In this episode of the Big Technology Podcast, Box CEO Aaron Levie discusses the current state of AI adoption in businesses, addressing a recent MIT study claiming that 95% of organizations see no return on their AI investments. Levie strongly disagrees with the study’s conclusions, emphasizing that the reality is far more nuanced. He explains that AI adoption is still in its early stages, with many companies experimenting through pilots and proof of concepts, which naturally have a high failure rate. He highlights that companies attempting to build AI solutions entirely in-house tend to struggle more than those leveraging external, purpose-built AI applications tailored to specific use cases.
Levie points out that successful AI implementations often require significant change management and workflow re-engineering. AI is not a plug-and-play solution; businesses must adapt their processes to fully harness AI’s potential. He uses AI coding as an example, where engineers now act more as managers of AI agents that perform much of the coding work, requiring new workflows and oversight. This shift means that employees will increasingly become reviewers and orchestrators of AI-generated work, rather than relying on AI to seamlessly integrate into existing processes without adjustment.
The conversation also touches on the widespread personal use of AI tools like ChatGPT, even when official enterprise adoption lags behind. Levie sees this as empirical evidence of AI’s value, noting that individuals are already benefiting from AI in their daily work. He predicts that AI agents—autonomous systems that perform multi-step tasks—will become pervasive in knowledge work over the coming decade. These agents will be integrated into business workflows, automating complex tasks such as document analysis, contract review, and report generation, thereby significantly boosting productivity.
Levie explains Box’s recent AI initiatives, including Box Automate, a drag-and-drop workflow builder that allows businesses to embed AI agents into their processes. These agents leverage enterprise content to perform tasks like extracting data from contracts or invoices and automating workflows around that data. He stresses that AI’s effectiveness depends heavily on providing the right context and trustworthy data, which helps minimize errors or hallucinations in AI outputs. While AI models like GPT-5 have shown incremental improvements, Levie believes the real breakthrough lies in how these models are applied within enterprise workflows.
Finally, Levie discusses the economics of the AI industry, acknowledging the massive cash burn by companies like OpenAI but framing it as a strategic investment in what he calls the “third industrial revolution”—the automation of knowledge work. He argues that the potential productivity gains across sectors such as healthcare, law, and engineering justify the high costs. Levie also contrasts the slower progress in consumer AI products like Alexa with the more focused, prescribed use cases in business, attributing the difference to execution challenges and the need for reliable, scalable solutions. Overall, he is optimistic about AI’s transformative impact on the economy and encourages businesses to embrace the technology early to avoid being left behind.
