Box CEO and tech thought leader Aaron Levie says he recently met with 20 enterprise AI and IT leaders and came away with insights into what everyone, especially the stock market, wants to know: how—and how fast—large U.S. companies are adopting AI for core business functions. In a post on X, he outlined the main themes he heard.
Here’s a closer look at those key themes.
Agents move from hype to production
“Agents are clearly the big thing,” Levie wrote. “Enterprises [are] moving from talking about chatbots to agents, though we’re still very early. Coding is still the dominant agentic use-case being adopted thus far, with other categories . . . across knowledge work starting to emerge. Lots of agentic work moving from pilots and PoCs into production, and some enterprises had lots of active live use-cases.”
Recent models from Anthropic and OpenAI, including Claude Opus 4.5 and 4.6 and GPT-5.2 and 5.3, have pushed AI coding agents beyond simple code generation toward something closer to operating like junior software engineers. As trust in these tools grows among developers, enterprise decision-makers appear increasingly eager to deploy them within software teams.
From coding copilots to company-wide agents
“Agentic use-cases span every part of a business, from back office operations to client facing experiences from sales to customer onboarding workflows,” Levie wrote on X. “General feeling is that agentic workflows will hit every part of an organization, often with [the] biggest focus on delivering better for customers, getting better insights and intelligence from data and documents, speeding up high ROI workflows with agents, and so on. Very limited discussion on pure cost cutting.”
AI companies have long argued that the capabilities powering coding tools—planning, reasoning, and tool use—can extend across knowledge work. Based on Levie’s conversations, enterprise leaders are starting to act on that idea. What works in software engineering may translate to marketing, finance, and HR. That raises the specter of job displacement, but Levie suggests companies are prioritizing improved customer experience over head-count reduction.
Governance becomes the bottleneck
“Data and AI governance still remain core challenges,” Levie added. “Getting data and content into a spot that agents can securely and easily operate on remains a huge task for more organizations. Years of data management fragmentation that wasn’t a problem now is an issue for enterprises looking to adopt agents. And governing what agents can do with data in a workflow [is] still a major topic.”
One of the big lessons from the OpenClaw agent craze is that the more autonomy agents have, the greater the chance they’ll get in trouble. Within the enterprise this could mean exposing sensitive data to the wrong people or exposing data to hackers. And when agents are asked to retrieve data from different data stores in different clouds, the risk increases.
Who gets access, and how much
Levie said identity and access control are quickly becoming central concerns as companies deploy more agents. “Can the agent have access to everything you have? In a world of dozens of agents working on [your] behalf,” he wrote, “potentially too much data exposure and scope for the agents. How do we manage agents with a partitioned level of access to your information?”
You’ll increasingly see a new software layer (like Credo AI’s AI Agent Registry or ServiceNow’s AI Control Tower) that tracks all agents used in an organization—including homegrown and third-party agents—and manages their activities, connections, access levels, and security protections.
The token economy hits the balance sheet
Levie said companies are starting to grapple with how to allocate and control spending on AI usage across teams: “This is going to become a bigger part of OpEx over time, and probably won’t make sense to be considered an IT budget anymore. Likely needs to be factored into the rest of operating expenses.”
AI apps and agents are powered by generative AI models, and access to those models is measured by the number of tokens sent back and forth between the app and the model. As AI agents proliferate across an organization, these tokens become the fuel for an organization’s intelligence engine. Now big companies are asking whether it still makes sense, from an accounting perspective, to pay for AI tokens in the same way they might pay for, say, cloud access or software licenses.
No single platform will win
“Interoperability is key,” Levie continued. “Every enterprise is deploying multiple AI systems right now, and it’s unlikely that there’s going to be a single platform to rule them all. Customers are getting savvier on how to handle agent interoperability, and this will be one of the biggest drivers of an AI stack going forward.”
The grand vision of the AI industry is that enterprises will deploy agents across departments in an organization, and the agents will interact with each other and with third-party agents from partners and suppliers. All of these agent exchanges will rely on technical protocols (Anthropic’s MCP, for example) and trust assurances. The work of developing such standards is just beginning.
The real challenge is change management
“Lots more takeaways than just this, but needless to say the momentum is building but equally enterprises are acutely aware of the change management and work ahead,” Levie concluded. “Lots of opportunity right now.”
Roughly a third of the U.S. stock market’s value is tied up in a relatively small group of companies, hyperscalers, AI firms, data center builders, and chipmakers, all betting that corporate America is ready to shift quickly from traditional software to AI-driven systems. The hype around that transition has been relentless. But the clearest signal comes from the executives who have to approve the spending and justify it to their boards. That is the group Levie spoke with.
His most telling takeaway may be that enterprises are “acutely aware of the change management and work ahead.” In many core business functions, the technology itself is rapidly approaching high performance. The slower process will be organizational: humans working alongside agents, training them, correcting them, and gradually building trust. That transition is likely to take years.
