Box CEO: AI Agents Will Force Companies to Build Software That Agents Want

Box CEO Aaron Levie argues that the rise of personal AI agents will force companies to build software that agents can navigate and operate, shifting competition from user attention to agent attention. For AI builders, this means rethinking APIs, data access, and interface design to support agent-driven workflows.

The Argument: AI Agents Become the New Consumer

In a post on X, Levie said agents need to be able to "use your MCP or CLI, easily navigate your site, be able to transact, and more" to complete tasks end-to-end on behalf of users. He predicts that competition will shift from winning human attention to winning agent attention, calling this "the biggest opportunity and shakeup in consumer tech since the App Store itself." Benzinga He highlighted food ordering, e-commerce, flight bookings, and data interactions as early battlegrounds.

Why This Matters for Product and API Design

For AI builders, product design can no longer focus only on human UX. If an agent cannot navigate your service, book a flight, or place an order, it will choose a competitor that supports agent-driven workflows. This directly impacts how you design APIs, authentication flows, and data access patterns. Agents need programmatic entry points, not just polished frontends.

The shift also ties into broader industry moves around AI safety and governance. Meta CEO Mark Zuckerberg delayed the Muse AI agent for months over safety concerns, and Microsoft CEO Satya Nadella warned that persistent agents could create insider risks, calling for third-party evaluations and human oversight. Benzinga As agents gain more autonomy, building auditable and secure agent access becomes a product requirement, not an afterthought.

What Builders Should Do Differently

Start designing for agent attention now. That means exposing MCPs and CLIs, ensuring your service can be navigated and transacted by an automated agent. Context engineering, the ability to feed agents the right proprietary data and institutional knowledge, is becoming a competitive differentiator. CareerofBusiness Levie has also noted that AI will commoditize expertise, so companies that own unique data and build agent-friendly interfaces will win.

For enterprise tooling, this means rethinking how agents access internal systems. Instead of building for human-only workflows, consider how an agent acting on behalf of a user will authenticate, query data, and execute actions. The demand for APIs that grant controlled, auditable access to enterprise software will rise.

What Remains Unclear

Levie's vision is a compelling thesis, but it is not yet a proven market reality. Exact timelines for this shift, adoption rates, and the scope of the opportunity are not quantified. Safety and governance concerns, as highlighted by Zuckerberg and Nadella, could slow deployment. Additionally, the claim that this is the biggest opportunity since the App Store is a vendor perspective, not an independent forecast. Builders should treat this as a directional signal rather than a guaranteed roadmap.