May 28, 2026 · 1h 13m · mad
State of Enterprise AI 2026: Aaron Levie on Tokenmaxxing, Rise of Headless, and AI-Proofing Your Job
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, host Matt Turck interviews Box Co-Founder and CEO Aaron Levie to explore the realities of enterprise AI adoption in 2026, addressing soaring token costs, architectural paradoxes, data governance, and strategies for future-proofing knowledge worker careers.
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 10.5% of the talking time here. How this is scored →
Matt as informed peer 4.5 Guest teaching 5.8 Guest disagreement 1.7 Matt pushing back 3.1
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 6:19 Rejection of host's cynical CIO framing
Levie directly counters Matt's premise that enterprises are fatigued and cynical, asserting that his direct conversations with over 200 CIOs show statistical optimism due to real coding gains.
Hardest push from Matt ▶ 30:31 Host insists on a 10-year enterprise adoption horizonMatt refuses to accept fast rollout assumptions, drawing a direct parallel to the decade-long cloud migration to argue enterprise AI will take at least 10 years.
Biggest teaching moment ▶ 24:30 Systematic breakdown of AI coding vs knowledge workLevie educates the host on why AI coding success cannot easily be replicated across general enterprise workflows, citing verifiable code tests vs complex permission and context barriers.
Matt holds his own ▶ 37:12 Identifying the semantic layer rebrandMatt demonstrates high domain expertise by calling out Levie's description of 'new' agentic data problems as merely the classic 20-year-old semantic layer and ontology challenge.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| The Gap Between Silicon Valley and Global 2000 Enterprise AI | 4 | 5 | 2 | 2 | Matt sets up the segment by asking Levie to compare Bay Area AI adoption with traditional Global 2000 enterprises like GE and Procter & Gamble. Levie reframes the contrast as engineering vs non-engineering knowledge work rather than pure geography, outlining how agentic deployment differs from standard chat systems. | |
| CIO Sentiment and the Realities of Agentic Deployment | 5 | 5 | 4 | 5 | Matt introduces a cynical premise that enterprises are fatigued by failed chat pilots and hesitant about agent hype. Levie directly pushes back on this framing, arguing that based on his sample of hundreds of CIOs, enterprise sentiment is statistically far more optimistic due to visible coding productivity gains. | |
| Soaring Token Costs and Enterprise AI Budget Overruns | 6 | 5 | 3 | 5 | Matt brings timely reporting regarding Microsoft and Uber CTO token billing concerns to challenge the narrative that token costs always fall. Levie politely corrects the press spin on Microsoft while confirming that agent contextual complexity is creating unprecedented enterprise billing shock. | |
| Shifting AI Spend to Line-of-Business Budgets & FinOps Challenges | 2 | 6 | 1 | 1 | Levie monologues on the structural shift of AI costs moving out of fixed IT budgets into line-of-business OPEX, highlighting the lack of enterprise FinOps tooling for tokens. Matt largely listens, offering brief humorous commentary on startup opportunities. | |
| Enterprise Model Strategy: Dedicated Capacity and the Multi-Model Mosaic | 3 | 5 | 1 | 2 | Matt asks if new lab pricing structures will stabilize enterprise costs. Levie explains the emerging mosaic of frontier models alongside lower-cost models for repetitive tasks, while joking with Matt about classic software still having a place. | |
| Venture Capital Subsidies and Exploiting Token Pricing | 4 | 7 | 1 | 2 | After banter regarding venture capital token subsidies, Levie delivers a comprehensive breakdown contrasting AI coding with general knowledge work. He educates on why knowledge work diffusion is far harder due to fragmented context, access control issues, and lack of verifiable outputs. | |
| The 10-Year Horizon and the Reference Architecture Paradox | 5 | 6 | 1 | 4 | Matt challenges aggressive adoption timelines by suggesting enterprise AI deployment could easily take a decade like cloud computing did. Levie agrees and articulates the paradox where rapid frontier model breakthroughs make existing reference architectures obsolete before deployment finishes. | |
| Data Integrity, Semantic Layers, and Internal FDE Roles | 8 | 6 | 2 | 6 | When Levie frames enterprise AI readiness around data modeling and access controls, Matt brilliantly interjects that this is simply the classic 20-year-old semantic layer problem rebranded as an ontology. Levie readily concedes the point while explaining why democratized AI agents make data integrity even more critical. | |
| Headless Software, Box APIs, and Evolving Enterprise Business Models | 5 | 6 | 1 | 4 | Matt probes whether headless software will completely replace graphical interfaces. Levie outlines why GUIs and APIs will co-exist in a dual seat-plus-consumption business model, referencing Box's API-first architecture. | |
| Job Evolution, Human-in-the-Loop Demands, and Jevons Paradox | 4 | 7 | 2 | 2 | Matt asks how enterprise org charts will adapt to AI automation fears. Levie presents a strong, detailed counter-argument to AI job destruction, citing human-in-the-loop demands like legal liability, Jevons paradox, and classic division of labor principles. | |
| Career Advice: AI-Proofing Your Job Through Hands-On Mastery | 4 | 6 | 1 | 1 | Matt asks how individual enterprise workers can AI-proof their careers. Levie advises hands-on experimentation with tools like Perplexity and Codex, framing AI as an unlimited chief of staff that drives new work creation. |
Statements from this episode (21)
Levie: AI divide is Silicon Valley engineering versus everyone else, not geography
“It's sort of not just Silicon Valley versus everybody else. It's sort of Silicon Valley engineering versus everybody else.”
