Are 95% of Businesses Really Getting No Return on AI Investment? — With Aaron Levie
Timestamps are as accurate as they can be but may be slightly off. We encourage you to listen to the full context.
Podcast Summary
In this episode of Big Technology Podcast, host Alex Kantrowitz interviews Box CEO Aaron Levy fresh off the company's BoxWorks AI event. The conversation centers on the controversial MIT study claiming 95% of organizations get zero return on their AI investment, which Levy strongly challenges based on his firsthand experience with Box customers. (02:15)
- Main Theme: Debunking AI ROI misconceptions while exploring the practical reality of AI agents transforming business workflows and the decade-long transformation ahead
Speakers
Aaron Levy
CEO of Box, the leading cloud content management platform serving over 120,000 customers worldwide. Levy has positioned Box at the forefront of enterprise AI adoption, developing AI agents that extract structured data from documents and automate business workflows.
Alex Kantrowitz
Host of Big Technology Podcast and author who covers the intersection of technology and business. Kantrowitz provides nuanced analysis of tech industry trends and conducts in-depth interviews with technology leaders.
Key Takeaways
Companies Must Avoid DIY AI Approaches
The MIT study revealed that internal AI builds fail at double the rate of external partnerships. (04:00) Levy explains that companies attempting to build their own AI infrastructure often end up managing 10-15 different software components before a single user can interact with AI. This creates unnecessary complexity and dramatically increases failure rates. The key insight is that most organizations should focus on applied solutions rather than building foundational AI technology themselves, similar to how most companies don't build their own databases from scratch.
AI Requires Workflow Reengineering, Not Drop-In Replacement
One of the most critical realizations is that AI won't simply adapt to existing workflows - businesses must modify their processes to fully leverage AI capabilities. (10:07) Levy uses AI coding as an example, where engineers become "managers of AI agents" rather than writing code line-by-line. This fundamental shift in how work gets done is essential for achieving the promised productivity gains, and companies that resist this change will miss out on significant ROI opportunities.
Context Engineering Is Key to AI Trustworthiness
The reliability of AI outputs depends heavily on providing high-quality, grounded context rather than relying on the model's general knowledge. (20:26) When AI agents work with existing enterprise data as source material - like PowerPoint templates, customer information, and company documents - accuracy rates can reach 99%. This approach nearly eliminates hallucinations and makes AI outputs trustworthy enough for professional use with minimal review time.
Early Adopters Are Achieving Massive Productivity Multipliers
Despite broader skepticism, innovative startups are demonstrating extraordinary productivity gains through AI adoption. (22:06) Levy describes a nine-person startup that estimates it operates with the capacity of a 100-person company, with individual engineers producing the output of 5-20 traditional engineers. These companies work fundamentally differently, spending more time on specifications and architecture while letting AI agents handle implementation and focusing their effort on reviewing and managing agent outputs.
The Decade of Agents Is Beginning Now
2025 marks the first year where AI agents can be seriously deployed at scale, representing the start of a decade-long transformation similar to the mobile revolution. (34:22) Unlike previous hype cycles, we now have the foundational architecture that works - the "iPhone moment" for agents has already arrived. However, like autonomous vehicles, it will require years of engineering refinement and real-world testing before reaching mainstream adoption across all industries.
Statistics & Facts
- The MIT study found that 95% of organizations get zero return on their AI investment despite enterprise investment of $30-40 billion into generative AI. (02:35) However, Levy challenges this finding, noting it primarily reflects early-stage pilot projects rather than mature AI implementations.
- Official LLM purchases cover only 40% of firms, yet 90% of employees use personal AI daily. (13:51) This reveals a significant gap between individual adoption and enterprise deployment, suggesting strong revealed preference for AI tools among workers.
- Internal AI builds fail at double the rate of external partnerships according to the same MIT study. (08:17) This validates Levy's argument that companies should focus on applied solutions rather than building AI infrastructure themselves.
