0:00 / 0:00
research
AI Dev 26 x SF | Jerry Liu: My Agent Can't Read a PDF?
DeepLearning.AI16 June 2026Watch on YouTube
Part of series
Ep. 9 · Agent Document Build
View the seriesDescription
The future of automating knowledge work depends on AI agents that can reliably read and understand documents — but today's agents struggle with complex layouts, tables, and visual elements. This talk by LlamaIndex' Jerry Liu explores why document parsing remains a critical bottleneck for agentic workflows and introduces new open-source innovations to address it, including ParseBench, a benchmark for evaluating document OCR quality for AI agents, and LiteParse, a fast VLM-free parser. It also covers LlamaParse, purpose-built to deliver the best agentic understanding of complex documents at scale.
What you'll learn
- Document parsing is a critical bottleneck for AI agents performing automated knowledge work
- ParseBench provides a benchmark to evaluate the OCR quality of document parsing for AI agents
- LiteParse and LlamaParse are open-source, purpose-built solutions to reliably read complex document layouts, tables, and visual elements
Frequently asked questions
Why do AI agents struggle with reading documents?
AI agents struggle with complex document layouts, tables, and visual elements that are difficult to parse. This represents a significant bottleneck for reliable knowledge work automation.
What is ParseBench and what does it do?
ParseBench is a benchmark tool specifically designed to evaluate and measure the OCR quality of document parsing for AI agents.
What are the differences between LiteParse and LlamaParse?
LiteParse is a fast parser without VLM dependencies, while LlamaParse is a purpose-built solution designed to process complex documents at scale with optimal agentic understanding.
Why is reliable document parsing important for the future of automation?
The future of automated knowledge work depends on AI agents that can reliably read and understand documents. Without effective document parsing, agents cannot perform knowledge work correctly.