Levie: AI coding agents have reached escape velocity
“AI coding agents that we know have totally, you know, reached escape velocity.”
Levie: Enterprise CIOs see real productivity gains from AI coding assistants
“When we talk to the CIO, they know that their engineering teams are using cloud code and codex and cursor, and they're seeing the productivity gains come out of those teams.”
Levie: CIO sentiment on AI remains optimistic, avoiding trough of disillusionment
“I think the tone is actually remarkably optimistic and excited and positive as opposed to, you know, there's a sort of, You know, typical trough of disillusionment, you know, from Gartner and the hype cycle or whatnot.”
Levie: Token costs and budgeting comprise one-third of top AI issues
“When we go and talk to organizations right now about where they are with agents tokens, the cost of tokens and budgeting and budget planning and all of this probably is at least one third of the hottest button issues that relate to AI.”
Levie: A single coding agent task can consume $1,000 in compute
“One, you know, coding agent could be consuming, you know, a thousand dollars of compute on a single task. So clearly like you can't lump that all into a 20 dollar per user per month fee.”
Levie: AI compute cost reductions will take 5 to 10 years
“The data center providers, the labs, et cetera, have pricing power. They don't need to lower their prices on anything. So you're not seeing the typical things that drive down the cost of compute. I'm highly optimistic that that happens over the next five to 10…”
Levie: Corporate IT spending is constrained to 3% to 7% of revenue
“IT spend is basically somewhere between like three to seven percent of revenue in a company, sometimes lower, sometimes higher, but like, it's kind of trapped at that.”
Levie: Enterprise AI spend will shift from IT to line-of-business budgets
“And so if AI is, you know, truly adding this productivity gain to your engineering team or your client onboarding process or your marketing team, then clearly you don't want to be trapped by this sort of three to seven percent in the business. It's going to es…”
Levie: A $5B startup will be built around AI compute ERP
“You're going to need, you know, new, new pieces of software. Probably there's probably a, you know, a five billion dollar startup waiting to happen just in like ERP for your AI compute.”
Levie: Average enterprise will use half a dozen AI models
“I think what's going to happen is you're going to have a mosaic of models in the enterprise. I think the average enterprise will certainly be using, you know, half a dozen models in their organization.”
Levie: Savvy startups can exploit temporary VC subsidies for AI compute
“Cause like, if you were really savvy, there's probably some parts of the market where you keep like, oh, I could somehow use this LP capital to do work for me as my startup. And there's like a window where you can find those exploits.”
Levie: Knowledge work AI is harder than coding AI due to fragmented context
“The code base has so much of the context in coding. Whereas in the rest of knowledge work, the context lives across like 20 different things. Some digital and some very not digital, you know, kind of mediums.”
Levie: Rapid AI breakthroughs paradoxically slow corporate AI adoption
“The technology is getting so advanced that it makes obsolete the prior thing that you implemented, which actually means that the rollout takes longer because we have no stable, there's no stable environment to roll things out in.”
Levie: Enterprises are refusing multi-year contracts with AI labs
“Nobody's signing up for more than like one year deals with the labs.”
Levie: AI doomers ignore permanent jobs required for agent maintenance
“As an asterisk, it's actually why the doomers are also wrong about jobs, because this is actually going to be a very real sustaining job. That is not like a one time you implement the agent and you upgrade the system and then it kind of works forever.”
Levie: Headless AI queries will be 100x larger than interface work
“So I think it's just going to be this sort of dual, dual model with the one nuance being probably by like you know, database queries, headless will just be a hundred times larger than the interface driven, you know, way of doing work.”
Levie: Enterprise software will combine seat and consumption models within three years
“So I think any enterprise software company in three years from now, that sort of, that, that gets through this AI transformation period, It will have a seat business model, assuming it has an end user component, and it'll have a consumption business model.”
Levie: Perplexity Computer is the best agent for web search tasks
“Perplexing computer, I find does a better job than any other computer based agent for just being a workhorse. For going through websites and doing search related things where you have to click on the page and you have to read the page and all that.”
Levie: AI agents will drive hiring as executives capture new value
“You will actually see, interestingly, if you're like an executive and you start to do this, you'll see lots of areas actually where you should hire more people because you're like, oh my God, this thing is spitting out, you know, incredible goldmine of value, …”
Levie: AI labs won't displace vertical startups without massive specialized headcount
“Unless the labs build out literally the equivalent of hundreds or thousands of people for every single vertical and every single line of business, that means that there's actually a lot of opportunity in that kind of bridge area of the work.”