← back
Building an Agentic Platform — Ben Kus, CTO Box
Takeaway
For enterprises sitting on mountains of unstructured content, off-the-shelf generative models plus light preprocessing have already beaten the legacy IDP industry at structured extraction.
Summary
- Ben Kus (CTO Box) describes Box's 2023 GenAI journey serving 115K enterprise customers and 2/3 of the Fortune 500 with an exabyte of unstructured content.
- Focuses on data extraction (pulling structured fields from contracts/proposals) — historically a brittle, multi-billion-dollar IDP industry requiring custom ML models.
- Off-the-shelf GenAI plus OCR preprocessing now outperforms specialized legacy IDP models with single-shot prompting.
- Frames agentic capabilities as applying well beyond chatbot UX — especially to structured extraction workflows over enterprise unstructured data.
enterprise-aidata-extractionagentic-platform
Original description
Explore the technical evolution of metadata extraction at Box and how it shaped the foundation of our AI platform. We’ll walk through our transition to an agentic-first design—why it was necessary, how we approached the rebuild, challenges we encountered along the way, and the advantages it unlocked. Timestamps 00:00 Box's Content Platform and Enterprise Focus 01:50 Initial AI Deployment in 2023 02:54 The Challenge of Unstructured Data in Enterprises 03:56 Limitations of Pre-Generative AI Data Extraction 04:54 First Version: LLM-Based Extraction 07:05 Challenges with the Pure LLM Approach 08:58 Despair and the Need for a New Architecture 09:30 Introducing Agentic Architecture 10:04 AI Agent Reasoning Framework 10:45 Agentic Routine for Data Extraction 12:28 Advantages of Agentic Architecture 14:05 Key Lesson Learned: Build Agentic Architecture Early 18:37 Approach to Fine-tuning and Model Support Ben Kus CTO Ben Kus is the Chief Technology Officer at Box and is responsible for developing Box’s technology vision and strategy and ensuring that technological resources are aligned with the company's business needs. Previously Ben was the VP of Product Management at Box. Before joining Box, Ben was the Co-Founder and CTO of Subspace, Inc., an enterprise security solution that was acquired by Box. Ben has held various leadership positions, including the role of Chief Architect for IBM, and Senior Director of Technology for BigFix, Inc. Ben studied Computer Science at the University of California, Berkeley.