Box CEO Aaron Levie joins Matthew Berman to discuss the real future of AI agents, the evolution of enterprise software, and why the job of the future isn’t doing the work—it’s managing the agents that do.
From bold takes on white-collar jobs to a deep dive into Box’s transformation into an AI-first company, Aaron Levie lays out one of the most compelling visions yet of how work will change in the age of AI.
Key Moments from the Interview
00:00 – The 50% Prediction
Why half of all white-collar jobs could disappear, and what comes next.
01:50 – What “AI-First” Really Means
Box’s strategy to expand capabilities, not just cut costs.
06:04 – How Box Deploys AI Internally
From Box Hubs to internal demos: building an AI-native company culture.
13:20 – Jevons Paradox for Labor
Why automation leads to more work, not less.
18:33 – Rethinking Work
AI flips the model: it’s not here to help you work faster, you’re here to make it productive.
22:55 – Managing the Agent Workflow
Why future jobs are about instruction, orchestration, and review.
31:00 – Memory, Portability, and Ownership
What happens when your agent knows you better than your coworkers?
35:47 – Agent Responsibility
Who’s liable when an agent screws up?
39:50 – The Real Bottleneck
Tech moves fast. People don’t. Change management is the drag coefficient.
43:20 – Who Owns the Agent Ecosystem?
Why there won’t be one winner—and why standards matter.
47:20 – The Application Layer Collapse
Does the SaaS UI disappear? Aaron agrees… sort of.
51:00 – Agents vs. Browsers vs. APIs
The future of the web might not have humans in the loop.
54:40 – Where Companies Get AI Wrong
Underestimating workflow change. Overestimating automation.
Full Interview: Aaron Levie on Agents, SaaS, and the Future of Work
In His Own Words
The 50% Prediction (00:30)
Will half of white-collar jobs disappear in five years? Aaron’s take is nuanced—but bold.
The AI agent isn’t here to help us be a little more productive. It’s our job to make the AI agent productive.
AI-First at Box (01:50)
Box is going all in on AI—not to replace people, but to expand what’s possible.
We’re not trying to cut headcount. We’re trying to do way more than we ever could before.
How Box Deploys AI (06:04)
Weekly demos, Box Hubs with embedded AI, and org-wide experimentation.
Every week, someone demos an agent they built. That’s how fast the culture is changing.
Jevons Paradox for Labor (13:20)
If you make a task more efficient, demand for it often increases. AI may not replace jobs—it might explode new ones.
A small business that used to need 10 people might soon need 50—because they can finally grow.
Rethinking Work (18:33)
The old mental model: AI helps you type faster. The new model: You manage a fleet of tireless agents.
Enterprise software must now be designed to make AI productive—not just the user.
Managing the Agent Workflow (22:55)
Why the new skill isn’t execution—it’s decomposition, orchestration, and review.
You don’t want to over-specify the task. You want to get just enough abstraction that an agent can run.
Memory, Portability, and Ownership (31:00)
What happens when your personal agent knows everything—and you change jobs?
We need a spec for personal AI memory. Like an agent resume on steroids.
Agent Responsibility (35:47)
AI is probabilistic. So when it fails—who’s at fault?
For the foreseeable future, the responsibility lies with the human. There’s no way around that.
The Real Bottleneck (39:50)
AI is moving fast. Humans aren’t.
We overestimate how quickly people can change. It’s not a tech problem—it’s a change management problem.
Who Owns the Agent Ecosystem? (43:20)
Model providers, SaaS companies, and infra platforms will all own pieces of the stack.
The world wants choice. It won’t be winner-take-all.
The Application Layer Collapse (47:20)
Satya Nadella said SaaS UIs will collapse into agents. Aaron agrees… with a twist.
Yes, agents will run in the background—but GUIs aren’t going away. Dashboards exist for a reason.
Agents vs. Browsers vs. APIs (51:00)
Will agents browse the web—or use APIs to access it?
I’m bullish on both. The web’s last mile is messy. Agents will need both toolsets.
Where Companies Get AI Wrong (54:40)
It’s not about doing the same work faster. It’s about changing how you work.
If you flip your mental model, you’ll find five times more ways to use AI.
Key Takeaways
Box is transforming into an AI-first company with human-first incentives.
AI agents won’t just help you work, they’ll do the work.
Workflow design, memory portability, and orchestration are the future of work.
SaaS companies won’t vanish, but their role will shift.
The change won’t happen overnight. It’ll take a decade of human adaptation.
Enjoyed this conversation?
For more in-depth interviews with the people shaping AI, follow us on X and subscribe to our YouTube channel.
