Aaron Levie: AI Agents Will Expand Enterprise Work Into Previously Unaffordable Areas

Summary

Box CEO Aaron Levie argues that within five years, about 95% of the tokens used by AI agents may support tasks humans have not previously performed. After meeting with roughly 30 enterprises over two days and a dinner, he said a recurring opportunity was applying automated work to areas that companies could not afford or scale with human labor. Levie contrasts this with the common expectation that AI will simply enter existing workflows and make them about 20% more efficient. He says companies often do far less than they could because talent is expensive or limited. Examples include reviewing every lease in real estate for trends and opportunities, accelerating drug discovery or finding data-quality problems in life sciences, analyzing past deals in financial services, and handling contracts or legal work that was previously unprofitable. He adds that the same pattern could extend to software projects and security work that companies previously left undone. His recommendation is for AI-agent builders and enterprise adopters to ask what work they would pursue if labor became 100 times cheaper and faster.