Agents of Scale Box’s Ben Kus on Building AI for Millions of Users and Billions of Files
12 snips
Oct 30, 2025 
Ben Kus, Chief Technology Officer at Box, joins Wade Foster to discuss the evolution of AI in handling enterprise unstructured data. They delve into how AI transforms extensive data into actionable insights, making tasks manageable without needing extensive data science expertise. Ben explains the importance of trust in AI adoption, emphasizing the need for strong security and governance practices. The conversation also highlights the role of agents in streamlining workflows, ultimately transforming how businesses leverage their data effectively.
AI Snips
Chapters
Transcript
Episode notes
Extract Structure, Then Answer Questions
- Do extract structured attributes from documents (dates, terms, client IDs) to make content useful for workflows.
- Implement retrieval-augmented Q&A so users ask questions and receive concise answers instead of raw documents.
Agents Are Models That Manage Work
- Agents differ from single-shot models by deciding when a task is done and how to progress through workflows.
- Enterprise workloads will increasingly rely on agentic systems that reason, loop, and validate outputs.
Let Agents Do The Tedious Reviews
- Use agents for first-pass reviews (legal, compliance, brand) to automate tedious checks and surface issues.
- Automate low-value reviews so humans focus on final decisions and higher-value work.