Apr 17, 2025 · 1h 15m · mad
Box’s Big AI Leap: Aaron Levie on Agents & the Future of Work
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 MAD Podcast, Box CEO Aaron Levie joins host Matt Turck to discuss enterprise AI strategy, agent architectures, public market dynamics, and the organizational culture required to lead a major tech transformation.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 12.4% of the talking time here. How this is scored →
Matt as informed peer 4.0 Guest teaching 4.7 Guest disagreement 1.8 Matt pushing back 2.2
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 50:28 Rejection of Proprietary Enterprise Data Advantage
Aaron directly rejects Matt's premise that unique enterprise data justifies model training, arguing that corporate documents across sectors simply mirror horizontal LLM training tokens.
Hardest push from Matt ▶ 46:28 Challenging Stance on Fine-TuningMatt pushes back on Aaron's refusal to build or fine-tune models by citing direct competitors Snowflake and Databricks releasing Arctic and DBRX.
Biggest teaching moment ▶ 50:28 Reframing Enterprise Data LLM DynamicsAaron breaks down why enterprise content like scripts, clinical trials, and financial plans does not constitute a unified dataset requiring a proprietary foundation model.
Matt holds his own ▶ 46:28 Citing Competitor Model BenchmarksMatt demonstrates sharp market awareness by pressing Aaron on specific custom model initiatives launched by rival data platforms.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| Welcome and Reflections on Current Macroeconomic Uncertainty | 2 | 3 | 2 | 1 | Matt sets up the macro topic and offers light banter regarding tariffs. Aaron speaks at length on macroeconomic supply chain risks and political negotiation dynamics without tension. | |
| Being a Public Company CEO and Private Market Dynamics | 4 | 4 | 1 | 2 | Matt raises key market dynamics such as adverse selection in public offerings and SPAC risks. Aaron explains why late-stage private capital changed IPO incentives while defending public market governance. | |
| The Founding of Box and Early Financing Story | 3 | 2 | 1 | 1 | Matt prompts Aaron to recount the origin story of Box and draws a connection to Mark Cuban's investment pattern with Synthesia. Aaron shares a biographical narrative of early rejections and initial funding. | |
| Comparing the Cloud Shift to the AI Wave | 3 | 5 | 1 | 1 | Matt asks Aaron to contrast the cloud shift with the current AI wave. Aaron educates on the structural difference between ten years of CIO enterprise resistance in cloud versus immediate pull in AI. | |
| Enterprise AI Adoption and the Crossing the Chasm Framework | 3 | 6 | 2 | 2 | Matt brings up the PoC versus enterprise deployment debate. Aaron reframes enterprise AI adoption by applying Moore's Crossing the Chasm framework subcategory by subcategory. | |
| Inside Box's AI Platform Architecture, Hubs, and Ecosystem | 4 | 5 | 1 | 2 | Matt inquires about Box's AI platform architecture and whether expanding into document workflows makes them a vertical SaaS player. Aaron details the technical stack behind Hubs and explains Bill Joy's ecosystem philosophy. | |
| Multi-Model Ecosystem and Model Selection Strategy | 6 | 6 | 3 | 5 | Matt pushes back on Aaron's model-agnostic stance by pointing to Snowflake's Arctic and Databricks' DBRX models. Aaron strongly rejects the idea of training or fine-tuning custom base models for non-hyperscalers. | |
| The Reality of Enterprise Data for Model Training | 5 | 7 | 4 | 5 | Matt challenges Aaron by suggesting Box's access to proprietary enterprise data could justify fine-tuning. Aaron refutes the premise, explaining that diverse enterprise document tokens mirror the horizontal datasets used by frontier LLMs. | |
| Challenges, Search Quality, and Risks of AI Agents | 5 | 5 | 1 | 3 | Matt highlights technical bottlenecks like compounding errors in agent chains. Aaron outlines practical limitations in search quality, enterprise permissions, and document review accuracy limits. | |
| Inter-Agent Communication Protocols and the Role of MCP | 5 | 5 | 2 | 2 | Matt brings up Anthropic's Model Context Protocol (MCP) and inter-agent standards. Aaron draws parallels to the adoption curve of REST APIs while warning against using GPU-intensive MCP handoffs for tasks simple APIs solve. | |
| Driving Internal AI Transformation and Cultural Alignment at Box | 4 | 4 | 1 | 2 | Matt asks if driving AI transformation required 'founder mode' or faced internal resistance. Aaron describes the internal change management required to align executive leadership and engineering teams. | |
| Fostering an AI Culture and Developer Productivity Tools | 4 | 4 | 1 | 1 | Matt cites Tobi Lütke's memo on mandatory AI adoption at Shopify. Aaron discusses Box's internal AI tools usage and highlights why delegating maintenance drudgery to AI increases developer productivity. | |
| Monetization Strategy and the Future of Enterprise Software | 4 | 5 | 3 | 2 | Matt asks about monetization expectations and customer traction. Aaron debunks Wall Street's expectation of an 'AI pricing premium,' clarifying that revenue growth comes from expanding addressable use cases. |
Statements from this episode (25)
Levie: Proposed US tariffs will cause a total economic disaster
“If as of, you know, Tuesday afternoon, you know, the tariffs that have been proposed, if those roll through you know, this is going to be a total disaster.”
