Key Takeaways

  • Aaron Levie, founder of Box, argues that AI agents won't replace core 'system of record' enterprise software like accounting or CRM platforms.
  • Instead, these agents will increase usage of existing software by requiring access to deterministic data, established permissions, and robust workflow designs within these systems.
  • This new interaction model creates an "agentic upside" for incumbent SaaS companies, enabling more sophisticated and automated workflows directly within trusted applications.
  • The rise of AI agents will likely shift software monetization towards consumption-oriented or "headless" usage models rather than traditional user licenses.
  • Levie dismisses the notion that "vibe-coded" AI solutions can handle mission-critical enterprise tasks, emphasizing the need for reliable, deterministic software.

AI Agents: The Unlikely Usage Catalyst for Existing SaaS

When Aaron Levie looks at the future of AI in the enterprise, he sees something most founders miss: a usage boom for existing software, not a total rewrite. For years, the conversation has centered on AI replacing jobs, or even entire software categories. But Levie, the veteran CEO of Box, flips that script. He contends that the most critical enterprise applications—the "systems of record" that manage things like accounting, customer relationships, and financial documents—are simply too important and too complex to be supplanted by what he calls "vibe-coded" AI solutions.

Instead, these core systems, built over decades with layers of security and specific logic, become indispensable to AI agents. “What happens when you have agents that are running around and they need to go do all this useful work in your enterprise,” Levie explained, “Well, the useful work they're going to do is going to require access to data that's inside these systems.” Agents need reliable data, clear permissions, and predictable workflows. They can't just operate in a vacuum of AI magic. They need deterministic software as their operational ground, complete with “the right walls, the right data access, the right permissions, the right workflow design that's largely going to come from existing software.”

This means that rather than seeing a decline, Box and other enterprise SaaS giants might see an increase in their product's usage. AI agents become power users, constantly querying, processing, and interacting with the data stored and managed by these core platforms.

The "Agentic Upside" and New Consumption Models

Levie's vision points to what he calls an "agentic upside" for established SaaS players. Imagine AI agents automating expense reports, updating CRM records, or managing compliance documents — all tasks that require interacting with specific, permissioned systems. "We actually see an increase in usage because agents are now roaming around accessing all of this data and you want them to access the same data that the user has access to," Levie said. This is a far cry from the doomsday scenarios some predict, where large language models simply bypass traditional software interfaces.

This shift isn't just about more clicks or API calls; it also suggests a transformation in how software is monetized. Levie forecasts a move towards consumption-oriented models, where companies pay based on the volume of agent interactions or data processed, rather than a fixed per-user license. “Mostly it's mostly be like a consumption-oriented model... a little bit on this more more this headless approach,” he noted. “But I think there's going to be a ton of usage of software as a result of the agent, you know, kind of adoption piece.” It reframes the argument from “Claude wins so SAS loses” to a more symbiotic relationship where AI intelligence enhances, and is enhanced by, the deterministic software it relies on.

What to Do With This

If you're building enterprise software today, audit your product's API surface and internal data models. Identify where an AI agent could naturally access, update, or analyze data within your system, and begin designing for those "headless" use cases. Start experimenting with consumption-based pricing for specific features or API calls, preparing for a future where agents, not just humans, drive the bulk of your usage.