Enterprise AI transformation: Box CEO Aaron Levie on agents and AI-first strategy

Executive overview

Enterprises are moving past AI experimentation into real deployment, but most leaders lack a mental model for where value actually comes from. AI agents unlock work that was previously unaffordable to do — not just automation of existing tasks.

The core shift: AI doesn't replace work that was already happening; it makes previously unfunded work economically viable for the first time.

Where enterprise AI value is actually coming from

• The biggest use cases aren't replacing existing jobs — 80% of customer conversations are about work the company never did before
• Reviewing 50,000 contracts for upsell propensity, mining historical sales decks for new employee onboarding — tasks that never had headcount assigned
• AI agents make previously unaffordable analysis cost-effective (e.g., $5,000 for a task that would have required dozens of people)
• Insights from AI agents generate downstream work for humans, driving growth rather than cutting it

What AI-first actually means

• Frame AI as compressing the timeline of work, not reducing the total mass of work
• Box's policy: prove you can use AI to get headcount — the inverse of Duolingo's "justify why AI can't do this"
• Avoid perverse incentives where non-AI use is rewarded with more budget
• Productivity gains should be reinvested into growth, not banked as profit — competitors will force this anyway

The agent moment for enterprises

• Developer tools (Cursor, Windsurf) had their agent moment 12–18 months ahead of the general market
• Deep research tools (ChatGPT, Perplexity, Grok) are the first agentic use case for knowledge workers
• Enterprises are entering their ChatGPT-equivalent agent moment now — but full adoption will be a decade-long journey
• Cloud computing is still being adopted 20 years later; AI will move faster but expect the same long tail

Keeping up with relentless AI change

• Five to six credible AI model providers are leapfrogging each other on a monthly basis
• Box evaluates new models on weekends to have benchmarks ready for Monday customer conversations
• Staying current requires treating it as genuinely exciting — exhaustion comes from seeing it as a threat
• No hobbies; sleep is still possible

AI policy and talent

• Trump administration: positive on open source AI and light regulation; neutral-to-negative on talent
• Tariffs and regulation both artificially raise costs — Levie opposes both on the same economic grounds
• China's AI is less threatening than its hacking and IP issues; winning AI matters economically, not geopolitically
• Best policy lever: stamp green cards on US computer science diplomas; talent that leaves builds AI elsewhere