to log in as them. You don't get to oversee t…”
Levie: AI agent security risks are 1,000x greater than human employees
“The risk is like a thousand times greater. Like these people, like they will just leak your information whenever they want. Like they will happily just go and send some email to somebody because they got prompt ejected.”
Levie: Enterprise AI must assume context window data can be leaked
“And so then thus, if anything can ever enter that context window, because they have access to a resource, then in theory you should assume it can be, you know, prompt ejected out of the context window, and I don't know that we know of a way to solve that at th…”
Sinofsky: Enterprise AI security fears will let startups outpace big companies
“Part of what's going to happen is you're, we're going to go through this phase where like the enterprise customers are just going to like close everything off. Until there's some sense of sanity in all of this, and then, but in the meantime, the individual, an…”
Levie: Enterprise AI diffusion will take longer than Silicon Valley expects
“The diffusion of AI capability is going to take longer than people in Silicon Valley realize.”
Sinofsky: Vibe coding cannot recreate complex enterprise systems like SAP
“It's just absurd to think you're going to vibe code your way to like SAP.”
Sinofsky: Persistence of SAP will slow AI adoption on enterprise data
“SAP isn't going anywhere, so then that's going to slow the diffusion of AI on that particular data source, independent of whether or not it's Agentified AI that's doing stuff or just read-only reporting on stuff?”
Levie: Future SaaS success depends on building high-quality APIs for AI agents
“Everybody that built a SaaS business or a software business is like, the game is, can you build really, really high quality APIs? Can you have a way of monetizing that? You know, do you have a way of handling the identities and all of the access controls for a…”
Levie: AI agents will increase file volume, boosting Box's business model
“Every agent really loves working with files, so there'll probably be more files in the future than there was going to be before, and so, you know, can we build a platform that, like, makes it really easy for agents to work with that data? You know, we, we're b…”
Casado: AI agents select software based on core performance, not interfaces
“And at the end of the day, like, it's the semantics that end up mattering a lot more, right? And so, like, the agents, in my recollection, or in my experience, are very, very good at picking the right Back end for whatever they're doing. So they don't, they're…”
Sinofsky: AI agents will create fragmented, unofficial systems of record
“One of the risks with the model you're describing is that the agents themselves will spin up what becomes like a de facto new system of record.”
Sinofsky: J.P. Morgan will be slowest to adopt autonomous AI agents
“JP Morgan is gonna be the slowest, At doing this, and the startups are going to be the fastest, but the delta is huge, but even the startup one is a little far off, because even startups do need some systems of record at some point.”
Casado: Enterprise software layers will persist despite AI agent proliferation
“I don't think that we're, I think, like, systems are going to continue to be used in fairly similar ways. Maybe there's more agents using them, but I don't think they're going to evolve as much.”
Levie: AI-native service firms will disrupt incumbents before hitting normal corporate limits
“I do think that will be relatively disruptive You know, for some time until the bigger incumbents can kind of, you know, get out of the way on this. And that will at least create, you know, some precedent or case studies of what this new sort of corporation co…”
Levie: AI agents with spending budgets will unlock new internet business models
“You do give these agents, you know, a budget and a protocol to work with, and all of a sudden you're like, oh, Like, on the fly, they can go get medical research for some deep research tasks they're doing, and I'll pay, like, three dollars for that, and the ag…”
Casado: 50 a16z infrastructure portfolio companies recently saw asymptotic growth
“I you know, I've been in for investing for, 10 years now, I probably have a portfolio of 240 companies that work with. I have visibility that was saying that 50 of them, these are all infrastructure companies. Some historically have done well, some not so well…”
Sinofsky: Enterprise AI software will rely on bulk licenses, not micropayments
“Where they always think that, like, you'll be able to get, like, a fraction of a penny, but in the end, especially in the enterprise, like, people are just going to consume things, it's just cheaper and easier to buy, like, a bulk license for a bunch of stuff.”
Casado: High AI token costs are forcing software toward usage-based pricing
“And because tokens are such a significant part of COGS right now, it is pushing the industry to do usage base in the way that we have. Like, I remember when we went from like perpetual to recurring, and that required like a bunch of huge changes. Like, we're l…”
Levie: Enterprise AI compute is unique because users can spin up resources
“I don't think we've ever had a point where the end you, every end user in an organization has sort of a completely elastic ability to spin up a resource on their behalf.”
Sinofsky: Local compute engines will act as a release valve for AI costs
“Also don't, I keep thinking, do not discount the local compute engine as being a release valve for all of this.”
Sinofsky: Startups will burn capital pretending compute spend isn't a problem
“First, like the startups are going to burn through available capital pretending like it's not a problem. And they are going to do that.”
Sinofsky: Large companies will freeze AI spend, driving employee rogue purchases
“And a lot of big companies are going to be so terrified. They're just going to freeze and not do anything. And then people are going to actually start buying it on their own.”
Levie: Only 10% of traditional engineers ever managed cloud infrastructure spend
“Only, only like 10% of your engineering had to think about cloud infrastructure spend.”
Levie: Software teams should intentionally waste AI tokens to maximize experimentation
“Like, for me right now, I'm like, yeah, we should probably waste a lot of tokens, because that means that we're, like, trying new things.”
Sinofsky: AI token budgeting constraints will completely disappear
“The thing is, is this is all going to go away. There's absolutely no doubt that this just goes away.”
Sinofsky: AI compute will soon experience a breakthrough transistor moment
“And, like, we are gonna have a transistor moment with all of this. It might just be more supply the way we think of it, But it also might be an actual algorithmic fundamental change. It could be a change in the hardware. There's a lot of stuff that can happen.…”
Sinofsky: AI token prices will face steep deflation like IBM mainframes
“People were on MIPS and then one day the reality was IBM was selling more MIPS for fewer dollars every year. And didn't even realize it. And they were still pricing their mainframes by MIPS until it got pointed out to them that they were on a decreasing curve …”