"If we were building Box today, what would we do?” Ben Kus (CTO @ Box) deconstructs their playbook for enterprise AI innovation. We cover their journey to reimagine & reorient the company to a new technical vision, how they run a “multi-speed” org that balances startup agility and & enterprise-grade stability, and their “platform first” approach to build AI features. Ben also explains why security/compliance was foundational from "day negative one" in their AI strategy, the evolution of agentic AI, determining the right guardrails for AI agents & the future of multi-agent systems, enterprise trends & more.
"If we were building Box today, what would we do?” Ben Kus (CTO @ Box) deconstructs their playbook for enterprise AI innovation. We cover their journey to reimagine & reorient the company to a new technical vision, how they run a “multi-speed” org that balances startup agility and & enterprise-grade stability, and their “platform first” approach to build AI features. Ben also explains why security/compliance was foundational from "day negative one" in their AI strategy, the evolution of agentic AI, determining the right guardrails for AI agents & the future of multi-agent systems, enterprise trends & more.
Ben Kus is the Chief Technology Officer at Box, where he leads technology and AI strategy to help enterprises securely unlock insights from their unstructured data. Ben’s career spans engineering, product leadership, and startup innovation—including co-founding Subspace (acquired by Box) and being an early employee at BigFix (acquired by IBM), where he later served as Chief Architect of Mobile Security. Ben holds a degree in Computer Science from UC Berkeley.
ToolHive Unlocks the Full Value of MCP & Your AI Agents
So you’ve invested in AI agents for code generation, but they’re limited to experiments or even stuck on the shelf. To do real, valuable work, those AI agents need access to your data and systems.
ToolHive helps you confidently connect the pieces by making it simple and secure for you to use the Model Context Protocol (MCP).
ToolHive includes a pre-vetted registry of MCP servers, containerizes every MCP server for consistency and leans on built-in security to keep your secrets safe.
Leaders trust ToolHive to put MCP into production and put their AI agents to work.
ToolHive is open source, so get started for free at toolhive.dev
Join us at ELC Annual 2025
ELC Annual is the premier event for engineering leaders. This is our biggest event of the year: 1,000+ CTOs, VPs & Directors in San Francisco @ ELC Annual 2025 for two days of leadership breakthroughs, tactical peer learning & curated connections!
SHOW NOTES:
- How Generative AI sparked Box’s reoriented vision, by unlocking the value of unstructured data (3:29)
- Using GenAI to create structure out of unstructured data (5:28)
- Internal & external conversations that inspired Box’s new direction (7:18)
- Box’s “platform first” approach to building a secure and scalable foundation for all future AI features (10:02)
- Why security and compliance must be built in from "day negative one" not added on later (12:40)
- How to set a technical vision that can respond to future developments you can't yet predict (14:46)
- The “multi-speed” business model: Using a small, fast-moving internal group to test ideas before they enter the normal, slower development cycle (17:47)
- Example of a successful project: AI-driven data extraction & the evolution to critical feature (20:26)
- The story of an abandoned project and the challenge of knowing which ideas are revolutionary versus which aren’t worth continuing (22:17)
- Ben’s long-term vision for AI agents and why he believes they are an incredibly powerful technology paradigm (23:58)
- State diagrams & the journey behind building Box’s initial agentic AI systems (26:50)
- “Context Engineering”: The new paradigm of programming and the mental model shift required for engineers to adopt it (29:07)
- The future of AI benchmarks: Measuring what a person can accomplish with an agent, not just the agent’s performance alone (31:03)
- How to balance development speed with security risks, especially when agents can take actions and change the environment (34:00)
- Key questions to ask to determine the right guardrails for AI agents, including thinking about the worst-case scenario (36:41)
- Enterprise technology trends to watch, and why multi-agent systems will become the new “org chart” (37:56)
- Rapid fire questions (39:03)
LINKS AND RESOURCES
- Harry Potter and the Methods of Rationality - Eliezer Yudkowsky’s alternate-universe Harry Potter fan-fiction wherein Petunia Evans has married an Oxford biochemistry professor and young genius Harry grows up fascinated by science and science fiction. When he finds out that he is a wizard, he tries to apply scientific principles to his study of magic, with sometimes surprising results.
This episode wouldn’t have been possible without the help of our incredible production team:
Patrick Gallagher - Producer & Co-Host
Jerry Li - Co-Host
Noah Olberding - Associate Producer, Audio & Video Editor https://www.linkedin.com/in/noah-olberding/
Dan Overheim - Audio Engineer, Dan’s also an avid 3D printer - https://www.bnd3d.com/
Ellie Coggins Angus - Copywriter, Check out her other work at https://elliecoggins.com/about/