How Box Evolved from Simple AI to Agentic Systems for Enterprise | LangChain Interrupt

Box's CTO, Ben Kus, shares how their AI document extraction system hit a wall when customers brought real-world complexity—300-page documents, 500+ data fields, and OCR challenges that broke their simple AI approach. After initially deploying to 10 million pages successfully, Box completely re-architected their system using a multi-agent agentic approach that separated complex problems into specialized sub-agents. This transformation not only solved their scaling issues but made their system dramatically easier to evolve and update for enterprise needs.

Watch all of our recorded sessions from Interrupt here: https://interrupt.langchain.com/video/?utm_medium=social&utm_source=youtube&utm_campaign=q2-2025_interrupt-2025_co