Sep 18, 2025 · 44m · big-technology
Are 95% of Businesses Really Getting No Return on AI Investment? — With Aaron Levie
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 in-depth discussion, Box CEO Aaron Levie joins Alex Kantrowitz to debunk claims of widespread enterprise AI failure, explaining how purpose-built tools, context engineering, and autonomous agentic workflows are driving massive productivity gains. Levie analyzes the macroeconomic impact of generative AI as an industrial revolution for cognitive knowledge work and outlines how modern organizations must re-engineer business processes to stay competitive.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Alex holds 19.4% of the talking time here. How this is scored →
Alex as informed peer 5.4 Guest teaching 5.3 Guest disagreement 3.0 Alex pushing back 4.0
speaking balance: gold is Alex, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 1:52 Levie rejects MIT study premise
Aaron rejects the MIT study's 95% failure conclusion on seven dimensions and characterizes Wall Street's reaction to AI adoption as completely schizophrenic.
Hardest push from Alex ▶ 10:19 Pushback on pilot framing vs enterprise realityAlex refuses to let Aaron dismiss enterprise failures as early-stage pilots, stressing that the study surveyed entire organization-wide implementations.
Biggest teaching moment ▶ 27:22 Crossing the Chasm framework in enterprise AIAaron reframes Alex's consumer vs enterprise question by explaining Geoffrey Moore's adoption chasm, showing that early enterprise adopter enthusiasm does not yet equal mainstream adoption.
Alex holds their own ▶ 22:36 Kantrowitz corroborates solo AI developersAlex leverages his direct reporting on Anthropic's Dario Amodei and independent developer interviews to substantiate the conversation on solo developer leverage.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Alex as informed peer | Guest teaching | Guest disagreement | Alex pushing back | Why |
|---|---|---|---|---|---|---|
| Evaluating the MIT Study on Generative AI ROI | 5 | 5 | 4 | 3 | Alex cites findings from an MIT study indicating 95% of enterprise AI investments yield zero return. Aaron forcefully dismisses the headline, explaining the disparity between failed DIY internal builds and high-ROI targeted software workflows. | |
| Workflow Re-Engineering, AI Pilots, and Enterprise Adoption Pitfalls | 6 | 6 | 4 | 6 | Alex challenges Aaron's repeated framing of failures as mere pilots, noting the study evaluated organization-wide deployments. Aaron counters by explaining how centralized surveys miss unmanaged shadow AI usage and fail to measure deep workflow re-engineering. | |
| Shadow AI, Claude File Generation, and Practical Enterprise Diffusion | 5 | 5 | 3 | 5 | Alex questions whether Claude's newly announced document generation is merely an impressive party trick given rigid corporate formatting templates. Aaron explains that enterprise integration will directly inject company data into native presentation templates. | |
| Context Engineering, Hallucinations, and the Rise of AI-First Startups | 6 | 4 | 2 | 2 | Alex validates Aaron's observations about solo engineers scaling software companies by referencing his own reporting on Anthropic. Aaron details how context engineering and inverted review models allow small teams to output the work of large engineering divisions. | |
| Consumer Tech Challenges vs. Crossing the Enterprise AI Chasm | 4 | 6 | 3 | 3 | Alex asks why consumer AI assistants like Alexa and Apple Intelligence struggle if enterprise AI is progressing rapidly. Aaron corrects the premise by applying Geoffrey Moore's Crossing the Chasm framework, clarifying that enterprise AI remains in the early adopter stage. | |
| Defining AI Agents and Box Automate Workflow Integration | 5 | 5 | 2 | 3 | Alex asks for an exact definition of AI agents given industry hype and questions whether Box's workflow features are generally available. Aaron provides a concrete architectural definition of looping models with memory and details Box Automate's multi-agent processes. | |
| GPT-5 Expectations, AI Economics, and Automating Knowledge Work | 7 | 6 | 3 | 6 | Alex confronts Aaron with staggering capex numbers, citing OpenAI's projected $115B cash burn and massive compute commitments. Aaron mounts an economic defense, arguing that automating post-industrial knowledge work justifies massive upfront capital outlays. |
Statements from this episode (14)
Levie: Companies Should Buy Purpose-Built AI Rather Than Building Internally
“You need to have purpose-built solutions that, that solve sort of tailored use cases. Those can be very big use cases, like all of AI coding, but you probably don't want to be in a position where you have to kind of bootstrap this or build it all out yourselve…”
Levie: Enterprise AI ROI Is So High Employees Hide It From Boards
“We've talked to customers where where they have had, had colleagues that, that can't actually they can't present the actual ROI savings to their board. The actual kind of expected ROI savings to the board because the board won't believe how, how they won't bel…”
Levie: Non-tech enterprises building custom open-source AI is a recipe for disaster
“Open source is insanely valuable, but not in the sense where a law firm should go off and build their own AI project using an open source model. Like, That, that is just a recipe for disaster if, you know, we think that, that every single company on the planet…”
Levie: Dropping AI into existing workflows without re-engineering fails to produce gains
“This is not a panacea type of solution where you could take an existing workflow, drop AI directly into it, and then all of a sudden that workflow will be, you know, three X better. You usually do have to re-engineer the work to take advantage of AI.”
Kantrowitz: MIT study shows 90% personal AI use vs 40% corporate purchases
“Official LLM purchases cover only 40% of firms, yet 90% of employees use personal AI daily, at least those surveyed”
Levie: Claude is the first AI to reliably generate high-quality documents
“Claude this week announced a new capability that will generate files for you, and even though we're two and a half years, you know, nearly three years into the ChatGPT moment, it's the first time Where an AI system can, I believe, generate reliably a kind of h…”
Levie: Proper context engineering nearly eradicates AI hallucinations
“As long as you are really good about what context you're giving the AI and how you are effectively grounding the AI in
trustworthy data with the right kinds of prompts and a high enough quality model.
you can nearly eradicate the, you know, all of, if not th…”
Levie: Real productivity gains come from humans managing and reviewing AI agents
“We will be the reviewers of the AI agents work.
We will be the editors.
We will be the managers.
We'll be the orchestrators.
And that's actually how you then get the productivity gains.”
Levie: Pfizer And Eli Lilly Won't Be Disrupted By Delayed AI Adoption
“Like if Pfizer or Eli Lilly took a little bit longer to adopt AI as a result of, you know, wanting to be more pragmatic, that'll be totally fine.
They're not going to get disrupted.
Like they have enough of market position.
They have enough distribution.
They …”
Levie: Apple and Amazon Can Still Catch Up in AI
“The companies just mentioned, like, I don't think the, those spaces have been so utterly disrupted that, that that they can't catch up once they land on a final architecture.”
Levie: Early Adopter Enthusiasm Predicts Nothing About Mainstream Tech Adoption
“What happens is the early adopters, the people that, you know, we all hang out with and talk to all day, they're going to try everything. We're going to try these crazy goggles, and we're going to put, you know, magnets on our head, and we're going to do the c…”
Levie: AI agents do not require a new breakthrough architecture
“We don't need any kind of new breakthrough architecture. We have the, we have an architecture that, that already kind of works as the core scaffolding for agents.”
Levie: AI brings scalable automation to knowledge work tuned by compute
“For the first time ever with AI, we can bring automation to effectively all of that work. And that automation can kind of be tuned based on just how much compute we throw at the problem.”
Levie: AI lab losses are a strategic choice to subsidize market dominance
“And the losses are a choice, to be clear. Like, that's very obvious. Like, they're choosing to lose that money. They're doing it, ah, for a strategic reason. You know, that, that's at least their decision. The strategic reason is, is that this is such a valuab…”