Aaron Levie, CEO of BOX (NYSE:BOX), discussed the uneven pace of AI adoption across enterprises, arguing that the biggest barrier is not model capability but the need for companies to redesign workflows to make AI agents effective. In a wide-ranging conversation that touched on “agent-to-agent” economies, the future of SaaS, and evolving pricing models, Levie emphasized that organizations will have to adapt how work gets done—and how context is captured, governed, and delivered—if they want to realize meaningful ROI.
AI adoption is accelerating in engineering, lagging elsewhere
Levie described AI as a “tale of two cities,” pointing to software engineering as the clearest example of rapid adoption. He said it is becoming difficult for engineers to build software without AI assistance, and referenced public examples where “100% of the software is written by AI.” In that environment, development shifts toward directing an agent, reviewing output, and shipping code—an approach he said looks “completely different than how you would’ve built it a year ago.”
- Verification is harder in functions like legal and contracts, where outcomes aren’t instantly testable the way software is.
- Permissions and access controls are more complex for knowledge work, where information visibility varies widely across roles and teams.
- Work isn’t consistently digitized—context is frequently created in meetings and calls and not captured in systems agents can reliably use.
- Documentation norms differ, with engineering more accustomed to producing artifacts for others to reference.
Enterprises must adapt to agents—not the other way around
A central theme of Levie’s remarks was that many organizations are assuming agents will fit neatly into existing processes. Instead, he said enterprises will need to “adapt to how agents work,” building systems and routines that produce the right context for agents to operate effectively.
He framed “context” as a defining challenge for the next decade of enterprise agent deployment. Organizations will need to supply information to agents in a way that is efficient, accurate, and comprehensive, while also avoiding pitfalls like “context rot,” where performance can degrade as too much information is provided. Levie also stressed that permissioning must be authoritative and up to date so that agents do not access data they shouldn’t.
Customer use cases: unlocking value from unstructured content
From Box’s perspective, Levie said much of a company’s operational context lives in unstructured enterprise content—research materials, memos, marketing assets, contracts, and financial documents. He argued that enterprises historically stored this content for decades but couldn’t “tap into that value at scale,” whereas agents make it possible to extract and apply it.
As an example, Levie described customers processing “10 million medical records per year,” where each record may contain critical information for care delivery, insurance claims, or patient understanding. Previously, organizations either devoted significant human effort to review data or designed workflows that effectively ignored it. He said customers are now deploying agents to review information and surface insights through other software interfaces to improve decisions, routing, and automation—though he noted this still requires significant change management and workflow re-engineering.
What “good” looks like: a barbell approach to productivity and workflow wins
Looking ahead, Levie suggested a bimodal—or “barbell”—strategy for enterprise adoption. On one side are broad productivity tools that may yield “one to two hours” of daily gains through general assistance like content generation. On the other side are deeper, targeted workflow reinventions where organizations choose specific processes to redesign and automate.
He said that achieving “three to five wins” over the next year or two—focused on select high-impact workflows—would be key. He gave illustrative examples such as automating contract processes, improving financial processes to generate better customer insights for upsell opportunities, or streamlining onboarding workflows (including a wealth management example) to improve competitiveness.
SaaS and pricing: “SaaS is not dead,” but models will shift
Addressing the claim that “SaaS is dead,” Levie said he expects the opposite: software will become cheaper to build, which should lead to more software and more competition, potentially putting downward pressure on prices. However, he said he does not believe enterprises will broadly “vibe code” their own CRM or ERP systems, arguing companies have finite resources and should focus on what customers value rather than rebuilding core infrastructure.
Levie also contended that systems of record and deterministic workflows may become more valuable in an agent-heavy world. If enterprises eventually run “100 times or 1,000 times more agents than people,” he said the need for systems that “traffic cop” what agents can do—what data they access, add, or produce—will increase because the opportunities for value creation and risk multiply.
On commercial models, Levie anticipated pricing pressure on traditional software as vendors deliver more features without charging proportionally for every incremental capability. At the same time, he expects consumption-based models to expand for agent-driven work, likening it to paying for service providers—such as paying per contract reviewed. He suggested customers may prefer consumption-based pricing during experimentation, then shift toward locking in rates at scale to avoid quarter-to-quarter volatility.
In closing, Levie encouraged CIOs and CISOs to view AI as augmentation—not only cost reduction—and to use agents to pursue more ambitious initiatives that were previously too expensive or time-consuming to attempt. He argued that as agents reduce the cost of doing work, enterprises can build and test ideas faster, potentially changing how organizations prioritize innovation.
About BOX (NYSE:BOX)
Box, Inc is a leading provider of cloud content management and file sharing solutions designed to support enterprises in securely managing, accessing and collaborating on digital content from anywhere. The company offers a unified platform that enables organizations to store, share and automate workflows across various departments, enhancing productivity and ensuring governance over sensitive information. Box’s services are tailored to meet the needs of industries such as healthcare, financial services, government and media, where compliance and data security are paramount.
The core offerings of Box include its Content Cloud platform, which provides content collaboration, workflow automation, data classification and secure file sharing.