Levie: US trade friction will allow international entrepreneurs to outcompete US businesses
“If you're an international entrepreneur right now and you're watching this, you're like, I can now finally outcompete the U S because they've just going to create friction for their own businesses.”
Levie: AI will expand Box's TAM without disrupting seat pricing
“Like, we think the TAM of our category goes up because of AI and the business model doesn't get like severely disrupted in the sense of we used to be selling seats and now all of a sudden something totally different, it's seats and then something in addition t…”
Levie: Regulatory scrutiny on public companies is net positive for shareholders
“I don't think we made it too hard to go public. I think actually it's good to have a, of an, a heightened degree of scrutiny and regulatory pressure on being a public company. It is, is, absolutely net positive for the average shareholder that we have all of t…”
Levie: Late-stage capital allows tech companies to stay private indefinitely
“Where we're at is we now have this new innovation, which is late stage capital that can basically keep you private for as far as we can tell, maybe forever, because you can kind of outrun the problem of converting these companies into burn companies to cashflo…”
Box raised its first $80k round at a $240k or $320k valuation
“We rated it 80,000 dollars for a post money of I believe this is the post money. It was either two 40 or three 20, but the post money was either 240 K Or 320 K. We raised 80, 80,000.”
Aaron Levie raised hundreds of thousands from Mark Cuban via cold email
“I had emailed Mark Cuban. He got back to us in, like, 10 minutes and said, interesting company. We started working on, like, a partnership, weirdly in in one of his in one of his companies, and then that sort of parlayed into an investment, so we ended up rais…”
Mark Cuban had Travis Kalanick conduct due diligence on Box in 2005
“As a part of the due diligence for the company he had us meet Travis Kalanick. So he invested in Travis at this company called Red Swoosh. And so so I had met Travis in LA and Travis was basically like Mark asked me to meet you know, see if everything's above …”
Levie: Box launched a year before AWS, forcing self-built infrastructure
“We actually had to build out infrastructure for the first decade and a half because merely we started the company six months too early. So we started the company and our first launch was on these co-located servers. And Amazon web services didn't launch until …”
Levie: Enterprise AI faces no philosophical resistance unlike early cloud
“The difference is the energy is very different where I think most companies I meet with and happen to be in New York and meeting lots of banks. The energy is like absolutely in the category of how many use cases can I apply this to? How do I start to lean in m…”
Levie: The AI wave would not have happened without ChatGPT
“Without chat GPT, none of this would have happened.”
Levie: AI coding tools have reached mainstream enterprise adoption
“AI coding is very clearly in the pragmatists. You know, get up co-pilot now, I don't know, three years old cursor's flying off your shelves, wind service flying off the shelf, you know, replit, all these guys are kind of being, you know, used everywhere. So yo…”
Levie: Box officially supports Gemini, Anthropic, and OpenAI models
“We officially support Gemini family. We support anthropic, you know, variety of cloud models and then support open AI and the GPTs.”
Levie: Box supports open-source models for select unannounced customers
“Yes for a couple of particular customers that we just haven't announced yet, but, you know, coming soon.”
Levie: Box considered training a foundation model for '10 minutes'
“We considered it for 10 minutes, and we said no effing way are we gonna get in this war?”
Levie: Abandon custom AI scaffolding when foundational models advance natively
“If anybody has a breakthrough in the model space that we have built scaffolding or plumbing around to mitigate the lack of that breakthrough previously, we should just like kill our stuff and just adopt whatever that breakthrough is.”
Levie: AI agents' biggest impact is unlocking previously unfeasible software use cases
“The really big impact is that is the expansion of now what people can do with software and start to solve the use cases that we just never Ended up prioritizing in most of these categories.”
Levie: Solving enterprise AI search and governance will take years
“So, so this is the part that will take years to fully figure out and get right.”
Levie: Developers should not vibe code into production yet
“You probably shouldn't vibe code into production yet.”
Levie: Single-shot AI extraction drops to 75% accuracy on 200-page documents
“If you ask it to pull out data from a short document, it's going to basically be a hundred percent accurate on that. If you ask it to pull out data from a 200 page document, And you only give it one shot to do it. It's going to get 75% accuracy.”
Levie: Agent interoperability will follow the REST API adoption curve
“I think it'll, it'll follow probably the same curve that, like, REST APIs did 20, 25 years ago, 23 years ago or whatever.”
Levie: All enterprise software will interoperate via AI agents within two years
“I think we could easily be two years from now where you just feel totally comfortable that like all the software you use can talk to all the other software in an agentic way.”
Levie: Box’s AI team is the largest group in the company
“Cause now our AI team is probably the biggest team. Yeah. Easily the biggest team in the whole company in terms of kind of any one group.”
Levie: Box is enabling Cursor for its VS Code developers
“We're enabling cursor for the VS Code crowd to be able to use.”
Levie: Software companies will not get an AI price premium
“You will grow faster because you're entering new markets, but you won't grow faster because there's some kind of like magical AI premium on your software. And those are two things to kind of decouple, which is like the price that you can charge because you hav…”